Dynamic Analysis of Structures PDF

Dynamic Analysis of Structures
@CivilMethod
Dynamic Analysis of
Structures
John T. Katsikadelis
Professor Emeritus of Structural Analysis
School of Civil Engineering
National Technical University of Athens
@CivilMethod
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
© 2020 Elsevier Inc. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our
arrangements with organizations such as the Copyright Clearance Center and the Copyright
Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional practices, or medical
treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating
and using any information, methods, compounds, or experiments described herein. In using
such information or methods they should be mindful of their own safety and the safety of others,
including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of any methods, products,
instructions, or ideas contained in the material herein.
Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
ISBN 978-0-12-818643-5
For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals
Publisher: Matthew Deans

Editorial Project Manager: Joshua Mearns
Production Project Manager: Nirmala Arumugam
Cover Designer: Greg Harris
Typeset by SPi Global, India
@CivilMethod
Dedication
To my wife Efi
for her loving patience and support,
to my children Stefan and Christina,
and to
my granddaughter Katharina
for patiently enduring and sharing the years of
preparation of the book with me.
This work is also dedicated to my numerous students

whose attendance of the Structural Dynamics
course over the past 40 years encouraged an
international publication of the book.
@CivilMethod
Preface
The statement of the laws of motion by Newton 334 years ago (1686)a was a
milestone in the evolution of mechanics and modern engineering. The relation
between force (cause) and motion (effect) was quantified as a relation between
the linear momentum of the body and the force exerted on it. Thus, this relation
from a subject of philosophy up to that time turned out to be a valuable tool of
science for the study of the natural world. The subsequent developments in the
sciences were rapid. Astronomy, mathematics, mechanics of fluid and deform-
able bodies, and in general, mechanics of continuous media reached their peaks
in the centuries that followed, with immense applications to all engineering dis-
ciplines. Nevertheless, the laws of motion, which were stated as an axiom
(Axiomata sive Leges Motus) by Newton because, apparently, he could not jus-
tify their derivation, was a consequence of the discoveries of great scientists
who preceded him such as Galileo,b Kepler,c Hook, etc.
The implementation of the laws of motion leads to mathematical models
described by differential equations, ordinary or partial, whose solution effort
has given a great impetus to the development of mathematics. Unfortunately,
analytical solutions are limited to simple problems such as vibrations of discrete
systems with a few degrees of freedom; linear vibrations of beams, membranes,
plates, and shells with simple geometry; and simple support conditions made
from materials, mostly with a linear behavior. These solutions, while useful
for extracting qualitative conclusions about the dynamic response of structures,
are not capable of solving realistic problems in engineering, where the geometry
and loads are complicated while the response is generally nonlinear. Although it
has been shown that Newton’s law of motion is of an integer-order derivative,d
in recent years, the fractional derivatives have been proven more suitable for
modeling the actual structures. However, the use of fractional calculus has
not been employed in mathematical physics for three centuries because the
a. I. Newton, Philosophiae Naturalis Principia Mathematica, Royal Society Press, London, 1686.
b. J.T. Katsikadelis, Derivation of Newton’s law of motion using Galileo’s experimental data, Acta
Mech. 226 (2015) 3195–3204, doi:10.1007/s00707-015-1354.
c. J.T. Katsikadelis, Derivation of Newton’s law of motion from Kepler’s laws of planetary motion,
Arch. Appl. Mech. 88 (2018) (2017) 27–38, doi:10.1007/s00419-017-1245-x.
d. J.T. Katsikadelis, Is Newton’s law of motion really of integer differential form? Arch. Appl.
Mech. 89 (2019) 639–647, doi:10.1007/s00419-018-1486-3.
xv
@CivilMethod
xvi Preface
fractional derivative, contrary to the integer-order derivative, has no direct

physical or geometrical meaning. Thus, it was not adopted by the famous sci-
entists of earlier centuries who developed modern science and engineering, as
they could not visualize the significance of the fractional derivative. Thus, we
dare say that in a certain sense, Newton’s Principia has delayed the development
of mechanics and engineering.
Structural dynamics became a tool for the dynamic analysis of structures
only with the advent of computers in the middle of the last century. Nowadays,
the cheap, fast, and large-capacity computers as well as the development of effi-
cient numerical methods to solve the equations of motion allow the safe study of
complicated structures under dynamic loads. Aircrafts, large bridges, ships,
high-rise buildings, dams, fast trains, etc., are now being safely analyzed and
designed. The finite element method (FEM) permits the dynamic analysis of
continuous systems by modeling them with equivalent discrete systems. The
boundary element method (BEM) analyzes with great accuracy the dynamic
response of systems described by ordinary or partial differential equations.
Besides, the coupling of the two methods paves the way to solving very difficult
dynamic problems such as the interaction of the structure with soil or fluids as
well as the large deformation analysis of systems with nonlinear constitutive
equations. Recently, meshless methods, a numerical tool beyond FEM, have
been able to solve efficiently the equations describing the dynamic response
of continuous systems.
The dynamic analysis of structures today is mainly done by using profes-
sional computer codes such as NASTRAN, SAP2000, ETABS, ADINA, etc.,
which have relatively unlimited capabilities. While these programs are
undoubtedly a useful and indispensable tool in the hands of the engineer, they
do not allow for an understanding of the dynamic behavior of the structures or
the limits of the validity of the results. This can be achieved only with a deep
theoretical background, which is constantly expanding with the development of
the computational methods in mechanics. The engineer without theoretical
knowledge is unable to formulate the structural model. The modeling of struc-
tures cannot be done by the computer but only by the engineer, who is the only
one responsible for performing proper dynamic analysis. The belief that theo-
retical knowledge is not necessary today because there are ready-to-use com-
puter codes is, at the very least, dangerous. The engineer should never
accept the results of a computer program unless he/she can check their validity
and accuracy. Apparently, this is possible only if the engineer has a deep the-
oretical knowledge of the dynamics of structures. We emphasize that theoretical
knowledge today is more necessary than ever. Areas that a few decades ago
were covered by applied mathematics and theoretical or applied mechanics
have become the subject of the modern engineering praxis.
Dynamics of Structures for engineers, particularly for civil engineers, was
included as a regular course in the curricula of many universities long ago,
@CivilMethod
Preface xvii
particularly after the advent of computers. Therefore, many books have appeared
on this subject. However, these books provide a means only to understand the
response of simple and mostly unrealistic structures when subjected to dynamic
loads, especially to ground motion. The principles of dynamics are illustrated by
applying them to very simple models, which cannot describe actual structures
and therefore cannot be employed for dynamic analysis and design. We should
have in mind that the dynamic analysis of actual structures requires their model-
ing, the formulation of the governing equations of motion, their solution under
any dynamic load, and the physical interpretation of the results.
In the last 30 years, almost all seismic codes have encountered earthquake
ground motion as an effective dynamic load. The advent of computers in the
early 1960s encouraged engineers to develop methods of dynamic analysis
of structures, modeled first by the FEM and later by other advanced numerical
methods. Today, these methods constitute a powerful tool for dynamic engi-
neering analysis. Thanks to the availability of cheap computer power, every
engineer can use them. The essential ingredients of a book on Dynamics of
Structures for Civil Engineers should be:
(a) The basic concepts and principles of structural dynamics as they are
applied to particles as well as rigid and deformable bodies, enabling the
student or the engineer to formulate the equations of motion of any struc-
ture, no matter how complex, once the dynamic model has been adopted.
(b) Realistic modeling of actual structures under dynamic loads.
(c) Analysis of the dynamic response of the structure represented by its
model under any specified load. The analysis should include single- and
multiple-degree-of-freedom systems for linear and nonlinear response
under any dynamic excitation.
(d) Approximation of real structures using computational methods such as the
FEM, which replaces the actual structure (distributed parameter system)
with an approximate discrete system for which analysis methods can be
applied.
(e) Effective present-day numerical methods for dynamic analysis, including the
numerical solution of eigenvalue problems and the direct solution of the equa-
tions of motion of large systems, namely, systems with a large number of
degrees of freedom such as those resulting from the employed discretization.
Students attending a course on Dynamics of Structures should be exposed at
least to the above subjects. However, not all of them can be found in a single
book. Therefore, people interested in structural dynamics should refer to more
than one book in order to retrieve the required knowledge. Apparently, these
books cannot be used as integrated textbooks in the sense described above.
The student should be acquainted with different symbols and approaches, which
complicate the acquisition of knowledge, an approach that is, at least, educa-
tionally inappropriate.
@CivilMethod
xviii Preface
Knowledge of the dynamics of structures is particularly necessary for engi-

neers, who are studying the response of their structures subjected to seismic
ground motion. Modern earthquake codes, which as mentioned treat the ground
motion as a dynamic load, require a deep theoretical understanding of the
dynamics of structures. The aim of this book is to give the student as well as
the professional engineer the minimum knowledge necessary to understand
the dynamic response of the structures. The author has taught structural dynam-
ics, both as an undergraduate and a graduate course, at the School of Civil Engi-
neering of the National Technical University of Athens (NTUA) continuously
since 1981, when it was first introduced as a formal course by the late Prof. A.E.
Armenákas. The author’s teaching experience together with his long research in
this area have contributed to the presentation of material in a simple and com-
prehensible way. The book contains numerous illustrative examples that help to
understand the theory. Particular emphasis is given to numerical methods,
which are presented methodically and given in the form of pseudo-codes so that
interested readers can write their own computer codes in the language of their
choice. The list of the programs in FORTRAN and MATLAB are given in elec-
tronic form on this book’s companion website. Nevertheless, the numerical
results in the Examples have been obtained using the MATLAB programs
because it is easy for the students to master this language. The programs can
be used to solve many problems of engineering praxis.
The book is divided into two parts.
The first part deals with the single-degree-of-freedom (SDOF) systems
and contains nine chapters. In particular, Chapter 1 presents the general
concepts and principles of dynamics as they apply to structural dynamics.
Chapter 2 deals with the free vibrations of the SDOF systems while
Chapter 3 deals with their forced vibrations. Particular emphasis is given
to the study of the resonance phenomenon. Chapter 4 presents the most effi-
cient numerical methods for solving the equation of motion. Chapter 5 deals
with the nonlinear response of the SDOF systems. Chapter 6 presents the
response of structures due to ground motion. Chapter 7 deals with the damp-
ing of structures while Chapter 8 approximates the continuous systems by
generalized SDOF systems and studies the dynamic response of beams trea-
ted as continuous systems. Finally, Chapter 9 describes the analysis of SDOF
systems in the frequency domain.
The second part deals with the multi-degree-of-freedom (MDOF) systems
together with the well-known computational methods for their analysis. It con-
tains seven chapters. In particular, Chapter 10 describes the various methods of
modeling of the MDOF systems, which are classified into categories that facil-
itate the selection of the appropriate method for formulating the equations of
motion. Chapter 11 presents in detail the FEM for the dynamic analysis of skel-
etal structures (trusses, frames, grids) as an extension of the static matrix struc-
tural analysis to dynamics. The reduction of degrees of freedom due to the
omission of axial deformations in specific structural members as well as the
dynamic response of flexible structures containing rigid bodies are also
@CivilMethod
Preface xix
discussed. The derivation of the equivalent element nodal quantities, that is,
mass and stiffness matrices and forces, are derived by using the Lagrange equa-
tions instead of the principle of virtual works. Although the principle of vir-
tual works offers a handy tool for the derivation of these quantities, the use of
the Lagrange equations was preferred here. The reason is that the Lagrange
equations not only offer a straightforward method for the derivation of the
equivalent nodal quantities for all types of elements, especially in complex
systems with a nonlinear response, but they also allow understanding of their
physical significance. Chapter 12 studies the free vibrations of MDOF sys-
tems without and with damping. The linear eigenvalue problem is presented
from the mathematical point of view, aiming at drawing useful conclusions
about the eigenfrequencies and the mode shapes of the physical systems.
Chapter 13 presents the numerical methods for the computation of the eigen-
frequencies and mode shapes, especially for systems with a large number of
degrees of freedom such as those derived from the application of finite ele-
ments. Chapter 14 studies the forced vibrations of MDOF systems. A large
part of this chapter is devoted to the mode superposition method. It also dis-
cusses the use of Ritz vectors to reduce the degrees of freedom. Particular
emphasis is given to the response spectrum method. The response of linear
systems when they are subjected to a synchronous and an asynchronous
motion of the supports are also discussed. This chapter ends with the presen-
tation of the numerical methods, giving the respective pseudocodes for the
time integration of linear and nonlinear systems of equations of motion.
Chapter 15 discusses the approximation of multistory buildings by skeletal
structures and presents methods of formulating their equation of motion.
Finally, Chapter 16 discusses the response of seismically isolated buildings.
This chapter is introductory to the subject and aims primarily at understand-
ing the impact of base isolation on structures.
The book is supplemented by an appendix. Therein, the basic theory of rigid
body dynamics is presented for large and small displacements and the relevant
equations are derived, which are employed in the development of the material of
the book.
In closing, the author wishes to express his sincere thanks to his former stu-
dent and coworker Dr. A. J. Yiotis for carefully reading the manuscript as well
as for his suggestions, constructive recommendations and his overall contribu-
tion to minimizing the oversights in the text. Warm thanks also belong to Dr.
Nikos G. Babouskos, also a former student and coworker of the author, not only
for his careful reading of the manuscript and his apposite suggestions for
improvement of the book but also for his assistance in checking the computer
programs and in producing the numerical results of examples therein. Finally,
thanks belong to Dr. G. Dasios, professor of mathematics at the University of
Patras as well as to his former students, Dr. G. Tsiatas, associate professor of
mathematics at the University of Patras, and Dr. P. Tsopelas, associate professor
of mechanics at NTUA, for reading certain sections of the book and making
constructive suggestions.
@CivilMethod
xx Preface
It is a pleasure to make grateful acknowledgment of many helpful sugges-

tions contributed by my students who have attended the course over the past
39 years.
J.T. Katsikadelis
Athens
April 2020
@CivilMethod
Chapter 1
General concepts and principles

of structural dynamics
Chapter outline
1.1 Introduction 3 1.7 Hamilton’s principle 42
1.2 Types of dynamic loads 6 1.8 Lagrange’s equations 54
1.3 Dynamic degrees of freedom 9 1.8.1 Derivation of
1.4 Dynamic model and Lagrange’s equations 54
formulation of the equation 1.8.2 Lagrange multipliers 64
of motion of SDOF systems 11 1.8.3 Small displacements 68
1.5 Derivation of the equations 1.8.4 Raleigh’s dissipation
of motion using d’Alembert’s function 72
principle 15 1.9 Influence of the gravity loads 73
1.6 Principle of virtual 1.10 Problems 74
displacements 38 References and further reading 82
1.1 Introduction
Apart from static loads, engineering structures may be subjected to dynamic
loads, that is, loads whose magnitude as well as direction of action and/or posi-
tion vary with time. The analysis of stresses and deflections developed in a
given structure undergoing dynamic loads is the fundamental objective of the
dynamic analysis of structures. Between static and dynamic analysis of struc-
tures, there exist two substantial differences:
(a) In static analysis, the loads are assumed time-invariant, and the resulting
response is unique, at least in linear theory. On the other hand, in dynamic
analysis the loads are time-varying and the deformations and stresses
depend on time, that is, at each instant the response of the structure is
different.
(b) In dynamics analysis, the material points of the structure change position
with the time, hence they have velocity and acceleration. Inasmuch as
the structure has a mass, inertial forces are produced due to the accelerations
of the material points. These inertial forces constitute an additional loading
that cannot be ignored. To make it tangible, we consider the cantilever beam
of Fig. 1.1.1a. The beam has a mass per unit length m and a flexural rigidity
EI , both assumed constant along the length, and it is subjected to the time-
Dynamic Analysis of Structures. https://doi.org/10.1016/B978-0-12-818643-5.00001-7

© 2020 Elsevier Inc. All rights reserved. 3
@CivilMethod
4 PART I Single-degree-of-freedom systems
varying position-dependent distributed loading pðx, t Þ (Fig. 1.1.1a). We can

derive the equation of motion by considering the dynamic equilibrium of a
beam element of length dx (Fig. 1.1.1b). In static consideration, the element
would be in equilibrium under the action of the external load pðx, t Þdx and
the internal forces, that is, the shear force Q, the bending moment M at the left
cross-section of the element, and the shear Q + dQ and moment M + dM at
its right cross-section.
(a)
(b)
(c)
FIG. 1.1.1 Vibrating cantilever beam.
@CivilMethod
General concepts and principles of structural dynamics Chapter 1 5
The transverse deflection is a function not only of the spatial variable x but
also of time t, namely it is u ¼ u ðx, t Þ. As the element has a mass mdx, an inertial
€ arises, which, according to d’Alembert’s principle (see Section 1.5)
force m udx
opposes the motion, that is, if the positive transverse displacement u ðx, t Þ in the
beam is directed downward, the inertial force is directed upward (see Fig. 1.1.1b
and c). Similarly, due to angular acceleration ∂u€ðx, t Þ=∂x of the cross-section, an
inertial moment is also developed, which we may neglect [1]. Thus, referring to
Fig. 1.1.1b, we obtain the equation of dynamic equilibrium of the beam element
in the y direction as
€ ¼0
Q + Q + dQ + pðx, t Þdx m udx
or
∂Q ∂2 u
¼ pðx, t Þ + m 2 (1.1.1)
∂x ∂t
From the Euler-Bernoulli beam theory we have
∂3 u
Q ¼ EI (1.1.2)
∂x 3
Substituting Eq. (1.1.2) into Eq. (1.1.1) yields
∂4 u ∂2 u
EI + m 2 ¼ pðx, t Þ (1.1.3)
∂x 4 ∂t
Eq. (1.1.3) is known as the equation of the dynamic equilibrium or the equa-
tion of motion of the vibrating beam. It is apparent that if we omit the inertial
term m∂2 u=∂t 2 in Eq. (1.1.3), we obtain the equation of the deflection of the
beam under static loading, that is,
d4u
EI ¼ pðx Þ (1.1.4)
dx 4
Fig. 1.1.1c shows the beam subjected to the inertial forces. These forces
resist the accelerations and they need to be accounted for in the solution. This
is the most important characteristic of the dynamic problem. Obviously, the
magnitude of the inertial forces depends on the magnitude of the acceleration.
When the produced accelerations are very small, as in the case of slow motion,
the inertial forces are very small too, and they can be neglected. In this case,
the time appears in the equation as a parameter and the response at any instant
@CivilMethod
can be obtained by the static structural analysis, even though the load and the
response are time-varying. This response is pseudodynamic and is referred to
as quasistatic. The inertial forces appear in the equation of motion of the struc-
ture with the second derivatives of the displacements with respect to time.
Therefore, the equations that must be solved in dynamic analysis in order
to establish the deformations and stresses in the structure are differential
equations, contrary to static analysis where the governing equations are alge-
braic. For this reason, the solution procedure in dynamic analysis is essentially
different from that used in static analysis.
1.2 Types of dynamic loads

As already mentioned, dynamic loads are time-varying. Such loads are those
due to an unbalanced rotating machinery, the motion of vehicles on structures,
blast loads, wind loads, etc. The motion of the support of a structure, even in the
absence of external dynamic loads, produces a dynamic response, too. This is
the case of seismic ground motion (see Chapter 6).
FIG. 1.2.1 Periodic load.
Dynamic loads can be classified into two great groups that characterize the
approach of evaluating the structural response: The deterministic dynamic loads
and the nondeterministic or random dynamic loads. In the first group are the
dynamic loads whose time variation is fully determined, regardless of the
complexity of their mathematical presentation. They are also referred to as pre-
scribed dynamic loads. They can be represented by an analytic or a generalized
function (Dirac or Heaviside) as well as numerically by a set of their values at
discrete time instances. The second group includes the loads, whose time var-
iation is not completely known but it can be defined in a statistical sense. In this
book, the dynamic response of structures only under deterministic loads is
studied.
@CivilMethod
5
2
H = me sin t
2.5
0
H(t)
–2.5
T T T
–5
0 2 4 6 8 10
t
FIG. 1.2.2 Harmonic loading due to an unbalanced rotating mass.
From the analytical point of view, it is convenient to divide the deterministic

loads into two basic categories, periodic and nonperiodic loads. Periodic loads
are those whose time variation profile repeats continually at regular time inter-
vals T . Mathematically, they can be represented by a periodic function
pðt Þ ¼ pðt + nT Þ (1.2.1)
where n is a natural number. The time interval T is called the period of the load.
A general type of periodic load is shown in Fig. 1.2.1, which also identifies the
period of the load. A usual type of periodic load is the harmonic load caused by
an unbalanced rotating machine (Fig. 1.2.2); H ¼ mew2 cos wt is the horizontal
component of the centrifugal force. Loads that do not show any periodicity are
called nonperiodic loads. They may be of long duration, such as those resulting
from an earthquake. Nonperiodic loads of short duration are called impulsive
loads. A windblast striking a building and the pressure of a bomb explosion
on a structure are typical impulsive loads (Fig. 1.2.3). The earthquake produces
a special type of load, which is due to the excitation of the support of the
@CivilMethod
structure and, as we will see later in this book (Chapter 6), it can be reduced to an
effective dynamic load if the accelerogram of the ground motion is known
(Fig. 1.2.4).
FIG. 1.2.3 Nonperiodic load due to explosion.
400
–400
0 10 20 30
FIG. 1.2.4 Effective dynamic load pðt Þ ¼ m u€g ðt Þ due to seismic ground motion.
If we examine static loading closer, we will see that even what we call static
loads are actually dynamic in nature. They are applied starting from a zero value
until the final prescribed value is reached within a time span. That is, they are
time-varying, thus dynamic. However, the duration of the application of the
static load is longer than the period of vibration of the structure. This produces
negligible accelerations and consequently the response under a “static load”
@CivilMethod
could be considered as a special case of the dynamic problem with negligible

accelerations, that is, quasistatic.
1.3 Dynamic degrees of freedom

The displacement method is the most suitable method for the dynamic analysis
of structures. In this method, the unknowns are the displacements. For structures
with distributed mass, the displacements are functions of the spatial coordinates
and of the time as well. Their dynamic response is described by partial differen-
tial equations of the hyperbolic type, which must be solved in order to determine
these displacements. The solutions of such equations belong to the most difficult
problems of mathematics. The available solutions refer to simple structures, for
example, beams with a constant cross-section, which are not adequate to study
the dynamic response of engineering structures. Thanks to the development of
modern computational methods such as FEM, BEM and Meshless Methods, the
actual structure is approximated by discrete models in which the mass is local-
ized at a finite number of points (nodal points). These models are adequate to
represent the effects of all significant inertial forces of a structure. At each
instant, the deformed configuration of the structure is determined from the dis-
placementsa of the nodal points, which are functions only of time. The response
of the discretized structure is governed by ordinary differential equations, which
are easy to solve analytically or at least numerically. The number of the indepen-
dent nodal displacements required to determine the deformed shape of the mov-
ing structure is called the number of degrees of freedom. It is apparent that
continuous systems have an infinite number of dynamic degrees of freedom.
Structures with one degree of freedom are called single-degree-of-freedom
(SDOF) systems. Accordingly, we have two-degree-of-freedom (2 DOF) sys-
tems, three-degree-of-freedom (3 DOF) systems, and generally multi-degree-
of-freedom (MDOF) systems. Fig. 1.3.1 shows SDOF systems. Fig. 1.3.1a rep-
resents the idealization of a silo. It consists of two massless columns and a square
rigid plate of mass m. With the assumption that the axial deformation of the col-
umns is negligible, the horizontal displacement u ðt Þ is adequate to completely
determine the motion of the system. Hence, the system has one degree of free-
dom. Likewise, under the same assumptions for the columns, the motion of
the water tower of Fig. 1.3.1b can be determined from the angle fðt Þ.
Fig. 1.3.1c represents the typical model of a SDOF. Fig. 1.3.2a represents the
model of the two-story shear frame. To determine its motion, it is necessary to
establish the two independent horizontal displacements u1 ðt Þ and u2 ðt Þ.
Fig. 1.3.2b represents a cantilever column with a mass atop. This may be
considered as the idealization of a water tower. The whole mass is lumped at
the top while the column is massless. During the motion, the mass undergoes
horizontal displacement u ðt Þ and rotation fðt Þ. These two geometrical quanti-
ties are independent. Hence, the system has two degrees of freedom and thus
a. The term displacent denotes both translation and rotation.
@CivilMethod
two differential equations of motion are necessary to determine these displace-

ments. If, however, the mass m of the system is assumed to be concentrated at a
point, its rotational inertia I is equal to zero. Hence, the inertial moment I f€ðt Þ is
zero, too, and one of the equations of motion becomes algebraic. This permits
the elimination of the rotational displacement, leading to only one equation of
motion for u ðt Þ. Consequently, the system has only one dynamic degree of free-
dom, even though it has two static degrees of freedom. Apparently, the number
of static degrees of freedom is not necessarily equal to the number of dynamic
degrees of freedom. Concluding, we can say that in an MDOF system, the num-
ber of dynamic degrees of freedom is equal to the number of independent dif-
ferential equations of motion that must be formulated to establish the dynamic
response of the system.
Massless
Massless
columns
columns
(a) (b)
(c)
FIG. 1.3.1 Systems with one degree of freedom (SDOF).
Rigid
Rig
(a) (b)
FIG. 1.3.2 Systems with two degrees of freedom (2 DOF).
@CivilMethod
FIG. 1.3.3 Multi-degree-of-freedom (MDOF) system.
FIG. 1.3.4 System with infinite degrees of freedom. Continuous system.
The lumped mass idealization provides a simple means of reducing the num-
ber of degrees of freedom. Fig. 1.3.3 represents the discrete model of a canti-
lever column, whose mass has been localized at three points. Neglecting the
axial deformation of the column and considering plane motion, the system
has six degrees of freedom, the three transnational ui ðt Þ and the three rotational,
fi ðt Þ. If the masses are fully concentrated so that their rotational inertia can be
ignored, the inertial moments Ii f€i are zero and the number of dynamic degrees
of freedom reduces to three. Obviously, the number of degrees of freedom
increases with the number of nodal points, where the mass of the structure is
lumped. As the number of points becomes infinitely large, the discretized struc-
ture approaches the continuous system (Fig. 1.3.4).
1.4 Dynamic model and formulation of the equation

of motion of SDOF systems
The modeling of the real structure plays a fundamental role in the dynamic anal-
ysis of structures. It is the most difficult task in dynamic analysis because in this
stage of analysis, the experience and theoretical background of the engineer
intervene critically in approximating the structural response.
@CivilMethod
Spring
Damper
Frictionless rollers
FIG. 1.4.1 Model of a SDOF system.
Fig. 1.4.1 shows a simple dynamic model of a SDOF system. It consists of a

rigid body of mass m constrained to move along the x axis in the plane of the
paper, a weightless spring connecting the mass to the firm support, and a damper.
Center of mass
FIG. 1.4.2 Forces applied to the free body.
The forces applied to the body at time t are shown in the free body diagram
of Fig. 1.4.2. These are
(a) The external load pðt Þ
(b) The elastic force fS
(c) The damping force fD
(d) The inertial force fI .
The spring force fS depends on the displacement u ðt Þ and it is generally
expressed by a nonlinear function, fS ¼ fS ðu Þ. For linear response of the struc-
ture, the force fS is proportional to the displacement and is given by
fS ¼ ku (1.4.1)
where k is the constant that represents the spring stiffness coefficient, that is, the
force required to change the length of the spring by a unit. The force fS repre-
sents the elastic force of the structure that resists the motion and tends to bring
the body to its initial undeformed position.
The damping force fD also resists the motion. It represents the energy loss
due to internal or external dissipative forces. Damping forces are complex in
nature. Their exact expression in terms of the parameters of motion and of
the geometrical and material properties of the structure is complicated and dif-
ficult to determine. The simplest form of damping is linear viscous damping.
This produces damping forces, which are the easiest to handle mathematically
@CivilMethod
and provide analytical results for the response of a system close to the experi-
mental ones. The linear viscous damping mechanism is indicated by a dashpot,
as shown in Fig. 1.4.1. In viscous damping, the resisting force is proportional to
the velocity
fD ¼ cu_ (1.4.2)
where c is a constant that can be established experimentally. Inasmuch as the
work done by this force is converted to heat, the damping force is a nonconser-
vative force. It is the force that makes the amplitude of a vibrating
structure decay.
The inertial force fI depends on the mass m of the body and its acceleration
€ It also resists the motion. It is given by Newton’s second law of motionb
u.
fI ¼ m u€ (1.4.3)
A simple example of a structure that can be modeled as SDOF is the
one-story, one-bay frame of Fig. 1.4.3. It consists of two identical weightless
columns fixed on the ground and having height h, cross-sectional moment of
inertia Ι, and modulus of elasticity E. The cross-sectional moment of inertia
of the horizontal beam is assumed infinitely large. This means that the beam
behaves like a rigid body of mass m and hence the cross sections of the columns
at the roof level cannot rotate when the frame deforms. The frame is subjected to
an external horizontal force pðt Þ, as shown in Fig 1.4.3a, which forces the frame
to move. Neglecting the axial deformation of the beam and columns, an allow-
able assumption for frames, the only possible movement is the displacement
u ðt Þ at the roof level. The rotation of the beam as a rigid body is excluded
because this would cause a change in the length of columns.
(a)
(b) (c)
FIG. 1.4.3 Two-column shear frame.
b. Actually, this form of Newton’s law of motion is attributed to L. Euler, who defined it indepen-
dently as a mechanical principle [2, 3]. This law was recently derived from Kepler’s laws of plan-
etary motion [4].
@CivilMethod
Referring to Fig. 1.4.3b, we see that the elastic forces are the shear forces Q
at the top cross-sections of the columns. These forces are given by the known
relation of statics
12EI
Q¼ u ðt Þ (1.4.4)
h3
The quantity 12EI =h 3 represents the translational stiffness of the column.
This is the force required to produce a unit relative displacement between
the end cross-sections of the column. These shear forces tend to restore the
frame to the undeformed position. Therefore, they play the role of the spring
in the SDOF model with a stiffness coefficient
12EI
k ¼2 (1.4.5)
h3
The inertial force is given by fI ¼ m u€ while the damping force by fD ¼ cu. _
Another convenient model to represent the single-story frame is shown in
Fig. 1.4.3c. It consists of a mass m placed at the top of a column with transla-
tional stiffness equal to the sum of the translational stiffness coefficients of the
columns of the frame. During the motion, the top cross-sections of columns
undergo only the translational displacement u ðt Þ. Models of this type are also
suitable to idealize multistory shear frames (see Fig. 1.4.4), in which the masses
are placed at the floor levels and the girders are assumed rigid.
FIG. 1.4.4 Four-story shear frame and its model without damping.
(a) (b)
FIG. 1.4.5 Two-story, two-bay shear frame and its model without damping.
@CivilMethod
Fig. 1.4.5a shows another two-story shear frame. The columns are assumed
weightless. Fig. 1.4.5b shows its dynamic model. The column 1-2 is represented
by a spring of stiffness k ¼ 12EI =h 3 . The stiffness coefficients k1 and k2 include
only the stiffness of the columns with heights h1 and h2 , respectively.
Given the dynamic model of the structure, the equation of motion of the sys-
tem is formulated. For the SDOF system, the equation of motion can be formu-
lated using Newton’s second law of motion as it is applied for the motion of a
particle
m u€ ¼ F (1.4.6)
where
F ¼ pðt Þ fS fD (1.4.7)
is the resultant of the external forces. Using Eqs. (1.4.1), (1.4.2), (1.4.7),
Eq. (1.4.6) is written
m u€ + cu_ + ku ¼ pðt Þ (1.4.8)
Eq. (1.4.8) is the equation of motion of the SDOF system. The equation of
motion represents the dynamic equilibrium of the system. It is an ordinary dif-
ferential equation of the second order with respect to the unknown variable u ðt Þ.
The solution of this equation yields the displacement as a function of time. For
MDOF systems, the number of equations of motion that must be formulated is
equal to the number of dynamic degrees of freedom. The use of Newton’s law of
motion is not always well suited to formulate the equations, especially for
MDOF systems or complex SDOF systems. It requires advanced knowledge
of the dynamics of the rigid and deformable body as well as mastering various
special methods. Generally, the equations of motion can be formulated using:
(a) d’Alembert’s principle or method of equilibrium of forces.
(b) Principle of virtual work.
(c) Hamilton’s principle.
(d) Lagrange’s equations.
These methods will be presented in the following and will be demonstrated by
appropriate examples. The acquaintance with the application of these methods
constitutes a fundamental presupposition for the analysis of the dynamic
response of structures.
1.5 Derivation of the equations of motion using

d’Alembert’s principle
Actually, d’Alembert’s principle is a different interpretation of Newton’s
second law of motion. Suppose that we write it in the form
F m€
u¼0 (1.5.1)
@CivilMethod
where F is the resultant of all external forces acting on the particle of mass m and
€ is its acceleration with respect to an inertial frame of reference.c If we consider
u
that the term m€ u is another force, known as inertial force, then Eq. (1.5.1) states
that the vector sum of all forces, external and inertial, is zero during the motion.
But this is the necessary and sufficient condition for the static equilibrium of the
particle. Thus, in a sense, the dynamic problem is reduced to a problem of statics
according to the following statement, known as d’Alembert’s principle.
The laws of static equilibrium can be applied also to a dynamic system with
respect to an inertial frame of reference if the inertial forces are considered as
applied forces on the system together with the actual external forces.
The motion of a rigid body of mass m with respect to an inertial frame of
reference X, Y ,Z is decomposed into a translational motion of its center of
mass, where the whole mass is considered to be concentrated, and a rotational
motion about it (Fig. 1.5.1).
FIG. 1.5.1 Rigid body moving with respect to the inertial. frame X,Y , Z .
If R ¼ X ðt Þi + Y ðt Þj + Z ðt Þk is the position vector of a particle A of

the body with respect to the inertial system of axes XYZ and
r ¼ x ðt Þe1 + y ðt Þe2 + z ðt Þe3 the position of the same point with respect to
the nonrotating system of axes xyz through the center of mass C (see
Fig. 1.5.1), then the equations of motion of the body can be written as
€c
F ¼ mR (1.5.2a)
_c
Mc ¼ H (1.5.2b)
where F ¼ Fx i + Fy j + Fz k is the resultant of the external forces,
€ c ¼ X€ ci + Y€ c j + Z€c k is the acceleration of the center of mass,
R
Mc ¼ Mx e1 + My e2 + Mz e3 is the resultant moment of the external forces with
c. In classical dynamics, an inertial frame of reference is a frame of reference in which a body with
zero force acting upon it is not accelerating; that is, the body is at rest or it is moving at a constant
velocity in a straight line [5].
@CivilMethod
respect to the center to mass, and H_ c is the rate of change of the angular momen-
tum Hc of the body with respect to the same point given as
ZZZ
H_ c¼ r r€rdV (1.5.3)
V
in which r ¼ rðx, y, z Þ is the mass density of the body.

Eq. (1.5.2a) is the equation of the translational motion while Eq. (1.5.2b) is
the equation of the rotational motion.
For a plane body moving in its plane, Eqs. (1.5.2a), (1.5.2b) become (see
Appendix)
Fx ¼ m X€ c (1.5.4a)
Fy ¼ m Y€ c (1.5.4b)
Mc ¼ Ic w_ (1.5.4c)
where w is the angular velocity of the rotational motion about the center of mass
and Ic the polar moment of inertia of the body about the same point.
Path of P
FIG. 1.5.2 Plane body moving in the XY plane. The system of xy axes moves with P without
rotating.
It is often convenient to study the motion with reference to an arbitrary point

P, which is not the center of mass of the body (see Fig. 1.5.2). Special attention
should be paid in this case because Eqs. (1.5.2a), (1.5.2b) take the form
€ p ¼ m€rc
F mR (1.5.5a)
€p¼H
Mp rc m R _p (1.5.5b)
where Rp is the position vector of point P moving with the body and rc the posi-
_ p are the moment of the
tion vector of the center of mass with respect to P. Mp , H
external forces and the rate of change of the angular momentum with respect to
P, respectively.
@CivilMethod
When small displacements are considered, as in the theory of linear vibra-

tions, Eqs. (1.5.5a), (1.5.5b) become (see Appendix).

Fx ¼ m X€ p yc w_ (1.5.6a)

Fy ¼ m Y€ p + xc w_ (1.5.6b)

Mp ¼ Ip w_ + m xc Y€ p yc X€ p (1.5.6c)
The kinetic energy of a plane body moving in its plane is given
(a) with respect to the center of mass
1 2 1
T ¼ m X_ c + Y_ c + Ic w2
2
(1.5.7)
2 2
(b) with respect to an arbitrary point P of the body (K€onig’s theorem)
1 2 1
T ¼ m X_ p + Y_ p + Ip w2 + m xc Y_ p yc X_ p w
2
(1.5.8)
2 2
We shall write now Eqs. (1.5.4a)–(1.5.4c) in terms of the displacement
vector. Apparently, the displacement vector from the beginning of the motion
is defined as
u ¼ Rðt Þ Rð0Þ ¼ u ðt Þi + v ðt Þj (1.5.9)
where
u ¼ X ðt Þ X ð0Þ, v ¼ Y ðt Þ Y ð0Þ (1.5.10)
Hence, X€ ¼ u,
€ Y€ ¼ v€. Moreover, if fðt Þ represents the change of the rota-
_ w_ ¼ f,
tion in the same time interval and set w ¼ f, € Eqs. (1.5.4a)–(1.5.4c) are
written in terms of displacements as
Fx ¼ m u€c (1.5.11a)
Fy ¼ m v€c (1.5.11b)
Mc ¼ Ic f€ (1.5.11c)
or in matrix form
€c
Fc ¼ m c U (1.5.12)
where
8 9 8 9 2 3
< Fx = < uc = m 0 0
Fc ¼ Fx , Uc ¼ vc , mc ¼ 4 0 m 0 5 (1.5.13)
: ; : ;
Mc f 0 0 Ic
are the force vector, the displacement vector, and the mass matrix of the body,
respectively.
@CivilMethod
Similarly, Eqs. (1.5.6a)–(1.5.6c) are written

Fx ¼ m u€p yc f€ (1.5.14a)

Fy ¼ m v€p + xc f€ (1.5.14b)

Mp ¼ Ip f€ + m xc v€p yc u€p (1.5.14c)
or in matrix form
€p
Fp ¼ m p U (1.5.15)
where
8 9
>
< Fx >
=
Fp ¼ F x (1.5.16a)
>
: >
;
Mp
8 9
< up >
> =
Up ¼ vp (1.5.16b)
: >
> ;
f
2 3
m 0 my c
mp ¼ 4 0 m mx c 5 (1.5.16c)
my c mx c Ic
Note that the mass matrix is not diagonal when the point of reference is not
the center of mass.
Finally, Eqs. (1.5.7), (1.5.8) are written as
1 1
T ¼ m u_ 2c + v_ 2c + Ic f_
2
2 2
(1.5.17)
1_T _c
¼ U mc U
2 c
1 1
T ¼ m u_ 2p + v_ 2p + Ip f_ + m xc u_ p yc v_ p f_
2
2 2
(1.5.18)
1_T _p
¼ U mp U
2 p
The set of equations with reference to point P can also be derived from the
set of equations with reference to point C by transforming the displacements
and the forces from point C to P (see Section 10.7).
Example 1.5.1 Equation of motion of an elastically supported body
Consider the rigid plate of constant thickness and total mass m shown in
Fig. E1.1a. The plate is hinged at O and elastically supported at A. Formulate
@CivilMethod
the equation of motion of the system for small amplitude motion using the
method of equilibrium of forces.
Solution
The only possible motion of the plate is the rotation in its plane about the point
O. Hence, the system has one degree of freedom. The motion can be described
either by the rotation fðt Þ about O or by the translational displacement of
a point, for example, the displacement u ðt Þ of point B, which is related
to fðt Þ as
u ðt Þ ¼ a tan fðt Þ afðt Þ, u ðt Þ ¼ BB 0 cos f BB 0 (1)
because we have assumed small displacements. Moreover, AA0 ¼ BB 0 =2 ¼ u=2

The applied forces are shown in the free body diagram in Fig. E1.1b. These
are:
The weight of the body:
W ¼ mg (2a)
The spring force:
2
fS ¼ k ðAA0 Þ ¼ ku (2b)
3
The inertial force at the center of mass along x
d2 1 b
fIx ¼ m ðCC 0 Þx ¼ m u€ (2c)
dt 2 2 a
The inertial force at the center of mass along y
d2 1
fIy ¼ m 2
ðCC 0 Þy ¼ m u€ (2d)
dt 2
The inertial moment about the center of mass
u€
MIc ¼ IC f€ ¼ IC (2e)
a
The external force

pðt Þ (2f)
The quantities ðCC 0 Þx and ðCC 0 Þy are the horizontal and the vertical dis-
placements of the center of mass C due to rotation, respectively. They are
obtained from Fig. E1.1b as
@CivilMethod
(a)
(b)
(c)
FIG. E1.1 Rigid plate in Example 1.5.1.
1b
ðCC 0 Þx ¼ ðOC Þf sin b ¼ u (3a)
2a
1
ðCC 0 Þy ¼ ðOC Þf cos b ¼ u (3b)
2
The equation of motion results from the dynamic equilibrium of moments
with respect to point O. Thus, we obtain
a 2a b a
W fS fIx fIy MIc + pðt Þa ¼ 0 (4)
2 3 2 2
@CivilMethod
which by virtue of Eq. (2) becomes

" #
a 2 b2
m + + IC
2 2 4 W
u€ + ku ¼ + pðt Þ (5)
a2 9 2
Using Steiner’s formula, we have
" #
a 2 b2 a 2 + b2
IO ¼ m + + IC ¼ m (6)
2 2 3
Hence, Eq. (5) becomes

a 2 + b2 4 W
m 2
u€ + ku ¼ + pðt Þ (7)
3a 9 2
Eq. (7) can be also obtained if we consider the motion with reference to point
O and employ Eq. (1.5.14c) for
u€ 2a a
u€O ¼ v€O ¼ 0, f€ ¼ IO , MO ¼ fS + W + pðt Þa, MIo ¼ IO f€
a 3 2
The weight W can be eliminated from Eq. (7), if the total displacement u ðt Þ
is expressed as the sum of the static displacement ust caused by the weight plus
the additional dynamic displacement uðt Þ, as shown in Fig. E1.1c, that is,
u ðt Þ ¼ ust + uðt Þ (8)
The static equilibrium of moments with respect to point O, when the plate is
loaded only by the weight yields (see Fig. E1.1c)
4 a
kau st ¼ W (9)
9 2
Noting that u€st ¼ 0 because ust is a constant, and using Eqs. (8), (9), Eq. (7)
becomes
m ∗ u€
+ k ∗ u ¼ p ∗ ðt Þ (10)
where
a 2 + b2 4
m∗ ¼ m 2
, k ∗ ¼ k, p ∗ ðt Þ ¼ pðt Þ
3a 9
Eq. (10) has the form of Eq. (1.4.8) and represents the equation of motion of
the system. The quantities m ∗ , k ∗ , having dimensions of mass and translational
stiffness, respectively, are referred to as the generalized mass and the general-
ized stiffness of the SDOF system.
If the rotation fðt Þ, measured from the position of static equilibrium, is
taken as the parameter of motion in place of uðt Þ, the equation of motion results
from Eq. (10) using the relation u ¼ fa. Thus, we have
4
IO f€ + ka2 f ¼ apðt Þ (11)
9
@CivilMethod
Example 1.5.2 Equation of motion of a frame with a rigid column

Formulate the equation of motion of the plane frame shown in Fig. E1.2a for
small amplitude motion. The mass of the horizontal beam CD is negligibly small
while the column of height L and nonnegligible width h ¼ L=4 is assumed rigid
The elastic stiffness of the ground is simulated by the
with total mass m ¼ mL.
spring CR while its damping by the two dashpots with damping parameters c.
(a)
(b)
FIG. E1.2 Frame with a rigid column in Example 1.5.2.
Solution
The only possible motion of the system is the rotation fðt Þ of the column as a
rigid body about the hinged support at point A of its base. Because the rotation is
small, we have:
sin f f, cos f 1, f2 0
Hence
h h
u ¼ L sin f ¼ Lf, d ¼ sin f f
2 2
@CivilMethod
h h
ð1 2Þ ¼ sin f + L cos f f + L,
2 2
h h
ð3 4Þ ¼ L sin f + cos f Lf + ,
2 2
h h
ð5 6Þ ¼ cos f
2 2
The forces applied on the column are shown in Fig. E1.2b. These are:
The elastic moment at the corner C MS ¼ ð1:5L
6EI
Þ2
4EI
d + 1:5L f
The elastic moment due to the rotational spring MR ¼ CR f ¼ EI
L f
MIA ¼ IA f€ ¼ mL €
2
The moment of inertia of the mass m 3 f
The elastic shear force at the beam end C QS ¼ ð1:5L

12EI
Þ3
6EI
d + ð1:5L Þ2
f
The damping forces fD ¼ c h2 f_
The external load pðt Þ
The equilibrium of moments with respect to point A yields
MIA + MS + MR + QS ð3 4Þ + 2fD ð5 6Þ pðt Þ ð1 2Þ ¼ 0
which after substituting their exressions becomes

mL2 € cL2 _ 79EI pL 28EI 2
f+ f+ f+ f ¼ Lpðt Þ (1)
3 32 18L 8 9L
Further, linearizing (f2 0) gives

mL2 € cL2 _ 79EI pL
f+ f+ f ¼ Lpðt Þ (2)
3 32 18L 8
If the displacement u ¼ Lf at the level of the beam is taken as the parameter

of the motion, the equation of motion becomes

m cL 79EI p
u€ + u_ + u ¼ pðt Þ (3)
3 32 18L3 8L
Example 1.5.3 Equation of motion of a system of rigid bodies

The rigid body assemblage shown in Fig. E1.3a consists of the rigid bar AF of
total mass m hinged at A, and the rigid square plane body supported rigidly at F.
The dynamic excitation of the bar is due to the uniformly distributed transverse
load pðt Þ. The motion is constrained by a spring at B and the damper at G. For-
mulate the equation of motion of the system for small amplitude displacements
using the method of equilibrium of forces. The mass per unit length of the bar is
m ¼ m=3L and the surface mass density of the body is g ¼ 2m=L2
@CivilMethod
(a)
(b)
FIG. E1.3 System with two rigid bodies in Example 1.5.3.
Solution
As the bar AF is rigid, the only possible motion is its rotation about A. Hence,
the system has a SDOF. Its motion can be described either by the angle of rota-
tion fðt Þ about the hinge at A or by the transverse displacement of any point
along the axis of the bar. We choose the upward displacement u ðt Þ at point
B as the parameter of the motion. For small amplitude motion, the forces acting
on the system are shown in Fig. E1.3b. These are:
The elastic force fS at B: As it opposes the motion, it is directed downward
and is expressed as
fS ¼ ku (1)
The damping force fD at G: It is directed also downward and is expressed as
d d
fD ¼ c ðGG 0 Þ ¼ c ð1:625u Þ ¼ 1:625cu_ (2)
dt dt
The inertial force fIK and the inertial moment MIK at the center of mass K of
the bar due the distributed mass m are
d2
Þ
fIK ¼ ðm3L ðKK 0 Þ ¼ 0:75m u€ (3)
dt 2
MIK ¼ IK f€
@CivilMethod
or taking into account that

ð3LÞ3 m ð3LÞ3

IK ¼ m ¼ ¼ 0:75mL2
12 3L 12
u u u€
f ¼ ¼ 0:5 , f€ ¼ 0:5
2L L L
we obtain
MIK ¼ 0:375mLu€ (4)
The inertial force fIG and the inertial moment MIG at the center of mass G of
the rigid body due to the mass gL2 =2:

L d2 2m L2 d 2
fIG ¼ gL 2
ðGG 0 Þ ¼ 2 ð1:625u Þ ¼ 1:625m u€ (5)
2 dt L 2 dt 2
" #
€ LðL=2Þ3 ðL=2ÞL3 u€
M I ¼ IG f ¼ g
G
+ 0:5 ¼ 0:052mLu€ (6)
12 12 L
The external load is 2Lpðt Þ.

The equilibrium of the moments about A yields the equation of motion of the
system. Thus, we have
fS ð2LÞ fD ð3:25LÞ fIK ð1:5LÞ fIG ð3:25LÞ

(7)
MIK MIG + pðt Þ ð2LÞ L ¼ 0
or inserting Eqs. (1)–(6) into Eq. (7) we obtain
m ∗ u€ + c ∗ u_ + k ∗ u ¼ p ∗ ðt Þ (8)
where
m ∗ ¼ 6:833m, c ∗ ¼ 5:281c, k ∗ ¼ 2k, p ∗ ðt Þ ¼ 2L
pðt Þ (9)
The quantities defined by Eq. (9) are referred to as the generalized mass, the
generalized damping, the generalized stiffness, and the generalized load,
respectively.
Once the dynamic displacement u ðt Þ is established from the solution of
Eq. (8), the vertical reaction RA can be evaluated from the dynamic equilibrium
of forces in the direction of the y axis. This yields
RA + pðt Þ2L fIK fIG fS fD ¼ 0
or using Eqs. (1)–(3), (5) we obtain
RA ¼ ku + 1:625cu_ + 2:375m u€ 2L

pðt Þ
@CivilMethod
Example 1.5.4 Equation of motion of a single-story shear building

Formulate the equation of motion of the single-story building shown in Fig. E1.4a.
The damping is neglected. The columns are fixed on the ground, are inextensible,
and their mass is assumed to be lumped at their ends. Moreover, the roof plate
is assumed rigid. The material of the structure is reinforced concrete having spe-
cific weight g ¼ 24kN=m3 and modulus of elasticity E ¼ 2:1 107 kN=m2 . The
total load of the plate (dead and live) is 20kN=m2 . The force pðt Þ acts in the direc-
tion of the x axis and is given by pðt Þ ¼ 20sin 13t. The acceleration of gravity is
g ¼ 9:81m=s2 and the dimensions of the rectangular cross-sections of columns are
k1 : 30 30cm2 and k2 : 30 20cm2 .
(b)
(a)
FIG. E1.4 Single-story shear building in Example 1.5.4.
Solution
Taking into account that the structure is symmetric with respect to the x axis,
the columns are inextensible, and the load pðt Þ acts on the axis of symmetry, the
only possible motion of the plate is the horizontal displacement u ðt Þ in the
direction of the x axis. The SDOF model of the structure is shown in Fig. E1.4b.
The total mass of the system is due to the load of the plate and to half the
weight of the columns
5 10 20 + ð4 0:3 0:3 + 2 0:3 0:2Þ 2 24
m¼ ¼ 104:285
9:81
The stiffness of the system is equal to the sum of the translational stiffness
coefficients of all columns, which are given as
12EI i
ki ¼
hi3
where Ii is the moment of inertia of the cross-section of the i column with
respect to the y axis through its mass center and hi its height. Thus, we have
@CivilMethod
Columns 30 30:
0:304
12 2:1 107
k3030 ¼ 12 ¼ 2657:8kN=m
43
Columns 30 20:
0:303 0:20
12 2:1 107
k3020 ¼ 12 ¼ 1771:9kN=m
43
Therefore the stiffness of the system is
k ¼ 4 2657:8 + 2 1771:9 ¼ 14175:0kN=m
The equation of motion results from the equilibrium of the forces shown in
Fig. E1.4b. This yields
fI fS + pðt Þ ¼ 0
or
m u€ + ku ¼ pðt Þ
Substituting the numerical values for m, k and the expression for pðt Þ, the
above equation of motion becomes
5:21u€ + 708:75u ¼ sin 13t
Example 1.5.5 Equation of motion of a two-story shear frame

Formulate the equations of motion of the two-story shear frame shown in
Fig. E1.5a using the method of equilibrium of forces. The damping is ignored.
Solution
The system has two degrees of freedom because the girders are rigid and the
axial deformation of columns is ignored. The model of the structure is shown
in Fig. E1.5b. The masses are lumped at the story levels. The motion of the
system can be fully determined from the horizontal displacements u1 ðt Þ and
u2 ðt Þ of the masses m1 and m2 , respectively.
Rigid
Rigid
(a) (b) (c)

FIG. E1.5 Two-story shear frame in Example 1.5.5.
@CivilMethod
The equations of motion result from the dynamic equilibrium of forces

applied to the masses m1 and m2 . These forces are shown in the free body dia-
grams of Fig. E1.5c. Thus, we obtain
m1 u€1 + k1 ðu1 u2 Þ ¼ p1 ðt Þ (1)
m2 u€2 k1 ðu1 u2 Þ + k2 u2 ¼ p2 ðt Þ (2)
Eqs. (1), (2) are written in matrix form as
m€
u + ku ¼ pðt Þ (3)
where

u1 m1 0 k1 k1 p1 ðt Þ
u¼ , m¼ , k¼ , pðt Þ ¼
u2 0 m2 k1 k1 + k2 p2 ðt Þ
Example 1.5.6 Equation of motion of a general single-story shear building

The rigid horizontal plate is supported by K columns as shown in Fig. E1.6. The
columns are fixed on the ground as well as on the plate. Their principal axes have
arbitrary directions in the xy plane. Formulate the equation of motion of the
plate when the plate is loaded by the horizontal load P ðt Þ through the point A.
Solution
We choose O xy as the system of reference of the motion, whose origin coin-
cides with point O at the beginning of motion. Let xi , yi represent the coordi-
nates of the center of mass of the cross-section of i column and fi the angle
between its principal x axis and the x axis. The axes xy will be referred to
as the global axes of the system while the axes xy as the local axes of the
column.
Inasmuch as the axial deformation of columns is ignored, the only possible
motion of the plate is inside its plane, which can be determined by the two
translational displacements of a point and the rotation of the plate. We study
the motion of the plate with reference to point O and let U , V represent its
translational components with respect to the global axes xy, which are
assumed fixed in the plane, and W the rotation of the plate. As a consequence
of this motion, the cross-section of the i column at the level of the plate
undergoes the displacements u i , v i , wi , with respect to its base. These displace-
ments generate elastic forces X i , Y i , M i , which act on the plate. Thus,
we define the following vectors and matrices that will be used in the subse-
quent analysis.
@CivilMethod
FIG. E1.6 General single-story shear building in Example 1.5.6.
(a) In global axes:

8 9
< U =
¼ V displacements of point O of the plate
U
:;
W
8 9
< ui =
i ¼ vi displacements of i column
D
: i;
w
8 i9
< X =
i ¼ Y i elastic forces of i column
F S
: i;
M
(b) In local axes
8 9
< ui =
D ¼ v i displacements of i column
i
: i;
w
8 9
< Xi =
FiS ¼ Y i elastic forces of i column
: i;
M
@CivilMethod
The transformation matrix for the vector quantities related to i column from the
global axes to the local axes is given as
2 3
cos fi sin fi 0
Ri ¼ 4 sin fi cos fi 0 5
0 0 1
Hence, the vectors are transformed from one system of axes to the other as
i
D i ¼ Ri D (1a)

i ¼ Ri T D i
D (1b)
i
FiS ¼ Ri F (2a)
S

i ¼ Ri T F i
F (2b)
S S
where
2 3
cos fi sin fi 0
R ¼ 4 sin fi
i T
cos fi 0 5
0 0 1
i 1 i T
is the transpose of R . Note that R
i
¼ R because Ri is orthonormal.
The elastic forces X , Y , M are related to the displacements u i , v i , wi by
i i i
12EI y i
Xi ¼ u (3a)
h3
12EI x i
Yi ¼ v (3b)
h3
GI t i
Mi ¼ w (3c)
h
where Ix ,Iy are the principal moments of inertia of the column cross-section and
It is the torsional constant, E and G are the material constants, and h is the
height of the column.
Setting
12EI y 12EI x GI t
i
k11 ¼ i
, k22 ¼ i
, k33 ¼ (4)
h3 h3 h
Eqs. (3a)–(3c) can be written in matrix form as
8 9 2 i 38 9
< Xi = k11 0 0 < u i =
Y i ¼ 4 0 k22 i
0 5 vi
: i; i : i;
M 0 0 k33 w
or
Fi ¼ k i D i (5)
@CivilMethod
The matrix
2 i
3
k11 0 0
6 7
ki ¼ 4 0 i
k22 0 5
i
0 0 k22
is the stiffness matrix of i column.

Eq. (5) is transformed in global axes using Eqs. (2b), (1a). Thus, we have

i ¼ Ri T F i
F S S
T
¼ Ri k i D i
T
i
¼ Ri k i R i D
or
F i D
i ¼ k i (6)
where

i ¼ Ri T k i R i
k (7)
is the stiffness matrix of the column in global axes, which becomes after per-
forming the matrix multiplications
2 i 3
k11 k12 0
i
6 7
i ¼ 6 ki ki 0 7
k (8)
4 21 22 5
0 0 k
i
33
where
9
k11 ¼ k11 >
i i
cos 2 fi + k22
i
sin 2 fi >
>
>
>
>
=
k22 ¼ k11
i i
sin 2 fi + k22
i
cos 2 fi
i (9)
k12 ¼ k21 ¼ k11 sin fi cos fi >
i i
k22
i >
>
>
>
>
;
i
k 33 ¼ k33
i
Inasmuch as the plate is rigid, the displacements ui , vi , wi of the i column
of point O. The geometrical rela-
depend on the plate displacements U , V , W
tions result from the following consideration.
@CivilMethod
FIG. E1.7 Displacements of point i due to rotation of plate.
The point i of the plate undergoes translational displacements due to

(a) The translational displacements of point O
i
u t ¼ U
i
v ¼ V
t
(b) The rotation of the plate about O. Referring to Fig. E1.7 and observing that
¼ wi , we obtain
cos ai ¼ xi =ri , sin ai ¼ yi =ri , W
i
u r ¼ ri W sin ai ¼
yi W
i
v r ¼ ri W cosai ¼ xi W

Thus, we have

ui ¼ ui t + ui r ¼ U yi W (10a)

vi ¼ vi t + vi r ¼ V + xi W (10b)

wi ¼ W (10c)
The previous equations are written in matrix form as
8 9 2 38 9
< ui = 1 0 y i < U =
vi ¼ 4 0 1 xi 5 V (11)
: i; :;
w 0 0 1 W
or setting
2 3
1 0
yi
ei ¼ 4 0 1 xi 5
0 0 1
@CivilMethod
we can write Eq. (11) as

i ¼ ei U
D (12)
The matrix e defined by Eq. (12) is referred to as the translation matrix or
i
transformation matrix of the i column.

The equations of motion result from Eqs. (1.5.14a)–(1.5.14c) if they are
employed for point O, with P O, up ¼ U , vp ¼ V , f ¼
W. The external force
ðt Þ ¼ P x ðt Þ, P y ðt Þ T . Thus,
F is equal to the sum of all elastic forces FiS plus P
we have
X
K
P x ðt Þ X ¼ m U€
i €
yc W (13a)
i¼1
X
K
P y ðt Þ Y ¼ m V€
i €
+ xc W (13b)
i¼1
K
X
ðt Þ
M xi Y yi X + M
i i €
i ¼ m xc V€ yc U€ + Io W (13c)
i¼1
where M ðt Þ with respect

ðt Þ ¼ xA P y ðt Þ yA P x ðt Þ is the moment of the force P
to O, m is the mass of the plate, and Io its moment of inertia with respect to O.
Eqs. (13a)–(13c) are written in matrix form as
8 9 2 38 i 9 2 32 32 38 9
€

> = XK 1 0 0 > < X >= 1 0 0 m 0 0 yc >
1 0 
=
6 7 6 76 76 7
P y ðt Þ 4 0 1 0 5 Y i
¼ 4 0 1 0 54 0 m 0 54 0 1 xc 5 V€
>
: >
; i¼1 >
: i> ; >
: € >
;
M ðt Þ y i xi 1
M y c xc 1
0 0 Ic 0 0 1 W
or
X
K
ðt Þ
P
T i
ei F €
¼ ðec ÞT mðec ÞU (14)
i¼1
where
8 9


> =
ðt Þ ¼ P y ðt Þ
P (15a)
: >
> ;
M ðt Þ
2 3
m 0 0
m ¼ 40 m 0 5 (15b)
0 0 Ic
Finally, using Eqs. (6), (12), we obtain the equation of motion
M € + K
U U ¼P
ðt Þ (16)
@CivilMethod
where
2 3
m 0 m yc
¼ ðec ÞT mec ¼ 4 0
M m m xc 5 (17a)
m yc m xc Io
X
K
¼
K ei
T T
Ri ki Ri ei (17b)
i¼1
are the mass and stiffness matrices of the structure, respectively.

The equation of motion (16) can be transformed with reference to the center
of mass by working as follows.
Using Eq. (12), we relate the displacements of the center of mass C to the
displacements of O. Hence, we have
U
c ¼ ec U (18)
which can be inverted to give
¼ ðec Þ1 U
U c (19)
We can readily show that
2 3
1 0 yc
ðec Þ1 ¼ 4 0 1 xc 5 (20)
0 0 1
Substituting Eq. (19) into Eq. (16) and premultiplying it byðec ÞT , we
obtaind
€ + K
c
mU cU ðt Þc
c ¼P (21)
where
c ¼ ðec ÞT K
K ðec Þ1 (22a)

c ðt Þ ¼ ðec ÞT P
P ðt Þ (22b)
Eq. (22a) represents the transformed stiffness matrix of the structure from
point O to the center of mass C .
The stiffness matrix with respect to an arbitrary point O is, in general, a full
3 3 matrix, namely
2 3
k11 k12 k13
¼6
K
7
4 k21 k22 k23 5 (23)
k31 k32 k33
T
d. The notation eT ¼ ðe1 Þ is employed.
@CivilMethod
The off-diagonal terms cause coupling between the elastic force in one
direction and the displacement in another direction. For example, the element
k12 represents the force acting in the direction of the x axis when the plate
undergoes a unit displacement in the direction of the y axis. Similarly, the ele-
ment k31 represents the moment acting on the plate about the z axis, if the plate
undergoes a unit displacement, U ¼ 1, in the direction of the x axis. The elastic
center or center of resistance of the plate is defined as the point of the plate
where an applied force in any direction does not produce rotation. This implies
the vanishing of the elements k13 and k23 (hence also their symmetric k31 and
k32 ) in the stiffness matrix (23). This point can be established as follows.
The stiffness matrix K is transformed from point O to the sought elastic cen-
ter E ðxE , yE Þ according to Eq. (22a), if ec is replaced by eE . Namely
2 32 32 3
1 0 0 k11 k12 k13 1 0 yE
T 1
E ¼ eE
K eE
K ¼ 40 1 0 54 k21 k22 k23 54 0 1 xE 5
yE xE 1 k31 k32 k33 0 0 1
or after performing the matrix multiplications
2 E E E3
k11 k12 k13
6 7
6 E E E7

K ¼ 6 k21 k22 k23 7
E 6
7
4 5
E E E
k 31 k 32 k 33
2 3
k11 k12 k11 yE k12 xE + k13
6 7
6 7
¼ 6 k21 k22 k21 yE k22 xE + k23 7
4 5
k11 yE k21 xE + k31 k12 yE k22 xE + k32 2 2
k 13 yE + k 23 xE + k 33
The vanishing of the elements k13 and k23 yields
E E
k11 yE k12 xE + k13 ¼ 0

k21 yE k22 xE + k23 ¼ 0
from which we establish the coordinates of E

k 11 k 13
k k
xE ¼
21 23 (24a)
k 11 k 12
k21 k22

k 12 k13
k k23
yE ¼
22 (24b)
k 11 k12
k21 k22
@CivilMethod
Thus, the stiffness matrix with respect to the elastic center takes the form
2 E E 32 3
k11 k12 0 k11 k12 0
6 7
E ¼ 6 kE kE 0 76
K 7
4 21 22 54 k 21 k 22 0 5

0 0 k 13 yE + k 23 xE + k 33
0 0 k
E 2 2
33
The coupling is now limited between the translational displacements and

the corresponding elastic forces. They can also be decoupled if the matrix is
transformed into a new system of axes x 0 Ey 0 , resulting from xE
y by rotation
through an angle q and demanding the off-diagonal elements to vanish. The
stiffness matrix in the new system is obtained using Eq. (7) for ki ¼ K
E and
2 3
cos q sin q 0
R ¼ 4 sin q cos q 0 5
0 0 1
Thus, we have
K ER
^ E ¼ RT K (25)
or after performing the matrix multiplications
2 3
k11 k22
6 k cos 2 q + k sin 2 q k sin 2q
sin 2q + k 12 cos2q 0 7
6 11 22 12 2 7
^ 6 7
K ¼ 6 k 11 k 22
E 7
6
4 sin 2q + k12 cos 2q k11 sin q + k22 cos q + k12 sin 2q 0
2 2 7
5
2
0 0 k13 y2 + k23 x 2 + k33
E E
The vanishing of the off-diagonal elements yields
2k12
tan 2q ¼ (26)
k 22 k11
The axes defined by angle q are referred to as the principal directions of stiff-
ness of the structure. The stiffness matrix becomes now diagonal
2 3
k^11 0 0
KÊ ¼6 4 0 k^22 0 5
7
0 0 k^33
where
k^11 ¼ k11 cos 2 q + k22 sin 2 q k12 sin 2q (27a)
k^22 ¼ k11 sin 2 q + k22 cos 2 q + k12 cos 2q (27b)
k^33 ¼ k13 y2E + k23 x2E + k33 (27c)
@CivilMethod
The previous analysis allows us to draw the following conclusions:

(a) In static analysis, the concepts of the elastic center and the principal direc-
tions permit the uncoupling of the three equations of static equilibrium and
give a better insight into the deformation of the structure. These concepts,
however, do not have any advantage in the dynamic analysis because, in
general, the center of mass does not coincide with the elastic center and
therefore the equations of motion remain coupled through the second deriv-
atives of the displacements. Apparently, this fact complicates the study of
the dynamic analysis.
(b) The equations of motion can be decoupled with respect to the physical dis-
placements U , V , W only if the center of mass coincides with the elastic
center. This uncoupling should not be confused with that achieved via
modal coordinates (see Chapter 12).
1.6 Principle of virtual displacements

D’Alembert’s principle allows the application of the principle of virtual dis-
placements to formulate the equations of motion of structural systems, espe-
cially for complex assemblages comprising a number of interconnected
particles or rigid bodies of finite size. The principle of virtual displacements
or virtual works can be expressed as follows:
The necessary and sufficient condition for the dynamic equilibrium of a
system is the vanishing of the total work done by the set of all externally applied
forces (actual and inertial) when the system is subjected to a virtual displace-
ment, that is, a displacement pattern compatible with the geometrical con-
straints of the system.
Thus, the equations of motion of the dynamic system can be derived by first
identifying all forces (imposed external forces, elastic forces, damping forces,
and inertial forces) acting on the masses. Then by introducing a virtual displace-
ment corresponding to each degree of freedom, the equations of motion are
obtained by setting the virtual work produced by all forces equal to zero. A
major advantage of this approach is that the contribution of the work done
by the reactions of nonyielding bilateral supports as well as the internal inter-
action forces on the separated masses do not appear explicitly in the equations.
Moreover, the quantities we have to deal with are scalar and not vectorial, thus
they can be algebraically manipulated. An important provision for the applica-
tion of the principle of virtual displacements is that the masses of the system are
subjected to small displacements. This implies that the geometry of the structure
remains essentially unchanged after the action of the displacements.
Example 1.6.1 Equation of motion of a complex SDOF system
Formulate the equation of motion of the system shown in Fig. E1.8a by using the
principle of virtual displacements for small amplitude motion. It is assumed that
the cable can undertake compression.
@CivilMethod
(a)
(b)
FIG. E1.8 System in Example 1.6.1.
Solution
Because the cable is inextensible, the displaced configuration of the system can
be specified either by the angle of rotation of one of the bars or by the transverse
displacement of a point on it. Thus, the system has only one degree of freedom.
If the upward transverse displacement u ðt Þ of point C is taken as the basic
parameter of the motion, then all other displacements can be expressed in terms
of it. Fig. E1.8b shows the deformed system with all forces applied to it.
The elastic forces fS1 and fS2 are due to the deformation of the springs k1 and
k2 . They are directed downward as they oppose the motion. The force fD is
due to the viscous damping mechanism and is directed upward as it also
opposes the motion. The inertia moments MIA , MIO , and MIE are due to the rota-
tion of the masses about A, O , and E, respectively. All forces are expressed in
terms of the single displacement u ðt Þ
fS1 ¼ k1 ðBB 0 Þ ¼ ku=2, fS2 ¼ k2 ðCC 0 Þ ¼ 2ku
d
fD ¼ c ðDD 0 Þ ¼ cu_
dt
ð2LÞ3 u€
m
MIA ¼ IA f€1 ¼ 2 u€
¼ 1:333mL
3 2L
ð1:5LÞ3 u€
m
MIE ¼ IE f€2 ¼ 2 u€
¼ 0:750mL
3 1:5L
ð0:8LÞ2 u€
MIO ¼ IO f€3 ¼ mL
2 u€
¼ 0:200mL
8 0:4L
@CivilMethod
If point C is given a virtual displacement du, the forces ride the following
displacements
d ðCC 0 Þ ¼ du, dðBB 0 Þ ¼ du=2, dðDD0 Þ ¼ du
df1 ¼ du=2L, df2 ¼ du=1:5L, df3 ¼ du=0:4L
du ðx Þ ¼ xdf1 ¼ xdu=2L
The work done by the forces acting on the system due to the virtual displace-
ment should be set equal to zero, that is,
fS1 d ðBB 0 Þ fS2 dðCC 0 Þ fD dðDD 0 Þ MIA df1
Z L
(1)
MIE df2 MIO df3 + pðt Þdu ðx Þdx ¼ 0
0
Using the expressions for the forces and the displacements in terms of the
basic displacement derived previously, Eq. (1) yields

0:25ku 2ku cu_ 1:333mL
2 u=2L
€ 2 u=1:5L
0:750mL €
0:200mL €
2 u=0:4L + pðt ÞL=4du ¼ 0
or, inasmuch as du 6¼ 0, the expression within the square brackets should vanish.
This yields the equation of motion
m ∗ v€ + c ∗ v_ + k ∗ v ¼ p ∗ ðt Þ (2)
where

m ∗ ¼ 1:667mL, c ∗ ¼ c, k ∗ ¼ 2:25k, p ∗ ðt Þ ¼ 0:25
pðt ÞL
Example 1.6.2 Equation of motion of a rigid body assemblage
Formulate the equations of motion of the rigid body assemblage shown in-
Fig. E1.9a by using the principle of virtual displacements on the basis of small
amplitude motion.
Solution
Due to the spring k1 , the rigid bars can rotate independently from each other
about their hinged supports at A and F. Hence, the system has two degrees
of freedom. Its motion can be specified by the transverse downward displace-
ments u1 ðt Þ and u2 ðt Þ of points C and E, respectively. The forces applied to the
displaced system are shown in Fig. E1.9b. They are
The elastic force fS1 ¼ k1 ðCC 0 Þ ¼ k ðu2 u1 Þ
The elastic force fS2 ¼ k2 ðDD 0 Þ ¼ 4ku 2
The damping force fD ¼ c dtd ðBB 0 Þ ¼ c u_21
The inertial moment MIA ¼ IA f€1 ¼ IA 2a u€1 2
¼ 4ma
3 u €1
€
The inertial moment M ¼ IF f2 ¼ IF ¼
F u€2 2
8ma
u€2
I a 3
The system is given a virtual displacement pattern du1 and du2 corresponding
to the two degrees of freedom. The forces ride the following displacements
du1 du1
d ðBB 0 Þ ¼ , dðCC 0 Þ ¼ du1 , df1 ¼
2 2a
@CivilMethod
(a)
(b)
du2
d ðDD 0 Þ ¼ 2du2 , d ðEE 0 Þ ¼ du2 , df2 ¼
a
According to the principle of virtual displacements, the work done by the
applied forces must be equal to zero, that is,
MIA df1 fD dðBB 0 Þ + fS1 dðCC 0 Þ fS1 dðEE 0 Þ fS2 dðDD 0 Þ

(1)
MIF df2 + pðt Þd ðCC 0 Þ ¼ 0
Introducing the expressions of the forces and virtual displacements into

terms of the basic displacements in Eq. (1) yields

2ma u_ 1
u€1 c + k ðu2 u1 Þ + pðt Þ du1
3 4

(2)
8ma
+ u€2 k ðu1 u2 Þ 8ku 2 du2 ¼ 0
3
Inasmuch as the quantities du1 and du2 are arbitrary, Eq. (2) is valid
only if

2ma u_ 1
u€1 + c k ðu2 u1 Þ pðt Þ ¼ 0 (3a)
3 4

8ma
u€2 + k ð9u2 u1 Þ ¼ 0 (3b)
3
@CivilMethod
Eqs. (3a), (3b) are the equations of motion of the system. In matrix form they
are written as
2 3 2 c 3( ) "

2ma #( ) ( )
6 3 0 7 u€1 0 u_ 1 k k u1 pðt Þ
4 44 5 ¼
5 u€2 +
8ma u_ 2
+
k 9k u2 0
(4)
0 0 0
3
1.7 Hamilton’s principle

The development of dynamics and generally of mechanics has been accom-
plished through two different approaches. The first is based on Newton’s laws
of motion. These laws deal with the motion of a body under the action of forces
acting on it. The involved quantities are (i) the imposed forces, which may be
externally applied forces, forces of interaction between the masses, and reactions
of constraints and (ii) the momentum or the quantity of motion as defined by
Newton. Because both quantities are vector quantities, this approach of mechan-
ics is called vectorial mechanics. The analysis of complicated systems by direct
application of Newton’s laws of motion becomes increasingly difficult. The prin-
cipal reason is that the equations are vectorial in nature and the forces and accel-
erations are often difficult to determine. Moreover, the reactions of the constraints
and the interaction forces between bodies must be explicitly accounted in the
equations of motion and have to be evaluated even when there is no interest to
evaluate them. In addition, each problem seems to require its own particular insi-
ghts and there are no general procedures for obtaining the equations of motion.
The second approach is based mainly on the work of Lagrange and Hamilton
and is called analytical mechanics. In this approach, the involved quantities are
scalar functions, and therefore the fundamental equations, in contrast to vectorial
mechanics, do not depend on the choice of the coordinates. Also, it is not neces-
sary to include explicitly the forces of the constraints and the interaction forces. It
will be shown that this approach circumvents to some extent the difficulties found
in the direct application of Newton’s law of motion to complicated systems. Fur-
thermore, the equations of motion are formulated in a standard convenient form.
Analytical dynamics is based on Hamilton’s principle and Lagrange’s equations.
Hamilton’s principle is presented in this section. Lagrange’s equations are pre-
sented in the next section resulting directly from Hamilton’s principle.
One of the most important principles of dynamics is Hamilton’s principle,
named after the famous Irish mathematician and physicist Sir William Rowan
Hamilton (1805–65). Inertial and elastic forces are not explicitly involved in
this principle; instead, variations of the kinetic and potential energy are utilized.
This formulation has the advantage of dealing only with purely scalar quanti-
ties. In the procedure of virtual displacements, even though the works them-
selves are scalar quantities, vector quantities, displacements, and forces are
utilized to represent them. Hamilton’s principle is presented here for discrete
parameter systems.
@CivilMethod
FIG. 1.7.1 Particle moving in space.
Consider a particle of mass m moving in space under the action of a force

Fðt Þ as shown in Fig. 1.7.1. If r ¼ rðt Þ ¼ x ðt Þi + y ðt Þj + z ðt Þk represents the
position vector of the particle at time t, then according to Newton’s second
law of motion, the Newtonian path of the particle is governed by the differential
equation
d 2r
m F¼0 (1.7.1)
dt 2
We confine our attention to an interval of time during which the particle
moves from point 1 at t ¼ t1 to point 2 at t ¼ t2 . We consider now a varied path,
specified by rðt Þ + drðt Þ, adjacent to the actual one. We will refer to the quan-
tity drðt Þ ¼ dx ðt Þi + dy ðt Þj + dz ðt Þk as the variation of r. The only restriction is
that the two paths coincide at time t ¼ t1 and t ¼ t2 . This implies that the var-
iation dr ¼ drðt Þ vanishes at these instants, that is,
drðt1 Þ ¼ drðt2 Þ ¼ 0 (1.7.2)
The first step to derive Hamilton’s principle is to take the inner product of
the left side of Eq. (1.7.1) with the vector dr and to integrate from time t1 to time
t2 . This gives
Z t2
d2r
m dr F dr dt ¼ 0 (1.7.3)
t1 dt 2
Integrating by parts the first term in the above integral and knowing that the
operator d acts like the differential operator [6], we obtain
Z
t2 Z t2
t2
d 2r dr dr dr
m 2 drdt ¼ m dr m d dt
t1 dt dt t1 t1 dt dt
@CivilMethod
The term outside the integral is equal to zero because of Eq. (1.7.2). More-
over, we can write the integrand as
" #
dr dr 1 dr dr 1 dr 2 1 dr 2
m d ¼m d ¼m d ¼ d m ¼ dT
dt dt 2 dt dt 2 dt 2 dt
where
2
1 dr
T¼ m ¼ x_ ðt Þ2 + y_ ðt Þ2 + z_ ðt Þ2 (1.7.4)
2 dt
is the kinetic energy of the particle. Hence, the integral (1.7.3) takes the form
Z t2
ðdT + F drÞdt ¼ 0 (1.7.5)
t1
The variation dr is a virtual displacement that leads from the actual path to
the varied one. Hence the term F dr in Eq. (1.7.5) is the virtual work done
by the force Fðt Þ. Eq. (1.7.5) is a statement of Hamilton’s principle as it is
applied to a particle. This equation can be transformed into a more convenient
form if the force Fðt Þ is separated in its conservative and nonconservative
components, that is
Fðt Þ ¼ Fc ðt Þ + Fnc ðt Þ (1.7.6)
A potential function A ¼ Aðx, y, z, t Þ exists from which the conservative
force Fc ðt Þ is derived as its minus gradient

∂A ∂A ∂A
Fc ¼ i+ j+ k (1.7.7)
∂x ∂y ∂z
Hence

∂A ∂A ∂A
Fc dr ¼ dx + dy + dz
∂x ∂y ∂z
or
Fc dr ¼ dA (1.7.8)
Hence, Hamilton’s principle, Eq. (1.7.5), can be written as
Z t2 Z t2
d ðT AÞdt + dWnc dt ¼ 0 (1.7.9)
t1 t1
where
dWnc ¼ Fnc dr
represents the virtual work of the nonconservative force.
@CivilMethod
In the absence of nonconservative forces, Fnc ¼ 0, Eq. (1.7.9) becomes

Z t2
d ðT AÞdt ¼ 0 (1.7.10)
t1
The scalar quantity

L¼T A (1.7.11)
is termed the Lagrangian or the kinetic potential. We should emphasize that
Hamilton’s principle depends upon the energies of the system and is invariant
under the coordinate transformation.
Eq. (1.7.10) states that of all possible paths of motion of the particle during
an interval of time from t1 to t2 , the actual path is that for which the integral
Z t2
Ldt ¼ 0 (1.7.12)
t1
has a stationary value. In fact, it can be shown that this value is the minimum
value of the integral.
The derivation of Hamilton’s principle for a particle can be extended to
MDOF systems as well as to continuous systems. The potential energy usually
arises from the gravity field. However, it may also arise from other sources such
as electrical and magnetic fields. The strain energy U ðt Þ should be included as
an additional potential energy. Thus, we can write
Z t2 Z t2
d ðU T + AÞdt dWnc dt ¼ 0 (1.7.13)
t1 t1
Hamilton’s principle is rather utilized to derive the equations of motion of

continuous systems. The equations of motion of discrete parameter systems can
result directly from Lagrange’s equations.
Example 1.7.1 Equation of motion of the SDOF system
Formulate the equation of motion of the SDOF system shown in Fig. 1.4.1 using
Hamilton’s principle.
Solution
The potential energy is due to the strain energy stored in the spring during defor-
mation. It is expressed in terms of the spring stiffness coefficient k and the dis-
placement u as
1
U ¼ ku 2 (1)
2
The kinetic energy is due to the motion of the mass m and is given as
1
T ¼ m u_ 2 (2)
2
@CivilMethod
The damping force fD ¼ cu,_ as a dissipative force, is nonconservative. The

virtual work of this force is
D
dWnc _
¼ fD du ¼ cudu (3)
The negative sign results from the fact that fD is opposite to the virtual dis-
placement du.
The external force is also treated as nonconservative and it does the
virtual work
p
dWnc ¼ pðt Þdu (4)
Because no conservative external forces act on the system, it is A ¼ 0.
The variations dU and dT are obtained from Eqs. (1), (2)
_ u_
dU ¼ kudu, dT ¼ m ud (5)
Introducing Eqs. (3)–(5) into Hamilton’s principle, Eq. (1.7.13), yields
Z t2 Z t2
_ u_ Þdt
ðkudu m ud _ + pðt Þdu dt ¼ 0
½cudu (6)
t1 t1
The next step is to remove the variation d u_ of the velocity u_ from Eq. (6).
This is achieved using integration by parts as follows:
Z t2 Z t2
du
_ udt
m ud _ ¼ _
m ud dt
t1 t1 dt
Z t2
d
¼ m u_ ðdu Þdt (7)
t1 dt
Z t2
_ tt21
¼ ½m udu €
m ududt
t1
According to Hamilton’s principle it holds

du ðt1 Þ ¼ du ðt2 Þ ¼ 0
Thus, the term outside the integral vanishes and Eq. (7) becomes
Z t2 Z t2
_ udt
m ud _ ¼ €
m ududt (8)
t1 t1
Eq. (6) by virtue of Eq. (8) is written as

Z t2
½m u€ + cu_ + ku pðt Þdudt ¼ 0 (9)
t1
In order that the integral in Eq. (9) is equal to zero for any time interval
½t1 , t2 , its integrand should vanish, that is,
½m u€ + cu_ + ku pðt Þdu ¼ 0
@CivilMethod
Moreover, because du is arbitrary, it must be

m u€ + cu_ + ku pðt Þ ¼ 0
or
m u€ + cu_ + ku ¼ pðt Þ (10)
which is the equation of motion.
Example 1.7.2 Equation of motion of a two-story shear frame
Formulate the equations of motion of the frame in Example 1.5.5 using Hamil-
ton’s principle.
Solution
Referring to Fig. E1.5b and c in Example 1.5.5, we have
1 1
U ¼ k1 ðu1 u2 Þ2 + k2 u22
2 2
1 1
T ¼ m1 u_ 21 + m2 u_ 22
2 2
Their variations are
dU ¼ k1 ðu1 u2 Þðdu1 du2 Þ + k2 u2 du2
(1)
¼ k1 ðu1 u2 Þdu1 k1 ðu1 u2 Þdu2 + k2 u2 du2
dT ¼ m1 u_ 1 d u_ 1 + m2 u_ 2 d u_ 2 (2)
Integrating by parts the variation dT in the interval ½t1 , t2 yields
Z t2 Z t2
dTdt ¼ ðm1 u_ 1 d u_ 1 + m2 u_ 2 d u_ 2 Þdt
t1 t1
Z t2
¼ ½m1 u_ 1 du1 + m2 u_ 2 du2 tt21 ðm1 u€1 du1 + m2 u€2 du2 Þdt
t1
and taking into account that du1 ðt1 Þ ¼ du1 ðt2 Þ ¼ du2 ðt1 Þ ¼ du2 ðt2 Þ ¼ 0, we
obtain
Z t2 Z t2
dTdt ¼ ðm1 u€1 du1 + m2 u€2 du2 Þdt (3)
t1 t1
Moreover, it is
p
dWnc ¼ p1 ðt Þdu1 + p2 ðt Þdu2 and A ¼ 0 (4)
Introducing Eqs. (1), (3), (4) into Hamilton’s principle, Eq. (1.7.13), we
obtain Z t2
½k1 ðu1 u2 Þdu1 k1 ðu1 u2 Þdu2 + k2 u2 du2 + m1 u€1 du1
t1
+ m2 u€2 du2 p1 ðt Þdu1 p2 ðt Þdu2 dt ¼ 0
@CivilMethod
or
Z t2
f½m1 u€1 + k1 ðu1 u2 Þ p1 ðt Þdu1 + ½m2 u€2 k1 u1 + ðk1 + k2 Þu2
t1
p2 ðt Þdu2 gdt ¼ 0 (5)
Because Eq. (5) is valid for any interval ½t1 , t2 , its integrand must be equal
to zero, that is,
½m1 u€1 + k1 ðu1 u2 Þ p1 ðt Þdu1 + ½m2 u€2 k1 u1 + ðk1 + k2 Þu2 p2 ðt Þdu2 ¼ 0
(6)
Inasmuch as the quantities du1 and du2 are arbitrary, Eq. (6) is valid only if
the quantities in the square brackets are equal to zero, that is,
m1 u€1 + k1 ðu1 u2 Þ p1 ðt Þ ¼ 0 (7a)
m2 u€2 k1 u1 + ðk1 + k2 Þu2 p2 ðt Þ ¼ 0 (7b)
which give the equations of motion
m1 u€1 + k1 u1 k1 u2 ¼ p1 ðt Þ (8a)
m2 u€2 k1 u1 + ðk1 + k2 Þu2 ¼ p2 ðt Þ (8b)
Example 1.7.3 Equation of motion of a complex MDOF system

The system shown in Fig. E1.9 consists of the three rigid bars AB,BC , CD con-
nected by hinges at points B and C , and it is supported by a roller at point D and
a hinge at point A. The relative rotations of the bars at the hinges B and C are
restrained by moment-resisting rotational springs with stiffness coefficients
k3 ¼ k4 ¼ 4kL2 and by the rotational dashpots with damping coefficients
c3 ¼ c4 ¼ 2cL2 . In the transverse direction, the motion is restrained by the
two springs at points E and Q with stiffness coefficients k1 ¼ k, k2 ¼ 2k, and
the two dashpots at points F and G with damping coefficients c1 ¼ c and
c2 ¼ 3c. A constant axial force P is applied at point D. The system is set in
motion by the transverse load pðx, t Þ ¼ ðpx=LÞf ðt Þ, linearly distributed along
the bar CD. The mass per unit length of the bars AB and CD is m while the
bar BC is massless and supports the rigid body S at H having surface mass

density g ¼ m=L. Assuming small amplitude displacements, formulate the
equations of motion of the system using Hamilton’s principle.
@CivilMethod
Solution
Inasmuch as the bars are assumed rigid, this system has only two degrees of
freedom. The displaced configuration of the system can be determined from
the two transverse displacements u1 ðt Þ and u2 ðt Þ of points B and C . Referring
to Fig. E1.11, we have
FIG. E1.11 Deformed configuration of the system.
9
f1 ¼ u1 =4L =
f2 ¼ ðu2 u1 Þ=3L (1)
;
f3 ¼ u2 =3L
The displacements of points of application of the forces and the changes of
angles are expressed in terms of the basic quantities u1 and u2 as
9
EE 0 ¼ u1 =4, FF 0 ¼ u1 =2, GG 0 ¼ 3u1 =4 >>
=
HH 0 ¼ u1 + ðu2 u1 Þ=3, QQ 0 ¼ u2 =2
(2)
DfB ¼ f2 f1 ¼ ð4u2 7u1 Þ=12L >
>
;
DfC ¼ f3 + f2 ¼ ð2u2 u1 Þ=3L
The potential energy U due to the deformation of the springs is
1 1 1 1
U ¼ k1 ðEE 0 Þ + k2 ðQQ 0 Þ + k3 ðDfB Þ2 + k4 ðDfC Þ2
2 2
2 2 2 2
which by virtue of Eqs. (2) becomes
1 2 1 2 1 2
U¼ ku 1 + ku 2 + k ð4u2 7u1 Þ2 + k ð2u2 u1 Þ2
32 4 72 9
¼ 0:934ku 21 + 1:361ku 22 1:667u1 u2
Its variation is
dU ¼ k ð1:868u1 1:667u2 Þdu1 + k ð1:667u1 + 2:722u2 Þdu2 (3)
The kinetic energy consists of the kinetic energies T1 and T2 of the bars ΑΒ
and CD, and of the kinetic energy T3 of the rigid body S. Thus, we have

2
1 1 1 d 1
T ¼ IA f_ 1 + ID f_ 3 + m ðHH 0 Þ + IH f_ 2
2 2 2
(4)
2 2 2 dt 2
@CivilMethod
where
ð4LÞ3 ð3LÞ3 L3

m ¼ mL,
IA ¼ m
, ID ¼ m
, IH ¼ m (5)
3 3 6
Introducing Eqs. (1), (2), (5) into Eq. (4) yields
2 1 1 1
T ¼ m u_ 21 + m u_ 22 + m ðu_ 2 + 2u_ 1 Þ2 + m ðu_ 2 u_ 1 Þ2
3 2 18 108
¼ 0:898m u_ 21 + 0:565m u_ 22 + 0:204u_ 1 u_ 2
and its variation
dT ¼ m ð1:796u_ 1 + 0:204u_ 2 Þd u_ 1 + m ð0:204u_ 1 + 1:130u_ 2 Þd u_ 2

Rt
Using integration by parts in the integral t12 dTdt and taking into
account that
d
du1 ðt1 Þ ¼ du1 ðt2 Þ ¼ du2 ðt1 Þ ¼ du2 ðt2 Þ ¼ 0 and d u_ ¼ d du
dt ¼ dt ðdu Þ
we obtain
Z t2 Z t2
dTdt ¼ ½m ð1:796u€1 + 0:204u€2 Þdu1 + m ð0:204u€1 + 1:130u€2 Þdu2 dt
t1 t1
(6)
The nonconservative forces include the loading pðx, t Þ and the damping
forces. Their virtual work is expressed in terms of the basic quantities as follows:
Z 3L x
dWnc p
¼ pðx, t Þ 1 du2 dx
0 3L
Z 3L (7)
x x
¼ p f ðt Þ 1 du2 dx ¼ 1:5
pLf ðt Þdu2
0 L 3L
d d d
D
dWnc ¼ c1 ðFF 0 Þd ðFF 0 Þ c2 ðGG 0 ÞdðGG 0 Þ c3 ðDfB ÞdðDfB Þ
dt dt dt (8)
d
c4 ðDfC Þd ðDfC Þ
dt
Using Eq. (2) and taking into account that c1 ¼ c, c2 ¼ 3c, c3 ¼ c4 ¼ 2cL2 ,
we can write
D
dWnc ¼ cð2:840u_ 1 0:833u_ 2 Þdu1 + cð0:833u_ 1 1:111u_ 2 Þdu2
Hence, we have
dWnc ¼ cð2:395u_ 1 0:833u_ 2 Þdu1 + cð0:833u_ 1 1:111u_ 2 Þdu2 + 1:5
pLf ðt Þdu2
(9)
@CivilMethod
Finally, the potential A of the external conservative forces is due to the con-
stant axial force P . Hence it is
A ¼ P ðDD 0 Þ ¼ Pe
and
dA ¼ Pde (10)
The variation de is evaluated as follows.
Referring to Fig. E1.11, we have
e ¼ ðAD Þ ðAD0 Þ ¼ 10L 4L cos f1 3Lcos f2 3L cos f3
Therefore
de ¼ Lð4sin f1 df1 + 3 sin f2 df2 + 3sin f3 df3 Þ
(11)
¼ Lð4f1 df1 + 3f2 df2 + 3f3 df3 Þ
which is introduced into Eq. (10) to yield

7P P P 2P
dA ¼ u1 + u2 du1 + u1 u2 du2 (12)
12L 3L 3L 3L
Introducing the expressions for dU , dT , dWnc , and dA into Hamilton’s prin-
ciple, Eq. (1.7.13), we obtain the following equations of motion

7P
1:796m u€1 + 0:204m u€2 + 2:395cu_ 1 0:833cu_ 2 + 1:868k u1
12L

P
+ 1:667k + u2 ¼ 0
3L

P
0:204m u€1 + 1:130m u€2 0:833cu_ 1 + 1:111cu_ 2 + 1:667k + u1
3L

2P
+ 2:722k u2 ¼ 1:5
pLf ðt Þ
3L
or in the matrix form
" #( ) " #( )
1:796 0:204 u€1 2:395 0:833 u_ 1
m +c
0:204 1:130 u€2 0:833 1:111 u_ 2
" #( ) ( ) (13)
1:868 0:583l 1:667 + 0:333l u1 0
+k ¼
1:667 + 0:333l 2:722 0:667l u2 pLf ðt Þ
1:5
where l ¼ P=kL.
The elastic forces of the system are
fS1 ¼ k ð1:868 0:583lÞu1 + k ð1:667 + 0:333lÞu2
fS2 ¼ k ð1:667 + 0:333lÞu1 + k ð2:722 0:667lÞu2
@CivilMethod
They may become zero if the system of equations

1:868 0:583l 1:667 + 0:333l u1 0
k ¼ (14)
1:667 + 0:333l 2:722 0:667l u2 0
has a nontrivial solution. This occurs if

+1:868 0:583l 1:667 + 0:333l

1:667 + 0:333l +2:722 0:667l ¼ 0 (15)
Expanding the determinant yields

0:27797l2 1:7227l + 2:3058 ¼ 0
from which we obtain
l1 ¼ 1:9555 l2 ¼ 4:2419
The obtained values of l specify two critical values, Pcr1 ¼ 1:9555kL and
Pcr2 ¼ 4:2419kL, of the compressive axial force for which the structure exhibits
no resistance to deformation, that is, it has no stiffness and the structure buckles.
Therefore, these critical loads are the buckling loads of the structure (first and
second). The condition for buckling is the vanishing of the determinant of the
stiffness matrix, Eq. (15). The resulting equation is called the buckling equation.
It is apparent that if the axial force is tensile, the determinant cannot vanish for
real values of the parameter l. Concluding, we can state that tensile axial forces
increase the stiffness of the structure while compressive axial forces reduce it
and may lead to buckling.
Example 1.7.4 Equation of motion of the elastic cantilever beam
Derive the equation of motion of the cantilever beam shown in Fig. E1.12.
Solution
The mass of the beam element is equal to mdx. Its kinetic energy is
mdx ½∂u ðx, t Þ=∂t 2 =2, which is integrated along the beam length to yield the
kinetic energy of the beam, namely
FIG. E1.12 Cantilever in Example 1.7.4.
Z
1 L
∂u ðx, t Þ 2
T¼ m dx (1)
2 0 ∂t
@CivilMethod
The strain energy of the beam is obtained by integrating the strain energy
density over its volume V , namely
Z
1
U¼ sx ex dV (2)
2 V
From the beam theory we have
M ðx Þ sx ∂2 u ðx, t Þ
sx ¼ y, ex ¼ , M ðx Þ ¼ EI
I E ∂x 2
Substituting the previous equations into Eq. (2) and integrating over the
cross-section of the beam yield
Z 2 2
1 L ∂ u ðx, t Þ
U¼ EI dx (3)
2 0 ∂x 2
For the simplicity of the expressions, the differentiation with respect to time
t will be designated by an over-dot while that with respect to the spatial coor-
dinated x by a prime. Moreover, the arguments will be dropped for the same
reason. Hence, expressions (1) and (3) can be rewritten as
Z
1 L
T¼ m u_ 2 dx (4)
2 0
Z
1 L
EI ðu 00 Þ dx
2
U¼ (5)
2 0
Their variations are
Z L
dT ¼ _ udx
m ud _ (6)
0
Z L
dU ¼ EI u 00 du 00 dx (7)
0
Integrating twice by parts the integral representing dU yields

Z L
EI u 0000 dudx ½EI u 000 du 0 + ½EI u 00 du 0 0
L L
dU ¼ (8)
0
The boundary conditions of the beam are

At x ¼ 0 u ¼ u 0 ¼ 0, hence du ¼ du 0 ¼ 0
At x ¼ L M ¼ EI u 00 ¼ 0 Q ¼ EI u 000 ¼ 0
Therefore, the quantities outside the integral vanish and Eq. (8) becomes
Z L
dU ¼ EI u 0000 dudx (9)
0
@CivilMethod
Because no conservative loads act on the system, it is A ¼ 0. Moreover, the

virtual work of the external load is
Z L
dWnc ¼
p
pðx, t Þdudx (10)
0
Introducing Eqs. (5), (9), (10) into Hamilton’s principle, Eq. (1.7.13), we
obtain
Z t2
Z L Z L Z L
EI u 0000 dudx _ udx
m ud _ pðx, t Þdudx dt ¼ 0 (11)
t1 0 0 0
Interchanging the integration in the second term and performing integration

by parts with respect to time, yield
Z t2
Z L Z L
Z t2
_ udx
m ud _ dt ¼ €
½m ududt + ½mudu tt21 dx
t1 0 0 t1
Z t
Z L 2
(12)
¼ €
½m ududx dt
t1 0
On the basis of Eq. (12), Eq. (11) becomes

Z t 2
Z L
½EI u 0000 + m u€ pðx, t Þdudx dt ¼ 0 (13)
t1 0
Because Eq. (13) is valid for any interval ½t1 , t2 , the integrand must vanish,
namely
Z L
½EI u 0000 + m u€ pðx, t Þdudx ¼ 0 (14)
0
Moreover, because du is arbitrary, Eq. (14) is valid only if

EI u 0000 + m u€ pðx, t Þ ¼ 0 (15)
which yields the equation of motion of the cantilever
EI u 0000 + m u€ ¼ pðx, t Þ (16)
Apparently, Eq. 16 is identical to that obtained in Section 1.1.
1.8 Lagrange’s equations

1.8.1 Derivation of Lagrange’s equations
In a system with N degrees of freedom, the displaced configuration can be deter-
mined from a set of coordinates, which take a certain value at each instant. The
system of coordinates for the analysis of a given mechanical system is not nec-
essarily unique. Many coordinate systems are possible. Furthermore, the number
of coordinates may vary, but it cannot be less than N . Anyhow, if the number of
coordinates is greater than N , then additional equations, referred to as equations
@CivilMethod
of constraint, must relate the coordinates so that the number of coordinates is

equal to the number of degrees of freedom plus the independent equations of
constraint. The requirement that the equations of motion hold together with
the equations of constraint complicates the solution. For this reason, we seek,
if possible, to choose N independent coordinates, which can specify the config-
uration of the system. For example, we consider the simple pendulum shown in
Fig. 1.8.1a. The rod is rigid and weightless. Its length is L and it can rotate freely
about the hinge at O, such that the motion is confined in a single vertical plane.
The position of the mass can be specified by the angle q between the vertical
axis y and the rod. Hence, the system has a SDOF. However, the displaced
configuration can also be determined by the coordinates ðx, y Þ, which represent
the position of the mass m within the xy plane. These coordinates, however,
are not independent because they must satisfy the constraint equation
x 2 + y 2 ¼ L2
(a) (b)
FIG. 1.8.1 Simple (a) and double (b) pendulum.
Similarly, the configuration of the double pendulum of Fig. 1.8.1b can be

specified by the two angles q1 and q2 . Hence the system has two degrees of free-
dom. On the other hand, the position of the masses m1 , m2 can be determined by
the coordinates ðx1 , y1 Þ and ðx2 , y2 Þ, which, however, are not independent
because they must satisfy the following two constraint equations
x12 + y12 ¼ L21
ðx2 x1 Þ2 + ðy2 y1 Þ2 ¼ L22

The quantities q in the simple pendulum or q1 , q2 in the double pendulum,
which would determine the configuration of the system, could be considered as
coordinates in a more general sense. Any set of quantities that serves to specify
the configuration of the system is referred to as generalized coordinates. The
geometrical significance of the generalized quantities is not always cognizable.
For systems in motion, the generalized coordinates vary with time and are trea-
ted as algebraic variables. The process of obtaining one set of generalized coor-
dinates from another is known as a coordinate transformation.
@CivilMethod
We consider now a transformation from a set of N generalized coordinates

q1 ðt Þ, q2 ðt Þ,…, qN ðt Þ to a set K of ordinary (for example, Cartesian) coordinates
x1 ,x2 ,…, xK ðK N Þ. The transformation equations are of the form
x1 ¼ x1 ðq1 , q2 , …, qN Þ
x2 ¼ x2 ðq1 , q2 , …, qN Þ
(1.8.1)
⋯ ⋯
xK ¼ xK ðq1 , q2 , …, qN Þ
For example, the transformation equations of the generalized coordinates
q1 , q2 to the ordinary coordinates x1 , x1 , y2 , y2 of the double pendulum are
x1 ¼ L1 sin q1
y1 ¼ L1 cos q1
x2 ¼ L1 sin q1 + L2 sin q2
y2 ¼ L1 cos q1 L2 cos q2
The kinetic energy of a system with K degrees of freedom may also depend
on the generalized coordinates q1 , q2 , …, qN beside the generalized velocities
q_ 1 , q_ 2 ,…, q_ N , that is,
T ¼ T ðq1 , q2 , …, qN , q_ 1 , q_ 2 , …, q_ N Þ (1.8.2)
In conservative systems, the potential energy A depends only on the posi-
tion, namely, it is
A ¼ Aðq1 , q2 , …, qN Þ (1.8.3)
The work done by the forces derivable from the potential energy A, when the
generalized coordinates qi are given a virtual displacement dqi , is expressed as
dA ¼ Q1 dq1 + Q2 dq2 + ⋯ + QN dqN
where
∂A ∂A ∂A
Q1 ¼ ¼ , Q2 ¼ ¼ , …, QN ¼ (1.8.4)
∂q1 ∂q2 ∂qN
The quantity Qi dqi represents the work done through the displacement dqi .
Inasmuch as the quantity Qi may or may not represent a force, it is referred to as
generalized force. Hence, if qi represents a translational displacement then Qi is
a force, whereas if qi represents a rotation then Qi is a moment. In some prob-
lems, the quantities dqi may represent surfaces, volumes, etc. Therefore, the
nature of the corresponding Qi is defined so that the quantity Qi dqi has the phys-
ical dimension of work.
Lagrange’s equations may be derived by direct application of Hamilton’s
principle. Thus, when the applied forces are conservative, we write
Z t2
ðdT dAÞdt ¼ 0 (1.8.5)
t1
@CivilMethod
The variations associated with the kinetic energy and the potential energy
defined by Eqs. (1.8.2), (1.8.3), respectively, are of the form
∂T ∂T ∂T ∂T
dT ¼ dq1 + ⋯ + dqN + d q_ + ⋯ + d q_
∂q1 ∂qN ∂q_ 1 1 ∂q_ N N
∂A ∂A
dA ¼ dq1 + ⋯ + dqN
∂q1 ∂qN
Substituting these expressions into Eq. (1.8.5), integrating by parts the terms
including d q_ i and taking into account dq1 ¼ dq2 ¼ ⋯ ¼ dqN ¼ 0 at instants t1
and t2 , we obtain
Z t 2

∂T d ∂T ∂A ∂T d ∂T ∂A
dq1 + ⋯ + dqN dt ¼ 0
t1 ∂q1 dt ∂q_ 1 ∂q1 ∂qN dt ∂q_ N ∂qN
Because the time interval ½t1 , t2 as well as the virtual displacements dqi are
arbitrary, this previous equation results in the following equations

d ∂T ∂T ∂A
+ ¼ 0 ði ¼ 1, 2, …, N Þ (1.8.6)
dt ∂q_ i ∂qi ∂qi
which, using Eq. (1.8.4), become

d ∂T ∂T
¼ Qi ði ¼ 1, 2, …, N Þ (1.8.7)
dt ∂q_ i ∂qi
Eq. (1.8.6) or (1.8.7) are the Lagrange equations of motion.
When nonconservative forces act on the system in addition to the conserva-
tive forces, we can include them in Lagrange’s equations, if the work done by
the nonconservative forces riding the virtual displacements is expressed in
terms of the generalized forces, that is,
dWnc ¼ Q1 dq1 + Q2 dq2 + ⋯ + QN dqN (1.8.8)
Introducing Eq. (1.8.8) into Hamilton’s principle, Eq. (1.7.9), the Lagrange
equations (1.8.6) become

d ∂T ∂T ∂A
+ ¼ Qi ði ¼ 1, 2, …, N Þ (1.8.9)
The elastic force components, which are derivable from a potential U (strain
energy), can be also involved in Eq. (1.8.9). Noting that
U ¼ U ðq1 , q2 , …, qN Þ (1.8.10)
the associated variation is
∂U ∂U
dU ¼ dq1 + ⋯ + dqN
∂q1 ∂qN
Therefore, the components ∂U =∂qi express generalized elastic forces and
Lagrange’s equations become
@CivilMethod

d ∂T ∂T ∂V
+ ¼ Qi ði ¼ 1, 2, …, N Þ (1.8.11)
where
V ¼U +A (1.8.12)
is the total potential energy of the system.

The N generalized forces Qi can be evaluated from the set of K actual forces
Fk associated with the set of Cartesian coordinates. For this purpose, we con-
sider the work done by the set of forces Qi when the coordinates qi are given an
increment, that is, a virtual displacement dqi
X
N
dW ¼ Qi dqi (1.8.13)
i¼1
If dxk represent the ensuing virtual displacements of the coordinates xk then

the set of forces Fk do the work
X
K
dW ¼ Fk dxk (1.8.14)
k¼1
From physical consideration, the work done by the two sets of forces is the
same. The only difference is that they are expressed in two different coordinate
systems. Therefore, we can write
X
N X
K
Qi dqi ¼ Fk dxk (1.8.15)
i¼1 k¼1
or in matrix form
QT dq ¼ FT dx (1.8.16)
where
Q ¼ fQ1 Q2 ⋯ QN gT , dq ¼ f dq1 dq2 ⋯ dqN gT (1.8.17a)
F ¼ f F1 F2 ⋯ FK gT , dx ¼ f dx1 dx2 ⋯ dxK gT (1.8.17b)

The relation between dxk and dqi results from the transformation equations
(1.8.1) by considering the variation of dxk . Thus, we have
∂xk ∂xk
dxk ¼ dq1 + ⋯ + dqN (1.8.18)
∂q1 ∂qN
or in matrix form
dx ¼ Jdq (1.8.19)
@CivilMethod
where J is the Jacobian matrix of the transformation (1.8.1), that is,

2 ∂x ∂x ∂x1 3
1 1
⋯
6 ∂q1 ∂q2 ∂qN 7
6 7
6 ∂x ∂x ∂x2 7
6 2 2
⋯ 7
6 7
J ¼ 6 ∂q1 ∂q2 ∂qN 7 (1.8.20)
6 7
6 ⋮ ⋮ ⋱ ⋮ 7
6 7
4 ∂xK ∂xK ∂xK 5
⋯
∂q1 ∂q2 ∂qN
Substituting Eq. (1.8.19) into (1.8.16) yields
QT dq ¼ FT Jdq (1.8.21)
QT ¼ F T J (1.8.22)
or
Q ¼ JT F (1.8.23)
Apparently, Eq. (1.8.23) represents the sought relation between Q and F.
Example 1.8.1 Equation of motion of the double pendulum
Formulate the equations of motion of the double pendulum shown in
Fig. 1.8.1b.
Solution
Because the bars are inextensional, the displaced configuration of the moving
system can be specified by the generalized coordinates q1 and q2 . Referring to
Fig. 1.8.1b, the Cartesian coordinates of the masses m1 and m2 are expressed in
terms of q1 and q2 by the geometrical relations
x1 ¼ L1 sin q1
y1 ¼ L1 cos q1
(1)
x2 ¼ L1 sin q1 + L2 sin q2
y2 ¼ L1 cos q1 L2 cos q2
The kinetic energy of the system is
1 1
T ¼ m1 x_ 21 + y_ 21 + m2 x_ 22 + y_ 22
2 2
which by virtue of Eqs. (1) becomes
1 1
T ¼ ðm1 + m2 ÞL21 q_ 1 + m2 L1 L2 q_ 1 q_ 2 cos ðq1 q2 Þ + m2 L22 q_ 2
2 2
(2)
2 2
The potential energy is
A ¼ m1 gy 1 + m2 gy 2
@CivilMethod
or using Eqs. (1)

A ¼ ðm1 + m2 ÞgL1 cos q1 m2 gL2 cos q2 (3)
Differentiating Eqs. (2), (3) yields

∂T
¼ ðm1 + m2 ÞL21 q_ 1 + m2 L1 L2 q_ 2 cos ðq1 q2 Þ
∂q_ 1

d ∂T
¼ ðm1 + m2 ÞL21 q€1 + m2 L1 L2 q€2 cos ðq1 q2 Þ
dt ∂q_ 1

m2 L1 L2 q_ 2 sin ðq1 q2 Þ q_ 1 q_ 2
∂T
¼ m2 L1 L2 q_ 1 q_ 2 sin ðq1 q2 Þ
∂q1
∂A
¼ ðm1 + m2 ÞgL1 sin q1
∂q1
∂T
¼ m2 L1 L2 q_ 1 cos ðq1 q2 Þ + m2 L22 q_ 2
∂q_ 2

d ∂T
¼ m2 L1 L2 q€1 cos ðq1 q2 Þ m2 L1 L2 q_ 1 sin ðq1 q2 Þ q_ 1 q_ 2 + m2 L22 q€2
_
dt ∂q 2
∂T
¼ m2 L1 L2 q_ 1 q_ 2 sin ðq1 q2 Þ
∂q2
∂A
¼ m2 gL2 sin q2
∂q2
Applying Eq. (1.8.6) for i ¼ 1, 2 and q1 ¼ q1 q2 ¼ q2 , we obtain the equa-
tions of motion of the double pendulum

ðm1 + m2 ÞL1 q€1 + m2 L2 q€2 cos a + q_ 2 sin a + ðm1 + m2 Þg sin q1 ¼ 0 (4a)
2
L1 q€1 cos a + L2 q€2 L1 q_ 1 sin a + g sin q2 ¼ 0

2
(4b)
where
a ¼ q1 q2
Example 1.8.2 Equation of motion of the “soft” pendulum

Formulate the equations of motion of the simple pendulum shown in Fig. E1.13,
taking into account the axial deformation of the rod (soft pendulum). The unde-
formed length of the rod is L, its cross-sectional area A, and the modulus of
elasticity of the material E.
@CivilMethod
FIG. E1.13 “Soft” pendulum in Example 1.8.2.
Solution
Because the rod is no more inextensional, the system has two degrees of freedom.
Its displaced configuration can be specified either by the orthogonal coordinates x
and y of the mass or by the angle of the q and the axial deformation of the rod.
The kinetic energy of the system is
1
T ¼ m x_ 2 + y_ 2 (1)
2
The potential energy of the external force (gravitational force) is
A ¼ mgy (2)
and the potential of the elastic force
1
U ¼ ke2 (3)
2
where k ¼ EA=L is the axial stiffness of the rod and e its elongation. The latter
is expressed in terms of x and y as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
e ¼ x 2 + y2 L (4)
Introducing Eq. (4) in the expression for the axial stiffness, Eq. (3), yields
1 EA pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi 2
U¼ x 2 + y2 L (5)
2 L
Differentiating the energies, we obtain
!
d ∂T ∂A ∂U EA L
¼ m x€, ¼ 0, ¼ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x (6)
dt ∂x_ ∂x ∂x L x 2 + y2
!
d ∂T ∂A ∂U EA L
¼ m y€, ¼ mg, ¼ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi y (7)
dt ∂y_ ∂y ∂y L x 2 + y2
Introducing Eqs. (6), (7) into Lagrange’s equations (1.8.11), we obtain the
equations of motion of the soft pendulum
!
EA L
m x€ + 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x ¼ 0 (8a)
L x 2 + y2
@CivilMethod
!
EA L
m x€ + 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi y ¼ mg (8b)
L x 2 + y2
Example 1.8.3 Equation of motion of a general one-story shear building

Formulate the equations of motion of the one-story building in Example 1.5.6
using the method of Lagrange’s equations.
Solution
The system has three degrees of freedom. We choose the displacements U , V of
about the z axis as the gen-
point O in the plane of the plate and its rotation W
eralized coordinates of the system, namely

q1 ¼ U , q2 ¼ V , q3 ¼ W
Because the point O is not the center of mass of the plate, the kinetic energy
is given by the expression (1.5.18) (K€onig’s theorem)
1 1
_ 2 m y U_ W _
_ + m xc V W
T ¼ m U_ 2 + V 2 + Io W

c (1)
2 2
The potential energy U consists of the strain energy of all columns. For the i
column it is
1 h i i 2 i 2 i 2 i
Ui ¼ k11 u + k22i
v + k33i
w
2
or using matrix notation
i 8 9
k 0 0 < u i =
1 i i i 11 i 1 T
Ui ¼ u v w 0 k22 0 v i ¼ Di ki Di (2)
2 0 0 k i wi : ; 2
33
Taking into account that (see Eqs. 1a, 12 of Example 1.5.6)

i T
¼ Ri ei U,
D i ¼ Ri D D ¼U T e i T Ri T
Eq. (2) is written as
1 T i
Ui ¼ U kU
2
where

i ¼ ei T Ri T ki Ri ei
k
Thus, we have
XK
1
U¼
Ui ¼ UT KU
i¼1
2
(3)
1 + 2k23 V W
2 + 2k12 U V + 2k13 U W

¼ k11 U + k22 V + k33 W
2 2
2
@CivilMethod
ij are the elements of the matrix

where k
X
K X
K
¼
K i ¼
k ei
T T
Ri k i Ri e i (4)
i¼1 i¼1
Because there are no external conservative forces, it is

A¼0 (5)
The generalized forces result from Eq. (1.8.20). In this case it is
T
F ¼ P x ðt Þ P y ðt Þ 0
The transformation relations between the displacements u, v, w of a point
y of the plate and the displacements of point O are
x,
v ¼ V + xW,
u ¼ U yW, w¼W

which yields the Jacobian matrix, Eq. (1.8.20)

2 ∂u ∂u ∂u 3
6 ∂U ∂V ∂W
7 2 3
6 7 1 0 0
6 ∂v ∂v ∂v 7
J¼6
6 ∂U
7 ¼ 4 0 1 05
6 7
∂V ∂W 7
4 5 y x 1

∂w ∂w ∂w
∂U ∂V ∂W

Thus, for point A we have
T
Q ¼ JT F ¼ P x ðt Þ P y ðt Þ
y A P x ðt Þ + xA P y ðt Þ (6)
Differentiation of Eq. (1) yields

d ∂T €
¼ m U€
m yc W (7a)
dt ∂U_
∂T
¼0 (7b)
∂U
!
d ∂T €
¼ m V€
+ mx c W (7c)
dt ∂ V
∂T
¼0 (7d)
∂V

d ∂T € m y U€
+ m xc V€
¼ Io W (7e)
_
dt ∂W c
∂T
∂W ¼0 (7f)
Moreover, Eq. (3) yields
@CivilMethod
∂U
¼ k11 U + k12 V + k13 W (8a)
∂U
∂U
¼ k21 U + k22 V + k23 W (8b)
∂V
∂U
∂W ¼ k 31 U + k 32 V + k 33 W (8c)
Introducing Eqs. (6)–(8) into the Lagrange

equations (1.8.11) with N ¼ 3,
¼ 0, yields the equation of motion
and taking into account that A U , V , W
of the structure
M € + K
U U ¼P
ðt Þ (9)
where
2 3 2 3
m 0 m yc k 11 k12 k13
6 7 6 7
¼6 0
M m mx c 7 ¼ 6 k21
K k22 k23 7
4 5, 4 5,
my c mx c Io k31 k32 k33
8 9
> P ðt Þ >
< x =

P ðt Þ ¼
P y ðt Þ
>
: >
;
y A P x ðt Þ + xA P y ðt Þ

As was anticipated, Eq. (9) is identical to Eq. (16) of Example 1.5.6.
1.8.2 Lagrange multipliers

Lagrange’s equations result as a direct application of Hamilton’s principle pro-
vided that the energies (kinetic and potential) as well as the virtual work of the
nonconservative forces can be expressed in terms of the generalized coordinates
and velocities, as is indicated in Eqs. (1.8.2), (1.8.3), (1.8.8), (1.8.10).
Lagrange’s equations apply to linear as well as to nonlinear systems.
In certain cases, it is impossible or it is not convenient to choose N indepen-
dent coordinates. Then, we choose K > N coordinates and we introduce
n ¼ K N constraint equations, which in general have the form
g1 ðq1 , q2 , …, qK Þ ¼ 0
g2 ðq1 , q2 , …, qK Þ ¼ 0
(1.8.24)
… …
gn ðq1 , q2 , …, qK Þ ¼ 0
We distinguish two approaches to derive the equations of motion:
@CivilMethod
(a) Eq. (1.8.24) can be solved in terms of n ¼ K N coordinates. Then the

redundant coordinates can be eliminated from Eqs. (1.8.2), (1.8.3),
(1.8.8), (1.8.10), and the equations of motion are formulated using
Lagrange’s equations (1.8.11). If the constraint equations are linear, the
technique presented in Section 1.8.1 can also be employed to reduce the
number of equations to N .
(b) Eq. (1.8.24) cannot be solved in terms of n ¼ K N coordinates. In this
case, the equations of motion can be derived by using the method of
Lagrange multipliers.
The second approach preserves the symmetry of the problem because there are
no preferred coordinates while others are eliminated. Though the method of
Lagrange multipliers deals with more coordinates than the degrees of freedom
of the system, quite often this procedure results in simpler equations.
To illustrate the method of Lagrange multipliers, we consider the variations
of the constraint functions given by Eq. (1.8.24)
dg1 dg1
dg1 ¼ dq1 + ⋯ + dqK ¼ 0
dq1 dqk
dg2 dg2
dg2 ¼ dq1 + ⋯ + dqK ¼ 0 (1.8.25)
dq1 dqk
… …
dgn dgn
dgn ¼ dq1 + ⋯ + dqK ¼ 0
dq1 dqk
which we write as
X
K
aji dqi ¼ 0 ðj ¼ 1, 2, …, n Þ (1.8.26)
i¼1
where
dgj
aji ¼ (1.8.27)
dqi
If we assume that the constraints are frictionless, then no work is done by the
constraint forces Ri when they ride any virtual displacement dqi , that is,
X
K
Ri dqi ¼ 0 (1.8.28)
i¼1
Now we multiply Eq. (1.8.26) by a factor known as the Lagrange multiplier.

Thus, we obtain
XK
lj aji dqi ¼ 0 ðj ¼ 1, 2, …, n Þ (1.8.29)
i¼1
@CivilMethod
where we note that a separate equation is written for each of the constraints.
Next, we subtract the sum of equations of the form (1.8.29) from
Eq. (1.8.28) and, interchanging the order of summation, we obtain
!
XK X n
Ri lj aji dqi ¼ 0 (1.8.30)
i¼1 j¼1
from which, noting that dqi are arbitrary, we conclude that

X n Xn
dgj
Ri ¼ lj aji ¼ lj (1.8.31)
j¼1 j¼1
dqi
The constraint forces Ri constitute additional generalized forces, which

must be included in Lagrange’s equations. Thus Eq. (1.8.11) become
X n
d ∂T ∂T ∂V dgj
+ ¼ Qi + lj ði ¼ 1, 2, …, K Þ (1.8.32)
dt ∂q_ i ∂qi ∂qi j¼1
dqi
What we have accomplished by this procedure is the inclusion of the con-

straint reactions in the equations of motion as additional generalized forces.
Therefore, the number of unknowns becomes K + n, namely the K generalized
coordinates Qi ðt Þ and the n functions lj ðt Þ. The available equations are also
K + n, that is Eq. (1.8.24) plus Eq. (1.8.32).
Eq. (1.8.32) can be derived from Hamilton’s principle if the potential energy
of the external forces is replaced by
Xn
A ¼A lj gj (1.8.33)
j¼1
The function A is referred to as the reduced potential energy. For a more

advanced formulation, including nonholonomic constraints and a dynamic
treatment of the Lagrange multipliers, see Ref. [7].
Example 1.8.4 Equation of motion of the simple pendulum using Lagrange
multipliers
Formulate the equations of motion of the simple pendulum shown in Fig. E1.14
in terms of the Cartesian coordinates x ðt Þ, y ðt Þ, assuming that the rod is weight-
less and rigid.
FIG. E1.14 Pendulum in Example 1.8.4.
@CivilMethod
Solution
The kinetic and the potential energies of the system are
1
T ¼ m x_ 2 + y_ 2
2
A ¼ mgy
U ¼0
Because the rod is rigid, the coordinates must satisfy the constraint equation
g1 ¼ x 2 + y 2 L2 ¼ 0
Differentiating the quantities T and A we obtain

d ∂T ∂T ∂A ∂g1
¼ m x€, ¼ 0, ¼ 0, ¼ 2x, Q1 ¼ px
dt ∂x_ ∂qi ∂x ∂x

d ∂T ∂T ∂A ∂g1
¼ m y€, ¼ 0, ¼ mg, ¼ 2y, Q2 ¼ py
dt ∂y_ ∂y ∂y ∂y
Applying Eq. (1.8.32) for q1 ¼ x and q2 ¼ y we obtain the equations of
motion
m x€ ¼ px + 2xl (1a)
m y€ + mg ¼ py + 2yl (1b)
x 2 + y2 l 2 ¼ 0 (1c)
Eqs. (1) must be solved for the three unknowns x, y, and l. It should be
noted that two of these equations are differential and one algebraic and therefore
special care is required for their solution. A convenient method is to differen-
tiate the constraint equation twice with respect to time and then to solve the
€ y€ and the parameter
resulting linear system of equations for the accelerations x,
l. For the problem at hand, we obtain
2T
x x€ + y y€ ¼ (2)
m
Eqs. (1a), (1b), (2) are combined and written in matrix form
2 38 9 8 9
m 0 2x < x€ = < px =
4 0 m 2y 5 y€ ¼ py mg (3)
: ; : ;
x y 0 l 2T =m
which are solved to yield
px py mg
L2 x€ + x x_ 2 + y_ 2 ¼ y 2 xy (4a)
m m
2 px 2 py mg
L y€ + y x_ + y_ ¼ xy + x
2 2
(4b)
m m
@CivilMethod

xpx y py mg
L l¼
2
T (4c)
2 2
Eqs. (4a), (4b) are solved using techniques for nonlinear differential equa-
tions. Analytical solutions are in general out of the question. However, a numer-
ical solution is always feasible using the methods presented in Chapter 5. Once
the coordinates x ðt Þ, y ðt Þ and the Lagrange multiplier l have been established,
they are utilized in Eq. (1.8.31) to evaluate the constraint forces, which are the
components of the axial force of the rod. Thus, we have
∂g ∂g qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Rx ¼ l ¼ 2lx, Rx ¼ l ¼ 2ly, S ¼ R2x + R2y ¼ 2lL (5)
∂x ∂y
1.8.3 Small displacements

So far, we have not imposed any restrictions on the magnitude of the displace-
ments in our discussion on Lagrange’s equations. Therefore, the derived equa-
tions of motion hold equally for small and large displacements and they are in
general nonlinear differential equations. For small displacements, however,
about the position of static equilibrium, the equations of motions are highly sim-
plified as they become linear. This is very important in structural dynamics,
where we usually deal with small displacements and the resulting linear differ-
ential equations can be readily solved and predict the response of the structure.
1.8.3.1 Potential energy and stiffness matrix

We shall consider a system of particles that is in static equilibrium under the
action of a set of conservative forces. If its configuration is specified by ordinary
coordinates x1 ,x2 , …, x3 N , then the condition for static equilibrium is that the
virtual work done by the applied forces Fi is zero, that is,
X
3N
dW ¼ Fi dxi ¼ 0 (1.8.34)
i¼1
for all virtual displacements dxi consistent with the constraints, which are
assumed workless and bilateral.
Inasmuch as the forces are conservative, they are derivable from a potential
function V ¼ V ðx1 , x2 , …, x3N Þ, V ¼ U + A, according to the relation
∂V
Fi ¼ (1.8.35)
∂xi
Using Eq. (1.8.35), Eq. (1.8.34) is written as
X
3N
∂V
dW ¼ dxi ¼ 0 (1.8.36)
i¼1
∂xi
@CivilMethod
If the number of degrees of freedom is n < 3N , dx’s are not independent. It

is possible to find n independent generalized coordinates by considering trans-
formation equations, that is,
x1 ¼ x1 ðq1 , q2 , …, qn Þ
x2 ¼ x2 ðq1 , q2 , …, qn Þ
(1.8.37)
… …
x3N ¼ x3N ðq1 , q2 , …, qn Þ
Then we have
X
n
∂xi
dxi ¼ dqj (1.8.38)
j¼1
∂qj
Substituting the previous expression for dxi into Eq. (1.8.36), we obtain
3N X
X n
∂V ∂xi
dW ¼ dqj ¼ 0 (1.8.39)
i¼1 j¼1
∂xi ∂qj
Noting that
∂V X 3N
∂V ∂xi
¼ (1.8.40)
∂qj i¼1
∂xi ∂qj
and interchanging the order of summation, Eq. (1.8.39) becomes

X
n
∂V
dW ¼ dqj ¼ 0 (1.8.41)
j¼1
∂qj
Because dqj are assumed to be independent, the virtual work is zero only if
the coefficients of dqj are zero at the equilibrium condition, that is, if

∂V
¼ 0, j ¼ 1, 2, …, n (1.8.42)
∂qj 0
The subscript zero denotes that the derivatives refer to the equilibrium
position.
Let us expand now the potential energy function V ðq1 , q2 , …, qn Þ in a Tay-
lor series about the position of equilibrium
Xn n 2
∂V 1X n X
∂ V
V ¼ V0 + dqi + dqi dqj + ⋯ (1.8.43)
i¼1
∂qi 0 2 i¼1 j¼1 ∂qi ∂qj 0
We can arbitrarily set the potential energy at the reference position equal to
zero, that is,
V0 ¼ 0 (1.8.44)
@CivilMethod
If we now assume that the displacements about the equilibrium position are
small, we can neglect terms of order higher than the second in Eq. (1.8.43).
Thus, using Eqs. (1.8.41), (1.8.44) the expression for the potential energy is sim-
plified as
n 2
1X n X
∂ V
V¼ qi qj (1.8.45)
2 i¼1 j¼1 ∂qi ∂qj 0
or setting

∂2 V
kij ¼ kji ¼ (1.8.46)
∂qi ∂qj 0
we can write Eq. (1.8.45) in the form

1X n X n
V¼ kij qi qj (1.8.47)
2 i¼1 j¼1
The quantities kij defined by Eq. (1.8.46) are the stiffness coefficients of the
system. Thus we see that the potential energy is expressed by a homogeneous
quadratic function of the generalized coordinates qi if small motions about the
position of equilibrium are examined.
Eq. (1.8.47) is written in matrix form
1
V ¼ qT kq (1.8.48)
2
where
8 9 2 3
>
> q1 >
> k11 k12 ⋯ k1n
< = 6 k21 k22
q2 ⋯ k2n 7
q¼ , k ¼6
4 ⋮ ⋮
7 (1.8.49)
>
> ⋮> > ⋱ ⋮ 5
: ;
qn kn1 kn2 ⋯ knn
The matrix k is called the stiffness matrix of the system.
The expression for the potential energy given in Eq. (1.8.47) is an example
of a quadratic form. For a system whose reference equilibrium configuration is
stable, the potential energy V is positive for all possible values of qi , except
q1 ¼ q2 ¼ … ¼ qn ¼ 0. In this case, the function V is referred to as positive def-
inite. This condition, however, puts restrictions on the allowable values of kij . It
is clear that all diagonal elements must be positive. The necessary and sufficient
condition that V be positive definite is that
2 3
k11 k12 ⋯ k1n
k k 6 k21 k22 ⋯ k2n 7
k11 > 0, 11 12 > 0, …, 6
4 ⋮ ⋮ ⋱ ⋮ 5>0
7 (1.8.50)
k21 k22
kn1 kn2 ⋯ knn
@CivilMethod
1.8.3.2 Kinetic energy and mass matrix

The kinetic energy is
2
1X3N
∂xk
T¼ mk (1.8.51)
2 k¼1 ∂t
Differentiating Eq. (1.8.37) with respect to time yields
∂xk X n
∂xk
¼ q_ (1.8.52)
∂t j¼1
∂qj j
Consequently, it is about the position of the static equilibrium

2 Xn X n
∂xk ∂xk ∂xk
¼ q_ i q_ j (1.8.53)
∂t 0 i¼1 j¼1
∂qi 0 ∂qj 0
Introducing this expression into Eq. (1.8.51) we can write the kinetic energy
in the form
1X n X n
T¼ mij q_ i q_ j (1.8.54)
2 i¼1 j¼1
where it was set
X
3N
∂xk ∂xk
mij ¼ mji ¼ mk (1.8.55)
k¼1
∂qi 0 ∂qj 0
The quantities mij defined by Eq. (1.8.55) are the inertia coefficients of the
system.
1
T ¼ q_ T mq_ (1.8.56)
2
where
8 9 2 3
>
> q_ 1 >
> m11 m12 ⋯ m1n
< = 6 m21 m22
q_ 2 ⋯ m2n 7
q_ ¼ , m ¼6
4 ⋮
7 (1.8.57)
>
> ⋮> > ⋮ ⋱ ⋮ 5
: ;
q_ n mn1 mn2 ⋯ mnn
The matrix m is called the mass matrix of the system. The kinetic energy is a
positive definite quadratic function because it is the sum of positive quantities,
that is, the kinetic energies of the masses of the individual particles.
@CivilMethod
The equations of motion are obtained by applying Eq. (1.8.11). Differenti-

ating the kinetic energy, Eq. (1.8.54), and the potential energy, Eq. (1.8.47),
yields
∂T
¼0
∂qi
X n
d ∂T
¼ mij q€j
dt ∂q_ i j¼1
∂V X n
¼ kij qj
∂qi j¼1
Substituting into Lagrange’s equations, we obtain the following equations of

motion
X
n X
n
mij q€j + kij qj ¼ Qi ði ¼ 1, 2, …, n Þ (1.8.58)
j¼1 j¼1
or in matrix form
m€
q + kq ¼ pðtÞ (1.8.59)
where p(t)¼Q
The matrices m and k are symmetric. It is an advantage of the Lagrange for-
mulation of the equations of motion that it preserves the symmetry of the coef-
ficient matrices for those cases where T and V are represented by quadratic
functions of the velocities and displacements, respectively.
1.8.4 Raleigh’s dissipation function

The dissipative forces arising in a mechanical system are nonconservative
forces. Therefore, they are not derivable from a potential function. They are
involved in the Lagrange equations with their virtual work. When the dissipa-
tive forces are due to such sources as air resistance or internal friction, they are
usually assumed to depend linearly on the velocities along with the physical
coordinates and opposed to the motion, that is
X
n
fDj ¼ cij q_ j (1.8.60)
i¼1
where cij ¼ cji are the damping coefficients of the linear viscous damping.
Apparently, we can construct a quadratic function
1X n X n
R¼ cij q_ i q_ j (1.8.61)
2 i¼1 j¼1
@CivilMethod
which yields
∂R Xn
fDj ¼ ¼ cij q_ j (1.8.62)
∂q_ j i¼1
If these forces are introduced as generalized forces in Lagrange’s equations

(1.8.11) and shifted to the left side, we obtain

d ∂T ∂T ∂V ∂R
+ + ¼ Qi ði ¼ 1, 2, …, n Þ (1.8.63)
dt ∂q_ i ∂qi ∂qi ∂q_ i
The function R defined by Eq. (1.8.61) is known as Raleigh’s dissipation
function.
The equations of motion, Eq. (1.8.58), become now
X
n X
n X
n
mij q€j + cij q_ j + kij qj ¼ pi ði ¼ 1, 2, …, n Þ (1.8.64)
j¼1 j¼1 j¼1
where pi denotes the nonconservative external forces.

q + cq_ + kq ¼ p
m€ (1.8.65)
The matrix c is called the damping matrix of the system.
1.9 Influence of the gravity loads

We consider the system of Fig. 1.9.1a, which can move in the vertical direction.
Apparently, the weight of the body must be added to the external forces. If the
vertical displacement from the undeformed position is designated by u ¼ u ðt Þ,
the equation of motion will read
m u€ + cu_ + ku ¼ pðt Þ + W (1.9.1)
(a) (b)
FIG. 1.9.1 Influence of the gravity load.
@CivilMethod
The elongation ust of the spring under its own weight will be
ust ¼ W =k ¼ constant (1.9.2)
Further we set
u ¼ ust + uðt Þ (1.9.3)
where uðt Þ represents the vertical displacement measured from the position of
the static equilibrium.
Differentiating Eq. (1.9.3) yields
_ u€ ¼ u€
u_ ¼ u, (1.9.4)
Using Eqs. (1.9.3), (1.9.4), the equation of motion (1.9.1) becomes
m u€
+ cu_ + ku st + k u ¼ pðt Þ + W
or using Eq. (1.9.2) we obtain
m u€
+ cu_ + k u ¼ pðt Þ (1.9.5)
The conclusion drawn from Eq. (1.9.5) states that, in the study of the
dynamic response of a system undergoing small displacements, the loads due
to gravity can be neglected. Of course, the total displacements will result as
the sum of the static plus dynamic displacements. That is, the superposition
principle is valid.
1.10 Problems
Problem P1.1 The plane square rigid body B of side length L and surface
mass density g is supported by two identical inclined columns having
cross-sectional moment of inertia I , modulus of elasticity E, and negligible
mass. Derive the equation of motion neglecting the axial deformation of
the columns (Fig. P1.1).
FIG. P1.1 Structure in problem P1.1
@CivilMethod
Problem P1.2 Consider the structure of Fig. P1.2a. The square plate of constant
thickness h ¼ a=10 and mass density g is supported at its center by a flexible
column having a circular cross-section with diameter d ¼ a=10, height a, and
material constants E, n. The plate is loaded by a force P acting in the plane
of the plate at point (Aða=8, a=6Þ and in the direction ∡x, P ¼ b ¼ 30° as
shown in Fig. P1.2b. Derive the equations of motion of the plate when the mass
of the column is neglected.
(a) (b)
FIG. P1.3 Structural model in problem P1.3
Problem P1.3 The semicircular rigid plate of constant thickness and total mass
m is supported as shown in Fig. P1.3. Taking into account that the support at
point O is a hinge, formulate the equation of motion of the plate using (i)
the method of equilibrium of forces, (ii) the principle of virtual displacements,
and (iii) the method of the Lagrange equations.
Problem P.1.4 Consider the system shown in Fig. P1.4. The bars AD and EG
are rigid with masses m and m=3 , respectively. The mass at end D is concen-
trated. The elastic supports at points at B, E , and D are simulated by springs
with a stiffness k while the end G is supported by a viscous damper with a
damping coefficient c. The rod CE is weightless and rigid. Derive the equation
of motion using the principle of virtual displacements.
@CivilMethod
FIG. P1.4 System in problem P1.4
Problem P1.5 Consider the system shown in Fig. P1.5. The mass m is sup-
ported at the top of the flexible and massless column 2 3, which is supported
on the ground by means of the rigid body 1 2 of mass 2a m. The support 1 is
elastically restrained by the rotational spring CR . Formulate the equation of
motion of the structure using CR ¼ EI =2a, m ¼ m=a.
Problem P1.6 Consider the two-story frame of Fig. P1.6. The columns 1 2,
10 20 , and the beam 3 30 are rigid while the columns 2 3, 20 30 , and the
beam 2 20 are massless and flexible with cross-sectional moment of inertia
I and modulus of elasticity E. The supports at 1 and 10 are elastically restrained
by rotational springs with a stiffness CR . Formulate the equation of motion of
the structure taking CR ¼ EI =2a and m ¼ m=a.
FIG. P1.6 Two-story frame in problem P1.6

@CivilMethod
Problem P1.7 The system of Fig. P1.7 consists of the beam AB and the rigid
body S interconnected at B. The beam AB has a negligible mass, modulus of
elasticity E , and cross-sectional moment of inertia I . The beam is fixed at A
while the rigid body is elastically restrained at C by a rotational spring with
a stiffness CR ¼ EI =10L. The total mass m is uniformly distributed. The system
is loaded by the concentrated moment M ðt Þ at point B. Derive the equation of
motion of the system using Lagrange’s equations.
10 10
FIG. P1.8 Frame in problem P1.8
Problem P1.8 The frame of Fig. P1.8 consists of the rigid beam BD of total
mass m and the two massless and flexible columns AB and CD with a
cross-sectional moment of inertia I and modules of elasticity E. The two mass-
less cables FB and GD have cross-sectional area A and cannot undertake com-
pressive force. Derive the equation of motion of the structure taking
I =A ¼ a2 =25 and m ¼ m=5a.
Problem P1.9 Consider the two-story frame of Fig. P1.9. The columns of the
frame are rigid and have a surface mass density g. Their elastic support on the
ground is simulated by the rotational springs with a stiffness CR ¼ EI =10a.
The horizontal beams are massless and flexible with a cross-sectional moment
of inertia I and modulus of elasticity E. Derive the equation of motion when the
structure is subjected to the horizontal loads pðt Þ at the beam levels.
@CivilMethod
FIG. P1.9 Two-story frame in problem P1.9
Problem P1.10 The hinge O of the soft pendulum of Fig. P1.10 is elastically
restrained by the rotational spring with a stiffness CR ¼ EAL=10. The length of
the rod is L, its cross-sectional area A, and the modulus of elasticity E. Formu-
late the equation of motion of the pendulum.
FIG. P1.10 Soft pendulum in problem P1.10
Problem P1.11 The rigid bar AB of circular cross-section and mass density
m ¼ m=a is hinged at point A (Fig. P1.11). The cables DB,FB have cross-
sectional area A and modulus of elasticity E. They are assumed massless
and are prestressed so that they can undertake compressive forces. Formulate

@CivilMethod
the equation of motion of the structure taking into account that the load P is
removed suddenly at instant t ¼ 0. Evaluate the minimum prestressing force
of the cables DB, FB so that they can undertake compressive loads.
Problem P1.12 Consider the structure of Fig. P1.12. The column AC has a
circular cross-section and a mass per unit length m ¼ m=a; it is supported by
a spherical hinge on the ground and is kept in place by three elastic cables of
cross-sectional area A and modulus of elasticity E. The cables are assumed mass-
less and are prestressed so that they can undertake compressive force. Derive the
equation of motion of the structure when it is loaded by the horizontal force P ðt Þ
acting at the top of the column in the direction ∡x,P ¼ b (Fig. P1.12b).
(a) (b)
Problem P1.13 The silo of Fig. P1.13 is supported on its fundament by four
identical columns of a square cross-section. The silo is full of material of density
g. The ground yields elastically with a subgrade constant Ks . The silo and the
fundament are rigid. Derive the equation of motion of the structure when it is
loaded by the horizontal force P ðt Þ acting at the top of the silo in the direction
(a) (b)
FIG. P1.13 Silo on elastic subgrade. (a) Vertical section. (b) Plan form.
@CivilMethod
∡x,P ¼ b (Fig. P1.13b) using the following data: Side of the columns a=4;
thickness of the bottom and walls of the silo a=8; density of the material of
the silo 1:5g; and soil constant Ks ¼ EI =1500a 3 .
Problem P1.14 Consider the one-story building of Fig. P1.14. The rigid plate is
an equilateral triangular with a side a and it is supported by three columns of
height a, rectangular cross-section a=10 a=20, and modulus of elasticity E.
The columns are fixed at both ends. Derive the equation of motion of the plate
when a horizontal force P ðt Þ acts at point Að0, a=5Þ in the direction ∡x, P ¼ b.
The dead weight of the plate is included in p (kN=m2 ).
Rigid plate
FIG. P1.14 One-story building in problem P1.14
Problem P1.15 The two one-story buildings of Fig. P1.15 are connected with a
beam as shown in the figure. All columns have a square cross-section with a
moment of inertia Ic ¼ 2I . The connecting beam has a square cross-section with
moment of inertia Ib ¼ I . The structure is loaded by the horizontal force F(t) at the
level of the plates as shown in Fig. P1.15b. Formulate the equations of motion
using Lagrange’s equations. Assume: Torsion constant It ¼ 2:25d 4 =16, d ¼side
length of the square cross-section of the beam.
Rigid plate
(a)
beam
(b)
FIG. P1.15 Structure in problem P1.15. (a) vertical section, (b) plan form.
Problem P1.16 The system of Fig. P1.16 consists of the block of mass m1 ,
which can slide without friction on the inclined surface, and the pendulum of
@CivilMethod
length L and mass m2 , which is pivoted at the center of mass of the block. The
rod of the pendulum has a cross-sectional area A and modulus of elasticity E.
Assuming plane motion, derive the equation of motion of the system taking
EA=L ¼ 5k and m1 ¼ 5m2 .
.
Problem P1.17 Derive the equation of motion of the system shown in

Fig. P1.17. The mass of the case is m2 . Use m2 ¼ 5m1 , k2 ¼ 3k1 , c2 ¼ 1:5c1 .
Problem P1.18 Consider the crane of Fig. P1.18. The horizontal beam is
assumed rigid. The column is flexible with a cross-sectional moment of inertia
I and the cable axially deformable with cross-sectional area A. The mass of the
cable and column is negligible. Derive the equation of motion of the system
when it is loaded by the horizontal force pðt Þ in the plane of the structure using
I =A ¼ a 2 =100 and a common modulus of elasticity E.
@CivilMethod
FIG. P1.18 Crane in Problem P1.18.
References and further reading

[1] S. Timoshenko, D.H. Young, W. Weaver Jr., Vibration Problems in Engineering, fifth ed., John
Wiley, New York, 1990.
[2] R.L. Coelho, On the deduction of Newton’s second law, Acta Mech. 229 (5) (2018) 2287–2290.
[3] Euler, L. (1912). Decouverte d’un Nouveau Principe de Mecanique, Memoires de l’academie
des sciences de Berlin 6, 185–217, in: Opera Omnia, Series II, vol. 5, 81–108, Leipzig.
[4] J.T. Katsikadelis, Derivation of Newton’s Law of Motion from Kepler’s Laws of Planetary
Motion, Arch. Appl. Mech. 88 (2018) 27–38, https://doi.org/10.1007/s00419-017-1245-x.
[5] D.T. Greenwood, Principles of Dynamics, Prentice-Hall, Englewood Cliffs, NJ, 1965.
[6] J.T. Katsikadelis, The Boundary Element Method for Engineers and Scientists, Academic Press,
Elsevier, Oxford, UK, 2016.
[7] S. Natsiavas, E. Paraskevopoulos, A set of ordinary differential equations of motion for con-
strained mechanical systems, Nonlinear Dyn. 79 (2015) 1911–1938.
[8] E.N. Strømmen, Structural Dynamics, Springer Series in Solid and Structural Mechanics,
Springer, New York, 2014.
@CivilMethod
Chapter 2
Single-degree-of-freedom
systems: Free vibrations
Chapter outline
2.1 Introduction 83 2.3.3 Overdamped system 96
2.2 Free undamped vibrations 83 2.4 Conservation of energy in an
2.3 Free damped vibrations 91 undamped system 97
2.3.1 Critically damped system 91 2.5 Problems 99
2.3.2 Underdamped system 92 References and further reading 103
2.1 Introduction
In this chapter, the free vibrations of a single-degree-of-freedom system
(SDOF) are studied, that is, its response when it is not subjected to any external
force, pðt Þ ¼ 0, but it is excited by an initial displacement and/or initial velocity.
The dynamic model of the system is shown in Fig. 1.4.1 and the equation of
motion (1.4.8) takes the form
m u€ + cu_ + ku ¼ 0 (2.1.1)
Eq. (2.1.1) is an ordinary linear homogeneous differential equation of the
second order with constant coefficients and its solution can be obtained using
known mathematical methods. Inasmuch as we are interested in the physical
response of the system described by this equation, it is advisable to analyze
the free vibration response in two stages, first for c ¼ 0 and then c 6¼ 0. In the
first case, we speak of free undamped vibrations while in the second case we
speak of free damped vibrations. Illustrative examples analyzing the free vibra-
tions of SDOF systems are presented. The pertinent bibliography with recom-
mended references for further study is also included.
2.2 Free undamped vibrations

Although systems without damping do not exist in the real world, the undamped
response is studied because it provides an insight into the free vibration response
of damped systems. In the absence of damping, c ¼ 0, Eq. (2.1.1) becomes
m u€ + ku ¼ 0 (2.2.1)
@CivilMethod
We look for a solution in the form [1,2]
u ¼ elt (2.2.2)
where l is an arbitrary constant to be determined. Substitution of Eq. (2.2.2)
into Eq. (2.2.1) gives
2
ml + k elt ¼ 0 (2.2.3)
Because for t 0 it is elt 6¼ 0, Eq. (2.2.3) holds only if
ml2 + k ¼ 0 (2.2.4)
Eq. (2.2.4) is the characteristic equation of the differential equation (2.2.1).

Its solutions are
pffiffiffiffiffiffiffi
l1 ¼ iw, l2 ¼ iw, i ¼ 1 (2.2.5)
where
rffiffiffiffiffi
k
w¼ > 0: (2.2.6)
m
The general solution of Eq. (2.2.1) is
u ðt Þ ¼ A0 eiwt + B 0 eiwt (2.2.7)

where A0 and B 0 are arbitrary constants. Using Euler’s formula
eiwt ¼ cos wt i sin wt (2.2.8)
Eq. (2.2.7) is written as
u ðt Þ ¼ ðA0 + B 0 Þcos wt + i ðA0 B 0 Þsin wt (2.2.9)

or introducing the new arbitrary constants A ¼ A0 + B 0 and B ¼ i ðA0 B 0 Þ
we can write
u ðt Þ ¼ A cos wt + B sin wt (2.2.10)

The velocity is obtained by the differentiation of Eq. (2.2.10)
u_ ðt Þ ¼ wA sin wt + wB cos wt (2.2.11)

The arbitrary constants A and B can be determined if the displacement u ðt Þ
and the velocity u_ ðt Þ of the moving system are known at a certain instant t0 .
Usually, it is taken t0 ¼ 0 and the quantities u ð0Þ ¼ u0 and u_ ð0Þ ¼ u_ 0 are
referred to as the initial conditions of the motion.
@CivilMethod
Single-degree-of-freedom systems: Free vibrations Chapter 2 85
Eqs. (2.2.10), (2.2.11) for t ¼ 0 give

A ¼ u0 (2.2.12a)
u_ 0
B¼ (2.2.12b)
w
and Eq. (2.2.10) becomes
u_ 0
u ðt Þ ¼ u0 cos wt + sin wt (2.2.13)
w
Obviously, it is u ðt Þ ¼ 0, when u0 ¼ u_ 0 ¼ 0. Hence, the system is set
to motion only if it is given an initial displacement and/or an initial
velocity.
We set
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi
u_ 0
r ¼ ðu 0 Þ2 + (2.2.14)
w
and we write Eq. (2.2.13) in the form

u0 u_ 0
u ðt Þ ¼ r cos wt + sin wt (2.2.15)
r rw
Inasmuch as it is
2 2
u0 u_ 0 u0 u_ 0
1, 1 and + ¼1 (2.2.16)
r rw r rw
we can set
u0 u_ 0
¼ cos q and ¼ sin q (2.2.17)
r rw
u ðt Þ ¼ rcos ðwt qÞ (2.2.18)
where

u_ 0
q ¼ tan 1 (2.2.19)
wu0
Eq. (2.2.18) states that the motion of the system is a harmonic vibration
with amplitude ju ðt Þjmax ¼ r, angular velocity w, and phase angle q. The
@CivilMethod
geometrical meaning of Eq. (2.2.18) is shown in Fig. 2.2.1. The displacement

u ðt Þ of the system can be visualized as the projection OA0 of the radius r ¼ OA
on the horizontal diameter,
pffiffiffiffiffiffiffiffiffi when it rotates counterclockwise with a constant
angular velocity w ¼ k=m starting off at an angle q equal to the phase angle.
The quantity w is referred to as the natural frequency of the system. Its phys-
ical dimension is s1 and it is measured in radians per second. The natural
frequency is also called eigenfrequency as it results as an eigenvalue of the
eigenvalue problem of linear algebra (see Chapter 12).
FIG. 2.2.1 Geometrical meaning of the free undamped vibration.
The time required for the undamped system to complete one cycle of free
vibration is referred to as the natural period of vibration of the system, which
is denoted by T and measured in seconds. It is related to the natural frequency of
vibration through
2p
T¼ (2.2.20)
w
The inverse of the period
1
f¼ (2.2.21)
T
expresses the number of cycles that the system performs in 1 s. This is referred
to as the natural cyclic frequency. The unit of f is the hertz (Hz) (cycles per
second, cps) and it is related to w through
w
f¼ (2.2.22)
2p
The displacement versus time for a system with w ¼ 8s1 , u0 ¼ 0:05m, and
u_ 0 ¼ 1m=s is shown in Fig. 2.2.2.
@CivilMethod
0.2
(du/dt)0 T = 2p /w
0.15
0.1
0.05
u0
0
–0.05
-
–0.1
–0.15
T = 2p /w
–0.2
0 0.5 1 1.5 2
FIG. 2.2.2 Response of an undamped SDOF system.
Example 2.2.1 Free undamped vibrations of an one-story shear building

Consider the one-story shear building shown in Fig. E2.1. The columns are
assumed massless, inextensible, and fixed on the base. Moreover, the slab is
assumed uniform and rigid. The material constants are E ¼ 2.1 107 kN/m2
and n ¼ 0:2. The total load of the plate (dead plus live) is 70kN=m2 . The dimen-
sions of the column cross-sections are k1 30 30cm2 and k2 30 20cm2 . The
acceleration of gravity is g ¼ 9:81m=s2 . Determine:
(i) The natural frequencies of the building.
(ii) The motion of the slab as well as of the top cross-section of the columns
ki , if the horizontal force P ¼ 1:0 103 kN at point Að2:0, 1:0Þ of
the slab acting in the direction b ¼ ∡x,P ¼ p=6 is removed suddenly
at t ¼ 0.
(a) (b)
FIG. E2.1 One-story shear building in Example 2.2.1. (a) Vertical section. (b) Plan form.
@CivilMethod
Solution
The motion of the slab is described by the two displacements U ,V of its
center O along the x,y axes, respectively, and its rotation W about O. The
stiffness and mass matrices of the structure can be established using
Eqs. (17a), (17b) for the single-story building in Example 1.5.6. However,
taking into account that the structure is symmetric with respect to both axes,
the three components of the motion are uncoupled and an ad hoc solution can
be readily obtained.
(i) The stiffness of the columns in the x and y directions are:
12EI 1y
kx1 ¼ ky1 ¼ ¼ 1360:8kN=m (1a)
h3
12EI 2y
kx2 ¼ ¼ 907:2kN=m (1b)
h3
12EI 2x
ky2 ¼ ¼ 403:2kN=m (1c)
h3
Hence the respective stiffnesses of the structure are
Kx ¼ 4kx1 + 2kx2 ¼ 7257:6kN=m (2a)
Ky ¼ 4ky1 + 2ky2 ¼ 6249:6kN=m (2b)
The torsional stiffness KW is equal to the moment produced by the elastic
forces of the columns for unit rotation of the slab. Referring to Fig. E2.2,
we have
FIG. E2.2 Displacements of point i due to rotation of slab.
u i ¼ ri W sin ai ¼ yi W (3a)

v i ¼ ri W cos ai ¼ xi W (3b)
@CivilMethod
Hence, the elastic forces at the top cross-section are
fx i ¼ kxi yi W (4a)
fy i ¼ kyi xi W (4b)
Besides these forces, a torsional moment is produced due to the rotation

of the top cross-section of the column
GI it E
M i ¼ kWi W, kW i ¼ , G¼ (5)
h 2ð 1 + n Þ
The moment of the elastic forces of column i with respect O is

M i ¼ yi2 kxi + xi2 kyi + kWi W (6)
and its torsional stiffness results for W ¼ 1
kWi ¼ yi2 kxi + xi2 kyi + kWi (7)
Taking into account that [3]

It1 0:141 0:34 ¼ 1:142 103 m4
It2 0:196 0:3 0:23 ¼ 4:7040 104 m4
GI 1t GI 2
kW1 ¼ ¼ 1998:7kNm, kW 2 ¼ t ¼ 823:2kNm (8)
h h
x1 ¼ x6 ¼ 4:85m, x2 ¼ x5 ¼ 0, x3 ¼ x4 ¼ 4:85m
y1 ¼ y2 ¼ y3 ¼ 2:85m, y4 ¼ y5 ¼ y6 ¼ 2:85m
we obtain the torsional stiffness of the structure
X
KW ¼ yi2 kxi + xi2 kyi + kWi ¼ 1:9663 105 kNm (9)
i¼1
The mass of the slab and its moment of inertia with respect to O are
m ¼ 10 5 70=9:81 ¼ 356:78kNm1 s2 (10)
10 53 + 103 5
Io ¼ 70=9:81 ¼ 3716:44kNs2 (11)
12
Therefore, the three natural frequencies w1 ,w2 ,w3 are
@CivilMethod
rffiffiffiffiffiffi 9
Kx 1 >
w1 ¼ ¼ 4:510s > >
>
>
m >
>
rffiffiffiffiffiffi >
=
Ky 1
w2 ¼ ¼ 4:185s > (12)
m >
>
rffiffiffiffiffiffiffi >
>
>
KW 1 >
w3 ¼ ¼ 7:274s > ;
I0
(ii) The components of the force P are
Px ¼ P cos b ¼ 866:02kN Py ¼ P sin b ¼ 500:0kN (13)

and its moment with respect to O
MO ¼ yA Px + xA Py ¼ 1866:02kNm (14)
which produces the static displacements
9
U0 ¼ Px =Kx ¼ 0:1193m >
=
V0 ¼ Py =Ky ¼ 0:0800m (15)
>
;
2
W0 ¼ MO =KW ¼ 0:949 10 rad
The slab performs free vibrations with initial conditions U ð0Þ ¼ U0 ,
U_ ð0Þ ¼ 0, V ð0Þ ¼ V0 , V ð0Þ ¼ 0, Wð0Þ ¼ W0 , W_ ð0Þ ¼ 0. Hence, the dis-

placements of the center of the slab are

U ðt Þ ¼ U0 cos w1 t, V ðt Þ ¼ V0 cos w2 t, Wðt Þ ¼ W0 cos w2 t (16)
while the displacements of the i column
ui ðt Þ ¼ U ðt Þ yi Wðt Þ, vi ðt Þ ¼ V ðt Þ + xi Wðt Þ, wi ðt Þ ¼ Wðt Þ (17)
Fig. E2.3 presents the displacements of the column at the upper right
corner (x1 ¼ 4:85m, y1 ¼ 2:85m)
0.2
u1(t)
0.15 v1(t)
w1(t)
0.1
0.05
–0.05
–0.1
–0.15
–0.2
0 1 2 3 4 5
t
FIG. E2.3 Displacements u ðt Þ,v ðt Þ and rotation wðt Þ of the top cross-section of column at
1 ¼ 4:85m, y1 ¼ 2:85m in Example 2.2.1.
@CivilMethod
x
2.3 Free damped vibrations

In this case, it is c 6¼ 0 and the equation of motion (2.1.1) becomes
m u€ + cu_ + ku ¼ 0 (2.3.1)
We look again for a solution of the differential equation (2.3.1) in the form
of Eq. (2.2.2). This produces the following characteristic equation
ml2 + cl + k ¼ 0
or
c
l2 + l + w2 ¼ 0 (2.3.2)
m
The two roots of Eq. (2.3.2) are evaluated from the expression
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c c 2
l1,2 ¼ w2 (2.3.3)
2m 2m
The type of the root depends on the discriminant of the quadratic polyno-
mial, Eq. (2.3.2)

c 2
D¼ w2 (2.3.4)
2m
Eq. (2.3.2) has for:
D > 0 two discrete real roots
D ¼ 0 one double real root
D < 0 two complex conjugate roots
The type of root of Eq. (2.3.2) determines the form of the solution of the dif-
ferential equation (2.3.1), and consequently the physical response of the system.
Therefore, we distinguish three types of motion with damping, according to
whether the discriminant is positive, negative, or zero. It is expedient to discuss
first the case when the discriminant vanishes, which expresses the condition for
critical damping.
2.3.1 Critically damped system

The value of the damping coefficient c for which the discriminant vanishes is
called critical damping and it is denoted by ccr . The system under critical damp-
ing is called a critically damped system. This designation is justified because, as
we shall see below, this value represents the transition threshold from oscilla-
tory to nonoscillatory motion and vice versa. Eq. (2.3.4) for D ¼ 0 determines
this value as
c ¼ ccr ¼ 2mw (2.3.5)
For this value of c, Eq. (2.3.3) gives the double root
l ¼ w (2.3.6)
@CivilMethod
and the general solution of Eq. (2.3.1) is

u ðt Þ ¼ ðA + Bt Þewt (2.3.7)
The arbitrary constants are evaluated from the initial conditions u ð0Þ ¼ u0
and u_ ð0Þ ¼ u_ 0 . Thus, we obtain
A ¼ u0 , B ¼ wu0 + u_ 0 (2.3.8)
u ðt Þ ¼ ½u0 + ðwu0 + u_ 0 Þt ewt (2.3.9)
Fig. 2.3.1 shows the plot of the displacement given by Eq. (2.3.9) if
u0 ¼ 0:05m, u_ 0 ¼ 0:2ms1 , w ¼ 8s1 . We see that the motion of the critically
damped system is nonoscillatory, but the displacement vanishes exponentially,
that is, the system returns to static equilibrium in infinite time.
0.07
0.06 (du/dt)0
0.05
0.04
u(t)
0.03
u0
0.02
0.01
0
0 0.2 0.4 0.6 0.8 1
t
FIG. 2.3.1 Response of a system with critical damping.
2.3.2 Underdamped system

This case is the most interesting because in all structural systems and in the
majority of the mechanical systems, the damping is much less than the critical.
To facilitate the study of the underdamped systems, we introduce the damping
ratio x, which is defined as
c c
x¼ ¼ <1 (2.3.10)
ccr 2mw
@CivilMethod
Using Eq. (2.3.10), we write Eq. (2.3.3) as

qffiffiffiffiffiffiffiffiffiffiffiffi
l1,2 ¼ xw w x 2 1 (2.3.11)
or
l1,2 ¼ xw iwD (2.3.12)
where
wD ¼ w 1 x2 (2.3.13)
Further, the equation of motion (2.3.1) can be written as

u€ + 2xwu_ + w2 u ¼ 0 (2.3.14)
whose general solution is
u ðt Þ ¼ A0 eðxw + iwD Þt + B 0 eðxwiwD Þt
(2.3.15)
¼ exwt ðA0 eiwD t + B 0 eiwD t Þ
or using Eq. (2.2.8), we obtain
u ðt Þ ¼ exwt ðA cos wD t + B sin wD t Þ (2.3.16)
The arbitrary constants A,B are evaluated from the initial conditions
u ð0Þ ¼ u0 and u_ ð0Þ ¼ u_ 0 . These yield
u_ 0 + u0 xw
A ¼ u0 , B ¼ (2.3.17)
wD
and Eq. (2.3.16) is written as

u_ 0 + u0 xw
u ðt Þ ¼ exwt u0 cos wD t + sin wD t (2.3.18)
wD
We observe that in the absence of damping, that is, x ¼ 0, it is wD ¼ w and
Eq. (2.3.18), as anticipated, becomes identical to Eq. (2.2.13). Eq. (2.3.18) can
also be written in the form
u ðt Þ ¼ rexwt cos ðwD t qÞ (2.3.19)
where the amplitude r and the phase angle q are evaluated from the relations
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
u_ 0 + u0 xw 2 2
r¼ + ðu0 Þ (2.3.20a)
wD

1 u_ 0 + u0 xw
q ¼ tan (2.3.20b)
wD u0
Fig. 2.3.2 shows the plot of the displacement given by Eq. (2.3.19) with
u0 ¼ 1m, u_ 0 ¼ 1ms1 , w ¼ 6s1 , x ¼ 0:07. Moreover, Fig. 2.3.3 presents the
@CivilMethod
dynamic response of a SDOF system for various values of the damping ratio and
u0 ¼ 1, u_ 0 ¼ 7ms1 , w ¼ 6s1 . From the plot of Eq. (2.3.19), we conclude that
the motion of an underdamped system is harmonic vibration with frequency
wD and period T ¼ 2p=wD , whose amplitude, however, decays exponentially
with time and is bounded by the envelops rexwt . Hence, wD is called the
damped frequency and the respective period the damped period of the under-
damped system.
1.5 T = 2p /wD
T T
1
re−xw t
0.5 u0
u(t)
–0.5
−re-xw t
–1
–1.5
–2
0 1 2 3 4 5
t
FIG. 2.3.2 Response of an underdamped SDOF system.
2
x=0
1.5 x=0.03
x=0.06
x=0.1
1
0.5
u(t)
–0.5
–1
–1.5
–2
0 1 2 3 4 5
t
FIG. 2.3.3 Response of an underdamped SDOF system for various values of the damping ratio.
@CivilMethod
If we denote the displacement at time t by u ðt Þ, then the displacement at

time t + nT , that is, after n complete cycles, will be
u ðt + nT Þ ¼ rexwðt + nT Þ cos ½wD ðt + nT Þ q (2.3.21)
Because
2p
T¼ (2.3.22)
wD
we have

2p
xw t + n w
u ðt + nT Þ ¼ re D
cos ½ðwD t + 2npÞ q
p2npx
ffiffiffiffiffiffiffiffi
(2.3.23)
¼ rexwt e 1x2 cos ðw
D t qÞ
pffiffiffiffiffiffiffi2ffi
2npx
¼ u ðt Þe 1x

u ðt Þ 2npx
d ¼ ‘n ¼ pffiffiffiffiffiffiffiffiffiffiffiffi (2.3.24)
u ðt + nT Þ 1 x2
The quantity d defined by Eq. (2.3.24) is called the logarithmic decrement
and can be employed to determine the damping ratio x experimentally, when the
displacements u ðt Þ and u ðt + nT Þ between n consecutive cycles are measured.
Example 2.3.1 Free damped vibrations of a silo
Fig. E2.4 presents the idealization of a silo. It consists of two massless and inex-
tensional columns of cross-sectional area a a and a square rigid plate of mass
m. At time t ¼ 0, the silo is displaced horizontally by a force P. Then the force is
suddenly removed and the system starts to vibrate. For P ¼ 200 kN, h ¼ 5m,
a ¼ 0:3m, E ¼ 2.1 107 kN/m2, m ¼ 100kN=ms2 determine:
(i) The damping ratio x of the system, if the horizontal displacement is reduced
to u1 ¼ 0:1u0 after n ¼ 5 oscillations.
(ii) Determine the displacement at time t1 ¼ 2.
FIG. E2.4 Model of the silo in Example 2.3.1.
@CivilMethod
Solution
(i) The only possible motion of the structure is its horizontal displacement
u ðt Þ. Hence, the system has a single degree of freedom. The stiffness k
of the structure is due to the relative displacement of the column ends,
which yield the stiffness
12EI 12 2:1 107 0:34 =12

k ¼2 ¼ 2 ¼ 2721:6kN=m (1)
h3 53
The force P produces an initial displacement
u0 ¼ P=k ¼ 200=2721:6 ¼ 0:0735m (2)
Eq. (2.3.24) for n ¼ 5, u1 ¼ 0:1u0 gives x ¼ 0:0731.
pffiffiffiffiffiffiffiffiffi
(ii) The natural frequency is w ¼ k=m ¼ 5:2169s1 . The motion of the
silo is given by Eq. (2.3.18) with u0 ¼ 0:0735m, u_ 0 ¼ 0,
pffiffiffiffiffiffiffiffiffiffiffiffi
wD ¼ w 1 x2 ¼ 5:2029
u ðt Þ ¼ exp e0:3813t ½0:07349 cos 5:203t + 0:005386 sin 5:203t (3)
which for t ¼ t1 ¼ 2 gives u ðt1 Þ ¼ 0:0211m.
2.3.3 Overdamped system

When the damping is greater than the critical, that is, c > ccr , the system is said
to be overdamped. Hence
c c
x¼ ¼ >1 (2.3.25)
ccr 2mw
then Eq. (2.3.3) gives two distinct real roots
l1, 2 ¼ xw W, W ¼ w x2 1 x > 1 (2.3.26)
and Eq. (2.3.1) has the general solution

u ðt Þ ¼ exwt A0 eWt + B 0 eWt (2.3.27)
Taking into account that
eWt + eWt eWt eWt
cosh Wt ¼ , sinh Wt ¼ (2.3.28)
2 2
Eq. (2.3.27) can be written in the alternative form
u ðt Þ ¼ exwt ðA cosh Wt + B sinh Wt Þ (2.3.29)
@CivilMethod
The arbitrary constants A,B are evaluated from the initial conditions
u ð0Þ ¼ u0 and u_ ð0Þ ¼ u_ 0 . Then, Eq. (2.3.29) becomes

u_ 0 + u0 xw
u ðt Þ ¼ exwt u0 cosh Wt + sinh Wt (2.3.30)
W
Eq. (2.3.30) has been plotted in Fig. 2.3.4 with u0 ¼ 1m, u_ 0 ¼ 10ms1 , and
w ¼ 6s1 . It becomes evident that the motion of the overdamped system is
nonoscillatory.
1.5
x =1
x =1.5
x =2
x =2.5
1
u(t)
0.5
0
0 0.5 1 1.5 2
t
FIG. 2.3.4 Response of an overdamped SDOF system for various values of the damping ratio.
2.4 Conservation of energy in an undamped system

The free undamped vibrations of the SDOF system are governed by Eq. (2.2.1),
namely
m u€ + ku ¼ 0 (2.4.1)
Multiplication of the previous equation by the velocity u_ yields
m u€u_ + ku u_ ¼ 0 (2.4.2)
which may be written as
1 2 1 2
md u_ + kd u ¼ 0 (2.4.3)
2 2
@CivilMethod
and after integration over the interval ½0, t gives

1 1
m u_ 2 + ku 2 ¼ E (2.4.4)
2 2
where E denotes the constant
1 1
E ¼ m ½u_ 0 2 + k ½u0 2 (2.4.5)
2 2
The first term on the left side of Eq. (2.4.4) represents the kinetic energy
T of the system while the second term represents the potential energy U , which
in this case is the elastic energy of the spring. Thus, we may write
T +U ¼E (2.4.6)
and make the following statement:
In a system that performs free undamped vibrations, the total energy, kinetic
plus potential, remains constant during the whole duration of the motion and is
equal to the sum of the kinetic and potential energy given to the system at the start
of the motion.
Eq. (2.4.4) for u_ ¼ 0 gives the maximum value of the potential energy
1
Umax ¼ ku 2max ¼ E (2.4.7)
2
hence
rffiffiffiffiffiffi
2E
umax ¼ (2.4.8)
k
Similarly, the maximum value of the kinetic energy is obtained from the
same equation for u ¼ 0
1
Tmax ¼ m u_ 2max ¼ E (2.4.9)
2
hence
rffiffiffiffiffiffi
2E
u_ max ¼ (2.4.10)
m
Eqs. (2.4.7), (2.4.9) imply
Umax ¼ Tmax (2.4.11)
or
1 2 1
ku ¼ m u_ 2max ¼ E (2.4.12)
2 max 2
@CivilMethod
Using Eq. (2.2.18) the potential energy is written

1 1
U ¼ ku 2 ¼ kr2 cos 2 ðwt qÞ (2.4.13)
2 2
which takes its maximum value when cos ðwt qÞ ¼ 1. Thus, we have
1
Umax ¼ kr2 (2.4.14)
2
Similarly, the kinetic energy is written
1 1
T ¼ m u_ 2 ¼ mw2 r2 sin 2 ðwt qÞ (2.4.15)
2 2
which takes its maximum value when sin ðwt qÞ ¼ 1. Thus, we have
1
Tmax ¼ mw2 r2 (2.4.16)
2
Substituting Eqs. (2.4.14), (2.4.16) into Eq. (2.4.11) yields
1 1
mw2 r2 ¼ kr2 (2.4.17)
2 2
which provides the following relation for computation of the natural frequency
of the system
rffiffiffiffiffi
k
w¼ (2.4.18)
m
The method we just described for the evaluation of the natural frequency of a
system based on the conservation of energy is known as the Rayleigh method.
Actually, Eq. (2.4.18) is identical to Eq. (2.2.6). Obviously, the present method
does not seem to provide any advantage over the direct method. This is gener-
ally true. The Rayleigh method is particularly useful for the approximate deter-
mination of the natural frequency of continuous systems, for which an analytical
solution is either very difficult or impractical to obtain (see Chapters 8 and 12).
Nevertheless, modern numerical methods have significantly limited the use of
the Rayleigh method. Of course, this method is very useful in cases where we
want to check the correctness of numerical approaches in problems for which
analytical or other numerical solutions are not available for comparison.
2.5 Problems
Problem P2.1 The structure of Fig. P2.1 consists of two identical rigid bars BA
and BC , both having line density m (mass/length). The support A is a hinge
while the support C is a simple support. The bracing rod DF has a
@CivilMethod
cross-sectional area A, a modulus of elasticity E, and a negligible mass. A ver-

tical force P applied at point B is suddenly removed at t ¼ 0. Determine the
motion of the structure taking the horizontal displacement u ðt Þ of C as the
parameter of motion. Data: m ¼ 5kg=m, a ¼ 3m, E ¼ 2.1 108 kN/m2,
A ¼ 12cm2 .
FIG. P2.1 Structure in Problem 2.1.
Problem P2.2 An equipment of weight 15kN is supported on the horizontal

base by three identical systems, each consisting of a spring and a damper placed
at the three vertices of an equilateral triangle. The projection of the weight cen-
ter of the equipment coincides with that of the triangle. The weight of the equip-
ment produces a static deflection ust ¼ 2:5cm. The system performs free
vibrations. The dampers are regulated so that the amplitude of vibration reduces
to 1=10 of the initial deflection after five complete cycles. Determine the
damping coefficient c and compare the frequencies w and wD .
Problem P2.3 The wooden body of mass m1 is constrained by the spring k and
the damper c as shown in Fig. P2.3. A projectile of mass m2 ¼ 0:2m1 is fired
into the body and becomes implanted in it. If the speed of the projectile is v,
determine
(i) the maximum displacement if x ¼ 0.
(ii) the displacement u ðt Þ of the body if x ¼ 0:1.
FIG. P2.3 SDOF system in Problem P2.3.
Problem P2.4 A SDOF system of mass m and stiffness k performs free vibra-
tions. At the end of four complete cycles, the displacement is u ð0Þ=3. If the mass
@CivilMethod
of the system is increased by 50%, determine the amplitude of the vibrations

after four complete cycles.
Problem P2.5 The packing of a sensitive instrument is modeled by the system
of Fig. P2.5. During transportation, the box of mass m2 ¼ 10m1 falls vertically
from a height h. Assuming that the box does not bounce after the collision with
the ground, determine the motion of the instrument. What is its maximum
acceleration?
FIG. P2.5 Structural system in Problem P2.5.
Problem P2.6 The horizontal force P applied to the structure of Fig. P2.6 is
suddenly removed at t ¼ 0. Determine the motion of the structure when (i)
the cables are free of any pretension and (ii) have been prestressed to withstand
compression, and compute the minimum required pretension forces. In both
cases, the cables are assumed massless. Data: a ¼ 2m, P ¼ 100kN,
m ¼ 100 kNm1 s2 , E ¼ 2.1 107 kN/m2, I ¼ 880cm4 , A ¼ 5cm2
FIG. P2.6 Structure in Problem P2.6.
Problem P2.7 In the structure of Fig. P2.7, the rigid rod AB of circular cross-
section and line mass density m ¼ m=a is supported on the ground by a spher-
ical hinge and held in vertical position by three identical elastic massless cables
of cross-sectional area A and modulus of elasticity E. The cables are prestressed
@CivilMethod
so that they can withstand compression. Their anchor points D,G,F form an
equilateral triangle. Determine the motion of the structure if the horizontal
force P applied at the top of the rod in the direction of the y axis is suddenly
removed at t ¼ 0. Evaluate the minimum values of the prestress forces, which
ensure the capability of the cables to withstand compression. Data: a ¼ 2m,
m ¼ 100 kN m1s2, A ¼ 5cm2 , E ¼ 2.1 108 kN/m2, P ¼ 100kN.
(a) (b)
FIG. P2.7 Structure in Problem P2.7. (A) Vertical view. (B) Plan form.
Problem P2.8 In the pendulum of Fig. P2.8, the rigid rod suspending the con-
centrated mass m has a line mass density m ¼ 2m=L. The hinge O is elastically
restrained by the rotational spring with stiffness CR ¼ kL2 =2. The rod is
displaced by an angle q0 from the vertical position and then is left to move.
Considering small displacements, derive the equation of motion and compute
its period.
FIG. P2.8 Pendulum in Problem P2.8.
Problem P2.9 The rigid silo of Fig. P2.9 is supported on its fundament by four
identical columns of square cross-sections with a side length a=8. The silo is full
of material with density g. The bottom and the walls of the silo have a thickness
a=8 and material density 1:5g. Compute the frequencies and the periods of the
structure.
@CivilMethod
(a) (b)
FIG. P2.9 Silo on four columns. (a) Vertical section. (b) Plan form.

[1] F.B. Hildebrand, Advanced Calculus for Applications, Prentice-Hall, Inc, Englewood Cliffs,
NJ, 1962.
[2] E. Kreyszig, Advanced Engineering Mathematics, fourth ed., John Wiley & Sons, New York,
1979.
[3] S. Timoshenko, J.N. Goodier, Theory of Elasticity, McGraw-Hill, New York, 1951.
[4] S. Graham Kelly, Mechanical Vibrations, Schaum’s Outline Series, McGraw-Hill, New York,
1996.
[5] W.T. Thomson, Theory of Vibration with Applications, fifth ed., Prentice Hall, Upper Saddle
River, NJ, 1998.
[6] S.S. Rao, Mechanical Vibrations, fifth ed., Prentice Hall, Upper Saddle River, NJ, 2011.
@CivilMethod
Chapter 3
systems: Forced vibrations
Chapter outline
3.1 Introduction 105 3.5.1 Rectangular pulse load 126
3.2 Response to harmonic loading 106 3.5.2 Triangular pulse load 128
3.2.1 Response of undamped 3.5.3 Asymmetrical triangular
systems to harmonic pulse load 131
loading 106 3.5.4 Response to piecewise
3.2.2 Response of damped linear loading 135
systems to harmonic 3.6 Response to a periodic loading 137
loading 110 3.6.1 Periodic loads 137
3.3 Response to arbitrary dynamic 3.6.2 Fourier series 138
loading—Duhamel’s integral 113 3.6.3 Response of the SDOF
3.3.1 Undamped vibrations 113 system to periodic
3.3.2 Damped vibrations 116 excitation 143
3.4 Analytical evaluation of the 3.7 Response to unit impulse 146
Duhamel integral-applications 117 3.7.1 The delta function or
3.4.1 Response to step Dirac’s delta function 146
function load 117 3.7.2 Response to unit impulse 148
3.4.2 Response to ramp 3.7.3 Response to arbitrary
function load 120 loading 151
3.4.3 Response to step function 3.7.4 The reciprocal theorem
load with finite in dynamics 151
rise time 121 3.8 Problems 152
3.5 Response to impulsive loads 125 References and further reading 157
3.1 Introduction
In this chapter, the forced vibrations of the SDOF system are studied. The
dynamic model of the system is shown in Fig. 1.4.1 and the motion of the system
is governed by Eq. (1.4.8), namely
m u€ + cu_ + ku ¼ pðt Þ (3.1.1)
where pðt Þ represents an arbitrary function of time. First, the response under a
harmonic load is examined. This type of loading is particularly important in the
dynamic analysis of structures because it allows understanding the major

@CivilMethod
differences between the static and dynamic response and identifying phenom-
ena such as resonance that are not conceived by the static consideration. More-
over, the harmonic load analysis allows studying the response of SDOF systems
under a general periodic load using the Fourier series representation of a peri-
odic load. Then, the response of a SDOF system under an arbitrary load is stud-
ied using Duhamel’s integral. Finally, the response to a unit load is discussed by
exploiting the properties of the Dirac delta function. The chapter ends by pre-
senting the dynamic reciprocal theorem. Throughout the chapter, illustrative
examples analyzing the forced vibrations of SDOF systems are presented.
The pertinent bibliography with recommended references for further study is
also included. The chapter is enriched with problems to be solved aiming at
better understanding the theoretical issues.
3.2 Response to harmonic loading

In this section, we study the motion of SDOF systems subjected to harmonic
excitations whose magnitudes may be represented by a sine or cosine function
of time, that is,

pðt Þ ¼ p0 sin wt (3.2.1a)
or

pðt Þ ¼ p0 cos wt (3.2.1b)
where p0 is a constant representing the amplitude of the force and w is the fre-
quency known as the excitation frequency. As mentioned in Section 3.1, the study
of the response of a SDOF system under harmonic loads is particularly important
in the dynamic analysis of structures because it allows understanding the major
differences between the static and dynamic response and identifying phenomena
such as resonance that are not realized by the static consideration. In fact, any load
that is periodic can be treated by resolving it into its harmonic components using
the Fourier series representation of a periodic function. The response of the sys-
tem is obtained as the superposition of individual responses to the harmonic
components of the external excitation. We defer this discussion to Section 3.6.
3.2.1 Response of undamped systems to harmonic loading

We study first the response of the system to harmonic loading in absence of
damping. In this case, the equation of motion (3.1.1) for the sine load becomes

m u€ + ku ¼ p0 sin wt (3.2.2)
Eq. (3.2.2) is an ordinary linear nonhomogeneous differential equation of the
second order with constant coefficients and its solution can be obtained using any
established mathematical method [1, 2]. Here, we look for a solution in the form
u ¼ uh + up (3.2.3)
@CivilMethod
Single-degree-of-freedom systems: Forced vibrations Chapter 3 107
where uh is the solution of the homogeneous equation and up a particular solu-

tion of the nonhomogeneous equation, that is any function that satisfies
Eq. (3.2.2). The solution of the homogeneous equation was obtained in
Section 2.2, where the free undamped vibrations were discussed. It is given
by Eq. (2.2.10), namely
uh ¼ A cos wt + B sin wt, w ¼ k=m (3.2.4)
There are general methods for obtaining a particular solution [1, 2]. A simple
method that can be applied in this case is the method of undetermined coeffi-
cients. According to this method, we look for a solution up in the form

up ¼ C sin wt (3.2.5)
where C is a constant to be determined. Substituting Eq. (3.2.5) into Eq. (3.2.2)
yields

m w2 + k C p0 ¼ 0
p0 1
C¼ , (3.2.6a)
k 1 b2
w
b¼ (3.2.6b)
w
and the particular solution (3.2.5) becomes
p0 1
up ¼
sin wt (3.2.7)
k 1 b2
Hence the general solution of Eq. (3.2.2) reads
p0 1
u ðt Þ ¼ Acos wt + B sin wt +
sin wt (3.2.8)
k 1 b2
The arbitrary constants A and B are evaluated from the initial conditions.
We examine the case u ð0Þ ¼ u_ ð0Þ ¼ 0. The expression for the velocity is
obtained by differentiation of Eq. (3.2.8) with respect to time t
p0 w
u_ ðt Þ ¼ Aw sin wt + Bw cos wt +
cos wt (3.2.9)
k 1 b2
For the considered initial conditions, Eqs. (3.2.8), (3.2.9) give
p0 b
A ¼ 0 and B ¼
k 1 b2
p0 1
u ðt Þ ¼ b sin wt Þ
ð sin wt (3.2.10)
k 1 b2
@CivilMethod
Obviously
ust ¼ p0 =k (3.2.11)
denotes the static displacement that would be produced by a load p0 , equal to
the amplitude of the harmonic excitation, if it were to be applied statically.
The time-dependent quantity
u ðt Þ 1
Rðt Þ ¼ ¼ b sin wt Þ
ð sin wt (3.2.12)
ust 1 b 2
is called the response ratio. It is dimensionless and expresses the number that
must multiply the static displacement at time t to obtain the respective dynamic
displacement. The response ratio provides a measure of the influence of the
dynamic loading.
The extreme value
D ¼ max jRðt Þj (3.2.13)
is referred to as the dynamic magnification factor (DMF). It is a very useful
quantity in dynamic analysis because, if it has been established for a given
loading, the extreme state of deformation and stress can be obtained by static
analysis.
Eq. (3.2.12) for b ¼ 1 takes the indeterminate form
0
Rðt Þ ¼
0
whose limit can be determined using the L’H^
opital rule. Thus, on the basis of
Eq. (3.2.6b) we obtain
b sin wt
sin wt
‘im Rðt Þ ¼ ‘im
b!1 b!1 1 b2
sin bwt b sin wt
¼ ‘im
b!1 1 b2
(3.2.14)
wt cos bwt sin wt
¼ ‘im
b!1 2b
wt cos wt sin wt
¼
2
From the latter relation, it is concluded that when b tends to 1, that is, when
the excitation frequency w of the harmonic force approaches the natural fre-
quency of the system, the dynamic displacement grows indefinitely with time,
although the amplitude of the harmonic loading is finite. This phenomenon is
called resonance. The growth of the amplitude of the displacement with time
due to resonance is shown in Fig. 3.2.1. The response is periodic with a period
@CivilMethod
2p=w. A measure of the growth rate can be obtained by taking the difference
of the amplitudes of two consecutive peaks.
The time peak occurs when
20
w t/2
15
2p /w 2p /w
10
5
R(t)
–5
p
–10 p
–15
–w t/2
–20
0 1 2 3 4 5
t
FIG. 3.2.1 Response ratio of an undamped system at resonance, D ¼ max jRðt Þj ! 1
when t ! 1 (w ¼ 7, b ¼ 1).
dRðt Þ w2 sin wt
¼ ¼0 (3.2.15)
dt 2
or
np
t¼ n ¼ 1, 2, … (3.2.16)
w
Hence the difference between consecutive peaks is
np p np
R + R ¼ p cos np
w w w (3.2.17)
¼ p
the
When the excitation force is of the cosine type, pðt Þ ¼ p0 cos wt,
employed procedure yields the particular solution
p0 1
up ¼
cos wt (3.2.18)
k 1 b2
and the general solution for zero initial conditions, u ð0Þ ¼ u_ ð0Þ ¼ 0, is
obtained as
p0 1
u ðt Þ ¼ b cos wt Þ
ð cos wt (3.2.19)
k 1 b2
@CivilMethod
3.2.2 Response of damped systems to harmonic loading

In this case, the equation of motion (3.1.1) for the sine load becomes
m u€ + cu_ + ku ¼ p0 sin wt
(3.2.20)
The general solution of Eq. (3.2.20) can be sought again in the form
u ¼ uh + up (3.2.21)
where uh is the homogeneous solution and up a particular solution of the
nonhomogeneous equation. We will limit our discussion to underdamped sys-
tems. Thus, the homogeneous solution is given by Eq. (2.3.16), namely
uh ¼ exwt ðA cos wD t + B sin wD t Þ (3.2.22)
which was derived in Section 2.3.2. The particular solution is established using
the method of undetermined coefficients. According to this method, up is sought
in the form [1, 2]
+ C2 cos wt
up ¼ C1 sin wt (3.2.23)
which is introduced into Eq. (3.2.20) to yield
p0 1 b2 p0 2xb
C1 ¼ , C2 ¼ (3.2.24)
k 1 b 2 2 + ð2xbÞ2 k 1 b2 2 + ð2xb Þ2
Hence, the general solution of Eq. (3.2.20) becomes
u ðt Þ ¼ exwt ðA sin wD t + B cos wD t Þ

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Transient response
p0 1
+ 2
1 b2 sin wt 2xb cos wt (3.2.25a)
k 1 b2 + ð2xb Þ2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Steady-state response
the previous procedure yields

If the excitation force is p0 cos wt,
u ðt Þ ¼ exwt ðA sin wD t + B cos wD t Þ
Transient response
p0 1
+ 2
1 b2 2xb sin wt + 1 b2 cos wt (3.2.25b)
k 1 b2 + ð2xbÞ2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Steady-state response
It is clear that the first term in Eq. (3.2.25a) decays rapidly with time because
of the exponential term exwt , so its contribution becomes negligible after a
short time, which, of course, depends on the damping ratio. Thus, the response
of the system is governed by the second term. For this reason, we say that the
@CivilMethod
first term expresses the transient response while the second term expresses the
steady-state response of the system. This is shown in Fig. 3.2.2.
FIG. 3.2.2 Response of an underdamped system to harmonic loading ðw ¼ 5, w ¼ 6, p0 =k ¼ 1,

x ¼ 0:1, u0 ¼ 1, u_ 0 ¼ 10Þ.
Referring to Eq. (3.2.25a) we see that the steady-state response can be writ-
ten in the form
qÞ
u ðt Þ ¼ rsin ðwt (3.2.26)
where

p0 h
2 2 2
i1=2
1 2xb
r¼ 1 b + ð2xbÞ , q ¼ tan (3.2.27)
k 1 b2
Therefore, the steady-state response of the underdamped system subjected
to a harmonic loading is an undamped free vibration.
The DMF is
r h 2 i1=2
D ¼ max jRðt Þj ¼ ¼ 1 b2 + ð2xbÞ2 (3.2.28)
p0 =k
We observe that D ¼ D ðbÞ. Consequently the maximum value of D is
obtained when
dD
¼0 (3.2.29)
db
This condition gives
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b ¼ 1 2x 2 (3.2.30)
@CivilMethod
6
x=0
5
x=0.1
4
D (b)
3
x=0.2
2 x=0.3
ξ=0.7
x=0.5
1
ξ=1.0
0
0 0.5 1 1.5 2 2.5 3
Frequency ratio, b
FIG. 3.2.3 Variation of the amplitude of the response ratio Dðb, x Þ.
and
1
Dmax ¼ pffiffiffiffiffiffiffiffiffiffiffiffi (3.2.31)
2x 1 x2
Apparently, the maximum value of the dynamic factor does not occur for
b ¼ 1. Nevertheless, for a small value of x it is b 1, for example, for
x ¼ 0:05 Eq. (3.2.30) gives b ¼ 0:9975. Fig. 3.2.3 shows the variation of D ver-
sus the frequency ratio b for different values of the damping ratio x. Note that
if x ¼ 0, Dmax becomes infinite.
4
1/2x
0
R(t)
–2
–1/2x
–4
–6
0 5 10 15
t
FIG. 3.2.4 Response ratio of an underdamped system at resonance (w ¼ 5, x ¼ 0:1, b ¼ 1,
D ¼ max jRðt Þj ! 1=2x when t ! 1).
@CivilMethod
Let us study now the response of the system when b ¼ 1, which is the
conventional value for resonance. For this value of b it is w ¼ w and
Eq. (3.2.25a) becomes
p0 cos wt
u ðt Þ ¼ exwt ðAsin wD t + B cos wD t Þ (3.2.32)
k 2x
If we assume zero initial conditions, u ð0Þ ¼ u_ ð0Þ ¼ 0, Eq. (3.2.32) gives
" ! #
1 xwt x
Rðt Þ ¼ e pffiffiffiffiffiffiffiffiffiffiffiffi sin wD t + cos wD t cos wt (3.2.33)
2x 1 x2
The time variation of Rðt Þ is shown in Fig. 3.2.4.

We observe that the DMF D ¼ max jRðt Þj, for x 6¼ 0 tends asymptotically
to the value 1=2x. Namely, the presence of damping prevents the occurence
of infinite displacements when x 6¼ 0. The conclusion drawn from this analysis
is that damping limits considerably the consequences of resonance. However,
the resonance phenomenon should not be ignored in structural design because
it produces displacements much larger than the static ones.
3.3 Response to arbitrary dynamic loading—Duhamel’s

integral
3.3.1 Undamped vibrations
In this section, we study the response of undamped SDOF systems when the
external loading pðt Þ is neither harmonic nor periodic, but an arbitrary function
of time. In this case, the equation of motion is
m u€ + ku ¼ pðt Þ (3.3.1)
which we write in the form
1 pffiffiffiffiffiffiffiffiffi
u€ + w2 u ¼ pðt Þ, w ¼ k=m (3.3.2)
m
There are several methods to obtain the solution of Eq. (3.3.2). A con-
venient, straightforward, and rather simple method for solving ordinary dif-
ferential equations with constant coefficients is the Laplace transform
method [1, 3]. This method is based on the Laplace transform, which for a func-
tion u ðt Þ, t 0 is commonly denoted by L and defined as
Z 1
L½u ðt Þ ¼ U ðs Þ ¼ u ðt Þest dt (3.3.3)
0
where s is the variable of the transform.
@CivilMethod
The inverse Laplace transform is defined as the function u ðt Þ, which results

as the solution of the integral equation (3.3.3), when the function U ðs Þ is given.
It is denoted by
u ðt Þ ¼ L1 ½u ðt Þ (3.3.4)
and it is given by the Fourier-Mellin integral.

A method to evaluate the Laplace transform of a simple function as well as
its inverse is to use the Tables of Laplace transform [4]. On the other hand,
ready-to-use computer applications such as Wolfram ALFA, etc., are also avail-
able. These applications can be employed to obtain the Laplace transform and
its inverse of complicated functions.
Using integration by parts, we can derive the Laplace transform of the
derivatives of a function u ðt Þ. Thus, we have
L½u_ ðt Þ ¼ sU u ð0Þ (3.3.5a)
L½u€ðt Þ ¼ s2 U ½su ð0Þ + u_ ð0Þ (3.3.5b)
Application of the Laplace transform to both sides of Eq. (3.3.2) yields

1
L u€ + w2 u ¼ L½pðt Þ
m
or
1
L½u€ + w2 L½u ¼ L½pðt Þ (3.3.6)
m
which by virtue of Eq. (3.3.5b) becomes

1
s2 U ðs Þ ½su ð0Þ + u_ ð0Þ + w2 U ðs Þ ¼ P ðs Þ (3.3.7)
m
where P ðs Þ is the Laplace transform of pðt Þ.
Eq. (3.3.7) is an algebraic equation with respect to U ðs Þ giving
s 1 1 1
U ðs Þ ¼ u ð 0Þ + 2 u_ ð0Þ + P ðs Þ (3.3.8)
s2 +w 2 s +w 2 m s + w2
2
The function u ðt Þ is obtained by taking the inverse Laplace transform of

Eq. (3.3.8), namely
u ðt Þ ¼ L1 U ðs Þ
@CivilMethod
or

1 s 1 s 1 1 s
u ðt Þ ¼ u ð0ÞL + u_ ð0ÞL + L P ðs Þ
s 2 + w2 s 2 + w2 m s 2 + w2
(3.3.9)
From the table of the Laplace transforms we obtain

s
L1 2 ¼ cos wt (3.3.10a)
s + w2

1 1 sin wt
L 2 2
¼ (3.3.10b)
s +w w
Now we focus our attention on the last term on the right side of Eq. (3.3.9).
Its inverse Laplace transform can be obtained using the convolution theorem.
The convolution of two functions f ðt Þ and g ðt Þ denoted by f ðt Þ∗ g ðt Þ or
ðf ∗ g Þt is defined as
Z t
f ðt Þ∗ g ðt Þ ¼ f ðt τÞg ðτÞdτ
0
Z t (3.3.11)
¼ g ðt τÞf ðτÞdτ
0
The Convolution Theorem

Let f ðt Þ and g ðt Þ be functions with Laplace transforms F ðs Þ and G ðs Þ, respec-
tively, that is, L½f ðt Þ ¼ F ðs Þ and L½g ðt Þ ¼ G ðs Þ, then
L½ðf ∗ g Þðt Þ ¼ F ðsÞG ðsÞ (3.3.12a)
or equivalently
Z t
1
L ½F ðs ÞG ðs Þ ¼ ðf ∗g Þðt Þ ¼ f ðt τÞg ðτÞdτ (3.3.12b)
0
Eq. (3.3.12b) can be employed to obtain the inverse Laplace of the last term
on the right side of Eq. (3.3.9). Thus, for F ðs Þ ¼ 1=ðs2 + w2 Þ and G ðs Þ ¼ P ðs Þ,
we obtain
Z t
1 1
L1 2 P ðs Þ ¼ pðτÞ sin ½wðt τÞdτ (3.3.13)
s + w2 mw 0
Hence, substituting Eqs. (3.3.10a), (3.3.10b), (3.3.13) into Eq. (3.3.9) gives
the solution of Eq. (3.3.1)
u_ ð0Þ
u ðt Þ ¼ sin wt + u ð0Þcos wt
w Z
t
1 (3.3.14)
+ pðτÞsin ½wðt τÞdτ
mw 0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Duhamel’s integral
@CivilMethod
It obvious that the first two terms in Eq. (3.3.14) express the contribution of
the initial conditions to the motion of the system. The third term, which
expresses the contribution of the external loading, is known as the Duhamel
integral for the undamped system.
3.3.2 Damped vibrations

When damping is taken into account, the motion is governed by Eq. (3.1.1),
namely
m u€ + cu_ + ku ¼ pðt Þ (3.3.15)
which for an underdamped system is written as
1
u€ + 2xwu_ + w2 u ¼ pðt Þ (3.3.16)
m
For an arbitrary loading function pðt Þ, the solution of Eq. (3.3.16) is
obtained using the Laplace transform method following the procedure pre-
sented in the previous section. Thus taking the Laplace transform of
Eq. (3.3.16), we obtain
1
L½u€ + 2xwL½u_ + w2 L½u ¼ L½pðt Þ (3.3.17)
m
Using Eqs. (3.3.5a), (3.3.5b) and solving for U ðs Þ, we obtain
s 1
U ðs Þ ¼ u ð 0Þ + 2 ½2xwu ð0Þ + u_ ð0Þ
s2+ 2xws + w 2 s + 2xws + w2
(3.3.18)
1 P ðs Þ
+
m s 2 + 2xws + w2
The evaluation of the inverse Laplace transform using the table of the
Laplace transforms requires the factorization of the denominator polynomial
f ðs Þ ¼ s 2 + 2xws + w2 (3.3.19)
Its roots are
l1 ¼ xw + iwD , l2 ¼ xw iwD (3.3.20)
Hence
s 2 + 2xws + iw2 ¼ ðs l1 Þðs l2 Þ (3.3.21)
From the table of the Laplace transforms we obtain

1 1 1 l1 t
L ¼ e e l2 t
ðs l1 Þðs l2 Þ l1 l2
(3.3.22a)
1
¼ exwt sin wD t
wD
@CivilMethod

s 1 l1 t
L1 ¼ l1 e l2 el2 t
ðs l 1 Þð s l 2 Þ l1 l2 (3.3.22b)
xwt
¼e cos wD t
Z t
P ðs Þ 1
L1 ¼ pðτÞexwðtτÞ sin wD ðt τÞdτ (3.3.22c)
ðs l 1 Þ ð s l 2 Þ wD 0
The last expression was obtained using the convolution theorem.

By virtue of Eqs. (3.3.22a)–(3.3.22c), Eq. (3.3.18) gives
u ðt Þ ¼ L1 ½U ðsÞ

u_ ð0Þ + u ð0Þxw
¼ sin wD t + u ð0Þ cos wD t exwt
wD
Z t (3.3.23)
1
+ pðτÞexwðtτÞ sin ½wD ðt τÞdτ
mwD 0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Duhamel’s integral
It is obvious that the first term in Eq. (3.3.23) expresses the contribution of
the initial conditions to the motion of the system. The last term represents the
Duhamel integral for the underdamped motion. It is obvious that Eq. (3.3.23)
for x ¼ 0 yields Eq. (3.3.14) as anticipated. The Duhamel integral can also be
derived using the method described in Section 3.7.3.
3.4 Analytical evaluation of the Duhamel integral-applications

The analytical evaluation of the Duhamel integral presents considerable diffi-
culties, especially when damping is taken into account. In practical terms, the
analytical integration is recommended when the excitation function is simple.
In recent years, however, the development of symbolic languages, such as
MATHEMATICA and MAPLE, has boosted interest in the use of analytical
integration. This paragraph presents some applications of the Duhamel integral
that help in understanding the response of an SDOF system under certain char-
acteristic forms of dynamic loading. The integration is performed analytically.
3.4.1 Response to step function load

A step load is a nonperiodic load applied suddenly at t ¼ t0 and remains constant
during the whole duration of the motion. Mathematically, this load can be repre-
sented using the Heaviside step function
pðt Þ ¼ p 0 H ðt Þ (3.4.1)
where

0, t < t0
H ðt t0 Þ ¼ (3.4.2)
1, t > t0
@CivilMethod
is the Heaviside step function and p0 the magnitude of the load. Fig. 3.4.1 shows
the step function load applied at t ¼ 0.
FIG. 3.4.1 The step function load.
For u ð0Þ ¼ u_ ð0Þ ¼ 0 the displacement is obtained from the Duhamel integral
in Eq. (3.3.14)
Z t Z t
1 p0
u ðt Þ ¼ p0 sin ½wðt τÞdτ ¼ sin ½wðt τÞd ½wðt τÞ
mw 0 mw2 0
p0
¼ ½ cos wðt τÞt0
mw2
or
p0
u ðt Þ ¼ ð1 cos wt Þ (3.4.3)
k
Taking into account that ust ¼ p0 =k represents the static displacement, the
response ratio is
u ðt Þ
R ðt Þ ¼ ¼ ð1 cos wt Þ (3.4.4)
ust
and the DMF
D ¼ max jRðt Þj ¼ 2 (3.4.5)
Eq. (3.4.5) shows that the suddenly applied load produces a maximum dis-
placement that is twice as large than the displacement that the load p0 would
produce if it were applied statically (slowly). This is an elementary but impor-
tant result that illustrates the difference between static and dynamic loading of a
structure.
When the damping is taken into account, the displacement for zero initial
conditions, u ð0Þ ¼ u_ ð0Þ ¼ 0, is obtained from Eq. (3.3.23)
Z t
p0
u ðt Þ ¼ exwðtτÞ sin ½wD ðt τÞdτ (3.4.6)
mwD 0
The evaluation of the Duhamel integral is more complicated. Nevertheless,
using MAPLE we obtain
" ! #
p0 x
u ðt Þ ¼ 1 cos wD t + pffiffiffiffiffiffiffiffiffiffiffiffi sin wD t exwt (3.4.7)
k 1 x2
@CivilMethod
2.5
x=0
2
x=0.05
x=0.15
1.5
1
R(t)
0.5
0.
–0.5
0 0.5 1 1.5 2 2.5 3 3.5 4
t/TD (TD=2p /w D)
FIG. 3.4.2 Response ratio of a system under step load.
We observe that Eq. (3.4.7) for x ¼ 0 becomes identical to Eq. (3.4.3).

The response ratio results from Eq. (3.4.7)
" ! #
x xwt
Rðt Þ ¼ 1 cos wD t + pffiffiffiffiffiffiffiffiffiffiffiffi sin wD t e (3.4.8)
1 x2
Fig. 3.4.2 shows the plot of Eq. (3.4.8) for different values of the
damping ratio.
The extreme values of Rðt Þ occur when
" #
2
dR xwt ðwx Þ
¼e + wD sin wD t ¼ 0 (3.4.9)
dt wD
which gives
np
tn ¼ n ¼ 0,1, 2, … (3.4.10)
wD
Substituting tn into Eq. (3.4.8) yields

Rðtn Þ ¼ 1 ð1Þn enx= 1x
(3.4.11)
@CivilMethod
The maximum value results for n ¼ 1. Hence

D ¼ 1 + ex= 1x
(3.4.12)
that is, the maximum displacement occurs at the first peak, where t1 ¼ p=wD .
This is shown in Fig. 3.4.2. Eq. (3.4.12) for x ¼ 0 gives D ¼ 2 as anticipated.
3.4.2 Response to ramp function load

The ramp function load is a load that increases linearly with time. Mathemat-
ically, it is defined by
pðt Þ ¼ lt t0 (3.4.13)
where l is a nonzero constant denoting the rate of increase of the load

(Fig. 3.4.3).
FIG. 3.4.3 Ramp function load; l ¼ p0 =t1 .
For zero initial conditions, u0 ¼ u_ 0 ¼ 0, the response of an undamped SDOF

system to a ramp function load is obtained by substituting the load pðt Þ ¼ p0 t=t1
into the Duhamel integral in Eq. (3.3.14)
Z
p0 =t1 t
u ðt Þ ¼ τ sin ½wðt τÞdτ (3.4.14)
mw 0
which after integration gives

p0 t sin wt
u ðt Þ ¼ (3.4.15)
k t1 wt1
Fig. 3.4.4 shows the response ratio Rðt Þ ¼ u ðt Þ=ust , ust ¼ p0 =k, of the
undamped system to a ramp function load. We see that it oscillates about the
line p0 t=t1 .
@CivilMethod
30
25
20
R(t)
15
10
0
0 0.5 1 1.5 2 2.5 3
t/T
FIG. 3.4.4 Response ratio of a system under ramp function load (t1 ¼ T =10).
For damped systems, the analytical evaluation is rather complicated. How-

ever, it can be evaluated using a symbolic language.
3.4.3 Response to step function load with finite rise time. Static load
The step function load with finite rise time is a constant load that, however, is
not applied suddenly, but rises linearly up to a value p0 within a time t1 , the rise
time, and thereafter remains constant, as shown in Fig. 3.4.5. Mathematically it
is defined by
(p
0
t t t1
pðt Þ ¼ t1 (3.4.16)
p0 t > t1
FIG. 3.4.5 Step function load with finite rise time.
For zero initial conditions, u0 ¼ u_ 0 ¼ 0, the response of the system is

obtained by studying the motion in two phases:
@CivilMethod
Phase Ι: Forced vibration under a ramp function load.

This phase starts at t ¼ 0 and ends at t ¼ t1 . The response of the undamped sys-
tem is given by Eq. (3.4.15)

p0 t sin wt
u I ðt Þ ¼ , 0 < t t1 (3.4.17)
k t1 wt1
Phase ΙI: Forced vibration under a step function load.

This phase starts at t ¼ t1 with initial conditions uII ð0Þ ¼ uI ðt1 Þ, u_ II ð0Þ ¼ u_ I ðt1 Þ
and loading pðt Þ ¼ p0 . The solution is given by Eq. (3.3.14). The value of the
Duhamel integral is given by Eq. (3.4.3). Thus, we have
u_ I ðt1 Þ p0
uII ðt Þ ¼ sin wt+ uI ðt1 Þcos wt+ ð1 cos wtÞ (3.4.18)
w k
where t¼ t t1 > 0.

Note that for t1 ¼ nT , n ¼ 1, 2, …, Eq. (3.4.17) gives
p0
uI ðt1 Þ ¼ , u_ I ðt1 Þ ¼ 0 (3.4.19)
k
and substituting into Eq. (3.3.18) gives
p0
uII ðt Þ ¼ (3.4.20)
k
or
RII ðt Þ ¼ 1 (3.4.21)
which means that the motion in the constant load phase is not oscillatory but the
displacement is constant and equal to the static displacement.
Fig. 3.4.6 presents the response ratio Rðt Þ for different values of t1 .
We observe that for smaller values of t1 =T , the response is similar to that
of a step function load while for larger values, the response is similar to that of
a static load. Therefore, the loads in real structures should not be applied
suddenly but slowly rising, that is, in time much larger than the natural
period of the structure to avoid dynamic magnification effects. Due to this
property, the step function load with finite rise time is also called static
loading.
@CivilMethod
FIG. 3.4.6 Response ratio of a system under a step function load with finite rise time.
Example 3.4.1 Blast load on one-story building

A blast-induced pressure wave strikes the one-story building in Example 2.2.1
in the x direction. The time variation of the blast pressure is represented by
pðt Þ ¼ p0 ð1 t=t0 Þet=t0 (Fig. E3.1). Determine the stress resultants max Qx ,
max Mx of the columns. The peak positive pressure is p0 ¼ 12kN=m2 and
the duration of the positive phase t0 ¼ 1. The system is at rest at t ¼ 0.
FIG. E3.1 Blast-induced pressure in Example 3.4.1.
Solution
The mass of the structure, its stiffness in the x direction, and the corresponding
natural frequency were evaluated in Example 2.2.1
m ¼ 356:78kN m1 s2 , k ¼ Kx ¼ 7257:6kN=m, w ¼ w1 ¼ 4:51
@CivilMethod
If A represents the area of the surface struck by the blast-induced pressure,

the peak force applied at the floor level of the building is
1 1
P ¼ A p0 ¼ 5 5 12 ¼ 150kN=m2 (1)
2 2
Hence
ust ¼ P=k ¼ 150=7257:6 ¼ 0:02067m (2)
Because the system is at rest at t ¼ 0, the response is given by Eq. (3.3.14)
with u0 ¼ u_ 0 ¼ 0
Z t
P
u ðt Þ ¼ ð1 τÞexp ðτÞsin ½wðt τÞdτ (3)
mw 0
The analytical evaluation of the Duhamel gives
(
P 2w3 w2 ð w2 1 Þ
u ðt Þ ¼ sin wt coswt
k ðw2 + 1Þ2 ð w2 + 1Þ 2
)
w2 ð w2 1 Þ w2
+ exp ð t Þ t exp ðt Þ (4)
ð w 2 + 1Þ 2 ð w 2 + 1Þ
1.5
D = 1.265
0.5
R(t)
tmax = 0.563
–0.5
–1
–1.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t
FIG. E3.2 Response ratio in Example 3.4.1.
Fig. E3.2 shows the plot of the response ratio Rðt Þ ¼ u ðt Þ=ðP=k Þ. The DMF
was found at D ¼ max jRðt Þj ¼ 1:1265. It was determined as the maximum
value of the array used to plot Rðt Þ. Thus, we have
umax ¼ Du st ¼ 1:1265 0:02067 ¼ 0:02328m (5)
@CivilMethod
The extreme values of the shear forces and bending moments are:
Columns 30 30
max Qx ¼ kx umax ¼ 1360:8 0:02328 ¼ 31:68kN
h
max Mx ¼ max Qx ¼ 31:68 5=2 ¼ 79:20kNm
2
Columns 30 20
max Qx ¼ kx umax ¼ 907:2 0:02328 ¼ 21:12kN
h
max Mx ¼ max Qx ¼ 21:12 5=2 ¼ 52:80kNm
2
3.5 Response to impulsive loads

In this section, a special group of dynamic loads with similar characteristics will
be considered: the impulsive loads or shock loads. The main characteristics of
such loads are their short duration and their high intensity (large magnitude).
Their duration is of the order of the natural period of the structure. Impulsive
loads are of great importance in the design of certain structural systems, for exam-
ple, buildings subject to aboveground wind blasts or explosions. Because of the
short duration, the maximum response is reached before damping starts to absorb
much energy. As a result, the maximum response is not significantly affected by
the presence of damping. Therefore, we can ignore damping in the study of the
SDOF systems under impulsive loads in the analysis that follows. The time func-
tion representing the impulsive load is arbitrary, e.g., Fig. 3.5.1. Nevertheless,
important conclusions can be drawn by considering impulsive loads with a reg-
ular shape. The response of a system to an impulsive load is studied in two phases.
The first phase is a forced vibration under the impulsive load with zero initial
conditions while the second phase is a free vibration with initial conditions being
the displacement and velocity that the system attains at the end of the first phase.
p(t)
t
@CivilMethod
FIG. 3.5.1 Time variation of an impulsive load.
The response of the system to an impulsive load can be analyzed using the
methods presented in previous sections for the solution of the differential equa-
tion of motion under an arbitrary loading, that is, either by solving directly the
differential equation or by evaluating Duhamel’s integral. Another method to
obtain the response is to express the pulse as the superposition of two or more
simpler pulses for which the response solution is available or simple to deter-
mine (Fig. 3.5.2). Nevertheless, the analytical methods, especially for arbitrary
impulse loads, have lost their importance because of the development of effi-
cient numerical methods. For this reason, only two simple impulsive loads
are considered, the rectangular pulse load and the triangular pulse load.
p(t )
p(t ) p 0H (t ) if t t1
p0 p(t ) 0 if t t1
O t
t1
p(t )
p0 p(t ) p0 sin t if t t1
p(t ) 0 if t t1
O t
t1
FIG. 3.5.2 Impulsive loads represented by simple functions.
3.5.1 Rectangular pulse load

The rectangular pulse load is shown in Fig. 3.5.3. Mathematically, this load is
defined as
p0 t t1
pðt Þ ¼ (3.5.1)
0 t > t1
FIG. 3.5.3 Rectangular pulse load.
@CivilMethod
Phase Ι: Forced vibration

The response of the system is a forced vibration due to the suddenly applied load
p0 at t ¼ 0 with zero initial conditions, u0 ¼ u_ 0 ¼ 0. The displacement is given
by Eq. (3.4.3), namely
p0
uI ðt Þ ¼ ð1 cos wt Þ, 0 < t t1 (3.5.2)
k
Phase ΙΙ: Free vibration

The displacement is given by Eq. (2.2.13), namely
u_ I ðt1 Þ
uII ðt Þ ¼ sin wt+ uI ðt1 Þcos wt, t¼ t t1 0 (3.5.3)
w
where
p0
uI ðt1 Þ ¼ ð1 cos wt1 Þ (3.5.4a)
k
p0 w
u_ I ðt1 Þ ¼ sin wt1 (3.5.4b)
k
We now evaluate the DMF D. First, we assume that the maximum displace-
ment occurs in Phase I. Eq. (3.5.2) has a maximum when
p0 w
u_ I ðt Þ ¼ sin wt ¼ 0 or wt ¼ np n ¼ 1, 2, …
k
np n 2p nT
t¼
¼ ¼ (3.5.5)
w 2w 2
Therefore, the displacement can reach its maximum umax within Phase I
if t1 T =2. In this case, it is D ¼ 2.
We assume now that t1 < T =2, then the maximum displacement umax
occurs in Phase II. Hence, it will be
FIG. 3.5.4 Response spectrum for rectangular pulse load.
@CivilMethod
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u_ I ðt1 Þ 2
umax ¼ + ½ u I ðt1 Þ 2 (3.5.6)
w
Using Eqs. (3.5.4a), (3.5.4b) the previous relation becomes
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

p0 2p 2p 2p
umax ¼ 1 2cos t1 + cos 2 t1 + sin 2 t1
k T T T
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ffi
p0 2p
¼ 2 1 cos t1
k T
or

2p0 t1
umax ¼ sin p (3.5.7)
k T
and the DMF D will be given by

umax t1 T
D¼ ¼ 2sin p , t1 (3.5.8)
p0 =k T 2
Eq. (3.5.8) shows that the maximum response depends only on the ratio
t1 =T . The plot of the function D ¼ D ðt1 =T Þ shown in Fig. 3.5.4 is referred
to as the displacement response spectrum or simply the response spectrum of
the impulsive load (see also Chapter 6). It is evident that the response spectrum
of a pulse load serves to determine the maximum response of the system under
this load without solving the differential equation of motion for the particular
pulse load.
3.5.2 Triangular pulse load

The next impulsive load is the decreasing triangular pulse load shown in
Fig. 3.5.5a.
Mathematically, this load is defined as
8

< t
p0 1 t t1
pðt Þ ¼ t1 (3.5.9)
:
0 t > t1
(a) (b) (c)

FIG. 3.5.5 Decomposition of a triangular pulse load.
@CivilMethod
The response of the system is studied again in two phases:

Eq. (3.5.9) suggests that the pulse load can be considered as the superposition of
a rectangular pulse and a ramp function pulse (Fig. 3.5.5b and c) and use the
solutions we already found for the respective load functions, that is,
Eqs. (3.4.3), (3.4.15).
Thus, we obtain

p0 sin wt t
uI ðt Þ ¼ 1 cos wt + , 0 t t1 (3.5.10)
k t1 w t1
Phase ΙΙ: Free vibration

The displacement is given by Eq. (2.2.13)
u_ I ðt1 Þ
uII ðt Þ ¼ sin wt+ uI ðt1 Þcos wt, t¼ t t1 0 (3.5.11)
w
where

p0 sin wt1
u I ðt1 Þ ¼ coswt1 (3.5.12a)
k t1 w

p0 w cos wt1 1
u_ I ðt1 Þ ¼ + sin wt1 (3.5.12b)
k wt1 wt1
We now evaluate the DMF D.

The maximum displacement in Phase I is obtained when

cos wt 1
u_ I ðt Þ ¼ w sin wt + ¼0 (3.5.13)
t1 t1
which may be written as

t1 t t
2p sin 2p + cos 2p 1¼0 (3.5.14)
T T T
When t1 =T is specified, Eq. (3.5.14) is a nonlinear algebraic equation for the
ratio t=T . Solving this equation and introducing the obtained solution in
Eq. (3.5.10) yield the respective maximum displacement max uI . The curve
RI in Fig. 3.5.6 represents the function max uI =ust versus t1 =T .
@CivilMethod
In Phase II, the maximum response is obtained from the expression

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

u_ I ðt1 Þ 2
max uII ¼ + ½uI ðt1 Þ2 (3.5.15)
w
where uI ðt1 Þ and u_ I ðt1 Þ are given by Eqs. (3.5.12a), (3.5.12b), which yield
when expressed in terms of t1 =T
2
3
t1
sin 2p

p0 6 T t1 7
uI ðt1 Þ ¼ 6 4 cos 2p 7 (3.5.16a)
k t1 T 5
2p
T
2
3
t1
cos 2p

p0 w 6
6 T t1 1 7 7
u_ I ðt1 Þ ¼ + sin 2p (3.5.16b)
k 4 t1 T t1 5
2p 2p
T T
The curve RII in Fig. 3.5.6 represents the function max uII =ust versus t1 =T .
From this figure, we conclude that for t1 =T 0:4 the maximum response D of
the system to the triangular pulse load occurs in Phase II while for t1 =T > 0:4 it
occurs in Phase I.
R = maxu /ust
D(0.4)=1.0513
D(t1/T)
R = maxu /ust
t1/T
FIG. 3.5.6 DMF for the triangular pulse load.
@CivilMethod
3.5.3 Asymmetrical triangular pulse load

The next impulsive load is the asymmetrical triangular pulse load shown in
Fig. 3.5.7. The response can be determined in three phases using the solutions
obtained in Sections 3.4.2 and 3.5.2.

p0 sin wt
uI ðt Þ ¼ t , 0 t t1 (3.5.17)
k t1 w
Phase II: Forced vibration
u_ I ðt1 Þ
uII ðt Þ ¼ sin wet + uI ðt1 Þ cos wet
w

(3.5.18)
p0 e sin wet et
+ 1 cos wt + , 0 et ¼ t t1 < t2
k t2 w t2
Phase ΙΙΙ: Free vibration
u_ II ðt1 + t2 Þ
uIII ðt Þ ¼ sin wt+ uII ðt1 + t2 Þcos wt, t¼ t ðt1 + t2 Þ 0
w
(3.5.19)
FIG. 3.5.7 Asymmetrical triangular pulse load.
@CivilMethod
Table 3.5.1 summarizes the response ratios for various loadings.
TABLE 3.5.1 Response ratios Rðt Þ for various loading functions.

Loading pðt Þ Response ratio Rðt Þ
Rðt Þ ¼ 1 cos wt
Rðt Þ ¼ 1 cos wt, t t1

Rðt Þ ¼ coswðt t1 Þ cos wt, t t1

1 sin wt
Rðt Þ ¼ t , t t1
t1 w
1
Rðt Þ ¼ 1 + ½ sinwðt t1 Þ sin wt , t t1
wt1
sin wt t
Rðt Þ ¼ 1 cos wt + , t t1
wt1 t1
sinwt1 sinwðt t1 Þ
Rðt Þ ¼ cos wt, t t1
wt1

1 sin wt t1
Rðt Þ ¼ t , 0t
t1 w 2

2 1 t1 t1
Rðt Þ ¼ t1 t + 2sin w t sin wt , t t1
t1 w 2 2

2 t1
Rðt Þ ¼ sinwðt t1 Þ + 2sinw t sinwt , t t1
wt1 2
b sinwt
sin wt
Rðt Þ ¼
1 b2
Example 3.5.1 One-story building under an impulsive moment

The rigid slab of the one-story building of Fig. E3.3a,b is loaded by the impul-
sive moment M ðt Þ, whose time variation is shown in Fig. E3.3c. The columns
are assumed massless and fixed on the ground. The pulse data are:
M0 ¼ 300kNm, M1 ¼ 500kNm, t1 ¼ T , where T is the period of the structure.
The other data are the same as in Example 2.2.1.
@CivilMethod
y
k1 k2 k1
M (t ) x
O 5m
k1 k2 k1
10 m
(a) (b)
(c)
FIG. E3.3 One-story building under impulsive load.
Determine the stress resultants max Qx , max Qy , max Mx , max My , and

max Mf of the corner columns using time history analysis.
Solution
Because the structure is symmetric with respect to both axes x and y, the only
possible motion of the slab due to the moment M ðt Þ is the rotation fðt Þ about its
center O. Hence the equation of motion of the slab is
IO f€ðt Þ + Kf fðt Þ ¼ M ðt Þ (1)
where IO is the moment of inertia of the mass of the plate with respect to O and
Kf the torsional stiffness of the structure. These quantities have been computed
in Example 2.2.1. Thus we have
IO ¼ 3:7164 103 kNm s2 (2)
Kf ¼ 1:9663 105 kNm (3)
rffiffiffiffiffiffi
Kf
w¼ ¼ 7:2738 (4)
IO
2p
T¼ ¼ 0:869s (5)
w
t1 ¼ T ¼ 0:869s (6)
@CivilMethod
The motion is studied in two phases.

Phase Ι: 0 t t1 . Forced vibration
The solution is given by the Duhamel integral
Z t
1
fI ðt Þ ¼ M ðτÞsin wðt τÞdτ (7)
IO w 0
M1 M0
M ðt Þ ¼ M0 + t (8)
t1
we have
Z t Z
1 M1 M0 t
fI ðt Þ ¼ M0 sin wðt τÞdτ + τ sin wðt τÞdτ (9)
IO w 0 t1 0
Apparently, the response results as the superposition of a rectangular pulse

and a ramp function pulse. Hence, using Eqs. (3.4.2), (3.4.15), we obtain

1 M0 M1 M0 sin wt
fI ðt Þ ¼ ð1 cos wt Þ + t (10)
IO w w wt1 w
or

M0 M1 M0 sin wt
f I ðt Þ ¼ ð1 cos wt Þ + t (11)
Kf Kf t1 w
which for the adopted data becomes

sin wt
fI ðt Þ ¼ 1:525 103 ð1 cos wt Þ + 1:177 103 t (12)
w
Phase ΙI: t1 < t. Free vibration

The solution is
f_ I ðt1 Þ
fII ðt Þ ¼ sin wðt t1 Þ + fI ðt1 Þcos wðt t1 Þ (13)
w
Eq. (12) for t ¼ t1 gives
fI ðt1 Þ ¼ 1:017 103 and f_ I ðt1 Þ ¼ 0 (14)

and Eq. (13) becomes
fII ¼ 1:017 103 cos wðt t1 Þ (15)
@CivilMethod
The response of the structure is shown in Fig. E3.4.
–3
4 ×10
fmax = 3.594e-3
3
2
f(t)
0
tmax = 0.462
–1 t1
–2
0 0.5 1 1.5 2 2.5 3 3.5 4
t
FIG. E3.4 Response of the structure in Example 3.5.1.
The stress resultants of the columns are computed using Eqs. (4a), 4(b), (6)
of Example 2.2.1. Thus, for the upper right column we have
kx1 ¼ ky1 ¼ 1360:8kN=m, kf1 ¼ 1:9985 103 kNm
x1 ¼ 4:85m, y1 ¼ 2:35m
max Qx1 ¼ kx1 y1 fmax ¼ 11:49kN (16a)
max Qy1 ¼ ky1 x1 fmax ¼ 23:72kN (16b)
h
max Mx1 ¼ max Qx1 ¼ 28:73kNm (16c)
2
h
max My1 ¼ max Qy1 ¼ 59:30kNm (16d)
2
max Mf1 ¼ kf1 fmax ¼ 7:183kNm (16e)
3.5.4 Response to piecewise linear loading

Certain types of loading can be represented by straight-line segments within
generally unequal time intervals, as shown in Fig. 3.5.8. The accelerogram, that
is, the employed recording of the acceleration of the ground motion during an
earthquake, is a representative example of such loading.
@CivilMethod
FIG. 3.5.8 Piecewise linear loading.
The loading function in the i interval is given by the expression

pðt Þ ¼ ai + bi t, ti1 t ti , i ¼ 1, 2, …, n (3.5.20)
where
pi pi1
ai ¼ pi1 , bi ¼ , to ¼ 0 (3.5.21)
ti ti1
The response is obtained in n phases, where n is the number of intervals.
The solution in the interval ti1 t ti is given by Eq. (3.3.23). Hence
we have

u_ ðti1 Þ + u ðti1 Þxw
ui ðtÞ ¼ sin wD t+ u ðti1 Þcos wD t exwt
wD
Z t (3.5.22)
1 xwðtτÞ
+ ðai + bi τÞe sin ½wD ðt τÞdτ
mwD 0
where t¼ t ti1 , 0 t ti ti1 .
The Duhamel integral can be evaluated as a sum of two integrals: one due to
the constant term ai of the loading and the other due to the linear term. The inte-
ð1Þ ð2Þ
grals are denoted by Di ðt Þ and Di ðt Þ, respectively.
The first integral is obtained from Eq. (3.4.7) for p0 ¼ ai
" ! #
ð1Þ ai x xwt
Di ðtÞ ¼ 1
cos wD t + pffiffiffiffiffiffiffiffiffiffiffiffi sin wD t e (3.5.23a)
k 1 x2
The analytical evaluation of the second integral is rather complicated.
Therefore, the recourse to a symbolic language is inevitable. Thus, using
MAPLE, we obtain
(" # )
ð2Þ bi 2x2 1 2x 2x
Di ð t Þ ¼ sin wD t+ cos wD t exwt + t (3.5.23b)
k wD w w
Substituting Eqs. (3.5.23a), (3.5.23b) into Eq. (3.5.22) gives

ui ðtÞ ¼ Ai0 + Ai1 t+ exwt Ai2 cos wD t+ Ai3 sin wD t (3.5.24)
@CivilMethod
in which

1 xbi
Ai0 ¼ ai 2
k w
bi
Ai1 ¼ , Ai2 ¼ u ðti1 Þ Ai0 (3.5.25)
k
1
A3 ¼
i
u_ ðti1 Þ + xwAi2 Ai1
wD
Differentiating Eq. (3.5.24) with respect to time gives the velocity

u_ i ðt Þ ¼ Ai1 + exwt xwAi2 + wD Ai3 cos wD t xwAi3 + wD Ai2 sin wD t
(3.5.26)
The presented method is exact. However, numerical methods are more con-
venient to compute the response to a piecewise linear loading (see Chapter 4).
3.6 Response to a periodic loading

3.6.1 Periodic loads
A periodic load is one whose time variation profile repeats continually at reg-
ular intervals T . It can be represented by a periodic function
pðt Þ ¼ pðt nT Þ, n ¼ 0,1, 2, … (3.6.1)
The smallest constant T that satisfies Eq. (3.6.1) is called the period of the
periodic function. Fig. 3.6.1 shows examples of periodic functions. Many load-
ings in nature are periodic or can be approximated by periodic loads.
(a)
(b)
(c)
FIG. 3.6.1 Examples of periodic loads.
@CivilMethod
3.6.2 Fourier series

The periodic function can be represented by the Fourier series, that is, a
trigonometric series of the form [5]
X
1
pðt Þ ¼ a0 + ðan cos nw0 t + bn sin nw0 t Þ (3.6.2)
n¼1
where an ,bn are constant coefficients to be determined and w0 ¼ T =2p is the

fundamental frequency of the periodic function. The coefficients an , bn ,
n ¼ 0, 1, 2, … are known as the Fourier series coefficients. They can be deter-
mined using the orthogonality property of the sine and cosine functions
presented below.
In general, a set of functions F : ffn ðt Þg is called orthogonal in a interval
½t1 , t2 , if for any two functions fm , fn F, m, n N holds
Z t2
0 if m 6¼ n
fm ðt Þfn ðt Þdt ¼ (3.6.3)
t1 cn if m ¼ n
where cn is a constant.
The set F of functions fn is complete if no other function outside F exists
that satisfies the orthogonality condition (3.6.3).
It can be readily shown that the set
F ¼ f1, cos w0 t, cos 2w0 t, cos 3w0 t, …sin w0 t, sin 2w0 t, sin 3w0 t, …g (3.6.4)
is orthogonal in ½T =2, T =2 and complete. Indeed, have
Z T =2
1cos mw0 tdt ¼ 0 for each m (3.6.5a)
T =2
Z T =2
1sin mw0 tdt ¼ 0 for each m (3.6.5b)
T =2
Z T =2
0 if m 6¼ n
cos mw0 t cos nw0 tdt ¼ (3.6.5c)
T =2 T=2 if m ¼ n
Z T =2
cos mw0 t sin nw0 tdt ¼ 0 for each m, n (3.6.5d)
T =2
Z T =2
0 if m 6¼ n
sin mw0 t sin nw0 tdt ¼ (3.6.5e)
T =2 T =2 if m ¼ n
The coefficients an are evaluated by multiplying Eq. (3.6.2) by cos mw0 t,

integrating over the interval ½T =2, T =2, and using the orthogonality relations
(3.6.5a)–(3.6.5e). Thus, we obtain
@CivilMethod
Z T =2
1
a0 ¼ pðt Þdt (3.6.6a)
T T =2
Z T =2
2
an ¼ pðt Þcos nw0 tdt (3.6.6b)
T T =2
Similarly, the coefficients bn are evaluated by multiplying Eq. (3.6.2) by

sin mw0 t, integrating over the interval ½T =2, T =2, and using the orthogonal-
ity relations (3.6.5a)–(3.6.5e). Thus, we obtain
Z T =2
2
bn ¼ pðt Þ sin nw0 tdt (3.6.6c)
T T =2
The series (3.6.2) represents the function pðt Þ, that is, converges to pðt Þ for
n ! 1, provided that it satisfies the following conditions, known as Dirichlet
conditions:
(a) The function pðt Þ has a finite number of discontinuities in one period.
(b) The function pðt Þ has a finite number of maxima and minima in one period.
(c) The function pðt Þ is absolutely integrable over a period, that is,
Z T =2
jpðt Þjdt ¼ k < 1 (3.6.7)
T =2
We shall say that that the function pðt Þ is piecewise continuous in the finite
interval ½T =2, T =2, if it satisfies conditions (a) and (b). At the points of dis-
continuity, for example, point t1 in Fig. 3.6.1c, the Fourier series converges to
the mean value
1
p t1 + p t1+ (3.6.8)
2

where p t1 and p t1+ are the left and right limits of pðt Þ at t1 .
In practice, the periodic function pðt Þ is approximated by a finite number of
terms of the Fourier series, that is, by a finite Fourier series.
Let
X
k
Sk ð t Þ ¼ a 0 + ðan cos nw0 t + bn sin nw0 t Þ (3.6.9)
n¼1
be the sum of the first k + 1 terms of the Fourier series, which will represent the
function pðt Þ in the interval ½T=2, T =2. Then we will have
X
k
pðt Þ ¼ a0 + ðan cosnw0 t + bn sin nw0 t Þ + ek ðt Þ (3.6.10)
n¼1
@CivilMethod
where
ek ðt Þ ¼ pðt Þ Sk ðt Þ (3.6.11)
is the error between pðt Þ and its approximation. The mean square error Ek is
given by
Z
1 T =2
Ek ¼ ½ek ðt Þ2 dt
T T =2
Z " #2 (3.6.12)
1 T =2 Xk
¼ pðt Þ a0 ðan cos nw0 t + bn sin nw0 t Þ dt
T T =2 n¼1
The error Ek is a function of a0 , an and bn . Therefore, it is minimized when

∂Ek ∂Ek ∂Ek
¼ 0, ¼ 0, ¼ 0, ðn ¼ 1, 2, …, k Þ
∂a0 ∂an ∂bn
Differentiating Eq. (3.6.12) and interchanging differentiation with integra-
tion give
Z " #
∂Ek 2 T=2 Xk
¼ pðt Þ a0 ðan cos nw0 t + bn sin nw0 t Þ dt (3.6.13a)
∂a0 T T=2 n¼1
Z " #
∂Ek 2 T =2 Xk
¼ pðt Þ a0 ðan cos nw0 t + bn sin nw0 t Þ cos nw0 tdt
∂an T T =2 n¼1
(3.6.13b)
Z " #
∂Ek 2 T =2 X
k
¼ pðt Þ a0 ðan cos nw0 t + bn sin nw0 t Þ sin nw0 tdt
∂bn T T =2 n¼1
(3.6.13c)
Using the orthogonality relations (3.6.5a)–(3.6.5e), the integrals (3.6.13a)–
(3.6.13c) become
Z
∂Ek 1 T =2
¼ a0 pðt Þdt ¼ 0 (3.6.14a)
∂a0 T T =2
Z
∂Ek 2 T =2
¼ an pðt Þcos nw0 tdt ¼ 0 (3.6.14b)
∂an T T =2
Z
∂Ek 2 T =2
¼ bn pðt Þ sin nw0 tdt ¼ 0 (3.6.14c)
∂bn T T =2
We observe that the values of the coefficients a0 , an , bn obtained by
Eqs. (3.6.14a)–(3.6.14c) are identical to those obtained by Eqs. (3.6.6a)–
(3.6.6c). Therefore, we may conclude that if a function is approximated by a
finite Fourier series, the mean square error is minimized.
@CivilMethod
Moreover, when a function is approximated by a finite Fourier series, the

error in the discontinuity region is considerable, even if the number of terms
of the series is very large. This is known as the Gibbs phenomenon. A technique
to overcome this problem is to adapt a suitable function at the region of discon-
tinuity that restores continuity (Fig. 3.6.2a). Similarly, a concentrated load can
be replaced with a bell-shaped function (Fig. 3.6.2b). This, of course, can be
done if the physical problem under consideration is not altered by this substi-
tution. For example, the functions

p t1 + p t1+ p t1 p t1+ pðt t 1 Þ
pðt Þ ¼ sin , jt t1 j e (3.6.15a)
2 2 2e

P pðt t1 Þ
pðt Þ ¼ 1 + cos (3.6.15b)
2e 2e
where e is a small number can play this role.
(a) (b)
FIG. 3.6.2 Functions pðt Þ replacing pðt Þ in the interval ½t1 e, t1 + e.
Example 3.6.1 Fourier series expansion of a periodic function. Gibbs

phenomenon
Expand the periodic function of Fig. 3.6.1b in Fourier series if t1 ¼ T =2.
Solution
The function pðt Þ within a period is defined as

2p0 t=T if 0 t < T =2
pðt Þ ¼ (1)
0 if T=2 < t T
We readily prove that the Dirichlet conditions are satisfied. Hence, the func-
tion can be expanded in Fourier series. Eqs. (3.6.6a)–(3.6.6c) give
Z
1 T 1
a0 ¼ pðt Þdt ¼ p0 (2a)
T 0 4
Z
2 T cos np 1
an ¼ pðt Þcos nw0 tdt ¼ p0 (2b)
T 0 n 2 p2
@CivilMethod
Z T
2 cos np
bn ¼ pðt Þ sin nw0 t ¼ p0 (2c)
T 0 np
Hence
p0 p0 X
N
cos np 1 p0 X
N
cos np
pðt Þ ¼ + 2 cos nw 0 t + sin nw0 t (3)
4 p n¼1 n2 p n¼1 n
Fig. E3.5 shows the graphical representation of the finite Fourier series in
Eq. (3) for various values of N with p0 ¼ 1, T ¼ 1. We observe that the conver-
gence is very slow at point t ¼ T =2 due to the Gibbs phenomenon.
N = 20 N = 100
1.2 1.2
1 1
0.8 0.8
0.6 0.6
p(t)
p(t)
0.4 0.4
0.2 0.2
0 0
–0.2 –0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t t
N = 1000 N = 2000
1.2
1 1
0.8 0.8
0.6 0.6
p(t)
p(t)
0.4 0.4
0.2 0.2
0 0
–0.2 –0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t t
FIG. E3.5 Gibbs’s phenomenon in Example 3.6.1.
Example 3.6.2 Fourier series expansion of a periodic load

Expand the periodic load of Fig. E3.6 in Fourier series if t1 ¼ T =3.
FIG. E3.6 Periodic load in Example 3.6.2.
@CivilMethod
Solution
The function pðt Þ within a period T ¼ 3t1 can be represented using the delta
function.a Thus we obtain
pðt Þ ¼ P ½d ðt T =3Þ d ðt 2T =3Þ (1)
Eqs. (3.6.6a)–(3.6.6c) give
Z
1 T
a0 ¼ pðt Þdt ¼ 0 (2a)
T 0
Z

2 T 2P 2np 4np
an ¼ pðt Þcos nw0 tdt ¼ cos cos (2b)
T 0 T 3 3
Z

2 T 2P 2np 4np
bn ¼ pðt Þ sin nw0 t ¼ sin sin (2c)
T 0 T 3 3
Inserting these values of the coefficients in Eq. (3.6.2) we obtain
(
1

2P X 2np 4np
pðt Þ ¼ cos cos cos nw0 t
T n¼1 3 3

)
2np 4np
+ sin sin sin nw0 t (3)
3 3
3.6.3 Response of the SDOF system to periodic excitation

As was shown in the previous section, any periodic function can be expressed as
a sum of harmonic functions. Theoretically, an infinite number of harmonic
terms are required to represent accurately the function as a Fourier series. How-
ever, the contribution of the terms decreases with increasing n. Consequently, a
few terms are adequate to approximate the loading with acceptable accuracy.
For periodic loading, each term of the series represents harmonic excitation.
Thus, the total response of an SDOF system to a periodic loading can be
obtained as the sum of the responses of the individual harmonic terms, provided
that the system is linear, which allows the application of the superposition prin-
ciple. In subsequent analysis, we assume that the periodic excitation acts for a
sufficiently long time, by which the transient response due to the initial condi-
tions has decayed. Thus, only the steady-state response will be considered.
The steady-state response to the constant load pðt Þ ¼ p0 can be obtained
from Eq. (3.2.25a) by setting p0 ¼ a0 , x ¼ 0, w ¼ 0 and neglecting the transient
response. This yields
ao
u 0 ðt Þ ¼ (3.6.16a)
k
a. The delta function d ðt t0 Þ and its properties are discussed in Section 3.7.1.
@CivilMethod
For the sine term bn sin nw0 t the steady-state response results from
Eq. (3.2.25a) by setting p0 ¼ bn , w ¼ nw0 , and b ¼ b n ð¼ nw0 =wÞ
bn 1
unsin ðt Þ ¼ 2 1 b2n sin nw0 t 2xbn cos nw0 t
k 1b 2
+ ð2xbn Þ 2
n
(3.6.16b)
while for the cosine term an cos nw0 t the steady-state response is obtained from
Eq. (3.2.25b) by setting p0 ¼ an , w ¼ nw0 , and b ¼ b n ð¼ nw0 =wÞ
an 1
uncos ðt Þ ¼ 2xbn sin nw0 t + 1 b 2n cos nw0 t
k 1 b 2 2 + ð2xb Þ2
n n
(3.6.16c)
The steady-state response of the damped system to periodic loading results
as the superposition of responses to individual terms of the Fourier series
X
n X
n
u ðt Þ ¼ u0 ðt Þ + unsin ðt Þ + uncos ðt Þ (3.6.17)
n¼1 n¼1
Inserting Eqs. (3.6.16a)–(3.6.16c) into Eq. (3.6.17) gives

ao X 1
an 1
u ðt Þ ¼ + 2 2xbn sin nw0 t + 1 b2n cos nw0 t
k n¼1
k 1b 2
+ ð2xb n Þ 2
n
X
1
bn 1
+ 1 b2n sin nw0 t 2xbn cosnw0 t
k 1b 2 2
n¼1 n + ð2xb n Þ2
(3.6.18)
or
ao X 1
1 1
u ðt Þ ¼ + 2
an ð2xbn Þ + bn 1 b2n sin nw0 t
k n¼1
k 1 b 2 + ð2xb Þ2
n n

+ an 1 b n bn ð2xbn Þ cos nw0 t
2
(3.6.19)
Theoretically, there is no transient response when x ¼ 0. However, for
small values of damping, which is the usual case in our structures, the
steady-state response can be obtained from Eqs. (3.6.16b), (3.6.16c) for
x 0. This yields
bn 1
unsin ðt Þ sin nw0 t (3.6.20a)
k 1 b2n
an 1
uncos ðt Þ cos nw0 t (3.6.20b)
k 1 b 2n
@CivilMethod
which are actually approximations of the particular solutions (3.2.7) and

(3.2.18) of the undamped motion due to harmonic loading.
On the basis of Eqs. (3.6.20a), (3.6.20b), (3.6.16a), the approximate steady-
state response becomes
ao X 1
an 1 X1
bn 1
u ðt Þ + cos nw0 t + sin nw0 t (3.6.21)
k n¼1
k 1 b 2
n n¼1
k 1 b 2n
The response is a periodic function with a period T ¼ 2p=w0 . The contribu-

tion of each harmonic term in the series (3.6.19) depends on: (i) the amplitudes
an and bn of the harmonic components of the load, and (ii) the value of b n . The
terms with b n near 1 have the greatest influence on the response. It should be
noted that in the case bn ¼ 1, x ¼ 0 is meaningless.
Example 3.6.3 Response of an SDOF system subjected to periodic loading
Determine the response of a SDOF system subjected to the periodic loading of
Fig. 3.6.1b. Adopt t1 ¼ T =2, x ¼ 0, T ¼ 1, b ¼ 1:25.
Solution
The function pðt Þ within a period is defined as

2p0 t if 0 < t < 1=2
pðt Þ ¼ (1)
0 if 1=2 < t 1
The Fourier series coefficients are obtained using Eqs. (3.6.6a)–(3.6.6c).
They have already been evaluated in Example 3.6.1. Then the response to
the periodic loading is obtained from Eq. (3.6.19). Fig. E3.7 shows the response
ratio Rðt Þ ¼ u ðt Þ=ðp0 =k Þ (a) for different numbers of series terms and (b) for
n ¼ 50 and various values of damping ratio x. We observe that a small number
of terms gives an acceptable accuracy of the response while the response for
small values of x is very close to that with x ¼ 0, which justifies the
expression (3.6.21).
(a) (b)
FIG. E3.7 Response to periodic loading in Example 3.6.3.
@CivilMethod
3.7 Response to unit impulse

3.7.1 The delta function or Dirac’s delta function
In problems of mechanics, we often come across concentrated loads, that is,
actions that are applied to a very small region, theoretically at a point in space
or an instant in time. In this section, we present a mathematical tool, the delta
function or Dirac’s delta function, that allows us to handle impulsive loads in a
simple way and understand better the dynamic response of SDOF systems to
such loads and in general to arbitrary dynamic loads.
Let P ¼ 1 be a unit force that acts on the SDOF system at time t ¼ 0. First,
we consider that the force is not imposed instantaneously but it is distributed
over a small time interval ½e, +e and let pðt Þ denote its distribution. This dis-
tribution, which most probably has the shape shown in Fig. 3.7.1, is not known.
Nevertheless, we know that it is sufficiently “concentrated” about t ¼ 0 and that
Z +1
pðt Þdt ¼ 1 (3.7.1)
1
FIG. 3.7.1 Distribution of a concentrated impulsive force over the interval ½e, +e.
namely, the total force is equal to unity. If we skip the problem of determining
analytically the function pðt Þ, we may assume a priori a prescribed shape for
this function, for example,

k=2, jt j < 1=k
pk ðt Þ ¼ (3.7.2a)
0, jt j 1=k
or
k
pk ðt Þ ¼ (3.7.2b)
pð1 + k 2 t 2 Þ
where k is a positive number.
@CivilMethod
Fig. 3.7.2a and b show both functions pk defined in Eqs. (3.7.2a), (3.7.2b),
respectively. Moreover, they satisfy Eq. (3.7.1), which means that they are
equivalent to pðt Þ and can represent the actual force P.
(a) (b)
FIG. 3.7.2 Functions pk sufficiently concentrated for large values of k.
The distribution of pðt Þ becomes more “concentrated” as the value of k in

Eqs. (3.7.2a), (3.7.2b) increases. The limiting case for k ! 1 yields a fictitious
distribution per unit time, which is denoted by d ðt Þ and defined by
d ðt Þ ¼ ‘im pk ðt Þ (3.7.3)
k!1
The function d ðt Þ is known as the delta function or Dirac’s delta function.

When the concentrated force acts at instant t ¼ t1 , the delta function is denoted
by dðt t1 Þ. In mathematics, the delta function is treated in the theory of gen-
eralized functions [6, 7]. We give below the definition of the delta function as a
generalized function, and we mention some of its properties [8, 9].
For a point source applied at t ¼ 0, the one-dimensional delta function is
defined by the relation
Z +1
d ðt Þf ðt Þ dx ¼ f ð0Þ (3.7.4)
1
or by the relation
Z +1
dðt t1 Þf ðt Þ dt ¼ f ðt1 Þ (3.7.5)
1
for a point source applied at t ¼ t1 . The function f ðt Þ is continuous in a finite

interval containing the source point t ¼ 0 or t ¼ t1 . The one-dimensional delta
function can also be described by the relations

0, t 6¼ 0
d ðt Þ ¼ (3.7.6a)
1, t ¼ 0
@CivilMethod
and
Z +1 Z e
d ðt Þdt ¼ d ðt Þdt ¼ 1 (3.7.6b)
1 e
where e is a small positive number. According to this definition, the function

dðt Þ has zero value everywhere except at t ¼ 0, where it becomes infinite,
and satisfies Eq. (3.7.6b).
Eq. (3.7.4) may be obtained using the mean value theorem of integral cal-
culus. Thus, referring to Fig. 3.7.2a and choosing e ¼ 1=k, we obtain
Z 1 Z +e

1
f ðt Þd ðt Þdt ¼ ‘im f ðt Þpk ðt Þdt ¼ ‘im f ðt Þ
∗ 2e
1 e!0 e e!0 2e
(3.7.7)
¼ ‘im ½f ðt ∗ Þ ¼ f ð0Þ
e!0
where t* is a value of t in the interval (e, þe)

Moreover, using integration by parts we can show
Z b
d m d ðt t1 Þ m d f ðt1 Þ
m
f ðx Þ dt ¼ ð 1 Þ , a < t1 < b (3.7.8)
a dt m dt m
3.7.2 Response to unit impulse

In dynamics, loads pðt Þ distributed over a very short interval ½e, +e are
referred to as impulsive loads (see Section 3.5). The integral of the distribution
of the impulsive load over the interval it acts is defined as the impulse of the load
and is denoted by I , namely
Z +e
I¼ pðt Þdt (3.7.9)
e
If the impulsive load P is concentrated and acts at instant t ¼ t1 , it can be

denoted by
pðt Þ ¼ Pdðt t1 Þ (3.7.10)
Obviously, Eq. (3.7.5) for P ¼ 1 gives
Z +e
I¼ Pdðt t1 Þdt ¼ 1 (3.7.11)
e
From Newton’s second law of motion, we have (see Appendix)

d ðm u_ Þ
¼ pðt Þ (3.7.12a)
dt
or if the mass is constant
d u_
m ¼ pðt Þ (3.7.12b)
dt
@CivilMethod
which after integration over the interval ½e, +e gives

Z +e
+e
m ½u_ e ¼ pðt Þdt (3.7.13)
e
or for e ! 0
mDu_ ¼ 1 (3.7.14)
where

Du_ ¼ u_ t1+ u_ t1 (3.7.15)
Hence the impulsive load produces an abrupt change (discontinuity) of the
t1 . If the system is at rest before the action of the impulsive load,
velocity at time
then it is u_ t1 ¼ 0 and Eq. (3.7.14) gives
1
u_ ðt1 Þ ¼ (3.7.16)
m
where t1 designates t1+ . Therefore, Eq. (3.7.16) presents the initial velocity
given to the system by the impulsive load. However, the displacement remains
continuous, which means that

Du ¼ u t1+ u t1 ¼ 0 (3.7.17)
If the elastic and damping forces are taken into account, then Eq. (3.7.12b)
is written
d u_ du
m ¼ pðt Þ c ku (3.7.18)
dt dt
which after integration over the interval ½e, +e gives
Z +e Z +e
+e +e
m ½u_ e ¼ pðt Þdt c½u e k u ðt Þdt (3.7.19)
e e
Applying the mean value theorem of integral calculus to the integral of elas-
tic force, we write Eq. (3.7.19) as
Z +e
+e +e
m ½u_ e ¼ pðt Þdt c½u e ku ðt ∗ Þ2e, e < t ∗ < e (3.7.20)
e
which by virtue of Eqs. (3.7.15), (3.7.17) becomes if e ! 0

mDu_ ¼ 1 (3.7.21)
This means that the elastic and damping forces do not influence the change
of the velocity
when
an impulsive load is applied.
If u t1 ¼ u_ t1 ¼ 0, then Eqs. (3.7.16), (3.7.17) imply that the unit
impulse produces free vibrations with initial conditions u ðt1 Þ ¼ 0,
u_ ðt1 Þ ¼ 1=m, where t1 designates t1+ .
@CivilMethod
h(t-t)
t =1
t
FIG. 3.7.3 Response to unit impulse.
If τ denotes the instant t1 and h ðt τÞ u ðt Þ the produced displacement,

then we obtain from Eq. (2.3.18)
1 xwðtτÞ
h ðt τ Þ ¼ e sin ½wD ðt τÞ, t τ (3.7.22a)
mwD
Obviously, for the undamped system, x ¼ 0, it is
1
h ðt τÞ ¼ sin ½wðt τÞ, t τ (3.7.22b)
mw
Eqs. (3.7.22a), (3.7.22b) express the response of the SDOF system to a unit
impulse acting at time τ. Note that h ðt τÞ ¼ 0 if t < τ. Fig. 3.7.3 shows the
response to a unit impulse acting at t1 ¼ τ ¼ 1.
The derivation of the response function h ðt τÞ presented previously was
achieved by physical consideration. Mathematically, it expresses the solution
of the equation of motion under the external loading pðt Þ ¼ d ðt τÞ, that is,
1
u€ + 2xwu_ + w2 u ¼ d ðt τ Þ (3.7.23)
m
The solution of the above equation can be obtained from Eq. (3.3.23) for
zero initial conditions and pðt Þ ¼ dðt t1 Þ. Thus, we have
Z t
1
u ðt Þ ¼ dðτ t1 ÞexwðtτÞ sin ½wD ðt τÞdτ (3.7.24)
mwD 0
@CivilMethod
which by virtue of Eq. (3.7.5) gives

1 xwðtt1 Þ
u ð t Þ h ð t t1 Þ ¼ e sin ½wD ðt t1 Þ, t t1 (3.7.25)
mwD
or by setting t1 ¼ τ, we recover Eq. (3.7.22a).
Eqs. (3.7.22a), (3.7.22b) represent the Green’s function of the equation of
motion of the damped and undamped SDOF systems, respectively.
3.7.3 Response to arbitrary loading

The arbitrary loading pðt Þ can be visualized as a sequence of pulse loads of
infinitesimal duration with magnitude P ¼ pðτÞdτ in the interval ½ 0, t . The
response of the system to this pulse load is
du ¼ ½pðτÞdτh ðt τÞ (3.7.26)
Apparently, the response of the system at time t is the sum of all infinites-
imal responses from t ¼ 0 τ ¼ t. This sum is expressed by the integral
Z t Z t
u ðt Þ ¼ du ¼ pðτÞh ðt τÞdτ (3.7.27)
0 0
The integral (3.7.27) is known as the convolution integral of the functions

pðt Þ and h ðt Þ, to which we referred as the Duhamel integral in Section 3.2.2.
3.7.4 The reciprocal theorem in dynamics

In statics, the reciprocity is expressed by Betti’s theorem, also known as the
Maxwell-Betti reciprocal work theorem, introduced by Enrico Betti in 1872.
This theorem is valid for systems with linear behavior and reads
If two sets of loads FI and FII act separately on a linearly elastic structure, the
work WI , II done by the first set of loads in acting through the displacements UII
produced by the second set of loads is equal to the work WII , I done by the second
set of loads in acting through the displacements UI produced by the first set of
loads, namely
WI , II ¼ WII , I (3.7.28)
In dynamics of linear systems, the reciprocity is expressed by the dynamic
reciprocal theorem [10], known also as the dynamic Betti-Rayleigh theorem.
This theorem for SDOF systems reads
If two loadings pI ðt Þ and pII ðt Þ act separately on a linear dynamic system and
produce the responses uI ðt Þ and uII ðt Þ then the convolution CI , II of the loading
pI ðt Þ with the response uII ðt Þ is equal to the convolution CII , I of the loading pII ðt Þ
with the response uI ðt Þ
@CivilMethod
CI , II ¼ CII , I (3.7.29)
or using definition (3.3.11) for the convolution, we may write
pI ðt Þ∗ uII ðt Þ ¼ pII ðt Þ∗ uI ðt Þ (3.7.30a)
or in integral form
Z t Z t
pI ðτÞuII ðt τÞdτ ¼ pII ðτÞuI ðt τÞdτ (3.7.30b)
0 0
The dynamic reciprocal theorem is readily proved by taking into account

that the convolution satisfies the following properties [8]:
ðiÞ f ðt Þ ∗g ðt Þ ¼ g ðt Þ ∗f ðt Þ (3.7.31a)
ðiiÞ f ðt Þ∗ ½g ðt Þ∗ q ðt Þ ¼ ½g ðt Þ ∗f ðt Þ∗ q ðt Þ (3.7.31b)
where f ðt Þ, g ðt Þ, and q ðt Þ are arbitrary functions.
If h ðt Þ represents the response to the unit impulse, Eqs. (3.7.22a), (3.7.22b),
we may write on the basis of Eq. (3.7.25)
uI ðt Þ ¼ pI ðt Þ ∗h ðt Þ (3.7.32a)
uII ðt Þ ¼ pII ðt Þ∗ h ðt Þ (3.7.32b)
We may further write
pI ðt Þ ∗uII ðt Þ ¼ pI ðt Þ∗ ½pII ðt Þ∗ h ðt Þ
¼ pI ðt Þ∗ ½h ðt Þ ∗pII ðt Þ
¼ ½pI ðt Þ ∗h ðt Þ∗ pII ðt Þ (3.7.33)
¼ uI ðt Þ ∗pII ðt Þ
¼ pII ðt Þ∗ uI ðt Þ
which proves the dynamic reciprocal theorem.
The dynamic reciprocal theorem has many applications in mechanics, for
example, moving loads, the boundary integral equation method for dynamic
problems [10], etc.
3.8 Problems
Problem P3.1 A machine carrying a mass m0 is placed on the slab of the build-
ing of Fig. P3.1. The mass rotates eccentrically about the point ð2:5, 2:0Þ at a
distance s ¼ 1:0m with a frequency f ¼ 4Hz. The columns are massless, inex-
tensible, and fixed on the base while the slab is assumed uniform and rigid.
Determine the time history of the shear forces Qx , Qy , the bending moments
Mx , My , and the torsion moment Mw at the top of the columns. The material
constants are E ¼ 1:2 107 kN=m2 , n ¼ 0:2. The total load of the plate (dead
plus live) is q ¼ 12kN=m2 ; m0 ¼ m=5, where m is the total mass of the slab.
The acceleration of gravity is g ¼ 9:81m=s2 .
@CivilMethod
FIG. P3.1 Building in problem P3.1
Problem P3.2 A SDOF system with parameters m, k, x is subjected to

harmonic loading p ¼ p0 cos wt. Determine the expression of the steady-state
response and show that its amplitude is the same with that produced by the load-

ing p ¼ p0 sin wt.
Problem P3.3 An undamped SDOF system is subjected to the loading
When w ¼ w1 , the system is set in resonance. Subsequently, a mass
p ¼ p0 sin wt.
Dm ¼ 0:20kN m1 s is added and the system is set in resonance, when
w ¼ 0:6w1 . Determine the mass m of the system.
Problem P3.4 The one-story building of Fig. P3.4a is supported by three col-
umns placed at the vertices of an equilateral triangle. The slab is rigid and the
columns are fixed at both ends and have a rectangular cross-section, a modulus
of elasticity E, and negligible mass. The slab is subjected to the load
for 0 t t1 and pðt Þ ¼ 0 for t1 < t (Fig. P3.4b) acting in
pðt Þ ¼ p0 cos 3 wt
(a) (b)
@CivilMethod
the x direction. Plot the response ratio Rðt Þ of the structure and the stress resul-
tants at the base of the columns 1 and 3 using the following data: Height of col-
umns a ¼ 4m, cross-section of columns a=10 a=20, side length of the
triangular slab a, p0 ¼ 10kN, E ¼ 2:1 107 kN=m2 , load of the plate, including
the dead load, q ¼ 20kN=m2 , t1 ¼ p=2w, x ¼ 0:05, w ¼ 8s1 , acceleration of
gravity g ¼ 9:81m=s . 2
Problem P3.5 The structure of Fig. P3.5a consists of the rigid girder BC and
the two flexible columns AB and CD having a cross-sectional moment of inertia
I and a modulus of elasticity E. The cables FB and GC have cross-sectional area
A, cannot withstand compression, and are assumed massless. The structure is sub-
jected to the impulsive loads pðt Þ shown in Fig. P3.5b and c. Study the response of
the structure and determine the maximum error when the impulsive loads are
substituted by equivalent concentrated forces. Plot the function D ðt1 =T Þ for the
two load cases. For which value of the ratio t1 =T is the maximum error less than
2%? Data: a ¼ 1:5m, I ¼ 33,740cm4 (IPE450), E ¼ 2:1 108 kN=m2 ,
A ¼ 3cm2 , t1 ¼ 0:1, m ¼ 1:0kNm1 s2 =m, p10 ¼ 10kN. The value p20 is deter-
mined so that both loads have the same impulse.
(a)
(b) (c)
@CivilMethod
Problem P3.6 Consider the structure of Fig. P3.6. The rigid column AC of circu-
lar cross-section and mass per unit length m is supported by the three elastic cables
of cross-sectional area A and modulus of elasticity E. The support on the ground is
a spherical hinge. The cables have been prestressed so that they can withstand com-
pression. Three advertising panels are massless fixed at the top of the columns, as
shown in the figure. The structure is subjected to the wind blast load of Fig. P3.6c in
the y direction (see Fig. P3.6b). Determine the minimum prestress force of the
cables. The cables and the panels have negligible mass. Use the data a ¼ 5:0m,
p0 ¼ 4kN=m2 , m ¼ 0:5kNm1 s2 =m, E ¼ 2:1 108 kN=m2 , A ¼ 4cm2 ,
t1 ¼ 0:1s.
(a) (b)
(c)
@CivilMethod
Problem P3.7 The one-story building in Example 2.2.1 is subjected to a blast

pressure in the x direction. The time variation of the blast pressure is approx-
imated as shown in Fig. P3.7. Determine the stress resultants max Qx and
max Mx of the columns. The peak positive pressure is p0 ¼ 10kN and the
peak negative pressure is p1 ¼ 0:2p0 ; t1 ¼ 0:2s and t2 ¼ 5t1 . The system is
at rest at t ¼ 0.
10
p
0
4
2
t1 t2
0
p1
–2
0 0.5 1 1.5 2
FIG. P3.7 Blast pressure in problem P3.7
Problem P3.8 Show that an impulsive load generated by the load pðt Þ
and acting over the interval ½t1 , t2 can be represented as pI ¼ ½H ðt t1 Þ
H ðt t2 Þpðt Þ, where H ðt ti Þ is the Heaviside step function. Write a
MATLAB program that constructs the impulsive load.
Problem P3.9 Determine the dynamic response of a SDOF system subjected to
the sine periodic loading of Fig. P3.9. Assume: T ¼ p=w, w ¼ 1:1w,
x ¼ 0, and
x ¼ 0:1.
FIG. P3.9 Periodic load in problem P3.9

@CivilMethod
Problem P3.10 Determine the dynamic response of a SDOF system subjected

T ¼ p=w,
to the periodic loading of Fig. P3.10. Assume w ¼ 1:1w, x ¼ 0,
and x ¼ 0:1.
FIG. P3.10 Periodic load in problem P3.10

[1] F.B. Hildebrand, Advanced Calculus for Applications, Prentice Hall, Englewood Cliffs, New
Jersey, 1962.
[2] E. Kreyszig, Advanced Engineering Mathematics, fourth ed., John Wiley & Sons, New York,
1979.
[3] R.V. Churchill, Operational Mathematics, second ed., MacGraw-Hill Book Company, Inc.,
New York, 1958.
[4] F. Oberhettinger, L. Badii, Tables of Laplace Transforms, Springer-Verlag, Berlin/Heidelberg/
New York, 1973.
[5] H.P. Hsu, Fourier Analysis, Simon and Schuster, New York, 1967.
[6] G.F. Roach, Green’s Functions, Van Nostrand Reinhold Company, London, 1970.
[7] Μ. Greenberg, Application of Green’s Functions in Science and Engineering, Prentice Hall,
Englewood Cliff, NJ, 1971.
[8] T. Myint-U, L. Debnath, Linear Partial Differential Equations for Scientists and Engineers,
fourth ed., Birkh€auser, Boston, 2007.
[9] J.T. Katsikadelis, The Boundary Element Method for Engineers and Scientists, Academic
Press, Elsevier, Oxford, UK, 2016.
[10] J. Dominguez, Boundary Elements in Dynamics, Computational Mechanics Publications,
Southampton, Boston, 1993.
@CivilMethod
Chapter 4
Numerical integration of the

equation of motion
Chapter outline
4.1 Introduction 159 4.5.4 Stability of the central
4.2 The central difference method 160 difference method 180
4.3 The average acceleration 4.5.5 Stability of the average
method 164 acceleration method 182
4.4 The analog equation method 169 4.5.6 Stability of the analog
4.5 Stability of the numerical equation method 185
integration methods 174 4.6 Accuracy of the numerical
4.5.1 Errors in the numerical integration 185
integration 174 4.7 Problems 188
4.5.2 Difference equations 175 References and further reading 189
4.5.3 Difference equations and
stability of the numerical
integration methods 178
4.1 Introduction
The previous analysis shows that an analytical solution of the equation of
motion for an single-degree-of-freedom (SDOF) system is possible only if
the external force is described by a simple function. If the excitation force varies
arbitrarily with time or is given by a set of its values, an analytical solution is out
of the question. However, such problems can be tackled numerically by time
step integration methods for differential equations. The literature about these
methods is vast. Extensive chapters and whole books cover this subject. They
present the mathematical development of these methods, their computer imple-
mentation, and their accuracy, convergence, and stability. Several computer
packages include ready-to-use subroutines for the solution of the differential
equation of motion.
Some of these methods have been specially developed for the study of the
dynamic response of systems. A survey of these methods is given in [1, 2]. The
central difference method (CDM), Houbolt’s method, Wilson’s q-Method, and
Newmark’s method are the most well known among them [3, 4]. Nevertheless,
with the increase of cheap computer power, some of them have lost their impor-
tance while others have taken dominating places in the computational arena.

@CivilMethod
These methods determine the displacement u ðt Þ by numerical integration of

the equation of motion
m u€ + cu_ + ku ¼ pðt Þ (4.1.1)
The displacement u ðt Þ is computed step by step. The numerical methods
start from the values u ð0Þ and u_ ð0Þ, which are known at t ¼ 0, and subsequently
marching in the integration time, these quantities are computed at instants
Dt, 2Dt, 3Dt,… where Dt denotes a small time interval. The development of
these methods is based on two assumptions. The first assumption is that the equation
of motion (4.1.1) is satisfied at discrete time instants Dt apart, contrary to the ana-
lytical methods in which the equation is satisfied at any instant t. This means that the
equilibrium of all forces, namely inertial, damping, elastic, and external, is satisfied
only at discrete instants within the time interval of the solution. The second assump-
tion is that the parameters of the solution, namely the displacement, velocity, and
acceleration, vary within the time interval Dt according to a prescribed law. The
accuracy, stability, and efficiency of the solution method depend on the numerical
integration scheme as well as on the magnitude of Dt. In the following sections, only
some of these methods, which are usually employed for the study of the dynamic
response of systems, will be briefly discussed. We confine our presentation to the
central difference method (CDM), Newmark’s average acceleration method
(AAM) [5], and the analog equation method (AEM), a new efficient method based
on the principle of the analog equation [6].
The presentation that follows is intended to discuss only the basic concepts
underlying these methods and provide a few computational algorithms together
with the computer programs based on them. While this material would be suf-
ficient for many practical problems and research applications, the reader should
recognize that a wealth of knowledge exists on the subject.
4.2 The central difference method

In this method, the values of u ðt + Dt Þ and u ðt Dt Þ of the function u ðt Þ at
instants t + Dt and t Dt (Fig. 4.2.1) are approximated by their Taylor series
1 1
u ðt + Dt Þ ¼ u ðt Þ + Dt u_ ðt Þ + Dt 2 u€ðt Þ + Dt 3 u___ðt Þ + ⋯ (4.2.1)
2 6
1 1
u ðt Dt Þ ¼ u ðt Þ Dt u_ ðt Þ + Dt 2 u€ðt Þ Dt 3 u
___ðt Þ + ⋯ (4.2.2)
2 6
Subtracting Eq. (4.2.2) from Eq. (4.2.1) yields
2
u ðt + Dt Þ u ðt Dt Þ ¼ 2Dt u_ ðt Þ + Dt 3 u
___ðt Þ + ⋯ (4.2.3)
6
For small values of Dt, the terms of order higher than two can be neglected
and Eq. (4.2.3) yields the following relation for the approximation of the first
derivative at time t
u ðt + Dt Þ u ðt Dt Þ
u_ ðt Þ (4.2.4)
2Dt
@CivilMethod
Numerical integration of the equation of motion Chapter 4 161
FIG. 4.2.1 Discretization of the interval ½0, T into N equal intervals h ¼ T =N .
Further, adding Eqs. (4.2.1), (4.2.2) and neglecting the terms of order higher
than three, we obtain the following expression to approximate the second deriv-
ative of u ðt Þ at time t
u ðt + Dt Þ 2u ðt Þ + u ðt Dt Þ
u€ðt Þ (4.2.5)
Dt 2
Substitution of the derivatives u_ ðt Þ and u€ðt Þ into Eq. (4.1.1) with their
approximations (4.2.4) and (4.2.5) gives
u ðt + Dt Þ 2u ðt Þ + u ðt Dt Þ u ðt + Dt Þ u ðt Dt Þ
m +c + ku ðt Þ ¼ pðt Þ
Dt 2 2Dt
which is solved for u ðt + Dt Þ to yield
m
c 2m m c
+ u ð t + Dt Þ ¼ p ð t Þ k u ðt Þ u ðt Dt Þ
Dt 2 2Dt Dt 2 Dt 2 2Dt
(4.2.6)
The previous equation may be written as
K^ u ðt + Dt Þ ¼ P^ (4.2.7)
where
m c
K^ ¼ 2 + (4.2.8)
Dt 2Dt

2m m c
P^ ¼ pðt Þ k 2 u ðt Þ u ðt Dt Þ (4.2.9)
Dt Dt 2 2Dt
The quantities K^ and P^ are referred to as the effective stiffness and the
effective load, respectively. Obviously, Eq. (4.2.7) allows the evaluation of
the displacement at instant t + Dt, if the displacements at the two preceding
instants t and t Dt are known. Because u ð0Þ is known from the initial
@CivilMethod
conditions, the procedure starts at t ¼ Dt. Obviously, this requires the value of
u ðDt Þ, which is unknown in the first instance, but it can be determined from
Eq. (4.2.2) for t ¼ 0. Thus neglecting terms of order higher than two, we have
1
u ðDt Þ u ð0Þ Dt u_ ð0Þ + Dt 2 u€ð0Þ (4.2.10)
2
In the above equation, the quantities u ð0Þ and u_ ð0Þ are known from the ini-
tial conditions while u€ð0Þ can be computed from the equation of motion,
Eq. (4.1.1), for t ¼ 0. Thus, we obtain
u€ð0Þ ¼ ½pð0Þ cu_ ð0Þ ku ð0Þ=m (4.2.11)
The stability of the CDM requires that time step Dt is less than a certain
critical value, that is, qffiffiffiffiffiffiffiffiffiffiffiffi
Dt Dtcr ¼ T 1 x2 =p (4.2.12)
where T is the period of the system (see Eq. 4.5.30). Otherwise, the procedure
“blows up” with time and the solution makes no sense. This is discussed in
Section 4.5.
Because T is usually a small number, Dt should be small, which implies
that a large number of time steps are required to solve the equation of motion.
This has been a major drawback of the method, especially in older times when
the computer capabilities in terms of memory and speed were restricted. This
fact has led researchers to develop integration methods in which the size of
the time step is not restricted by a critical value. Table 4.2.1 presents the
TABLE 4.2.1 Central difference method (CDM).

A. Data
w (or k), m, x, u0 , u_ 0 , ttot , pðt Þ
B. Initial computations
B.1. k ¼ mw2 (or w ¼ k=m ), c ¼ 2mwx, T ¼ 2p=w, ust ¼ pmax =k
B.2. u€0 ¼ ðp0 cu_ 0 ku 0 Þ=m
B.3. Choose Dt < Dtcr ¼ T 1 x2 =p and compute
B.4. ao ¼ 1=Dt 2 , a1 ¼ 1=2Dt, a2 ¼ 2ao , a3 ¼ 1=a2
B.5. uDt ¼ u0 Dt u_ 0 + a3 u€0 , K^ ¼ ao m + a1 c
C. For each time step compute:
C.1. P^t ¼ pt ðk a2 m Þut ðao m a1 cÞutDt
C.2. ut + Dt ¼ P^t =K^ , Rt + Dt ¼ ut + Dt =ust
C.3. u_ t ¼ a1 ðut + Dt utDt Þ, u€ðt Þ ¼ ao ðut + Dt 2ut + utDt Þ
C.4. Increase time t ¼ t + D and check: If t ttot end. Else set utDt ¼ ut , ut ¼ ut + Dt
and go to C.1
@CivilMethod
algorithm for the numerical implementation of the CDM in pseudocode-type

notation so that the reader can write a computer code in the language of his/
her preference.
Adhering to the steps of Table 4.2.1, a computer program called centr_
diff_lin.m has been written in MATLAB for the numerical integration of the
equation of motion using the CDM. The program is available on this book’s
companion website. It computes the displacement u ðt Þ, the velocity u_ ðt Þ, the
acceleration u€ðt Þ, and the response ratio Rðt Þ ¼ u ðt Þ=ðpmax =k Þ and gives their
graphical representation. Moreover, it computes the dynamic magnification
factor D ¼ max jRðt Þj and the time tmax it occurs. The user of the program is
responsible for providing the function representing the excitation force.
Example 4.2.1 Response of a SDOF using the central difference method (CDM)
Using the CDM, determine the response of a SDOF system with
m ¼ 100kN m1 s2 , k ¼ 2500kN=m, x ¼ 0:05, and u ð0Þ ¼ u_ ð0Þ ¼ 0, and sub-
jected to the load pðt Þ ¼ po exp ð1 0:5t Þ, po ¼ 100kN.
Solution
The solution is obtained using the program centr_diff_lin.m with Dt ¼ 0:01.
Fig. E4.1 gives the response of the SDOF system. Moreover, Fig. E4.2 shows
Displacement Velocity
0.2 0.5
0.1
u,t(t)
u(t)
0
0
–0.1 –0.5
0 5 10 0 5 10
t t
Acceleration Response ratio
4 2
2
1
u,tt (t)
R(t)
0
0
–2
–4 –1
0 5 10 0 5 10
t t
FIG. E4.1 Response of the SDOF system in Example 4.2.1.
@CivilMethod
the displacement u ðt Þ as compared with the exact one together with the error
u ðt Þ uex ðt Þ. The exact solution was obtained by analytical evaluation of
Duhamel’s integral giving

4po ð1 2xwÞ
u ðt Þ ¼ e 10:5t
+ cos w D t + sin w D t e 1xwt
m ð1 4xw + 4w2 Þ 2wD
(1)
x 10 -4
0.2 1
u(t) computed
u(t) exact 0.8 u-uex
0.15 0.6
0.4
0.1 0.2
u(t)
0
0.05 –0.2
–0.4
0 –0.6
–0.8
-0.05 –1
0 2 4 6 8 10 0 2 4 6 8 10
t
FIG. E4.2 Computed solution and error in Example 4.2.1.
4.3 The average acceleration method

In 1959, N. M. Newmark developed a family of time step methods for the
numerical integration of the equation of motion [5] based on the approxima-
tion of the acceleration in each time step. In the following, we describe
only the well-known AAM (Average Acceleration Method), which is also
known as Newmark’s method with b ¼ 1=4 or Newmark’s trapezoidal rule
method. The AAM is the most widely employed numerical method in struc-
tural dynamics as it is simple to implement numerically as well as being
unconditionally stable and accurate (see Section 4.5.5). It can also be
employed to solve nonlinear equations of short duration motion [6, 7].
In this method, the acceleration in the interval t to t + Dt is assumed con-
stant and equal to its mean value (Fig. 4.3.1)
1
u€ðt + τÞ ¼ ½u€ðt Þ + u€ðt + Dt Þ 0 τ Dt (4.3.1)
2
Integrating with respect to τ yields
τ
u_ ðt + τÞ ¼ ½u€ðt Þ + u€ðt + Dt Þ + C1
2
@CivilMethod
FIG. 4.3.1 Variation of the acceleration, velocity, and displacement in the average acceleration
method.
where C1 is an arbitrary constant. For τ ¼ 0 we obtain C1 ¼ u_ ðt Þ, hence

τ
u_ ðt + τÞ ¼ u_ ðt Þ + ½u€ðt Þ + u€ðt + Dt Þ 0 τ Dt (4.3.2)
2
Integrating once more with respect to τ yields
τ2
u ðt + τÞ ¼ τu_ ðt Þ + ½u€ðt Þ + u€ðt + Dt Þ + C2
4
which for τ ¼ 0 yields C2 ¼ u ðt Þ. Thus we have

τ2
u ðt + τÞ ¼ u ðt Þ + τ u_ ðt Þ + ½u€ðt Þ + u€ðt + Dt Þ 0 τ Dt (4.3.3)
4
Eqs. (4.3.2), (4.3.3) for τ ¼ Dt give
Dt
u_ ðt + Dt Þ ¼ u_ ðt Þ + ½u€ðt Þ + u€ðt + Dt Þ (4.3.4)
2
Dt 2
u ðt + Dt Þ ¼ u ðt Þ + Dt u_ ðt Þ + ½u€ðt Þ + u€ðt + Dt Þ (4.3.5)
4
@CivilMethod
Setting
Du ¼ u ðt + Dt Þ u ðt Þ (4.3.6a)
Du_ ¼ u_ ðt + Dt Þ u_ ðt Þ (4.3.6b)
Du€ ¼ u€ðt + Dt Þ u€ðt Þ (4.3.6c)
Eqs. (4.3.4), (4.3.5) are written as
Dt
Du_ ¼ ½2u€ðt Þ + Du€ (4.3.7)
2
Dt 2
Du ¼ Dt u_ ðt Þ + ½2u€ðt Þ + Du€ (4.3.8)
4
€ we obtain
Solving Eq. (4.3.8) for Du,
4
Du€ ¼ ½Du Dt u_ ðt Þ 2u€ðt Þ (4.3.9)
Dt 2
Moreover, substituting Eq. (4.3.9) into Eq. (4.3.7) yields
2
Du_ ¼ Du 2u_ ðt Þ (4.3.10)
Dt
We shall now express Du in terms of u_ ðt Þ and u€ðt Þ. For this purpose, we

apply Eq. (4.1.1) at time t + Dt and t. This yields
m u€ðt + Dt Þ + cu_ ðt + Dt Þ + ku ðt + Dt Þ ¼ pðt + Dt Þ
m u€ðt Þ + cu_ ðt Þ + ku ðt Þ ¼ pðt Þ
which after subtracting give

mDu€ + cDu_ + kDu ¼ Dp (4.3.11)
where
Dp ¼ pðt + Dt Þ pðt Þ
K^ Du ¼ P^ (4.3.12)
where
2c 4m
K^ ¼ k + + (4.3.13)
Dt Dt 2

^ 4m
P ¼ Dp + + 2c u_ ðt Þ + 2m u€ðt Þ (4.3.14)
Dt
@CivilMethod
The quantities K^ and P^ are referred to as effective stiffness and effective

load, respectively.
Eq. (4.3.12) allows the computation of Du when u_ ðt Þ and u€ðt Þ are known at
time t. Then Du€ðt Þ and Du_ ðt Þ can be computed using Eqs. (4.3.9), (4.3.10).
Subsequently, the values of u ðt + Dt Þ, u_ ðt + Dt Þ, u€ðt + Dt Þ result from the
expressions
u ðt + Dt Þ ¼ u ðt Þ + Du (4.3.15a)
u_ ðt + Dt Þ ¼ u_ ðt Þ + Du_ (4.3.15b)
u€ðt + Dt Þ ¼ u€ðt Þ + Du€ (4.3.15c)
It may be more convenient to compute the acceleration u€ðt Þ using the
equation of motion (4.1.1) than Eq. (4.3.15c), that is,
u€ðt + Dt Þ ¼ ½pðt + Dt Þ cu_ ðt + Dt Þ ku ðt + Dt Þ=m (4.3.16)
Table 4.3.1 presents the algorithm for the numerical implementation of the
AAM in pseudocode-type notation so that the reader can write a computer code
in the language of his/her preference.
TABLE 4.3.1 Average acceleration method (AAM).

A. Data
w (or k), m, x, u0 , u_ 0 , ttot , pðt Þ
B.1. k ¼ mw2 (or w ¼ k=m ), c ¼ 2mwx, ust ¼ pmax =k
B.2. u€0 ¼ ðp0 cu_ 0 ku 0 Þ=m
B.3. Choose Dt (usually Dt ¼ 0:1T =p) and compute
B.4. K^ ¼ k + 2c=Dt + 4m=Dt 2
C. For each time step compute
C.1. Dp ¼ pðt + Dt Þ pðt Þ, P^ ¼ Dp + ð4m=Dt + 2cÞu_ ðt Þ + 2m u€ðt Þ
^ K^
C.2. Du ¼ P=
Du_ ¼ 2Du=Dt 2u_ ðt Þ
Du€ ¼ 4½Du Dt u_ ðt Þ=Dt 2 2u€ðt Þ
C.3. u ðt + Dt Þ ¼ u ðt Þ + Du
u_ ðt + Dt Þ ¼ u_ ðt Þ + Du_
u€ðt + Dt Þ ¼ u€ðt Þ + Du€
C.4. Increase time t ¼ t + Dt and check: If t ttot end. Else set u_ ðt Þ ¼ u_ ðt + Dt Þ,
u€ðt Þ ¼ u€ðt + Dt Þ and go to C.1
@CivilMethod
As will be shown, contrary to the CDM, the stability of the AAM does not
demand any restriction on the size of the time step Dt. The time step, however,
is influenced by the accuracy of the method and its capability to describe an
oscillatory motion. Therefore, it must be small enough. The selection of Dt
equal to 1/10 of the period of the system or of the period of the excitation force
produces accurate results.
Adhering to the steps of Table 4.3.1, a computer program called av_acc_lin.m
has been written in MATLAB for the numerical integration of the equation of
motion. The program is available on this book’s companion website. It computes
the displacement u ðt Þ, the velocity u_ ðt Þ, the acceleration u€ðt Þ, and the response
ratio Rðt Þ ¼ u ðt Þ=ðpmax =k Þ and gives their graphical representation. Moreover, it
computes the dynamic magnification factor D ¼ max jRðt Þj and the time tmax it
occurs. The user of the program is responsible for providing the function of the
excitation force.
Example 4.3.1 Response of an SDOF using the average acceleration

method (AAM)
Determine the response the SDOF system in Example 4.2.1 using the AAM.
Solution
The solution was evaluated using the program av_acc_lin.m with Dt ¼ 0:01.
Fig. E4.3 gives the graphical representation of the displacement together with
the error u ðt Þ uex ðt Þ. The computed error by the AAM is almost double the
error of the CDM. Moreover, Fig. E4.4 shows the response of the system under
the harmonic load p ¼ 2 sin 5t. Obviously, this excitation produces resonance
(w ¼ w ¼ 5).
x 10−4
0.2 2
u(t) computed
u(t) exact 1.5 u-uex
0.15
1
0.5
0.1
u(t)
0.05
–0.5
–1
0
–1.5
−0.05 –2
0 2 4 6 8 10 0 2 4 6 8 10
t t
@CivilMethod
Displacement Velocity
0.01 0.04
0.005 0.02
u,t(t)
u(t)
0 0
–0.005 –0.02
–0.01 –0.04
0 5 10 15 0 5 10 15
t t
Acceleration Response ratio
0.2 10
0.1 5
u,tt (t)
R(t)
0 0
–0.1 –5
–0.2 –10
0 5 10 15 0 5 10 15
t t
FIG. E4.4 Response of the system in Example 4.3.1 under harmonic load.
4.4 The analog equation method

The AEM (Analog Equation Method) is a general method for solving differen-
tial equations, ordinary or partial, linear or nonlinear, of the elliptic, parabolic,
or hyperbolic type. The AEM is based on the principle of the analog equation,
according to which a differential equation can be replaced by another equation,
the so-called analog equation, provided that the substitute equation retains the
principal mathematical symbol of the differential equation, that is, the highest
order derivative of the original differential equation. Thus, a differential equa-
tion whose solution cannot be obtained can be reduced to an equation with a
known solution. The AEM, which was first presented in 1994 [8], has been
employed as a computational method for the solution of numerous difficult
problems in engineering science and mathematical physics [9–11]. Here, the
AEM is presented for the solution of the equation of motion in structural dynam-
ics as developed in [6]. It is self-starting, unconditionally stable, accurate, and
conserves energy. It performs well when large deformations and long-time
durations are considered and it can be used as a practical method for the inte-
gration of the equations of motion in cases where widely used time integration
procedures, for example, Newmark’s AAM, become unstable [6, 7].
We consider the IVP (initial value problem) for the SDOF system
m u€ + cu_ + ku ¼ pðt Þ t ½0, ttot , ttot > 0 (4.4.1)
@CivilMethod
u ð0Þ ¼ u0 , u_ ð0Þ ¼ u_ 0 (4.4.2)

Let u ¼ u ðt Þ be the sought solution. Then, if the operator d =dt is applied to
2 2
it, we have
u€ ¼ q ðt Þ (4.4.3)
where q ðt Þ is a fictitious source, unknown in the first instance. Eq. (4.4.3) is the
analog equation of Eq. (4.4.1). It indicates that the solution of Eq. (4.4.1) can
be obtained by solving Eq. (4.4.3) with the initial conditions (4.4.2), if the q ðt Þ
is first established. This is achieved as follows.
Taking the Laplace transform of Eq. (4.4.3) we obtain
1 1 1
U ðs Þ ¼ u ð0Þ + 2 u_ ð0Þ + 2 Q ðs Þ (4.4.4)
s s s
where U ðs Þ,Q ðs Þ are the Laplace transforms of u ðt Þ,q ðt Þ, respectively. The
inverse Laplace transform of Eq. (4.4.4) gives the solution in integral from
Z t
u ðt Þ ¼ u ð0Þ + u_ ð0Þt + q ðτÞðt τÞdτ (4.4.5)
0
Thus, the IVP of Eqs. (4.4.1), (4.4.2) is transformed into the equivalent
Volterra integral equation for q ðt Þ.
Eq. (4.4.5) is solved numerically within a time interval ½0, T . The interval
½0, T is divided into N equal intervals, Dt ¼ h, h ¼ T =N , in which q ðt Þ is
assumed to vary according to a certain law, for example, constant, linear,
etc. In this analysis, q ðt Þ is assumed to be constant and equal to the mean value
in the interval h (Fig. 4.4.1). That is
qr1 + qr
qrm ¼ (4.4.6)
2
FIG. 4.4.1 Discretization of the interval ½0, T into N equal intervals h ¼ T=N .
Hence, Eq. (4.4.5) at instant t ¼ nh can be written as

un ¼ u0 + nh u_ 0
" Z Z Z #
h 2h nh
+ q1 m
ðnh τÞdτ + q2
m
ðnh τÞdτ + ⋯ + qnm ðnh τÞdτ
0 h ðn1Þh
(4.4.7)
@CivilMethod
which after evaluation of the integrals yields

X
n 1
un ¼ u0 + nh u_ 0 + c1 ½2ðn r Þ + 1qrm + c1 qnm
r¼1
(4.4.8)
X
n 1
¼ un1 + h u_ 0 + 2c1 qrm + c1 qnm
r¼1
where
h2
c1 ¼ (4.4.9)
2
The velocity is obtained by direct differentiation of Eq. (4.4.5) using
Leibnitz’ rule for integrals [12]. Thus, we have
Z t
u_ ðt Þ ¼ u_ ð0Þ + q ðτÞdτ (4.4.10)
0
Using the same discretization for the interval ½0, T to approximate the inte-
gral in Eq. (4.4.10), we have
X
n1
u_ n ¼ u_ 0 + c2 qrm + c2 qnm
r¼1 (4.4.11)
¼ u_ n1 + c2 qnm
where
c2 ¼ h (4.4.12)
P
n1
Solving Eq. (4.4.11) for qrm and substituting into Eq. (4.4.8) gives
r¼1
un ¼ un1 + h u_ n c1 qnm (4.4.13)

By virtue of Eq. (4.4.6), Eqs. (4.4.13), (4.4.11) are written as
c1 c1
qn h u_ n + un ¼ qn1 + un1 (4.4.14)
2 2
c2 c2
qn + u_ n ¼ u_ n1 + qn1 (4.4.15)
2 2
Moreover, Eq. (4.4.1) at time t ¼ nt is written as
mq n + cu_ n + ku n ¼ pn (4.4.16)
Eqs. (4.4.14), (4.4.15), (4.4.16) can be combined as
2 3 2 3
m c k 8 9 0 0 0 8 9 8 9
6 1 q
7< n = 6 1 7< qn1 = < 1 =
6 c h 1 7 u_ 6 c 0 1 7 u_
6 2 1 7 n ¼6 2 1 7 n1 + 0 p (4.4.17)
4 5: ; 4 5: ; : ; n
1 un 1 u 0
c2 1 0 c2 1 0 n1
2 2
@CivilMethod
Because m 6¼ 0, the coefficient matrix in Eq. (4.4.17) is not singular for suf-
ficient small h and the system can be solved successively for n ¼ 1, 2, … to yield
the solution un and the derivatives u_ n , u€n ¼ qn at instant t ¼ nh T . For n ¼ 1,
the value q0 appears in the right side of Eq. (4.4.17). This quantity can be readily
obtained from Eq. (4.4.1) for t ¼ 0. Thus, we have
q0 ¼ ðp0 cu_ 0 ku 0 Þ=m (4.4.18)
Eq. (4.4.17) can be also written as
Un ¼ AUn1 + bpn , n ¼ 1, 2, …,N (4.4.19)
in which
8 9
< qn =
Un ¼ u_ n (4.4.20a)
: ;
un
2 31 2 3
m c k 0 0 0
6 1 7 6 1 7
6 c1 h 1 7 6 c1 0 1 7
6
A¼6 2 7 6 7 (4.4.20b)
7 6 2 7
4 1 5 4 1 5
c2 1 0 c2 1 0
2 2
2 31
m c k 8 9
6 1 7 <1=
6 c h 1 7
b¼66 2
1 7
7 :0; (4.4.20c)
4 1 5 0
c2 1 0
2
The recurrence formula (4.4.19) can be employed to construct the solution
algorithm. However, the solution procedure can be further simplified. Thus,
applying Eq. (4.4.19) for n ¼ 1, 2, … we have
U1 ¼ AUo + bp1
U2 ¼ AU1 + bp2
¼ AðAUo + bp1 Þ + bp2
(4.4.21)
¼ A2 Uo + Abp1 + bp2
⋯ ¼ ⋯⋯⋯⋯⋯⋯⋯⋯⋯

Un ¼ An Uo + An1 p1 + An2 p2 + ⋯A0 pn b
Obviously, the last of Eq. (4.4.21) gives the solution vector Un at instant
tn ¼ nh using only the known vector U0 at t ¼ 0. The matrix A and the vector
b are computed only once.
Table 4.4.1 presents the algorithm for the numerical implementation of
AEM in pseudocode-type notation so that the reader can write a computer code
in the language of his/her preference.
@CivilMethod
Adhering to the steps of Table 4.4.1, a computer program called aem_lin.m

has been written in MATLAB for the numerical integration of the equation of
motion using the AEM. The program is available on this book’s companion
website. It computes the displacement u ðt Þ, the velocity u_ ðt Þ, the acceleration
u€ðt Þ, and the response ratio Rðt Þ ¼ u ðt Þ=ðpmax =k Þ and gives their graphical rep-
resentation. Moreover, it computes the dynamic magnification factor
D ¼ max jRðt Þj and the time tmax it occurs. The user is responsible for provid-
ing the function of excitation force.
TABLE 4.4.1 The analog equation method (AEM).

A. Data
w (or k), m, x, u0 , u_ 0 , ttot , pðt Þ, ttot
Choose: h ð¼ Dt Þ and compute ntot ,
Compute: k ¼ mw2 , c ¼ 2mwx
q0 ¼ ðp0 cu_ 0 ku 0 Þ=m, c1 ¼ h 2 =2, c2 ¼ h
Formulate: U0 :¼ fq0 u_ 0 u0 gT
2 31 2 3 2 31
m c k 0 0 0 m c k 8 9
6 c1 7 6 1 7 6 c1 7 <1=
6 h 1 7 6 c1 0 1 7 6 h 1 7
6
Compute: A ¼ 6 2 7 6 7 6 7
7 6 2 7, b ¼ 6 2 7 :0;
4 c2 5 4 5 4 c2 5 0
1 0 1 1 0
2 c2 1 0 2
2
C. Compute solution
for n :¼ 1 to ntot
Un ¼ AUn1 + bpn
end
Example 4.4.1 Response of an SDOF system using the analog equation method
Determine the response the SDOF system in Example 4.2.1 using the AEM.
Solution
The solution is obtained using the program aem_lin.m with Dt ¼ 0:01.
Fig. E4.5 gives the graphical representation of the displacement together with
the error u ðt Þ uex ðt Þ. Moreover, Fig. E4.6 shows the response of the system
under the static load p ¼ t, if 0 t 1 and p ¼ 1, if 1 < t.
@CivilMethod
x 10−4
0.2 2
u(t) computed
u-uex
u(t) exact 1.5
0.15
1
0.1 0.5
u(t)
0
0.05
–0.5
–1
0
–1.5
–0.05
0 2 4 6 8 10 –2
0 2 4 6 8 10
t
× 10
-4 Displacement × 10
-4 Velocity
6 10
4 5
du(t)
u(t)
2 0
0 –5
0 5 10 0 5 10
t t
× 10
-3 Acceleration Response ratio
2 1.5
0 1
ddu(t)
R(t)
–2 0.5
–4 0
0 5 10 0 5 10
t t
FIG. E4.6 Response of the system in Example 4.4.1 under static load.
4.5 Stability of the numerical integration methods

4.5.1 Errors in the numerical integration
The numerical integration of the differential equations, hence of the equation of
motion, introduces errors, which influence not only the accuracy of the obtained
solution but also its capability to produce a reliable solution, that is, a solution
that within certain accuracy represents the actual solution. Factors that may
@CivilMethod
contribute to errors in the results obtained from well-defined loadings are of

three types [13]:
(i) Round-off errors: They result from calculations being done using numbers
expressed by too few digits.
(ii) Truncation errors: They are introduced in representing u ðt + Dt Þ or
u_ ðt + Dt Þ by a finite number of terms in the Taylor series expansion.
(iii) Instability errors: They are introduced by amplification of the errors from
one step to the subsequent one during the computations. The stability of
any method is improved by reducing the length of the time step.
Round-off errors are random in nature and therefore must be treated by statis-
tical methods. In computer calculations, it is often possible to reduce the round-
off errors by using a higher precision in the computations. Truncation errors are
accumulated locally at each step. These errors provide a useful criterion for
measuring the accuracy of the various methods of numerical integration, pro-
vided that the employed numerical scheme is stable.
To study the stability of the numerical method, it is important to know the
influence of the error introduced at one step on the computations at the next step.
If the error tends to increase, then the solution soon becomes unbounded and
ceases to make sense. In this case, we say that the method is unstable. The study
of the stability of numerical integration methods of the differential equation of
motion and general of IVPs of differential equations is facilitated by the use of
difference equations, elements of which are discussed below.
4.5.2 Difference equations

If tn represents the time nh ðh ¼ Dt Þ, n ¼ 0, 1, 2… and un the displacement at
the same time, then we may write Eq. (4.2.6) as
un + 1 ¼ 2a1 un a2 un1 + gn (4.5.1)
where
2 h 2 w2 1 hwx pn h 2
a1 ¼ , a2 ¼ , gn ¼
2ð1 + hwx Þ 1 + hwx m ð1 + hwx Þ
Eq. (4.5.1) is a recursive relation, which permits the computation of un + 1
from its previous two values. Applying this relation successively for
n ¼ 0, 1, … we have
9
For n ¼ 0 u1 ¼ 2a1 uo a2 u1 + g0 >
>
>
>
For n ¼ 1 u2 ¼ 2a1 u1 a2 u0 + g1 =
For n ¼ 2 u3 ¼ 2a1 u2 a2 u1 + g2 (4.5.2)
>
⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ > >
>
;
For n ¼ n un + 1 ¼ 2a1 un a2 un1 + gn
@CivilMethod
We observe that un + 1 can be computed if the two initial values u1 and uo
are known. Eq. (4.5.1) may also be written as
un + 1 2a1 un + a2 un1 ¼ gn (4.5.3)
Eq. (4.5.3) is a difference equation of the second order [14] whose solution
can yield un + 1 using the recursive procedure (4.5.2). Because Eq. (4.5.3) is
linear with respect to un1 , un , and un + 1 , it is called a linear difference equa-
tion. If gn ¼ 0 the difference equation is called homogeneous while if gn 6¼ 0 it is
called nonhomogeneous. In general, an equation of the form
un + k + a1 un + k1 + a2 un + k2 + … + ak un ¼ gn (4.5.4)
where a1 ,a2 , …,ak are constants is a linear difference equation of order k

with constant coefficients. The solution of the difference equation is a
sequence of uk values for which it is true. The order of the difference equation
is the difference between the largest and the smallest value of the argument k
appearing in it.
First, we examine the homogeneous linear difference equation ðgn ¼ 0Þ. The
solution of this equation is sought in the form un ¼ rn , which is inserted in
Eq. (4.5.4) to give
rn + k + a1 rn + k1 + a2 rn + k2 + … + ak rn ¼ 0
or dividing by rn we obtain the characteristic equation of the difference

equation (4.5.4)
pðrÞ ¼ rk + a1 rk1 + a2 rk2 + … + ak ¼ 0 (4.5.5)
which is polynomial of order k.

If we assume that all roots r1 ,r2 ,…,rk of pðrÞ are distinct, then
rn1 ,rn2 ,…, rnk are solutions of Eq. (4.5.4). Moreover, because of the linearity
of the equation
un ¼ c1 rn1 + c2 rn2 + … + ck rnk (4.5.6)
is also a solution for all n for which the difference equation is defined with
c1 ,c2 ,…, ck being arbitrary constants. Eq. (4.5.6) is the general solution of
the difference equation (4.5.4). The arbitrary constants c1 ,c2 ,…, ck are deter-
mined from the k initial conditions.
If two roots of the polynomial (4.5.5) are complex conjugate, say r1 ¼
a + ib and r2 ¼ a ib, then we can write them in exponential form
r1 ¼ reiq , r2 ¼ reiq (4.5.7)
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r¼ a2 + b 2 , q ¼ tan 1 ðb=aÞ (4.5.8)
@CivilMethod
Using Euler’s identity (2.2.8), we obtain

rn1 ¼ r n einq ¼ r n ð cos nq + i sin nqÞ:
rn2 ¼ r n einq ¼ r n ð cos nq i sin nqÞ
thus we can write
c1 rn1 + c2 rn2 ¼ r n ½ðc1 + c2 Þcos nq + iðc1 c2 Þsin nq
(4.5.9)
¼ r n c10 cos nq + c20 sin nq
where c10 ¼ c1 + c2 and c20 ¼ iðc1 c2 Þ. We conclude from Eq. (4.5.9) that the
contribution of the complex roots to the solution is oscillatory.
Finally, we examine the case where the polynomial (4.5.5) has a double root,
say r1 . Then, the second solution will be nrn1 . This is verified by direct substi-
tution of un ¼ nrn1 into Eq. (4.5.4). Thus we obtain
ðn + k Þrn1 + k + a1 ðn + k 1Þrn1 + k1 + … + ak1 ðn + 1Þrn1 + 1 + ak nrn1

¼ nrn1 rk1 + a1 rk1 + … + ak1 r1 + ak
1

+ rn1 + 1 krk1
1 + a1 ðk 1Þrk2 + … + ak1
which yields
nrn1 pðr1 Þ + rn1 + 1 p0 ðr1 Þ ¼ 0
The latter relation results because not only pðr1 Þ ¼ 0 but also p0 ðr1 Þ ¼ 0.
The solution of the nonhomogeneous equation, gn 6¼ 0, is obtained as the
sum of the homogeneous solution uno and a particular solution of the nonhomo-
geneous equation unp , namely
un ¼ uno + unp (4.5.10)
In the special case where gn ¼ g ¼ constant, the particular solution is read-
ily obtained. Indeed, introducing un ¼ G ¼ constant in Eq. (4.5.4) yields
ð1 + a1 + a2 + … + ak ÞG ¼ g
g
G¼ (4.5.11)
1 + a1 + a2 + … + ak
provided that 1 + a1 + a2 + … + ak does not vanish.
Example 4.5.1 Solution of a second-order difference equation
Find the solution of the IVP
un + 2 2un + 1 + 2un ¼ 0 (1)
u0 ¼ 0, u1 ¼ 1 (2)
Solution
The characteristic equation of (1) is
r2 2r + 2 ¼ 0 (3)
@CivilMethod
Its roots are r1 ¼ 1 + i, r2 ¼ 1 i and Eq. (4.5.8) give

pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi
r ¼ 12 + 12 ¼ 2, q ¼ tan 1 ð1Þ ¼ p=4
Hence, Eq. (4.5.9) gives
np np
un ¼ 2n=2 c1 cos + c2 sin (4)
4 4
Applying the initial conditions gives c1 ¼ 0, c2 ¼ 1 and the solution becomes
np
un ¼ 2n=2 sin (5)
4
Example 4.5.2 Solution of a third-order difference equation

Find the solution of the IVP
un + 3 2un + 2 un + 1 + 2un ¼ 0 (1)
u0 ¼ 0, u1 ¼ 1, u2 ¼ 1 (2)
Solution
The difference equation is of the third order. Its characteristic equation is
r3 2r2 r + 2 ¼ 0 (3)
Its roots are þ1, 1, 2. Hence the general solution of (1) is
un ¼ c1 ð1Þn + c2 ð1Þn + c3 ð2Þn
(4)
¼ c1 + ð1Þn c2 + 2n c3
Applying the initial conditions gives the following system for the arbitrary
constants c1 ,c2 , c3
c1 + c2 + c3 ¼ 0
c1 c2 + 2c3 ¼ 1 (5)
c1 + c2 + 4c3 ¼ 1
which is solved to give c1 ¼ 0, c2 ¼ 1=3, c3 ¼ 1=3. Therefore, we obtain the
closed form solution for the difference equation
1 2n
un ¼ ð1Þn + (6)
3 3
4.5.3 Difference equations and stability of the numerical

integration methods
Along with the widespread use of computers for solving differential equations,
it was observed that some known integration schemes lead to errors in the
solution that are much greater than expected due to discretization. Moreover,
these errors for a certain value of the variable t increased, although the time step
@CivilMethod
Dt ¼ h was reduced. To understand this behavior, we examine the solution of

the differential equation
u_ ¼ 2u + 1, u ð0Þ ¼ 1 (4.5.12)
Using the central difference to approximate the derivative, we obtain the
difference equation
un + 1 + 4hu n un1 ¼ 2h (4.5.13)
where h is the constant time step of the integration.
The solution of the previous difference equation is
1
un ¼ c1 rn1 + c2 rn2 + (4.5.14)
2
where r1 and r2 are the roots of the characteristic equation
r2 + 4hr 1 ¼ 0
These roots are
r1, 2 ¼ 2h 1 + 4h 2
Expanding 1 + 4h 2 in the Taylor series and keeping only the linear terms,
the roots are expressed as
r1 ¼ 1 2h (4.5.15a)
r2 ¼ ð1 + 2h Þ (4.5.15b)
and Eq. (4.5.14) is written
1
un ¼ c1 ð1 2h Þn + c2 ð1Þn ð1 + 2h Þn + (4.5.16)
2
It is known from differential calculus that
‘im ð1 + eÞ1=e ¼ e
e!0
Using this relation and taking into account that n ¼ tn =h, we obtain for a
given tn
‘im ð1 + 2h Þn ¼ ‘im ð1 + 2h Þð1=2hÞð2tn Þ ¼ e2tn

h!0 h!0
In the same way, we obtain

‘im ð1 2h Þn ¼ e2tn
h!0
Consequently, for h ! 0 the solution (4.5.16) becomes

2tn 1
u n ¼ c1 e + + c2 ð1Þn e2tn (4.5.17)
2
@CivilMethod
On the other hand, the exact solution of Eq. (4.5.12) is
1
un ¼ c1 e2tn + (4.5.18)
2
Evidently, the first term in Eq. (4.5.17) is the exact solution. The second
term is spurious (extraneous) and results from the fact that the first-order dif-
ferential equation is substituted by a second-order difference equation. The
application of the initial conditions would give c2 ¼ 0 if the computations were
exact. In practice, however, errors are introduced, which are mainly due to the
rounding of numbers or the inaccuracy of the starting value. Therefore, the con-
stant c2 is not exactly zero and consequently, a small error is introduced in each
integration step. This is magnified because it is multiplied by the factor
ð1Þn e2tn , which increases exponentially. Because the first term of the solution
(4.5.17) diminishes exponentially, the introduced error due to the spurious solu-
tion dominates the exact solution and leads to a totally wrong result. We
describe a method as unstable if the error increases exponentially with tn .
For the first order differential equations, the one-step integration methods do
not exhibit instability for small values of h. The multistep methods, however,
which lead to difference equations of order greater than one, introduce spurious
solutions and they may be unstable either for all values of h or for a certain
region of the values of h. In order to decide whether a multistep method is sta-
ble, we work as follows.
If the multistep method leads to a difference equation of order k, we find the
roots of the characteristic equation. If ri (i ¼ 1,2, …,kÞ are these roots, the gen-
eral solution will be
un ¼ c1 rn1 + c2 rn2 + … + ck rnk
One of the solutions, say rn1 , will tend to the actual solution of the differential
equation. The remaining roots are spurious. We will say that a multistep method
is strongly stable if for h ! 0 the spurious roots satisfy the condition
jri j < 1, i ¼ 2, 3, …, k (4.5.19)
Because we do not know which is the actual solution, the above condition
should apply to all roots ri . Apparently, this condition ensures that the error
diminishes as n increases. On the contrary, the error increases exponentially
if jri j > 1.
4.5.4 Stability of the central difference method

As was shown in Section 4.5.2, the CDM leads to the difference equation
(4.5.3), namely
un + 1 2a1 un + a2 un1 ¼ gn (4.5.20)
@CivilMethod
where
2 h 2 w2 1 hwx pn h 2
a1 ¼ , a2 ¼ , gn ¼ (4.5.21)
2ð1 + hwx Þ 1 + hwx m ð1 + hwx Þ
The characteristic equation of Eq. (4.5.20) is
r2 2a1 r + a2 ¼ 0 (4.5.22)
whose roots are
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r1, 2 ¼ a1 a21 a2
The type of root depends on the sign of the discriminant D ¼ a21 a2 , which
may be written as
pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi
s2 s 2 1 x2 s + 2 1 x2
D ðs Þ ¼ , s ¼ hw
4ð1 + sxÞ2
pffiffiffiffiffiffiffiffiffiffiffiffis > 0, the sign of the discriminant depends only on the factor
Because
s 2 1 x2 . Hence, we distinguish the following two cases
qffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffi
2 1 x2 T
ðiÞ Dðs Þ > 0 s > 2 1 x 2
or h > ¼ 1 x2 (4.5.23)
w p
qffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffi
2 1 x2 T
ðiiÞ Dðs Þ 0 s 2 1 x 2
or h ¼ 1 x 2 (4.5.24)
w p
In case (i), the characteristic equation has two real roots, r1 , r1 , The stability
of the solution requires that jr1 j < 1 and jr2 j < 1. But it can be shown that
jr2 j > 1, hence jr1 j > 1. Consequently, the solution is unstable in this case.
In case (ii), the characteristic equation has two complex conjugate roots
r1, 2 ¼ a1 i a2 a21 (4.5.25)
This gives
sffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffi 1 sx
j r1 j ¼ j r2 j ¼ a2 ¼ 1 (4.5.26)
1 + sx
Consequently, the solution is stable in case (ii).

The stability criterion (4.5.26) is also explained by writing the complex roots
in exponential form
r1 ¼ reiqt , r2 ¼ reiqt (4.5.27)
where
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffi 1 hwx a2
r ¼ a2 ¼ 1 and q ¼ tan 1 1 (4.5.28)
1 + hwx a21
@CivilMethod
Then solution (4.5.20) is written as

pffiffiffiffiffi n
un ¼ ð a2 Þ ðc1 cos nq + c2 sin nqÞ + unp (4.5.29)
pffiffiffiffiffi
which is bounded because a2 1.
The previous investigation leads to the conclusion that the CDM is stable if
the time step satisfies the condition.
T T
Dt < 1 x2 (4.5.30)
p p
This condition imposes a constraint on the size of the time step. In such a
case, we say that the numerical integration method is conditionally stable. It
should be noted that this condition is never a restriction for SDOF systems
because the accuracy of the results requires choosing a much smaller time step,
usually Dt ¼ 0:1T =p is adequate.
4.5.5 Stability of the average acceleration method

We consider the equation of motion (4.4.1) at time t ¼ nt
u€n + 2xwu_ n + w2 un ¼ pn , pn ¼ pn =m (4.5.31)
and the approximate expressions (4.3.5) and (4.3.4) for the displacement and
velocity at the same instant, namely
h2
un ¼ un1 + h u_ n + ½u€n1 + u€n (4.5.32)
4
h
u_ n ¼ u_ n1 + ½u€n1 + u€n (4.5.33)
2
The previous three equations can be combined as
2 3 2 3
1 2xw w2 8 9 0 0 0 8 9 8 9
6 h2 7< u€n = 6 h 2 7< u€n1 = < 1 =
6 7 6 7
6 0 1 7 u_ 6 h 1 7 u_ n1 + 0 pn
6 4 7: n ; ¼ 6 4 7: (4.5.34)
4 h 5 un 4h 5 un1 ; : 0 ;
1 0 1 0
2 2
Solving for the vector Un ¼ f u€n u_ n u_ n gT , we obtain
Un ¼ AUn1 + b
pn , n ¼ 1, 2, …,N (4.5.35)
where
2 31 2 3
1 2xw w2 0 0 0
6 h2 7 6 h2 7
6 7 6 7
6
A¼6 4 0 1 7 6 h 1 7 (4.5.36a)
7 64 7
4 h 5 4h 5
1 0 1 0
2 2
@CivilMethod
2 31
1 2xw w2 8 9
6 7 >
> 1>
6 h2 7 < > =
6 1 7
b ¼ 6 4 0 7 0 (4.5.36b)
6 7 >
> >
4 h 5 : > ;
0
1 0
2
Applying Eq. (4.5.35) for n ¼ 1, 2, … we have
U1 ¼ AUo + b
p1
U2 ¼ AU1 + b
p2
¼ AðAUo + b
p1 Þ + b
p2
(4.5.37)
¼ A2 Uo + Ab
p1 + b
p2
⋯ ¼ ⋯⋯⋯⋯⋯⋯⋯⋯⋯

Un ¼ An Uo + An1 p1 + An2 p2 + …A0 pn b
The matrix A is known as the amplification matrix. The stability of the
method requires that An is bounded. This is true if the spectral radius of A sat-
isfies the condition [15]
rðAÞ ¼ max fjr1 j, jr2 j, jr3 jg 1 (4.5.38)
where ri (i ¼ 1, 2, 3) are the eigenvalues of the matrix A.
Using a symbolic language (here MATLAB) we find
4 s2 + 4s x2 1 4 s 2 4s x2 1
r1 ¼ , r2 ¼ , r3 ¼ 0 (4.5.39)
4 + 4sx + s 2 4 + 4sx + s 2
where s ¼ xw.
The type of the roots r1 , r2 depends on the sign of the discriminant
DðxÞ ¼ x2 1. Hence, we distinguish the following two cases
(i) If D ðx Þ > 0, both eigenvalues are real. It can be shown that jr1 j < 1, hence
jr2 j < 1. Therefore, the method is stable
(ii) If DðxÞ 0, the eigenvalues are complex conjugate
4 s 2 i4s 1 x2
r1,2 ¼
4 + 4sx + s 2
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð4 s 2 Þ2 + 16s2 1 x 2 ðs 2 + 4Þ2 16s 2 x2
j r1 j ¼ j r2 j ¼ ¼ 1 (4.5.40)
ð4 + 4sx + s 2 Þ ½ðs 2 + 4Þ + 4sx 2
The equality is valid for x ¼ 0.
@CivilMethod
The conclusion is that the AAM is stable without imposing any constraint on
the size of the time step. We say in this case that the integration method is
unconditionally stable.
Essentially, the procedure based on the condition (4.5.38) to prove the sta-
bility of the AAM is not different from that presented in Section 4.5.3, where the
stability results from the response of the difference equation. This is shown in
what it follows.
We write Eq. (4.5.35) for tn ,tn1 , tn2
Un AUn1 ¼ b
pn
Un1 AUn2 ¼ b
pn1 (4.5.41)
Un2 AUn3 ¼ b
pn2
or in matrix form
8 9
2 3> Un > 8 9
I A 0 >
0 < >
= < pn =
4 0 I A 0 5 Un1
¼ b pn1 (4.5.42)
> Un2 > : ;
0 0 I A >
: >
; pn2
Un3
Eq. (4.5.42), beside the displacements un ,un1 ,un2 , un3 , contains the
velocities u_ n , u_ n1 , u_ n2 , u_ n3 and the accelerations u€n , u€n1 , u€n2 , u€n3 . Reor-
dering these equations and eliminating the velocities and accelerations yield the
equation
un 2a1 un1 + a2 un2 + a3 un3 ¼ c1 pn + c2 pn1 + c3 pn2 (4.5.43)
where
4 w2 h 2 4 4sx + w2 h 2
a1 ¼ , a2 ¼ , a3 ¼ 0 (4.5.44)
4a 4a
1 + xwh 1 xwh
c1 ¼ , c2 ¼ , c3 ¼ 0 (4.5.45)
a a
where
w2 h 2
a ¼ 1 + xwh +
4
Eq. (4.5.43) is a difference equation whose characteristic equation is
r3 2a1 r2 + a2 r ¼ 0 (4.5.46)
The roots of Eq. (4.5.46) are
4 s 2 + 4s x 2 1 4 s 2 4s x2 1
r1 ¼ , r2 ¼ , r3 ¼ 0 (4.5.47)
4 + 4sx + s 2 4 + 4sx + s 2
that is, they are identical to the eigenvalues of the matrix A.
@CivilMethod
4.5.6 Stability of the analog equation method

The stability of the AEM is studied by considering the eigenvalues of the ampli-
fication matrix A given by Eq. (4.4.20b). Hence, the stability requires the valid-
ity of Eq. (4.5.38). Using a symbolic language (here MATLAB) we obtain the
three eigenvalues
4 s2 + 4s x2 1 4 s 2 4s x2 1
r1 ¼ , r2 ¼ , r3 ¼ 0 (4.5.48)
4 + 4sx + s 2 4 + 4sx + s 2
where s ¼ xw. Obviously, they are identical to those given by Eq. (4.5.39).
Consequently, we can state that the AEM is unconditionally stable.
4.6 Accuracy of the numerical integration

The methods for numerical integration of the equation of motion found in the
literature are numerous. They are either unconditionally stable or conditionally
stable. In the latter case, wDt must be less than a certain value, which is greater
than one, for example, in the method of central differences this value is
2 1 x2 . However, the accuracy of the results requires values of wDt much
smaller than one. Therefore, the limitation of ensuring the stability of the solu-
tion is not decisive for the choice of the integration step, at least for the SDOF
systems. Hence, the choice of method should be based on the relative accuracy
of the numerical results. Actually, the accuracy of the numerical method
depends on the truncation error in the Taylor series representing the derivatives
(for example, CDM) or the approximation of the second derivative within the
time step (for example, AAM, AEM).
For free vibrations, as a measure of the relative accuracy, we define the
quantities
T T
PE ¼ (4.6.1)
T
and
rn rn + 1
AD ¼ (4.6.2)
rn
where T and T are the exact and the approximate periods, respectively, and
rn ,rn + 1 are the amplitudes at the consecutive times tn ,tn + 1 .
The first relation expresses the period elongation (PE) while the second the
amplitude decay (AD) over time.
The numerical solution obtained by the AAM or the AEM can be written in
terms of the eigenvalues
un ¼ c10 rn1 + c20 rn2 ð r 3 ¼ 0Þ (4.6.3)
@CivilMethod
or
un ¼ r n ðc1 sin nq + c2 cos nqÞ
(4.6.4)
n + c2 cos wt
¼ r n ðc1 sin wt nÞ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where r ¼ a 2 + b2 , q ¼ tan 1 ðb=a Þ, a ¼ Reðr1 Þ, b ¼ Imðr1 Þ, w ¼ q=h,
tn ¼ nh.
The corresponding exact solution is
un ¼ exwtn ðc1 sin wD tn + c2 cos wD tn Þ, wD ¼ 1 x2 , tn ¼ nh (4.6.5)
Comparison of Eqs. (4.6.4), (4.6.5) could show the accuracy of the numer-
ical scheme. Thus, the period elongation is
T T h 1 x2
PE ¼ ¼ 1 (4.6.6)
T q
For the amplitude decay, we can define an equivalent damping ratio x from
the relation

n
r n ¼ exwt ¼ exqn (4.6.7)
which gives
x ¼ ln r=q (4.6.8)
The difference Dx ¼ x x can be employed as a measure for the amplitude
decay. The dependence of the period elongation and amplitude decay on h=T is
shown in Figs. 4.6.1 and 4.6.2, respectively. Obviously, for small values of h=T
the scheme is accurate. Note that for x ¼ 0 it is jr2 j ¼ jr3 j ¼ r ¼ 1 and Eq. (4.6.8)
yields x ¼ 0. That is, there is no amplitude decay.
0.25
x=0
0.2 x = 0.1
Period elongation %
x = 0.2
0.15
0.1
0.05
0
0 0.05 0.1 0.15 0.2 0.25 0.3
h/T
FIG. 4.6.1 Period elongation versus h=T for different values x.
@CivilMethod
−3
x 10
2
x=0
0 x = 0.1
x = 0.2
−2
Δx
−4
−6
−8
0 0.05 0.1 0.15 0.2 0.25 0.3
h/T
FIG. 4.6.2 Amplitude decay Dx ¼ x x versus h=T for different values of x.
Example 4.6.1 Stability of Houbolt’s method

Houbolt’s method for the numerical integration of the equation of motion in the
absence of damping requires the solution of the difference equation [4]

2 + s 2 un + 1 5un + 4un1 un2 ¼ 0, s ¼ wh (1)
Investigate the stability of the method.
Solution
The characteristic equation of Eq. (1) is

2 + s 2 r3 5r2 + 4r 1 ¼ 0 (2)
Its roots are

1 b 12a 25
2
r1 ¼ +5 (2)
3a 2 b
pffiffiffi
1 b 12a 25 3 b 12a 25
r2,3 ¼ + +5 i +2 (3)
3a 4 b 6a 2 b
where

a ¼ 2 + s2 (4)
pffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1=3
b ¼ 720a + 108a 2 + 1000 + 12 3a 104a + 100 + 27a 2 (5)
The stability condition requires that jr1 j < 1 and jr2 j ¼ jr3 j < 1. This
condition is satisfied, as is shown in Fig. E4.7.
@CivilMethod
1
|r1|
0.9
|r2|
0.8 |r3|
0.7
0.6
0.5
|r2|=|r3|
0.4
0.3
0.2
0.1
| r1| (spurious)
0
1 2 3 4 5 6 7 8 9 10
h/t
FIG. E4.7 Houbolt’s method. Roots of the characteristic equation.
4.7 Problems
Problem P4.1 Find the solution of the IVPs for the difference equations
un + 1 3un ¼ 5, u0 ¼ 1
un + 2 4un + 1 + 3un ¼ 2n , u0 ¼ 0, u1 ¼ 1
Hint: The particular solution of (ii) will be sought in the form gn ¼ 2n c,
where c is a constant.
Problem P4.2 A SDOF system is subjected to pulse the load

t
pðt Þ ¼ p0 1 1 + et=t1 , 0 t t1
t1
pðt Þ ¼ 0, t > t1
Study the response of the system using
(i) The central difference method.
(ii) The average acceleration method.
(iii) The analog equation method.
Compare the results with the exact solution.
Data: u ð0Þ ¼ u_ ð0Þ ¼ 0, m ¼ 50kNm1 s2 , x ¼ 0:05, w ¼ 5s1 , p0 ¼ 100kN,
t1 ¼ 0:5s and ttot ¼ 10s.
Problem P4.3 A SDOF system is subjected to the piecewise linear load pðt Þ
shown in Fig. P4.3. Use the three discussed numerical methods to establish
its response.
@CivilMethod
Compare the results with the exact solution. Data: u ð0Þ ¼ u_ ð0Þ ¼ 0,
m ¼ 50kN m1 s2 , x ¼ 0:05, w ¼ 5s1 , ti ¼ 0:01i, pi ¼ 20 1 + ð1Þi ði + 5Þ=

ði + 1Þ , i ¼ 0, 2, …,100.
FIG. P4.3 Piecewise linear load in Problem P4.3.

[1] M.A. Dokainish, K. Subbaraj, A survey of direct time-integration method in computational
structural dynamics. I. Explicit methods, Comput. Struct. 32 (1989) 1371–1386.
[2] K. Subbaraj, M.A. Dokainish, A survey of direct time-integration methods in computational
structural dynamics. II. Implicit methods, Comput. Struct. 32 (1989) 1387–1401.
[3] C.H. Norris, Structural Design for Dynamic Loads, McGraw-Hill, New York, 1959.
[4] K.J. Bathe, E.L. Wilson, Numerical Methods in Finite Elements, Prentice-Hall, Englewood
Cliffs, NJ, 1976.
[5] N.M. Newmark, A method of computation of structural dynamics, J. Eng. Mech. 85 (1959)
67–94.
[6] J.T. Katsikadelis, A new direct time integration method for the equations of motion in struc-
tural dynamics, ZAMM Z. Angew. Math. Mech. 94 (9) (2014) 757–774, https://doi.org/
10.1002/zamm.20120024.
[7] K.J. Bathe, Conserving energy and momentum in nonlinear dynamics: a simple implicit time
integration scheme, Comput. Struct. 85 (2007) 437–445.
[8] J.T. Katsikadelis, The analog equation method–a powerful BEM-based solution technique
for solving linear and nonlinear engineering problems, in: Transactions on Modelling and
Simulation, vol. 7, WIT Press, pp. 167–182, www.witpress.com, 1743-355X.
[9] J.T. Katsikadelis, The analog boundary integral equation method for nonlinear static and
dynamic problems in continuum mechanics, J. Theor. Appl. Mech. 40 (4) (2002).
[10] J.T. Katsikadelis, The Boundary Element Method for Plate Analysis, Academic Press,
[11] J.T. Katsikadelis, The Boundary Element for Engineers and Scientists, Academic Press,
[12] F.B. Hildebrand, Advanced Calculus for Applications, Prentice-Hall, Inc., Englewood Cliffs,
NJ, 1962.
[13] J.L. Humar, Dynamics of Structures, second ed., A.A. Balkema Publishers, Lisse, NL, 2002.
[14] F. Scheid, Numerical Analysis, second ed., Schaum’s Outline Series, McGraw-Hill, New
York, 1998.
[15] U.N. Faddeeva, Computational Methods in Linear Algebra, Dover Publications, New York,
1959.
@CivilMethod
Chapter 5
Nonlinear response:
systems
Chapter outline
5.1 Introduction 191 5.4 The analog equation
5.2 The central difference method 195 method 203
5.3 The average acceleration 5.5 Problems 211
method 197 References and further reading 215
5.1 Introduction
The equation of motion of a vibrating system expresses the equilibrium condi-
tion of all forces applied to the system, namely the external excitation force, the
inertial force, the damping force, and the elastic force. The equilibrium condi-
tion reads
fI ðt Þ + fD ðt Þ + fS ðt Þ ¼ pðt Þ (5.1.1)
The forces fI , fD , and fS depend on the physical properties of the system.
In the systems we analyzed, the physical properties are not time-dependent
and the dependence of these forces on the cause that produces them is linear,
that is,
fI ¼ m u€ (5.1.2a)
fD ¼ cu_ (5.1.2b)
fS ¼ ku (5.1.2c)
where m, c,k are constant quantities. Systems with such a physical response are
referred to as linear systems.
In general, however, Eq. (5.1.2a)-(5.1.2c) may be of the form
fI ¼ m ðt Þu€ (5.1.3a)
_ tÞ
fD ¼ fD ðu, u, (5.1.3b)
_ tÞ
fS ¼ fS ðu, u, (5.1.3c)

@CivilMethod
That is, the mass may vary with time and the damping and elastic forces may
be nonlinear functions of u, u, _ t Þ, fS ðu, u,
_ and t, that is, fD ðu, u, _ t Þ. In this case,
Εq. (5.1.1) takes the form
_ t Þ + fS ðu, u,
m ðt Þu€ + fD ðu, u, _ t Þ ¼ pðt Þ (5.1.4)
which is a nonlinear differential equation of the second order. A dynamic sys-
tem, whose response is described by Eq. (5.1.4), is referred to as nonlinear.
Although systems with variable mass are not unusual [1], our discussion will
be limited to systems with constant mass. Besides, the forces fD and fS will
be considered of the form fD ðu_ Þ and fD ðu Þ. Thus, Eq. (5.1.4) becomes
m u€ + fD ðu_ Þ + fS ðu Þ ¼ pðt Þ (5.1.5)
We distinguish two types of nonlinearity: the geometric nonlinearity, which
is due to large displacements implying large deformations of the structure, and
the material nonlinearity, which is due to nonlinear constitutive equations (e.g.,
hyperelastic or elastoplastic materials). Of course, both types of nonlinearity
can simultaneously characterize the response of a system.
The analytical solution of the nonlinear equations of motion is a difficult and
complicated mathematical problem. Exact solutions are available only for a few
cases and for differential equations of a specific form [2, 3]. The existing solu-
tions aim rather at a qualitative study of the response of the system described by
a nonlinear equation than at offering a computational means for practical ana-
lyses. The knowledge of the nonlinear response of the single-degree-of-freedom
(SDOF) systems comes from approximate methods, and mainly from numerical
methods. Therefore, the recourse to numerical methods to solve the nonlinear
equations of motion is inevitable. The step-by-step methods play a dominant
role. The Runge-Kutta methods, usually employed for the solution of nonlinear
equations, belong to these methods [4].
The dynamic response of nonlinear systems can be studied effectively by
demanding the fulfillment of equation motion (5.1.5) at discrete time instants
Dt apart by the use of the step-by-step integration methods we discussed in
Chapter 4. These methods as developed for nonlinear equations of motion
are presented directly below while for the analytical methods, the reader is
advised to look in the vast related literature [2, 3, 5].
Example 5.1.1 Systems with a geometrical nonlinearity
Derive the equation of motion of the system shown in Fig. E5.1. The supports at
A,B and the interconnection at C are hinges. The mass m at C is concentrated.
The system is set to motion by the initial conditions u0 , u_ 0 and/or the vertical
external force pðt Þ. The bars are assumed massless.
@CivilMethod
Nonlinear response: Single-degree-of-freedom systems Chapter 5 193
l l
C
A φ B
u(t )
E, A E, A
m
p(t )
fI
S fS S
p(t )
FIG. E5.1 SDOF system with nonlinear response.
Solution
During the motion, the force pðt Þ, the inertial force fI , and the elastic forces of
the bars are in equilibrium
The inertial force is given by the relation
fI ¼ m u€ (1)
The total elastic force is caused by the elongation of the bars and is given by
fS ¼ 2S sin f (2)
where
EA
S¼ d (3)
l
E is the modulus of elasticity of the material of the bar, A its cross-sectional area,
and d the elongation of the bars.
The elongation of the bars at time t is
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d¼ l 2 + u2 l (4)
and
u
sin f ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5)
l 2 + u2
@CivilMethod
Hence, the elastic force in Eq. (2) by virtue of Eqs. (3)–(5) is expressed as
2 3
u6 1 7
fS ðu Þ ¼ 2EA 41 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi5 (6)
l 2
1 + ðu=l Þ
and the equation of motion (5.1.1) in the absence of damping becomes

2 3
u6 1 7
m u€ + 2EA 41 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi5 ¼ pðt Þ (7)
l
1 + ðu=l Þ2
Obviously, the equation of motion is nonlinear. Fig. E5.2 shows the graph
of fS ðu Þ. Because d 2 fS =du 2 > 0 the curve is concave upward. In this system,
the slope kT increases continually, which implies that the elastic force
increases with increasing u. In this case, we say that the system exhibits hard-
ening, in contrast to other systems that exhibit softening. In the latter systems,
the curve fS ðu Þ is concave downward (d 2 fS =du 2 < 0), that is, the slope
kT decreases continually, which implies that elastic force decreases with
increasing u, for example, a system with stiffness fS ¼ 40u u 3 exhibits soft-
ening; see Fig. E5.3.
FIG. E5.2 System with hardening.
@CivilMethod
FIG. E5.3 System with softening.
5.2 The central difference method

We consider the case where the damping force depends linearly on the velocity
(viscous damping) and the elastic force depends nonlinearly on the displace-
ment, then Eq. (5.1.5) becomes
m u€ + cu_ + fS ðu Þ ¼ pðt Þ (5.2.1)
The central difference method (CDM) developed for linear equations can be
readily adjusted to solve this equation explicitly. Thus, using Eqs. (4.2.4),
(4.2.5) to replace the derivatives in Eq. (5.2.1), we obtain
un + 1 2un + un1 un + 1 un1
m 2
+c + fS ðun Þ ¼ pn (5.2.2)
h 2h
or solving for un + 1 gives
K^ un + 1 ¼ P^ (5.2.3)
where
m c
K^ ¼ 2 + (5.2.4)
h 2h
2m m c
P^ ¼ pn fS ðun Þ + 2 un 2 un1 (5.2.5)
h h 2h
Eq. (5.2.3) is used to compute un + 1 from the displacements at the two preceding
instants. The value of u1 is computed from Eq. (4.2.10), namely
1
u ðDt Þ u ð0Þ Dt u_ ð0Þ + Dt 2 u€ð0Þ (5.2.6)
2
@CivilMethod
in which u€0 results now from Eq. (5.2.1) for t ¼ 0

u€0 ¼ ½p0 cu_ 0 fS ðu0 Þ=m (5.2.7)
Obviously, the numerical algorithm presented in Table 4.2.1 can be employed
to write a computer code for the nonlinear response of the SDOF systems after
modifying appropriately to include the nonlinear restoring force fS ðu Þ. On the
basis of this algorithm, a computer program called centr_diff_nlin.m has been
written in MATLAB for the numerical integration of the nonlinear equation of
motion using the CDM. The program is available on this book’s companion
website.
Example 5.2.1 Simple pendulum. CDM solution
Solve the initial value problem (IVP) describing the motion of the simple pen-
dulum (Fig. E5.4) under its own weight, namely
g
q€ + sin q ¼ 0 (1)
l
qð0Þ ¼ q0 (2a)
q_ ð0Þ ¼ q_ 0 (2b)
where qðt Þ represents the angle of the pendulum from the vertical position, l is
its length, and g the acceleration of gravity. The equation of motion (1) can be
readily derived using the method of Lagrange’s equations (see Section 1.8).
Eq. (1) admits an exact solution [6]
npffiffiffi hpffiffiffiffiffiffiffi io
qðt Þ ¼ 2 sin1 k sn g=l ðt + T0 Þ; k (3)
where k ¼ sin 2 ðq0 =2Þ and T0 is the quarter of the period; sn represents the
sn-Jacobean elliptic function [7].
Solution
The response of the pendulum for l ¼ g, q0 ¼ 0:40p, and q_ 0 ¼ 0 is obtained using
the program centr_diff_nlin.m with Dt ¼ 0:01. It is shown in Fig. E5.5 as com-
pared with the exact one.
FIG. E5.4 Simple pendulum.
@CivilMethod
t t
FIG. E5.5 Response of the simple pendulum in Example 5.2.1.
5.3 The average acceleration method

The average acceleration method (AAM) presented in Section 4.3 can be mod-
ified to solve the nonlinear equation of motion, Eq. (5.2.1). Although this
method becomes unstable for long-duration motions [9], it is presented here
because it is still widely used for the solution of the nonlinear equation of
motion.
At time t + Dt this equation is written
m u€ðt + Dt Þ + fD ðt + Dt Þ + fS ðt + Dt Þ ¼ pðt + Dt Þ (5.3.1)
Subtracting Eq. (5.2.1) from the foregoing equation yields
mDu€ + DfD + DfS ¼ Dp (5.3.2)
(a) (b)
FIG. 5.3.1 Tangent and secant of (a) damping and (b) stiffness curve.
Expanding fD ðt + Dt Þ and fS ðt + Dt Þ in the Taylor series gives

DfD ¼ fD ðt + Dt Þ fD ðt Þ
¼ fD ðu_ + Du_ Þ fD ðu_ Þ
(5.3.3)
df 1 d 2 fD
¼ D Du_ + ðDu_ Þ2 + ⋯
d u_ 2 d u_ 2
@CivilMethod
DfS ¼ fS ðt + Dt Þ fS ðt Þ
¼ fS ðu + Du Þ fS ðu Þ
(5.3.4)
df 1 d 2 fS
¼ S Du + ðDu Þ2 + ⋯
du 2 du 2
For small values of Dt, the quantities Du, Du_ are also small. Thus, neglect-
ing the nonlinear terms in Eqs. (5.3.3), (5.3.4), we obtain
DfD cT Du_ (5.3.5)
DfS kT Du (5.3.6)
Obviously, cT and kT express the slope of the tangent to the curves fD ðu_ Þ
and fS ðu Þ, respectively, at time t.
Referring to Fig. 5.3.1a, we have
fD ðt + Dt Þ ¼ fD ðt Þ + DfD
However, the exact value of DfD is
DfD ¼ Du_ tan fc
where fc is the angle of the secant. Hence, approximating DfD by Eq. (5.3.5)
introduces the error (see Fig. 5.3.1a)
eD ¼ ðcT tan fc ÞDu_ (5.3.7)
because cT is the slope of the tangent.
Similarly, the use of kT to approximate DfS introduces the error (see
Fig. 5.3.1b)
eS ¼ ðkT tanfk ÞDu (5.3.8)
The errors eD and eS cannot be avoided because u_ ðt + Dt Þ and u ðt + Dt Þ
are not known at instant t + Dt. However, as we will show, they can be kept
under a given bound, which specifies the accuracy of the solution procedure.
By virtue of Eqs. (5.3.5), (5.3.6), Eq. (5.3.2) is written in incremental form
mDu€ + cT Du_ + kT Du ¼ Dp (5.3.9)
The previous equation is of the form (4.3.11). Hence, the AAM is suitable to
solve it. Thus, using Eqs. (4.3.9), (4.3.10) to express Du€ and Du,
_ we obtain
2
Du_ ¼ Du 2u_ ðt Þ (5.3.10)
Dt
4
Du€ ¼ ½Du Dt u_ ðt Þ 2u€ðt Þ (5.3.11)
Dt 2
Then Eq. (5.3.9) becomes

2cT 4m 4m
kT + + 2 Du ¼ Dp + + 2cT u_ ðt Þ + 2m u€ðt Þ (5.3.12)
Dt Dt Dt
@CivilMethod
or
k ∗ Du ¼ Dp∗ (5.3.13)
where
2cT 4m
k ∗ ¼ kT + + 2 (5.3.14)
Dt Dt
and

4m
Dp∗ ¼ Dp + + 2cT u_ n + 2m u€n (5.3.15)
Dt
The value of Du obtained from Eq. (5.3.13) is used in Eq. (5.3.10) to eval-
_ Then we obtain
uate Du.
un + 1 ¼ un + Du (5.3.16a)
u_ n + 1 ¼ u_ n + Du_ (5.3.16b)
The acceleration u€n + 1 is evaluated directly from Eq. (5.3.1)

1
u€n + 1 ¼ ½pn + 1 fD ðu_ n + 1 Þ fS ðun + 1 Þ (5.3.16c)
m
Apparently, the value u€n + 1 obtained from the above equation reduces the
error introduced by cT and kT . The study of the stability of the time step inte-
gration schemes for nonlinear differential equation of motion is an issue beyond
the scope of this book.
The described method gives good results if the slope of the tangent is close
to that of the secant. For this reason, it is recommended to check the following
errors at the end of each step

kT ksec cT csec

ek ¼ a, ec ¼ a (5.3.17)
kT cT
where
DfS DfD
ksec ¼ , csec ¼ (5.3.18)
Du Du_
and a is a small specified number defining the upper bound of the error, for
example, a ¼ 0:01. If it is
min fek , ec g > a (5.3.19)

then we must reduce the time step Dt and repeat the computations. Table 5.3.1
presents the algorithm for the numerical solution of the nonlinear equation of
motion (5.1.5) using the AAM.
@CivilMethod
TABLE 5.3.1 Average acceleration method. Nonlinear equation of motion.

A. Data
m, fD ðu_ Þ, fS ðu Þ, pðt Þ, u ð0Þ, u_ ð0Þ, ttot , a
u€ð0Þ ¼ fpð0Þ fD ½u_ ð0Þ fS ½u ð0Þg=m
Select: Dt and set t ¼ 0 and
C. In each step compute
1. kT ¼ ðdf S =du Þt cT ¼ ðdf D =d u_ Þt
2. k ∗ ¼ kT + 2cT =Dt + 4m=Dt 2 , Dp ¼ pðt + Dt Þ pðt Þ
3. p∗ ¼ Dp + ð4m=Dt + 2cT Þu_ ðt Þ + 2m u€ðt Þ
4. Du ¼ p∗ =k ∗ , Du_ ¼ 2Du=Dt 2u_ ðt Þ
5. u_ ðt + Dt Þ ¼ u_ ðt Þ + Du_
u€ðt + Dt Þ ¼ fpðt + Dt Þ fD ½u_ ðt + Dt Þ fS ½u ðt + Dt Þg=m

S =Du c Df =Du_
ek ¼ kT Df
kT , ec ¼ T cTD
6. Check: if max fec , ek g < a go to the next time step. Else set t ¼ t + Dt and check
if t > ttot end. Else set u ðt Þ ¼ u ðt + Dt Þ, u_ ðt Þ ¼ u_ ðt + Dt Þ, u€ðt Þ ¼ u€ðt + Dt Þ, and
go to C.1
When the damping force depends linearly on the velocity, then the response
is governed by Eq. (5.2.1). In this case, the previously presented incremental
method can be improved by employing an iterative procedure within each step,
which minimizes the error introduced by the tangent stiffness kT .
The starting point is Eq. (5.3.13), which we write as
k ∗ du ð1Þ ¼ Dp∗ (5.3.20)
The quantity du ð1Þ is the first approximation to Dun within the time step
from tn to tn + 1 . That is
Dunð1Þ ¼ du ð1Þ (5.3.21)
FIG. 5.3.2 Graph of the function p*(u).
The index n denotes the number of the step that brings us from the displace-
ment un to un + 1 . Fig. 5.3.2 presents the graph of the function p∗ ðu Þ. Apparently,
@CivilMethod
the displacement du ð1Þ resulting from Eq. (5.3.20) decreases Dp∗ by dpð1Þ .
Hence, there is a remaining force
DF ð2Þ ¼ Dp∗ dpð1Þ (5.3.22)
which must be equilibrated. The change dpð1Þ is computed using Eq. (5.3.20), if
the tangential slope kT is replaced with the slope of the secant
" #
ð1Þ fS un + du ð1Þ fS ðun Þ 2c 4m
dp ¼ + + du ð1Þ
du ð1Þ Dt Dt 2

2c 4m
¼ fS un + du ð1Þ fS ðun Þ + + 2 du ð1Þ
Dt Dt
which by virtue of Eq. (5.3.14) becomes

dpð1Þ ¼ fS un + du ð1Þ fS ðun Þ + ðk ∗ kT Þdu ð1Þ (5.3.23)
The remaining force DF ð2Þ produces an additional displacement du ð2Þ ,

which is computed from the relation
k ∗ du ð2Þ ¼ DF ð2Þ (5.3.24)
Hence, the new approximation to Dun is
Dunð2Þ ¼ du ð1Þ + du ð2Þ
(5.3.25)
¼ Dunð1Þ + du ð2Þ
Then, it is used to compute the new remaining force
DF ð3Þ ¼ DF ð2Þ dpð2Þ (5.3.26)
where

dpð2Þ ¼ fS un + Dunð2Þ fS un + Dunð1Þ + ðk ∗ kT Þdu ð2Þ (5.3.27)
The force DF ð3Þ produces the additional displacement du ð3Þ , which is com-
puted from the relation
k ∗ du ð3Þ ¼ DF ð3Þ (5.3.28)
Thus, the new approximation of Dun is
Dunð3Þ ¼ Dunð2Þ + du ð3Þ (5.3.29)
Consequently, for the i + 1 approximation it is
DF ði + 1Þ ¼ DF ðiÞ dpðiÞ (5.3.30)

dpðiÞ ¼ fS un + DunðiÞ fS un + Dunði1Þ + ðk ∗ kT Þdu ðiÞ (5.3.31)
@CivilMethod
k ∗ du ði + 1Þ ¼ DF ði + 1Þ (5.3.32)
Dunði + 1Þ ¼ DunðiÞ + du ði + 1Þ (5.3.33)
Note that for i ¼ 1 it must be set Dunð0Þ ¼ 0 and DF ð1Þ ¼ Dp∗ .
The iteration procedure is terminated after I iterations, if

du ði + 1Þ

e (5.3.34)
DunðI Þ
where e is a specified small number. Then we assume that the convergence has
been achieved. The value Dun ¼ DunðI Þ is considered exact and it is used to
compute un + 1 and u_ n + 1 , u€n + 1 . Subsequently, the procedure continues to
the next step. The iterative procedure within the time step from tn to tn + 1 is
summarized in Table 5.3.2. This procedure is known as the modified
Newton-Raphson method.
TABLE 5.3.2 Modified Newton-Raphson method for the minimization

of the error in the average acceleration method.
A. Initial values
ð0Þ ð0Þ
un + 1 ¼ un , fS ¼ fS ðun Þ, DRð1Þ ¼ Dp∗n
B. In each iteration i ¼ 1,2,3,… compute:
1. du ðiÞ ¼ DRðiÞ =k ∗
ði Þ ði1Þ
2. un + 1 ¼ un + 1 + du ðiÞ
ði Þ ði1Þ
3. DpðiÞ ¼ fS fS + ðk ∗ kT Þdu ðiÞ
ði + 1Þ
4. DR ¼ DR DpðiÞ
ði Þ
5. du ði + 1Þ ¼ DRði + 1Þ =k ∗

6. If du ði + 1Þ =DunðI Þ > e set i ¼ i + 1 and go to B.2
Adhering to the steps in Table 5.3.1, a computer program called av_acc_

nlin.m has been written in MATLAB for the solution of the nonlinear equation
of motion. The electronic version of the program is available on this book’s
companion website.
Example 5.3.1 Simple pendulum. AAM solution
Solve the IVP describing the motion of the simple pendulum (Fig. E5.4) under
its own weight. The motion is described by Eqs. (1), (2a), (2b) of Example 5.2.1.
Solution
The solution is obtained using the program av_acc_nlin.m with Dt ¼ 0:01. The
response is shown in Fig. E5.5 as compared with the exact one.
@CivilMethod
5.4 The analog equation method

The solution procedure developed in Section 4.4 for the linear equation of
motion can be straightforwardly extended to the nonlinear equation. The non-
linear IVP treated here is described by the equations
m u€ + fD ðu_ Þ + fS ðu Þ ¼ pðt Þ (5.4.1)
u ð0Þ ¼ u0 , u_ ð0Þ ¼ u_ 0 (5.4.2)
where fD ðu_ Þ and fS ðu Þ are nonlinear functions of their arguments.
Applying Eq. (5.4.1) for t ¼ nDt, we obtain
mq n + fD ðu_ n Þ + fS ðun Þ ¼ pn , qn ¼ u€n (5.4.3)
Moreover, Eqs. (4.4.14), (4.4.15) hold also, that is,
2 c1 3 2 c1 3
" #( ) " #( ) +
h 1 u_ n 0 1 u_ n1 6 27 6 2 7
¼ +4 5qn + 4
c2 5 n1
q (5.4.4)
1 0 un 1 0 un1 c2
+
2 2
TABLE 5.4.1 The analog equation method. Nonlinear equation of motion.

A. Data
Read: m, fD ðu_ Þ, fS ðu Þ, u0 , u_ 0 , pðt Þ, ttot
1. Select: h ¼ Dt and compute ntot
2. Compute: c1 ¼ h 2 =2, c2 ¼ h, q0 ¼ ½pðt Þ fD ðu_ 0 Þ + fS ðu0 Þ=m
C. Compute solution
for n ¼ 1 to ntot solve for fqn u_ n un gT the system of the nonlinear algebraic equations:
mq n + fD ðu_ n Þ + fS ðun Þ ¼ pn
2 c 3 2 c 3

+
1

1
h 1 u_ n 0 1 u_ n1 6 27 6 2 7
¼ + 4 c 5qn + 4 c 5qn1
1 0 un 1 0 un1 2 2
+
2 2
Eqs. (5.4.3), (5.4.4) constitute a system of three algebraic equations, one

nonlinear and the other two linear. They can be solved successively for
n ¼ 1, 2, … to yield the solution un and the derivatives u_ n , u€n ¼ qn at instant
t ¼ nh T . For n ¼ 1, the value q0 appears in the right side of Eq. (5.4.4). This
quantity is readily obtained from Eq. (5.4.1) for t ¼ 0. This yields
q0 ¼ ½pð0Þ fD ðu_ 0 Þ + fS ðu0 Þ=m (5.4.5)
Several procedures can be applied to solve this system of algebraic equa-
tions. For example, a simple procedure is to substitute un and u_ n from
Eq. (5.4.4) into Eq. (5.4.3) and solve the resulting nonlinear equation for qn .
This can be achieved by employing any ready-to-use subroutine for nonlinear
algebraic equations, for example, MATLAB function fsolve. The steps of the
@CivilMethod
solution procedure are presented in Table 5.4.1. Adhering to the steps of this
table, a computer program called aem_nlin.m has been written in MATLAB
for the numerical integration of the nonlinear equation of motion using the
AEM. The program is available on this book’s companion website. It computes
the displacement u ðt Þ, the velocity u_ ðt Þ, and the acceleration u€ðt Þ.
Example 5.4.1 The Duffing equation. AEM solution
Use the AEM to solve the IVP for the Duffing equation
u€ + 0:2u_ + u + u 3 ¼ pðt Þ (1)
u ð 0Þ ¼ 0 (2a)
u_ ð0Þ ¼ 1 (2b)
For
pðt Þ ¼ e0:1t ½ð0:01sin t 0:2cost sin t Þ 0:2ð0:1 sin t cos t Þ
+ sin t + e0:2t ð sin t Þ3
Eq. (1) admits an exact solution uexact ðt Þ ¼ e0:1t sin t.
Solution
The solution is obtained using the program aem_nlin.m with Dt ¼ 0:01. The
graph of the solution is shown in Fig. E5.6 as compared with the exact one.
1
Computed
0.8 Exact
error X103
0.6
0.4
0.2
−0.2
−0.4
−0.6
−0.8
0 5 10 15 20 25
t
FIG. E5.6 Solution u and error u uexact in Example 5.4.1.
@CivilMethod
Example 5.4.2 Response of a nonlinear system with hardening

Use the AEM to study the dynamic response of the SDOF system shown in
Fig. E5.1, when it is subjected to:
(i) An initial displacement u0 ¼ 0:05m from the position of static equilibrium.
(ii) External load p0 ¼ mg suddenly applied at time t ¼ 0.
(iii) External load pðt Þ ¼ 10 sin Wt.
Data: m ¼ 3:8722kN m1 s2 , E ¼ 2:1 108 kN=m2 , A ¼ 3:142 104 m2 ,
g ¼ 9:81m=s2 , l ¼ 3:00m.
Solution
(i) The static displacement ust produced by the weight of the body is calculated
from Eq. (7) of Example 5.1.1 for u€ ¼ 0 and pðt Þ ¼ mg, namely

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2EAðust =l Þ 1 1= 1 + ðust =l Þ2 ¼ mg (1)
which is solved for the data of the problem to yield ust ¼ 0:25m.
The external force is the constant weight of the body, that is, pðt Þ ¼ mg,
while the elastic force
fS ðust + u Þ (2)
where u ¼ u ðt Þ denotes the additional displacement due to the dynamic
response. Thus, the equation of motion becomes
m u€ + fS ðust + u Þ ¼ mg (3)
with initial conditions
u0 ¼ 0:05m, u_ 0 ¼ 0 (4)
It should be noted that due to the nonlinearity of the elastic force, the super-
position of the displacements does not apply. The computed response of the sys-
tem with Dt ¼ 0:01 is shown in Fig. E5.7.
(ii) In this case the equation of motion reads
m u€ + fS ðu Þ ¼ mg (5)
with initial conditions
u0 ¼ 0, u_ 0 ¼ 0 (6)
where u ¼ u ðt Þ denotes the total dynamic displacement from the undeformed
position. The computed response of the system with Dt ¼ 0:01 is shown in
Fig. E5.8.
(iii) In this case, the IVP becomes
m u€ + fS ðu Þ ¼ mg + 10sin Wt (7)
u0 ¼ 0, u_ 0 ¼ 0 (8)
The computed response with Dt ¼ 0:01 is shown in Fig. E5.9 for the ratio
W=w ¼ 0:1, where w ¼ EA=lm . In all cases, the results are compared with
those obtained by the CDM
@CivilMethod
u(t)
t
FIG. E5.7 Response in Example 5.4.2 (i).
u(t)
t
FIG. E5.8 Response in Example 5.4.2 (ii).
FIG. E5.9 Response in Example 5.4.2 (iii).
@CivilMethod
Example 5.4.3 Response of a nonlinear system with softening

Study the dynamic response of a SDOF system with the following data:
m ¼ 1kN m1 s2 , u ð0Þ ¼ 1m, u_ ð0Þ ¼ 1m=s, fS ¼ 2u 3 , and ttot ¼ 1s. Because
d 2 fS=du 2 ¼ 12u < 0 the system exhibits softening. The integration of the
equation of motion will be performed using all three methods, that is, CDM,
AAM, and AEM. The problem admits an exact solution u ðt Þ ¼ 1=ð1 + t Þ.
Solution
The IVP is
u€ 2u 3 ¼ 0 (1)
u0 ¼ 1, u_ 0 ¼ 1 (2)
The solution in the interval 0 t 1 for Dt ¼ 0:001 is shown in Fig. E5.10.
FIG. E5.10 Response of the SDOF system in Example 5.4.2.
@CivilMethod
Example 5.4.4 Response of an elastoplastic system

In this example, the dynamic response of a system with material nonlinearity is
studied. The one-story building (Fig. E5.11a) is modeled as a SDOF system
with mass m. The structure is subjected to the horizontal rectangular pulse load
pðt Þ of Fig. E5.11b. The material is elastoplastic in nature. This yields the force-
displacement relation shown in Fig. E5.12 due to the elastoplastic hinges pro-
duced by the horizontal load. Neglecting damping, study the response of the
system using both the AAM and AEM. Data: t1 ¼ 0:25s, fSy ¼ 69:68kN,
k ¼ 5360kN=m, m ¼ 36kN m1 s2 , and p0 ¼ 45kN.
(a) (b)
FIG. E5.11 System with a material nonlinearity in Example 5.4.4.
FIG. E5.12 Restoring force in Example 5.4.4.
Solution
It is
fSy 69:68
uy ¼ ¼ ¼ 0:013m ¼ yield displacement
k 5360:00
um ¼ maximum displacement where the velocity changes sign
uR ¼ um uy ¼ remaining plastic deformation
The equation of motion is
m u€ + fS ðu Þ ¼ pðt Þ
@CivilMethod
where fS is the stiffness of the structure. It is

fS ¼ ku if u < uy ¼ 0:013m
fS ¼ fSy y u u m
if u
fS ¼ fSy k ðum u Þ if um 2uy < u < um , u_ < 0
Moreover, we have
rffiffiffiffiffi
k 2p
w¼ ¼ 12:20s1 , T¼ ¼ 0:515s
m w
(i) Solution by the AAM.

We employ the AAM following the displacement step by step.
At instant t we have
pðt Þ fS
u€ ¼ ¼ 0:0278½pðt Þ fS
m
4m 4 36
k ∗ ¼ kT + 2 ¼ k + ¼ kT + 1, 440, 000, kT ¼ 0, u > uy
Dt 0:012
The numerical results obtained by the AAM with Dt ¼ 0:01 < T =10 are
shown in Table E5.1 for two different time steps.
(ii) Solution by AEM.
A computer program has been written in MATLAB for the evaluation of the
elastoplastic response of the SDOF system. The obtained numerical results
are shown in Table E5.1 for two different time steps.
Finally, the time history of response of the system as well as of the restoring
force are shown in Fig. E5.13. and Fig. E5.14.
TABLE E5.1 Numerical solution of elastoplastic system in Example 5.4.4.

u ðt Þ (cm)
Dt ¼ 0:01 Dt ¼ 0:001
t (s) AEM AAM AEM AAM
0.00 0.000 0.000 0.000 0.000

0.05 0.124 0.151 0.149 0.151
0.10 0.503 0.550 0.546 0.551
0.15 1.003 1.053 1.050 1.055
0.20 1.446 1.483 1.480 1.484
0.25 1.721 1.745 1.743 1.745
0.30 1.643 1.709 1.677 1.677

Continued
@CivilMethod
TABLE E5.1 Numerical solution of elastoplastic system in
Example 5.4.4.—cont’d
u ðt Þ (cm)
Dt ¼ 0:01 Dt ¼ 0:001
t (s) AEM AAM AEM AAM
0.35 1.134 1.234 1.177 1.175
0.40 0.378 0.487 0.423 0.421
0.45 0.353 0.262 0.312 0.313
0.50 0.796 0.744 0.763 0.763

0.55 0.791 0.786 0.768 0.766
0.60 0.341 0.374 0.324 0.321
0.65 0.393 0.343 0.408 0.412
0.70 1.147 1.107 1.164 1.167
0.75 1.649 1.642 1.671 1.673
0.80 1.718 1.757 1.747 1.747

0.85 1.330 1.410 1.363 1.362
0.90 0.624 0.726 0.659 0.658
0.95 0.145 0.049 0.111 0.113
1.00 0.701 0.637 0.663 0.663
FIG. E5.13 Displacement in Example 5.4.4.
@CivilMethod
FIG. E5.14 Restoring force in Example 5.4.4.
5.5 Problems
Problem P5.1 Show that the motion of the simple pendulum is governed by
the IVP
g
q€ + sin q ¼ 0, qð0Þ ¼ q0 , q_ ð0Þ ¼ q_ 0 (1)
l
Solve the equation of motion numerically when g=l ¼ 1, q0 ¼ 0:1p, q_ 0 ¼ 0
and compute the period T of the pendulum. Give the graphical representation
of the function T ¼ T ðq0 Þ, 0:1 q0 1 for the two time steps Dt ¼ 0:1 and
Dt ¼ 0:0001. Compare with the exact expression
sffiffiffi Z
l p=2 df sin ðq=2Þ
T ðq 0 Þ ¼ 4 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , k ¼ sin ðq0 =2Þ, sin f ¼ (2)
g 0 1 k 2 sin 2 f sin ðq0 =2Þ
Use numerical integration to evaluate the Legendre elliptic integral of first

kind in Eq. (2).
Problem P5.2 The support O of the simple pendulum in Fig. P5.2 is subjected
The rotation about the
to the horizontal harmonic motion u ðt Þ ¼ u0 sin wt.
support is elastically restrained by the spring CR . Study the motion of the
system if g=l ¼ 1, u0 w2 =l ¼ 5, CR ¼ kl 2 =2, q0 ¼ p=6, q_ 0 ¼ 0, u0 ¼ 0:1m,
w ¼ 5s1 .
@CivilMethod
FIG. P5.2 Pendulum in problem P5.2.
Problem P5.3 Study the response of the system shown in Fig. P5.3, when
(i) u ð0Þ ¼ 0:05m, u_ ð0Þ ¼ 0, pðt Þ ¼ 0. The initial displacement will be taken
from the position of the static equilibrium.
(ii) u ð0Þ ¼ u_ ð0Þ ¼ 0, pðt Þ ¼ mg ð1 t=t1 ÞH ðt1 t Þ:pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(iii) u ð0Þ ¼ u_ ð0Þ ¼ 0, P ðt Þ ¼ 15sin wt,
where w ¼ kmin =m .
(iv) Plot of the dependence of the period T as a function of u0 , T ¼ T ðu0 Þ, if
u_ ð0Þ ¼ 0 and pðt Þ ¼ 0. The initial displacement will be taken from the
position of the static equilibrium.
Data: t1 ¼ 1s, ttot ¼ 5s, m ¼ 10kN m1 s2 , E ¼ 2:1 108 kN=m2 , A ¼ 1cm2 ,
L ¼ 4:0m, and g ¼ 9:81 ms2 . The cables are assumed massless to avoid the
sag due to self-weight [10–12].
FIG. P5.3 System in problem P5.3.
Problem P5.4 Study the response of the system shown in Fig. P5.3 if the elastic
force is approximated by the first three terms of its Tailor series. Compare the
results with those in Problem P5.3 when pðt Þ ¼ mg ð1 t=t1 ÞH ðt1 t Þ and
u ð0Þ ¼ u_ ð0Þ ¼ 0.
Problem P5.5 The response of a nonlinear system is governed by the following
equation, known as the van der Pol equation

u€ m 1 u 2 u_ + u ¼ 0
Plot the solution u ðt Þ, if u ð0Þ ¼ 0, u_ ð0Þ ¼ 0:1 and for (i) m ¼ 0:2 (ii) m ¼ 1:2.
Problem P5.6 The horizontal beams of the frame in Fig. P5.6 have negligible
mass and are flexible while the shear walls having a uniform mass density with
@CivilMethod
specific weight g are assumed rigid. Their support on the ground is elastic and it
is expressed by the relation

1
MR ¼ CR f f2 (a)
4
The system is set in motion by the initial conditions u0 ¼ 5cm, u_ 0 ¼ 0. Plot
the dependence of the ratio T=T0 (T0 is the period of the structure resulting
when the nonlinear term in Eq. a is ignored) on the initial displacement and dis-
cuss the influence of the nonlinearity of the elastic supports on the response of
the structure. Data: a ¼ 5m, g ¼ 24kN=m3 , E¼2.1107kN/m2, CR ¼ EI =5a,
and cross-sectional dimensions of the beam 0:20 0:40m2 .
FIG. P5.6 Frame in problem P5.6.
Problem P5.7 The buoy of Fig. P5.7 consists of two massless cones with a base
dimeter 2R and a height h. A concentrated mass m attached at the bottom of the
body keeps the buoy floating at the position shown in the figure. Study the
FIG. P5.7 Floating buoy in problem P5.7.
@CivilMethod
vertical motion of the buoy if it is displaced vertically downward from the equi-
librium position by u0 . Data: m ¼ 10kNm1 s2 h ¼ 5m, R ¼ 2m,
u ð0Þ ¼ 0:30m, u_ ð0Þ ¼ 0, and specific weight of the liquid g ¼ 2kN=m3 .
Problem P5.8 The water tower of Fig. P5.8a is subjected to the blast load of
Fig. P5.8b. The response of the structuren is elastoplastic.
h The
iorestoring force in
the elastic branch is given by fS ¼ ku 1 + 1= 1 + ð10u Þ6 =2 (Fig. P5.8c).
The structure is modeled by a SDOF system. Study the response of the structure
in the interval of t 0 ¼ 1s. Data: m ¼ 50kNm1 s2 , k ¼ 2000kN=m, x ¼ 0:07,
p0 ¼ 25kN, t1 ¼ 0:1s, and yield displacement uy ¼ 0:1m.
(b)
(a)
{ }
(c)
FIG. P5.8 Water tower in problem P5.8.
@CivilMethod

[1] J.T. Katsikadelis, Derivation of Newton’s law of motion using Galileo’s experimental data,
Acta Mech. 226 (9) (2015) 3195–3204, https://doi.org/10.1007/s00707-015-1354-y.
[2] H. Nayfeh, D.T. Mook, Nonlinear Oscillations, John Wiley & Sons, Inc, New York, 1995.
[3] F. Verhulst, Nonlinear Differential Equations and Dynamical Systems, Springer, Springer-
Verlag, Berlin Heidelberg, 1996.
[4] J.C. Butcher, Numerical Methods for Ordinary Differential Equations, second ed., John Wiley
& Sons Ltd, 2008.
[5] H.T. Davis, Introduction to Nonlinear Differential and Integral Equations, Dover Publications
Inc, New York, 2010.
[6] A. Belendez, C. Pascual, D.I. Mendez, T. Belendez, C. Neipp, Exact solution for the nonlinear
pendulum, Rev. Bras. Ensimo Fis. 29 (4) (2007) 645–648.
[7] M. Abramowitz, I.A. Stegun (Eds.), Handbook of Mathematical Functions, Dover Publica-
tions, New York, 1970.
[8] K.J. Bathe, Conserving energy and momentum in nonlinear dynamics: a simple implicit time
integration scheme, Comput. Struct. 85 (2007) 437–445.
tural dynamics, Angew. Math. Mech. 94 (9) (2014) 757–774, https://doi.org/10.1002/
zamm.20120024.
[10] J.W. Leonard, Tension Structures, McGraw-Hill, New York, 1988.
[11] J.T. Katsikadelis, Finite deformation of cables under 3-D loading: an analytic solution, in:
D.E. Beskos, D.L. Karabalis, A.N. Kounadis (Eds.), Proc. of the 4th National Congress on
Steel Structures, Patras, May 24–25, vol. II, 2002, pp. 526–534.
[12] C.G. Tsiatas, J.T. Katsikadelis, Nonlinear analysis of elastic cable-supported membranes, Eng.
Anal. Bound. Elem. 35 (2011) 1149–1158.
@CivilMethod
Chapter 6
Response to ground motion

and vibration isolation
Chapter outline
6.1 Introduction 217 6.4 Vibration isolation 235
6.2 Equation of motion: Relative 6.4.1 Transmission of force 235
displacement 217 6.4.2 Transmission of motion 237
6.2.1 Response spectra 220 6.5 Problems 240
6.3 Equation of motion in terms References and further reading 243
of the total displacement 230
6.1 Introduction
Structural systems are often excited by the motion of their support. The
response of a structure to support excitation is dynamic even though no external
dynamic loads act on it. The seismic motion of the ground represents a typical
example of support excitation of structures. The study of the response of struc-
tures to earthquake-induced motion is a specific but very important subject of
structural dynamics. It is discussed in depth in books on earthquake engineering
as well as in books on structural dynamics, preparing engineers to design struc-
tures for earthquake-induced motion [1,2]. This book treats the dynamic
response of structures when the excitation force is known. Therefore, the dis-
cussion in this chapter is limited only to the study of the dynamic response
of the SDOF system due to support excitation. Besides, some basic concepts such
as the response spectrum concept, which facilitates the dynamic analysis of struc-
tures due to ground motion, are presented. The general problem of the support
excitation of structures will be examined later when the MDOF (multi-degree-
of-freedom) systems are studied. The transmission of vibrations from the
structure to the fundament and vice versa are also discussed. Illustrative examples
analyzing the response of SDOF systems due to ground motion are presented. The
pertinent bibliography with recommended references for further reading is also
included. The chapter is enriched with problems to be solved.
6.2 Equation of motion: Relative displacement

Fig. 6.2.1 shows a simplified dynamic model of a one-story shear building
whose support moves horizontaly according to a known law described by the

@CivilMethod
function ug ðt Þ. The usual method to analyze the motion of the system is to

decompose the total displacement, namely the displacement with respect to a
fixed reference frame, utot , in two components (Fig. 6.2.1)
utot ðt Þ ¼ ug ðt Þ + u ðt Þ (6.2.1)
where u ðt Þ is the relative displacement of the top of the column with respect to
its base.
The deformation of the structure is caused by the relative displacement u ðt Þ.
Therefore, the elastic force fS ðt Þ and the damping force fD ðt Þ depend only on
u ðt Þ and u_ ðt Þ, respectively. However, the inertial force fI ðt Þ depends on the
total acceleration u€tot ðt Þ. Thus, we will have
fI ðt Þ ¼ m u€tot ðt Þ, fS ðt Þ ¼ ku ðt Þ, fD ðt Þ ¼ cu_ ðt Þ (6.2.2)
utot
m ug u
Fixed axis of reference
c m c
k k
(a) ug (t ) (b)
m
fI mutot fD cu
fS ku
(c)
FIG. 6.2.1 SDOF system subjected to ground motion (a), Dynamic model (b), Forces on the
free body (c).
The dynamic equilibrium of the system (see Fig. 6.2.1c) requires

fI ðt Þ + fD ðt Þ + fS ðt Þ ¼ 0
m u€tot + cu_ + ku ¼ 0 (6.2.3)
or using Eq. (6.2.1) we have
m u€ + cu_ + ku ¼ m u€g (6.2.4)
@CivilMethod
Response to ground motion and vibration isolation Chapter 6 219
Eq. (6.2.4) is an equation of forced motion, where the excitation function

is m u€g . In this case, the excitation function is called effective load and is
denoted by
pðt Þ ¼ m u€g ðt Þ (6.2.5)
We observe that the influence of the ground motion on the structure does not
depend directly on the ground displacement ug ðt Þ, but on its acceleration u€g ðt Þ.
Therefore, the dynamic response of the structure due to ground motion can be
studied if the accelerogram of the seismic motion is known, namely if there is a
record of the function u€g ðt Þ during the earthquake (see Fig. 6.2.2). It must be
emphasized that the initial conditions u ð0Þ and u_ ð0Þ for the relative displace-
ment do not vanish if ug ð0Þ 6¼ 0 or u_ g ð0Þ 6¼ 0, even though the system is at rest
at time t ¼ 0. Indeed, if
utot ð0Þ ¼ 0, u_ tot ð0Þ ¼ 0
max ug,tt=258.59
ug,tt (cm/s2)
t
FIG. 6.2.2 Accelerogram from Athens earthquake, Sept. 7, 1999 (Recorded by ITSAK, Α399-1.
V2, longitudinal component, max u€g ¼ 258:59cm=s2 ).
then Eq. (6.2.1) gives

u ð0Þ ¼ ug ð0Þ, u_ ð0Þ ¼ u_ g ð0Þ (6.2.6)
Eq. (6.2.4) can also be written as
m u€ + 2mwx u_ + ku ¼ m u€g ðt Þ (6.2.7a)
or
u€ + 2xwu_ + w2 u ¼ u€g ðt Þ (6.2.7b)
@CivilMethod
Apparently, Eq. (6.2.7b) states that the deformation of the system due to
given ground acceleration u€g ðt Þ depends only on the natural frequency w, hence
on the natural period T ¼ 2p=w, and on the damping ratio x, that is,
u ¼ u ðt, T , x Þ. Consequently, two systems with the same natural period T
and the same damping ratio x will undergo the same displacement u ðt Þ under
the same ground motion, in spite of the fact that the two systems may have dif-
ferent masses or different stiffnesses.
The negative sign in the effective load pðt Þ ¼ m u€g ðt Þ affects only the
direction of the displacement and not its magnitude. In practice, this has little
significance inasmuch as the engineer is usually interested in the maximum
absolute value of u ðt Þ. Therefore, the sign can be omitted in this case. This
assumption allows us to write the Duhamel integral in the forma
1
u ðt Þ ¼ U ðt Þ (6.2.8)
wD
where
Z t
U ðt Þ ¼ u€g ðτÞexwðtτÞ sin wD ðt τÞdτ (6.2.9)
0
For a given ground motion and a fixed damping ratio, we can evaluate the
largest absolute value of the function U ðt Þ, hence of u ðt Þ, for an interval of
values of the natural period T of the damped SDOF system and plot the curves
u ðT , xÞ ¼ max t ju ðt, T , xÞj for discrete values of the damping ratio x. Fig. 6.2.3
shows the curves u ðT , xÞ ¼ U ðT , x Þ=wD , 0 < T 2 for different values of x,
when the ground motion is induced by the Athens earthquake, Sept. 7, 1999.
The respective accelerogram is shown in Fig. 6.2.2. The curves u ðT , x Þ were
obtained by direct solution of Eqs. (6.2.7a), (6.2.7b). The solution can be
obtained using either the analytic solution presented in Section 3.5.4 or numer-
ically using any of the methods presented in Chapter 4.
6.2.1 Response spectra

The curve u ðT , x Þ ¼ max t ju ðt, T , x Þj is called the response spectrum of the
relative displacement or the deformation response spectrum. The response
spectrum, introduced by M. A. Biot in 1932 [3], can be used as a practical means
to study the effect of the ground motion on structures. It is clear that the defor-
mation response spectrum permits the direct evaluation of the absolute maxi-
mum value of the relative displacement of the SDOF system for a specified
excitation without solving the equation of motion, if its natural period T and
damping ratio x are given.
a. In Eqs. (6.2.8) and (6.2.9), we can set wD ¼ w 1 x 2 w, because in real structures the value of
the damping ratio is small (x ¼ 3% 15%, hence x2 ≪1) and the error due to this approximation is
much smaller than that due to the uncertainty of the determination of u€g ðt Þ.
@CivilMethod
u T
T (s)
FIG. 6.2.3 Deformation response spectrum u ðT , x Þ for Athens earthquake, Sept. 7, 1999.
Similarly, we define the relative velocity spectrum and the acceleration

response spectrum as the graphs of the curves u_ ðT , xÞ ¼ max t ju_ ðt, T , x Þj
and u€tot ðT , xÞ ¼ max t ju€tot ðt, T, x Þj, respectively.
The expression of the relative velocity results by direct differentiation of
Eq. (6.2.9) with respect to time t. Because t appears as a parameter in the limits
of the integral, the derivative can be obtained using Leibnitz’s rule [4].
Thus, we obtain
Z t
u_ ðt Þ ¼ wxu ðt Þ + u€g ðτÞexwðtτÞ cos wD ðt τÞdτ (6.2.10)
0
which by virtue of Eq. (6.2.8) is written

x
u_ ðt Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi U ðt Þ + U ðt Þ (6.2.11)
1 x2
where it was set
Z t
U ðt Þ ¼ u€g ðτÞexwðtτÞ cos wD ðt τÞdτ (6.2.12)
0
Differentiation of the expression (6.2.10) gives the relative acceleration

and the total acceleration is obtained from Eq. (6.2.1), which gives
u€tot ðt Þ ¼ u€g ðt Þ + u€ðt Þ. However, the total acceleration can be obtained directly
from Eq. (6.2.7b), that is
u€tot ðt Þ ¼ 2xwu_ ðt Þ w2 u ðt Þ
@CivilMethod
which by virtue of Eqs. (6.2.8), (6.2.11) becomes

w
u€tot ðt Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi 2x2 1 U ðt Þ 2xwU ðt Þ (6.2.13)
1 x2
The graphs of the absolute maximum of the functions (6.2.11) and (6.2.13),
that is, u_ ðT , xÞ ¼ max t ju_ ðt, T , x Þj and u€tot ðT , xÞ ¼ max t ju€tot ðt, T , x Þj, in an
interval of the undamped natural period T for a fixed value of the damping ratio
x of the SDOF systems give the response spectra of the relative velocity and the
(total) acceleration. Figs. 6.2.4 and 6.2.5 show these response spectra for the
accelerogram of Fig. 6.2.2. The curves were obtained by direct solution of
Eq. (6.2.7b). It should be noted that the quantities u ðT , xÞ, u_ ðT , xÞ, and
u€ðT , xÞ do not occur at the same time instant.
The shear force at the base of the SDOF system of Fig. 6.2.1 is equal to the
elastic force, that is,
Qo ¼ fS ¼ ku
or taking into account that k ¼ mw2 and using Eq. (6.2.8) we obtain
w
Qo ¼ m pffiffiffiffiffiffiffiffiffiffiffiffi U ðt Þ (6.2.14)
1 x2
Obviously, the maximum value of the base shear force is
w
max Qo ¼ m pffiffiffiffiffiffiffiffiffiffiffiffi U ðT , xÞ (6.2.15)
1 x2
T (s)
FIG. 6.2.4 Response spectrum of the relative velocity u_ ðT , xÞ for the Athens Earthquake,
Sept. 7, 1999.
@CivilMethod
T (s)
FIG. 6.2.5 Response spectrum of the total acceleration u€tot ðT , xÞ for the Athens earthquake,
Sept. 7, 1999.
For small values of the damping ratio (say, 0 x 0:15), we may set x 2 0.
Thus, the last two equations become
Qo ¼ mwU ðt Þ (6.2.16)
max Qo ¼ mwU ðT , x Þ (6.2.17)
We observe that the quantity wU ðT , x Þ in Eq. (6.2.17) has dimensions of
acceleration. In earthquake engineering, this quantity is designated by
Spa ðT , x Þ and it is known as spectral pseudoacceleration. It is an important
quantity because it allows direct evaluation of the maximum elastic force (base
shear force) in the SDOF system from the graph of Spa ðT , xÞ.
The quantity Spv ðT , x Þ ¼ U ðT , x Þ has dimensions of velocity and it is
known as spectral pseudovelocity. Actually, the quantities Spv ðT , x Þ,
Spa ðT , x Þ are different from u_ ðT , xÞ and u€tot ðT , x Þ. Therefore, they should
not be confused. Nevertheless, it is Spa ðT , xÞ ¼ u€tot ðT , xÞ, if x ¼ 0. Indeed,
Eq. (6.2.3) for x ¼ 0 becomes
u€tot ¼ w2 u ¼ wU ðt Þ (6.2.18)
u€tot ðT , x Þ ¼ Spa ðT , x Þ (6.2.19)
The deviation of the pseudoacceleration from the extreme value of the total
acceleration is small for small values of x (say 0 x 0:1). Thus we may set
Spa ðT , x Þ u€tot ðT , x Þ. This is shown in Fig. 6.2.6.
@CivilMethod
(utot),tt(T,0)
Spa(T,0)
T (s)
FIG. 6.2.6 Response spectra Spa ðT , x Þ and u€tot ðT , x Þ, ðx ¼ 0Þ.
For the sake of uniform notation, we set Sd ðT , x Þ ¼ u ðT , xÞ. It is obvious

that if one of the quantities Sd ðT , x Þ, Spv ðT , xÞ, Spa ðT , xÞ is known, the remain-
ing quantities result immediately. Thus, if Sd ðT , xÞ is known then we obtain
Spv ðT , xÞ ¼ wSd ðT , x Þ (6.2.20)
Spa ðT , x Þ ¼ wSpv ðT , xÞ (6.2.21)
The above relations are valid because the resulting values refer to the same
time and even at the instant where the maximum displacement Sd ðT , xÞ occurs.
Eq. (6.2.17) can be written as
max Qo ¼ mS pa ðT , x Þ
Spa ðT , xÞ (6.2.22)
¼W
g
where W is the weight of the SDOF system and g the acceleration of gravity.
The ratio
Spa ðT , x Þ
e¼ (6.2.23)
g
referred to as the elastic seismic coefficient is a quantity specified in the earth-
quake codes and it is used to evaluate the maximum equivalent shear force at the
base of a SDOF shear building.
@CivilMethod
The response spectra of the displacement, pseudovelocity, and pseudoacce-

leration contain the same information for the dynamic response of the system
because each of them results from another by multiplying or dividing it by a
constant number. The reason for using the three spectra is that each of them
refers to a different natural quantity. We saw that Sd ðT , x Þ expresses the
extreme value of the relative displacement while Spa ðT , x Þ is employed to
evaluate the shear force at the base of the one-story shear building. Finally,
Spv ðT , x Þ is used to calculate the extreme value of the energy of the
system during the earthquake. Indeed, the maximum value of the energy of
deformation is
1
E ðT , xÞ ¼ kSd ðT , x Þ2
2
1 Spv ðT , xÞ2
¼ k (6.2.24)
2 w2
1
¼ mS pv ðT , xÞ2
2
Fig. 6.2.7 presents the graphs of Sd ðT , x Þ, Spv ðT , x Þ, and Spa ðT , xÞ for the
Athens earthquake, Sept. 7, 1999. In earthquake engineering, the three plots are
usually presented in a single graphical representation using a logarithmic scale
for the four axes [2].
All previous results have been obtained by direct solution of the equation of
motion. For this reason, a computer program has been written in MATLAB and
given the name response_spectrum_aem.m. The program evaluates the spec-
tra of the deformation, relative velocity, and acceleration and makes their
graphs. The program uses the numerical method presented in Section 4.4.
The user should provide the one-dimensional array containing the values of
the accelerogram at the respective times. The electronic version of the program
is given on this book’s companion website.
Example 6.2.1 Fig. E6.1a shows a cement silo of square plan form supported on
four identical columns. The silo is made of reinforced concrete. When the silo is
empty, a horizontal load P in the x direction is applied, which causes a horizon-
tal displacement of the silo equal to 5cm. Then, the force is suddenly removed
and the system starts to vibrate. After time to the silo has performed six com-
plete vibrations and the amplitude was reduced to 1cm. Determine:
(i) The damping ratio x.
(ii) The damping coefficient c, the damped natural frequency wD , and the
time to
(iii) The displacement after 40 and 60s.
(iv) The maximum shear force max Q and the maximum bending moment
max M of the columns, when the silo is full and is subjected to the support
of duration t1 ¼ 3s in the absence of damping.
excitation ug ¼ uo sin wt
Determine also the dynamic magnification factor D.
@CivilMethod
Sd
(a)
(b)
Spa
(c)
FIG. 6.2.7 Response spectra for the Athens earthquake, Sept. 7, 1999 (x ¼ 0:1).
@CivilMethod
5m
z
2.5 m
x utot
1m P ug u
m c
10 m
k
x
(a) (b)
FIG. E6.1 Silo in Example 6.2.1
Data: Specific weight of cement g c ¼ 17kN=m3 , specific weight of reinforced

concrete g b ¼ 24kN=m3 , uo ¼ 1:7mm, w ¼ 2:5s1 , modulus of elasticity of
the reinforced concrete E ¼ 2:1 107 kN=m2 , cross-sectional area of columns
0:35 0:35m2 , and thickness of the silo walls and bottom 0:20m. The axial defor-
mation of the columns is ignored while their mass is assumed lumped at their ends.
Solution
The walls of the silo do not deform inside their plane. Therefore, the silo
behaves in actual fact as a rigid body. Because the axial deformation of the col-
umns is neglected, the only possible motion of the structure in the xz plane is the
horizontal one along the x axis. This is shown in the dynamic model of the sys-
tem in Fig. E6.1b.
(i) Τhe damping ratio x is computed from the relation (2.3.24) for n ¼ 6, t ¼ 0,
u ð0Þ ¼ 5cm, u ðnT Þ ¼ 1cm. This yields

x 1 5
pffiffiffiffiffiffiffiffiffiffiffiffi ¼ ‘n
1x 2 2 6p 1
which gives x ¼ 4:265%.
(ii) The computation of c and wD requires the determination of the mass m and
the stiffness k of the structure
2 24
m¼ 5:0 4:62 2:5 + 4:62 0:2=0:928 + 4 0:352 10=2
9:81
¼ 40:6kNm1 s2
12EI 12 2:1 107 0:354 =12

k ¼4 ¼ 4 ¼ 1260:5kN=m
h3 103
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
w ¼ 1260:5=40:6 ¼ 5:57s1
@CivilMethod
Hence
c ¼ 2mwx ¼ 2 40:6 5:57 0:04265 ¼ 19:3kNm1 s
qffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
wD ¼ w 1 x 2 ¼ 5:57 1 0:042652 ¼ 5:56s1
2p
to ¼ 6T ¼ 6 ¼ 6:78s
wD
(iii) The system performs free damped vibrations. The displacement u ðt Þ is

computed from Eq. (2.3.18) for u ð0Þ ¼ 0:05m, u_ ð0Þ ¼ 0, x ¼ 0:04265,
w ¼ 5:57s1 , wD ¼ 5:56s1 . Thus we have
u ðt Þ ¼ e0:238t ð5cos 5:56t + 0:214 sin 5:56t Þ 102 (1)
which yields
u ð40Þ ¼ 0:282 103 cm, u ð60Þ ¼ 0:268 105 cm
(iv) When the silo is filled with cement, we have
17
m ¼ 40:6 + 4:62 2:5 + 4:62 0:8=3 ¼ 142:1kNm1 s2
9:81
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
w ¼ 1260:5=142:1 ¼ 2:97834s1
The motion is examined in two phases.

Phase Ι (0 t 3). The motion is forced vibration. The equation of motion
is given by Eq. (6.2.4) for c ¼ 0, that is
m u€ + ku ¼ pðt Þ (2)
The effective force is
¼ 142:1 1:7 103 2:52 sin wt
pðt Þ ¼ mu o w2 sin wt ¼ 1:5098 sin wt

The displacement is given by Eq. (3.2.8). Because the system is at rest at
t ¼ 0, it is utot ð0Þ ¼ u_ tot ð0Þ ¼ 0. Hence
u ð0Þ ¼ utot ð0Þ ug ð0Þ ¼ 0
and

o ¼ 2:5 1:7 103 ¼ 4:25 103 m=s
u_ ð0Þ ¼ u_ tot ð0Þ u_ g ð0Þ ¼ wu
These initial conditions yield
u_ ð0Þ Po b
A¼ and B ¼ 0
w k 1 b2
Hence
u_ ð0Þ po 1
u I ðt Þ ¼ sin wt + b sin wt Þ
ð sin wt
w k 1 b2
@CivilMethod
¼ 2:5=2:98 ¼ 0:8389, po ¼ 1:5098kN,

The above relation for b ¼ w=w
k ¼ 1260:5kN=m, and u_ ð0Þ ¼ 4:25 103 m=s yields
uI ðt Þ ¼ 103 ð4:8303 sin wt 4:0545sin wt
Þ
u_ Ι ðt Þ ¼ 102 ð1:4386 cos wt 1:0136 cos wt
Þ
Phase II (t 3). The motion is free vibration and the displacement is given
by Eq. (2.2.13). Thus we have
u_ I ð3Þ
uII ðt Þ ¼ sin wet + uI ð3Þ cos wet , et ¼ t 3 0 (3)
w
The initial conditions for this phase are
uI ð3Þ ¼ 1:5309 103 m, u_ I ð3Þ ¼ 1:620884630 102 ms1
which give
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
u_ ð3Þ 2
max uII ¼ r ¼ + u_ ð3Þ2 ¼ 5:65341 103 m (4)
w
Fig. E6.2 shows the time history of the response ratio Rðt Þ ¼ u ðt Þ=ðpo =k Þ.
We observe that the max u takes place in phase II. Hence the dynamic load fac-
tor is
max uII
D¼ ¼ 4:7199 (5)
p0 =k
R(t)
t (s)
FIG. E6.2 Time history of the response ratio Rðt Þ in Example 6.2.1.
@CivilMethod
and
12EI k
max Q ¼ 3
max u ¼ max uII ¼ 1:781kN (6a)
h 4
h
max M ¼ max Q ¼ 8:907kNm (6b)
2
6.3 Equation of motion in terms of the total displacement

In certain cases, it is convenient to formulate the equation of motion in terms of
the total displacement utot ðt Þ. Then, the elastic force and damping force will be
expressed as

fS ¼ k utot ug , fD ¼ c u_ tot u_ g (6.3.1)
The dynamic equilibrium yields the following equation of motion

m u€tot + c u_ tot u_ g + k utot ug ¼ 0 (6.3.2)
or
m u€tot + cu_ tot + ku tot ¼ pðt Þ (6.3.3)
where now
pðt Þ ¼ cu_ g + ku g (6.3.4)
Eq. (6.3.3) is suitable to study the dynamic response when the u€g does not
exist in the classical sense or the supports of the structure undergo different
excitations (see Example 6.3.2).
Example 6.3.1 The support of a SDOF system, whose dynamic model is shown
in Fig. E6.3a, undergoes a sudden constant displacement ug ¼ u0 . Determine the
dynamic response of the system when utot ð0Þ ¼ u_ tot ð0Þ ¼ 0.
(a) (b)
FIG. E6.3 SDOF system in Example 6.3.1
@CivilMethod
Solution
The function representing the ground motion is shown in Fig. E6.3b. The prob-
lem can be solved by the following two ways:
(a) Formulating the equation of motion in terms of the total displacement, that
is, Eq. (6.3.3). This yields
m u€tot + ku tot ¼ ku 0 (1)
This implies that an effective constant excitation force pðt Þ ¼ ku 0 is
suddenly applied. The solution of Eq. (1) is obtained from Eq. (3.4.3)
for p0 ¼ ku 0 . Hence
utot ðt Þ ¼ u0 ð1 cos wt Þ, w ¼ k=m (2)
We observe that
ðutot Þmax ¼ 2u0
The relative displacement is
u ðt Þ ¼ utot ug ¼ u0 cos wt (3)
which yields an elastic force
fS ¼ ku ¼ ku 0 cos wt (4)
and
max jfS j ¼ ku 0 (5)
(b) Formulating the equation of motion in terms of the relative displacement
u ðt Þ. In this case, we have
ug ð0Þ ¼ u0 , u_ g ð0Þ ¼ 0, u€g ð0Þ ¼ 0
Consequently
pðt Þ ¼ 0, u ð0Þ ¼ utot ð0Þ ug ð0Þ ¼ u0 , u_ ð0Þ ¼ u_ tot ð0Þ u_ g ð0Þ ¼ 0
and the equation of motion becomes
m u€ + ku ¼ 0 (4)
with initial conditions u ð0Þ ¼ u0 , u_ ð0Þ ¼ 0.
The solution of Eq. (4) is obtained from Eq. (2.2.13) as
u ðt Þ ¼ u0 cos wt (5)
which is identical with that given by Eq. (3).
Example 6.3.2 The supports of the columns of the one-story frame of Fig. E6.4
are subjected to the displacements ug1 ¼ uo sin ðwt qÞ and ug2 ¼ uo sin wt.
Determine the equation of motion of the structure and give the expressions
of the relative displacement u ðt Þ and the stress resultants Q ðt Þ, M ðt Þ at the
top cross-sections of the columns. The dead load of the rigid beam is included
@CivilMethod
utot (t ) utot (t )
p
EI
h
35 × 70 40 × 80
ug 1 ug 2
L
FIG. E6.4 One-story frame in Example 6.3.2
in the load p. The material of the columns is reinforced concrete. Data: specific
weight of concrete g ¼ 24kN=m3 , u0 ¼ 0:02m, x ¼ 0:05, w ¼ 2:5rad=s,
u ð0Þ ¼ 0, u_ ð0Þ ¼ 0, L ¼ 15m, h ¼ 7m, E ¼ 2:1 107 kN=m2 , and
p ¼ 200kN=m. The mass of the columns is assumed lumped at their ends.
Solution
The system has one degree of freedom. The equation of motion with respect to
the total displacement utot ðt Þ is

m u€tot + c1 u_ tot u_ g1 + c2 u_ tot u_ g2 + k1 utot ug1 + k2 utot ug2 ¼ 0 (1)
or
m u€tot + cu_ tot + ku tot ¼ pðt Þ (2)
where
c ¼ c1 + c2 , k ¼ k1 + k2 (3)
p ¼ c1 u_ g1 + c2 u_ g2 + k1 ug1 + k2 ug2 (4)
The mass of the system is
200 15 + ð0:35 0:70 + 0:40 0:80Þ ð7=2Þ 24
m¼ ¼ 310:6kNm1 s2
9:81
The stiffness of the columns
12EI 1 12 2:1 107 0:70 0:353
k1 ¼ ¼ ¼ 1837:5kN=m
h3 12 73
3EI 2 3 2:1 107 0:80 0:403
k2 ¼ ¼ ¼ 783:7kN=m
h3 12 73
Hence the stiffness of the system is
k ¼ k1 + k2 ¼ 2621:2kN=m
@CivilMethod
The natural frequency is

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
w ¼ 2621:2=310:6 ¼ 2:905s1
and the natural period
T ¼ 2p=w ¼ 2:163s
The damping coefficient c is
c ¼ 2mwx ¼ 2 310:6 2:905 0:05 ¼ 90:23kNm1 s
and the damped frequency
wD ¼ w 1 x2 ¼ 2:901s1
The damping coefficients c1 and c2 that contribute to the effective force can-
not be directly determined. However, they can be estimated by setting
c1 ¼ ac and c2 ¼ ð1 a Þc, 0 a 1
Apparently, the cases a ¼ 0 and a ¼ 1 are not realistic.
For the data of the problem, Eq. (2) becomes
310:6u€tot + 90:23u_ tot + 2621:2utot ¼ pðt Þ (5)
where
pðt Þ ¼ 36:75sin ðwt
qÞ + 15:67sin wt

(6)
+ 1:80w½a cos ðwt
qÞ + ð1 a Þ cos wt

Eq. (5) can be solved using the analytic method presented in Chapter 3.
Figs. E6.5 and E6.6 show the graphs of utot ðt Þ for various values of the phase
angle q and the allocation coefficient a. Fig. E6.7 shows the influence of a on
the maximum displacement max |utot ðt Þ|.
utot(t) (m)
t
FIG. E6.5 Total displacement for different values of the phase angle q in Example 6.3.2 (a ¼ 0:2).
@CivilMethod
utot(t) (m)
t
FIG. E6.6 Total displacement for different values of the allocation coefficient a in Example 6.3.2
(q ¼ 3p=4).
max|utot(t)|
a
FIG. E6.7 Extreme value max|utot ðt Þ| versus the allocation coefficient a in Example 6.3.2
(x ¼ 0:05).
From the study of the numerical results, we may draw the following
conclusion:
The percentage allocation of the damping to the two columns does not affect
significantly the dynamic response of the structure. Consequently, an arbitrary
but reasonable allocation, for example, 0:4 < a < 0:6, allows treating practical
cases of asynchronous support excitations.
@CivilMethod
The stress resultants Q ðt Þ and M ðt Þ at the top cross-sections of the columns

are evaluated from the relations.
12EI 1
Q1 ðt Þ ¼ 3
utot ug1 ¼ 1837:5 utot ug1
h
3EI 2
Q2 ðt Þ ¼ 3 utot ug2 ¼ 783:7 utot ug2
h
6EI 1
M1 ðt Þ ¼ 2 utot ug1 ¼ 6431:2 utot ug1
h
3EI 2
M2 ðt Þ ¼ 2 utot ug2 ¼ 5485:9 utot ug2
h
6.4 Vibration isolation

Very often, machines that generate vibrations are mounted on our structures.
These vibrations, which are transferred to the supporting structure (soil, foun-
dation), are annoying or even harmful and need to be minimized. This is
achieved by inserting spring and damping devices between the machine and
its foundation. An inverse problem arises when vibrations from the environment
(ground support) are transferred to the structure. This problem occurs when sen-
sitive instruments are placed on top or inside moving structures (earthquake
motion of a building, motion of a car on a rough road, spacecrafts, airplanes,
ships, etc.) or special structures near vibration generation sources (monuments).
In the first situation, we talk about the transmission of force while the second is
about the transmission of motion.
6.4.1 Transmission of force

Let us consider first the force isolation problem. Fig. 6.4.1 presents a SDOF
system consisting of a mass m mounted on the fundament by means of a system
of springs and dampers having total stiffness k and total damping c. A vertical
is applied to the mass. The transfer of force is
harmonic force p ¼ p0 sin wt
FIG. 6.4.1 Force transmitted to the foundation.
@CivilMethod
examined when the motion is in the steady state phase. Hence, the vertical dis-
placement is given by Eq. (3.2.26), that is,
qÞ
u ðt Þ ¼ rsin ðwt (6.4.1)
where
p0 h 2 i 1
2
r¼ 1 b2 + ð2xb Þ2 (6.4.2a)
k

2xb
q ¼ tan 1
, b ¼ w=w (6.4.2b)
1 b2
The total force transmitted to the foundation is
f ¼ fS + fD
¼ ku + cu_
(6.4.3)
qÞ + crw cos ðwt
¼ krsin ðwt qÞ
q fÞ
¼ fT sin ðwt
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
fT ¼ r k 2 + ðcwÞ2 (6.4.4)

1 cw
f ¼ tan (6.4.5)
k
Using Eqs. (6.4.2a), (6.4.2b) for the expression of r and c=k ¼ 2x=w, the
previous relations are written as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 1 + ð2xbÞ2
fT ¼ p0 t 2 (6.4.6)
1 b2 + ð2xb Þ2
f ¼ tan 1 ð2xb Þ (6.4.7)

The ratio TR ¼ fT =p0 is called the transmission ratio or transmissibility. It
provides a measure of the force transmitted from the vibrating machine to the
fundament. Fig. 6.4.2 shows the graph of the function TRðb, x Þ for various
values ofpffiffix.ffi We observe that the reduction of the transmitted force is achieved
for b > 2. It is also apparent that the reduction of the ratio TR increases with
decreasing x. Theoretically, the value x ¼ 0 gives the minimum value of the
transmission ratio. However, this is not realistic for two reasons: first, because
the actual isolation systems have damping and second, some damping is desir-
able to avoid adverse results at the startup of the machine as it passes through the
resonance frequency.
@CivilMethod
TR=fT /p0
FIG. 6.4.2 Transmissibility for harmonic excitation.
6.4.2 Transmission of motion

In this case, the problem is to determine the motion that is transmitted from the
ground to the machine and in general to the structure supported on the ground.
Fig. 6.4.3 shows a mass supported on the ground by means of a system of
springs and dampers with total stiffness k and damping c. The vertical ground
motion is harmonic

ug ¼ u0 sin wt (6.4.8)
FIG. 6.4.3 Ground motion transmitted to the structure.
The equation of motion with respect to the relative displacement is

m u€ + cu_ + ku ¼ m w2 u0 sin wt (6.4.9)
The relative displacement in the steady state is given by Eq. (3.2.25) for
p0 ¼ m w2 u0 . That is
@CivilMethod
m w2 u0 1
u ðt Þ ¼ 2 2xb cos wt
1 b2 sin wt
k 1 b + ð2xbÞ
2 2
(6.4.10)
u0
¼ 2xb3 cos wt
1 b2 b 2 sin wt
2 2 2
1 b + ð2xbÞ
and the total displacement
utot ¼ ug + u
u0
+
¼ u0 sin wt 2
2xb3 cos wt
1 b2 b 2 sin wt
1 b2 + ð2xbÞ2
u0 nh i o
¼ 2
1 b2 + ð2xb Þ2 sin wt
2xb 3 cos wt
1 b2 + ð2xbÞ2
Þ
¼ uT sin ðwt
(6.4.11)
where
u
u 1 + ð2xbÞ2
uT ¼ u0 t 2 (6.4.12)
1 b2 + ð2xb Þ2
!
1 2xb3
¼ tan (6.4.13)
1 b 2 + ð2xb Þ2
From Eqs. (6.4.8), (6.4.12), we obtain
u
ðu€tot Þmax ðutot Þmax uT u 1 + ð2xbÞ2
TR ¼ ¼ ¼ ¼ t 2 (6.4.14)
u€g max ug max u0 1 b2 + ð2xb Þ2
We observe that the transmissibility of the amplitude of the ground har-

monic motion is the same as that of the force transmissibility. The same is valid
for the extreme values of the accelerations.
Example 6.4.1 A sensitive instrument of weight W ¼ 30kN is installed at a
place where the environment exerts vertical harmonic motion with a cyclic fre-
quency f ¼ 15Hz and maximum amplitude u0 ¼ 0:02m. For the insulation of
the motion, the instrument is placed on an elastic layer with stiffness
k ¼ 28kN=m and damping x ¼ 0:1. Determine (i) the acceleration transmitted
to the instrument and (ii) If the smooth functioning of the instrument requires
that the acceleration applied to it is less than 0:1g, indicate how to ensure that by
using the same elastic layer.
Solution
(i) Computation of the transmission ratio TR
The mass and the natural frequency of the instrument are
@CivilMethod
W 30kN
m¼ ¼ ¼ 3:058kNm1 s2
g 9:81m=s2
rffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
k 28
w¼ ¼ ¼ 3:026s1
m 3:058
The circular frequency of the harmonic motion and the maximum
acceleration are
w ¼ 2pf ¼ 94:248s1

u€g max ¼ w2 u0 ¼ 177:653 ¼ 18:109g g ¼ 9:81ms2
hence
w
b ¼ ¼ 31:147
w
and
u
u 1 + ð2xbÞ2
TR ¼ t 2 ¼ 0:0065
1 b 2 + ð2xbÞ2
Consequently, the maximum acceleration transmitted to the

instrument is

ðu€tot Þmax ¼ TR u€g max ¼ 0:118g
(ii) Modification of the system so that ðu€tot Þmax < 0:1g

It must be
u
u 1 + ð2xbÞ2 ðu€tot Þ 0:1g
TR ¼ t ¼ max < ¼ 0:0055
2 2
1 b + ð2xbÞ 2 u€g max 18:109g
For the same elastic layer, the stiffness k and the damping coefficient c
are unaltered. Consequently, the transmission ratio TR can be modified if
the ratio b is changed. But because w is prescribed, this ratio can be chan-
ged, only if the natural frequency w is changed. This is possible if the mass
of the system (instrument) is changed by an increment Dm.
Let m 0 ¼ m + Dm be the new mass and x 0 , w0 the new damping ratio and
the frequency of the system (instrument), respectively. Then we will have
rffiffiffiffiffiffi
w0 m
¼ (1)
w m0
c ¼ 2mwx ¼ 2m 0 w0 x0
@CivilMethod
which yield
rffiffiffiffiffiffi
0 mw m
x ¼x 0 0 ¼x (2)
mw m0
Moreover, we have
rffiffiffiffiffiffi
0 w m0
b ¼ 0¼b (3)
w m
From Eqs. (2), (3) we obtain
x0 b0 ¼ xb ¼ 3:115
Consequently, it must be
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u u
u 1 + ð2x 0 b0 Þ
2
u 2
TR ¼ u
¼ u 1 +
ð2 3:115Þ < 0:0055
t 2 2
t 2 2
1 b0 + ð2x0 b 0 Þ 1 b0
2
+ ð2 3:115Þ2
which holds if
33:885 < b0 ¼ b m 0 =m or m 0 > 3:619kNm1 s2
Hence Dm 0:561kNm1 s2 .
6.5 Problems
Problem P6.1 The supports 1 and 2 of the structure shown in Fig. P6.1 are sub-
and ug2 ðt Þ ¼ u0 sin ðwt
jected to the motions ug1 ðt Þ ¼ u0 sin wt p=3Þ, respec-
tively. A plane square rigid body of side a and density g is attached to
node 3. The mass of the column and the beam is neglected. Determine the
response of the structure and the reactions at the supports as well as the forces
that produce the support excitations. Data: L ¼ 8m, a ¼ 2m, cross section of the
@CivilMethod
beam and the column IPE360, w ¼ 10s1 , u0 ¼ 3cm, g ¼ 10kN=m2 , and

E ¼ 2:1 108 kN=m2 .
Problem P6.2 A SDOF system with parameters m, x,k is subjected
to ground motion whose accelerogram is given by u€gi ¼ 100ð1Þi
½ði + 5Þ=ði + 1Þ ði + 1Þ=ði + 5Þ with duration t1 ¼ 2:5s (Fig. P6.2). Compute
and plot the response spectra of the displacement, velocity, and acceleration
for the interval of the natural period 0 < T 2 and x ¼ 0:05, 0:1.
Data: u ð0Þ ¼ u_ ð0Þ ¼ 0, u€g ð0Þ ¼ 0, m ¼ 4, and k ¼ 100.
ug,tt
(ug)i
(ug)i
t
FIG. P6.2 Accelerogram in problem P6.2
Problem P6.3 The rigid vertical column AC (Fig. P6.3) of a circular cross-
section, line density m is supported on the ground by means of a spherical hinge
at A and three elastic cables of cross-sectional area A and modulus of elasticity E.
(a) (b)
@CivilMethod
The cables have been prestressed so that they can undertake compressive forces.
The column carries three advertising panels of total mass 10ma, which are
arranged as in Fig. P6.3b. Their support on the column extends to a length
1:25a. The structure is subjected to ground motion in the y direction, whose accel-
erogram u€g ðt Þ is given in Problem P6.2. Determine the minimum prestressing
force of the cable GB using the results of Problem P6.2. The cables are assumed
massless. Data: a ¼ 5m, m ¼ 0:5kNm1 s2 =m, E ¼ 2:1 108 kN=m2 , and
A ¼ 4cm . 2
Problem P6.4 The one-story building of Fig. P6.4 is subjected to ground

motion whose accelerogram u€g ðt Þ is given in Problem P6.2. The motion takes
place in the direction of angle b with respect to the x axis. Determine the
dynamic response of the structure and plot the relative displacement u ðb Þ
of the top cross-section of columns 1 and 2 as a function of the angle b,
0 b 2p. Moreover, compute the extreme normal stresses of the same
cross-sections due to bending. Data: x ¼ 0:07 and E ¼ 2:1 107 kN=m2 . The
mass of the columns is neglected. The load q includes also the dead weight
of the slab.
Problem P6.5 The vertical columns of the frame in Fig. P6.5 have specific
weight g b and are assumed rigid. The elastic support on the ground is simulated
by the rotational springs CR . The horizontal beams are flexible with cross-
sectional moment of inertia I and modulus of elasticity E while their mass
and axial deformation are assumed negligible. The structure is subjected to
of total duration ttot ¼ 3s. Give
the horizontal ground motion ug ðt Þ ¼ u0 sin wt
the graph of the response spectrum of the rotation of the structure and compute
the extreme values of the shear force and the bending moment of the beams.
Data: Cross-sectional area of the beams A ¼ a=10 a=10, CR ¼ EI =a,
E ¼ 2:1 107 kN=m, g b ¼ 24kN=m3 , u0 ¼ 0:03m, w ¼ 3, 5 and 7s1 .
@CivilMethod
=∞
FIG. P6.5 Frame in problem P6.5
Problem P6.6 A vehicle traveling with a velocity v¼80km/h on a multispan

bridge (Fig. P6.6) is idealized by the SDOF system of Fig. P6.6a. The length
of each span is L ¼ 20m. The deck of the bridge has been permanently
deformed due to creep so that each span can be simulated by a sinusoidal curve
with amplitude h ¼ 5cm. The stiffness of the system is k ¼ 10kN=m and its
damping ratio x ¼ 0:15. The tires of the wheels are assumed undeformable.
The total weight of the vehicle is W ¼ 20kN. (i) Determine the vertical motion
of the vehicle, (ii) Compute the maximum acceleration applied to the passenger
of the vehicle, and (iii) Compute the velocity that causes resonance to the vehi-
cle and gives its extreme amplitude.
(a)
h
h
(b) L
FIG. P6.6 Vehicle traveling on a bridge in problem P6.6

[1] R.W. Clough, J. Penzien, Dynamic of Structures, second ed., McGraw-Hill, Inc., New York, 1993.
[2] A.K. Chopra, Dynamics of Structures: Theory and Applications to Earthquake Engineering,
Prentice Hall, Englewood Cliffs, NJ, 1995.
[3] M.A. Biot, Vibrations of buildings during earthquake, in: Transient Oscillations in Elastic
System, Aeronautics Department, Calif. Inst. of Tech., Pasadena, CA, 1932. Chapter II in
Ph.D. Thesis No. 259.
@CivilMethod
[4] F.B. Hildebrand, Advanced Calculus for Applications, Prentice Hall, Englewood Cliffs, NJ,
1962.
[5] R.R. Craig Jr., J. Andrew, A.J. Kurdila, Fundamentals of Structural Dynamics, second ed., John
Wiley, New Jersey, 2006.
[7] J.T. Katsikadelis, Finite deformation of cables under 3-D loading: an analytic solution, in:
D.E. Beskos, D.L. Karabalis, A.N. Kounadis (Eds.), Proc. of the 4th National Congress on Steel
Structures, Patras, May 24–25, vol. II, 2002, pp. 526–534.
@CivilMethod
Chapter 7
Damping in structures
Chapter outline
7.1 Introduction 245 7.6.1 Introduction 257
7.2 Loss of energy due to damping 246 7.6.2 The fractional derivative 258
7.3 Equivalent viscous damping 249 7.7 Measurement of damping 260
7.4 Hysteretic damping 250 7.7.1 Free vibration decay
7.5 Coulomb damping 252 method 261
7.5.1 Free vibrations with 7.7.2 Resonance amplitude
Coulomb damping 252 method 262
7.5.2 Forced vibrations with 7.7.3 Width of response curve
Coulomb damping 255 method 263
7.6 Damping modeling via 7.8 Problems 265
fractional derivatives 257 References and further reading 267
7.1 Introduction
In this chapter, the damping of structures is discussed. Damping appears in all
mechanical systems that perform vibrations. It is the dissipation of energy in a
vibrating structure. The type of energy into which the mechanical energy is
transformed depends on the system and the physical mechanism that causes
the dissipation. The energy is lost either in the form of heat or is radiated into
the environment. For example, the loss of energy in the form of heat is perceived
when an iron rod is subjected to alternating bending. The sound produced by a
body that is hit represents the loss of energy dissipated into the environment. In
the study of vibrations, we are interested in the damping related to the response
of the structure. Damping is due to different energy dissipation mechanisms act-
ing simultaneously. In spite of the age-long detailed studies on the damping of
structures, the understanding of damping mechanisms is quite primitive. A
well-known method to get rid of this problem is to use so-called viscous damp-
ing. This approach was first introduced by Rayleigh [1] via his famous dissipa-
tion function (see Section 1.8.4).
The loss of energy of a vibrating system reduces the amplitude of the free
vibration. When a system undergoing forced vibrations reaches the phase of
the steady-state response, the loss of energy is balanced by the energy input into
the system by the excitation force.
In vibrating systems, we distinguish different types of damping forces,
which may be due to the internal molecular friction, the sliding friction, or the

@CivilMethod
resistance of the fluid in which the system moves. Sophisticated mathematical

models of damping forces are complex and are not convenient to study the vib-
rations. For this reason, simplified damping models have been developed that
allow an adequate study of the dynamic response of structures. We have already
used the model of viscous damping, which leads to equations that can be solved
analytically. The presented examples facilitate the comprehension of all con-
cepts. The pertinent bibliography with recommended references for further
study is also included. The chapter is enriched with problems to solved.
7.2 Loss of energy due to damping

The energy loss is determined in the phase of the steady-state response. The graph-
ical representation of the relationship fD ðu Þ between the damping force and the
displacement varies greatly for different types of damping. In all cases, however,
the curve is closed in a complete oscillation and includes a region called the hys-
teresis loop, whose area is equal to the energy lost per cycle (see also Refs. [2–5]).
In general, the loss of energy is expressed by the integral
Z T
WD ¼ fD ðu Þdu (7.2.1)
0
where T ¼ 2p= w the period of the vibration in the steady-state response. The
quantity WD depends on several factors such as temperature, frequency, or
amplitude of the vibration.
In this section, we consider the simplest form of energy loss, that is, the loss
due to viscous damping. The damping force, in this case, is given by the relation
_ Moreover, the displacement in the steady-state phase, due to the
fD ¼ cu.
t, is given by Eq. (3.2.26), namely
harmonic force pðt Þ ¼ p0 sin w
t qÞ
u ðt Þ ¼ rsin ðw (7.2.2)
where
p0 1
r¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7.2.3)
k
2 2
1 b + ð2xbÞ2
2xb
tan q ¼ (7.2.4)
1 b2
Consequently
u_ ðt Þ ¼ r t qÞ
w cos ðw (7.2.5)
The integral (7.2.1) yields
Z T
WD ¼ cu_ 2 dt
0
Z 2p=
w
(7.2.6)
w2 r 2
¼ c cos 2 ðw
t qÞdt
0
wr2
¼ pc
@CivilMethod
Damping in structures Chapter 7 247
Of particular
pffiffiffiffiffiffiffiffiffi interest is the
pffiffiffiffiffiffienergy
ffi loss at resonance. Then it is
¼ w ¼ k=m , c ¼ 2mxw ¼ 2x km and Eq. (7.2.6) becomes
w
WD ¼ 2xpkr2 (7.2.7)
Moreover, the energy input into the system is due to the work that the force
t produces in a complete oscillation. Namely
pðt Þ ¼ p0 sin w
Z T
Wp ¼ pðt Þdu
0
Z 2p=
w
(7.2.8)
¼ t qÞ sin w
w cos ðw
p0 r tdt
0
¼ p0 rp sin q
Taking into account that (see Eq. 3.2.27)
2xb
sin q ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ¼ 2xbrk=p0 (7.2.9)
1 b2 + ð2xbÞ2
we can write Eq. (7.2.8) as

wr2 ¼ WD
Wp ¼ pc (7.2.10)
Indeed, during the steady-state response, the input into the system by the
excitation force is equal to the dissipated energy due to damping. We come
to the same conclusion if we examine the work produced by the elastic and
inertial forces.
The work produced by the elastic force in a cycle is
Z T
WS ¼ fS ðu Þdu
0
Z 2p=
w
¼ _
ku udt (7.2.11)
0
Z 2p=
w
2
r
¼ kw t qÞ cos ðw
sin ðw t qÞdt ¼ 0
0
Similarly, the work produced by the inertial force is

Z T
WI ¼ fI ðu Þdu
0
Z 2p=
w
¼ m u€udt
_ (7.2.12)
0
Z 2p=
w
w3 r2
¼ m t qÞ cos ðw
sin ðw t qÞdt ¼ 0
0
@CivilMethod
Thus, it must be
WI + WD + W S ¼ Wp (7.2.13)
which by virtue of Eqs. (7.2.11), (7.2.12) yields Eq. (7.2.10).
Eq. (7.2.5) yields
u_ ðt Þ2 ¼ w
2 r2 cos 2 ðw
t qÞ

¼w r 1 sin 2 ðw
2 2
t qÞ (7.2.14)

¼w 2 r2 u 2
hence
fD2 ¼ c2 u_ ðt Þ2
(7.2.15)
¼ c2 w
2 r2 u 2
which is readily transformed into
2
fD 2 u
+ ¼1 (7.2.16)
wr
c r
FIG. 7.2.1 Hysteresis loop in a system with viscous damping.
The above relation represents an ellipse with semiaxes r and c wr in the

plane u, fD , Fig. 7.2.1. The area of the ellipse is pc wr2 , that is, it is equal to
the energy lost in a cycle. Therefore, the hysteresis loop of a system with viscous
damping is an ellipse.
The properties of damping in the material systems are expressed in
different ways depending on the scientific area where they are applied. Here
we mention two units for measuring the relative dissipated energy that are most
commonly encountered: the specific damping capacity and the specific damp-
ing factor.
The specific damping capacity is defined as
WD
wD ¼ (7.2.17)
U
where U ¼ kr2 =2 is the elastic energy of the system.
@CivilMethod
The specific damping factor is defined as

WD
¼ (7.2.18)
2pU
For viscous damping the above quantities result as
wr2 2pc
pc w 2pð2mwx Þ
w
wD ¼ ¼ ¼ ¼ 4pxb (7.2.19)
kr2 =2 mw2 mw2
¼ 2xb (7.2.20)
7.3 Equivalent viscous damping

When the damping of the structure is not inherently viscous, the curve fD ðu Þ is
not generally an ellipse. It is, however, possible to determine an equivalent vis-
cous damping coefficient ceq by equating the area WD of the experimentally
obtained damping loop in a cycle of harmonic vibration (Fig. 7.3.1) with the
theoretical value of the area of the hysteresis loop representing dissipation of
energy due to viscous damping.
FIG. 7.3.1 Energy dissipated in a cycle of harmonic vibration determined experimentally.
On the basis of Eq. (7.2.6) we write
wr2
WD ¼ ceq p (7.3.1)
from which we define the equivalent damping coefficient

WD
ceq ¼ (7.3.2)
wr2
p
and the respective damping ratio
ceq
xeq ¼ pffiffiffiffiffiffiffi (7.3.3)
2 km
@CivilMethod
7.4 Hysteretic damping

When the structure is subjected to cyclic loading, the dissipation of energy
occurs in the interior of the material of the structure. Experiments performed
by many researchers have shown that in most structural materials, such as steel
and aluminum, the energy loss per cycle does not depend on the frequency of the
external excitation force, at least for a wide range of frequencies, but it is pro-
portional to the square of the amplitude of the oscillation. This damping is called
structural damping or hysteretic damping.
The equation of motion for forced vibrations with hysteretic damping is of
the form
m u€ + f ðu Þ ¼ pðt Þ (7.4.1)
where f ðu Þ is in general a nonlinear function of the displacement. The solution
of Eq. (7.4.1) can be obtained numerically using any of the methods presented in
Chapter 5.
The force f ðu Þ can be expressed as the sum of two forces, an elastic force
with a mean value fS ¼ ku and a damping force
k
fD ¼ u_ (7.4.2)
w
where is the constant of damping. This is the simplest model that can be used
to represent the hysteretic damping.
The loss of energy in a cycle due to hysteretic damping results from
Eq. (7.2.6), if the coefficient c is replaced with k=
w. This yields
WD ¼ kpr2h (7.4.3)
where rh is the amplitude of the vibration with hysteretic damping. We observe
that the dissipation of energy per cycle does not depend on the excitation fre-
quency w , but on the square of the amplitude of the vibration.
Apparently, the function f ðu Þ representing the resisting force can be written
k
f¼ u_ + ku (7.4.4)
w
and the equation of motion (7.4.1) for harmonic excitation takes the form
k
m u€ + u_ + ku ¼ p0 sin w
t (7.4.5)
w
We define the damping coefficient ch ¼ k=
w, which yields
x h ¼ ch =2mw ¼ =2b (7.4.6)
that is, the damping ratio changes with b.
The solution of Eq. (7.4.5) for the steady-state response is obtained from
Eq. (7.2.2), if x is replaced with xh . Thus
@CivilMethod
t qh Þ
u ðt Þ ¼ rh sin ðw (7.4.7)
where
p0 1
rh ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi (7.4.8)
k 2
1 b2 + 2

tan qh ¼ (7.4.9)
1 b2
Working as in Section 7.2, we can determine the geometry of the hysteresis
loop for this type of damping. Thus, we obtain from Eq. (7.4.7)
u_ ðt Þ2 ¼ r2h w
2 cos 2 ðw
t qh Þ
2 2

¼ rh w 1 sin 2 ðw t qh Þ (7.4.10)

¼w 2 r2h u 2
Moreover, it is
2
k
fD2 ¼ ðf ku Þ2 ¼ u_ 2 (7.4.11)
w
which by virtue of (7.4.10) is written as

ðf ku Þ2 ¼ 2 k 2 r2h u 2 (7.4.12)
or
2
f ku 2 u
+ ¼1 (7.4.13)
krh rh
Therefore, in the plane u, f , Eq. (7.4.13) represents a rotated ellipse
(Fig. 7.4.1). The area of the ellipse is pkr2h and expresses the loss of energy
in a cycle. We observe that Eq. (7.4.13) does not involve the excitation fre-
quency. This implies that the hysteresis loop can be determined experimentally
using a low excitation frequency, that is, quasistatic, by plotting the load-
displacement curve.
For f ¼ 0, we obtain the abscissa of the ellipse on the u axis

d ¼ rh pffiffiffiffiffiffiffiffiffiffiffiffiffi (7.4.14)
1 + 2
which can be used to evaluate the damping coefficient . Hence
d
¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7.4.15)
r2h d 2
@CivilMethod
FIG. 7.4.1 Hysteresis loop in a system with hysteretic damping.
7.5 Coulomb damping

As mentioned in Section 7.1, damping is due to various energy dissipation
mechanisms acting simultaneously. Although the model of viscous damping
or hysteretic damping in most cases can adequately approximate the behavior
of the structure, it is not suitable to describe the response of the structure when
forces due to Coulomb friction are present. These forces arise on the dry contact
interface of two bodies with a relative sliding. The friction force is expressed by
the relation F ¼ mN , where N is the force normal to the contact surface and m
the friction coefficient, which is determined experimentally and is greater
before the sliding. The direction of the frictional force is opposite to that of
motion, that is, its sign is opposite to the sign of the velocity. Therefore, we
can set
F ¼ signðu_ ÞmN (7.5.1)

where
signðu_ Þ ¼ u=
_ ju_ j (7.5.2)
7.5.1 Free vibrations with Coulomb damping

We consider the SDOF system of Fig. 7.5.1a, which is sliding on a dry surface.
When the body is moving to the right (Fig. 7.5.1b) the equilibrium of forces
yields the equation of motion
m u€ + ku ¼ F (7.5.3)
while when the body is moving to the left (Fig. 7.5.1c) the equilibrium of forces
yields the equation of motion
m u€ + ku ¼ F (7.5.4)
@CivilMethod
(a)
(b) (c)
FIG. 7.5.1 Motion with Coulomb friction.
On the basis of Eq. (7.5.1), Eqs. (7.5.3), (7.5.4) can be combined as

m u€ + ku ¼ F signðu_ Þ (7.5.5)
or
m u€ + F signðu_ Þ + ku ¼ 0 (7.5.6)
Eq. (7.5.6) is nonlinear because of the term of damping and can be solved by
any of the methods developed in Chapter 5. In the following, an analytical solu-
tion of Eq. (7.5.6) is presented with zero initial conditions, u ð0Þ ¼ u0 , u_ ð0Þ ¼ 0.
The body is displaced from the equilibrium position to the right by u0 and we
study the subsequent motion. The body will move to the left according to
Eq. (7.5.4). A particular solution to this equation is
F
up ðt Þ ¼ (7.5.7)
k
Hence the general solution of Eq. (7.5.4) is
F
u ðt Þ ¼ A cos wt + B sin wt + (7.5.8)
k
Applying the initial conditions
u ð0Þ ¼ u0 , u_ ð0Þ ¼ 0 (7.5.9)
gives
F
A ¼ u0 , B ¼ 0 (7.5.10)
k
and the displacement (7.5.8) becomes

F F
u ðt Þ ¼ u 0 cos wt + (7.5.11)
k k
@CivilMethod
Eq. (7.5.11) holds until the velocity vanishes, namely t ¼ p=w. At that
instant, the body is at the extreme left position, where the displacement is
u ðp=wÞ ¼ ðu0 2F=k Þ. The body will now start moving to the right with ini-
tial conditions
p
p
F
u ¼ u0 2 , u_ ¼0 (7.5.12)
w k w
The motion is now described by Eq. (7.5.3), which has a particular solution
F
up ðt Þ ¼ (7.5.13)
k
and general solution
F
u ðt Þ ¼ A cos wt + B sin wt (7.5.14)
k
The initial conditions (7.5.12) give
3F
A ¼ u0 , B¼0 (7.5.15)
k

F F
u ðt Þ ¼ u 0 3 cos wt (7.5.16)
k k
FIG. 7.5.2 Free vibrations with Coulomb damping.
@CivilMethod
Eq. (7.5.16) holds until the body reaches the extreme right position, namely
until the instant t ¼ 2p=w. At that time, the body has completed a full oscillation
and the displacement is u ð2p=wÞ ¼ ðu0 4F=k Þ. This solution procedure con-
tinues to obtain the response of the next oscillations. The graphical representa-
tion of the displacement versus time is shown in Fig. 7.5.2. The curve was
obtained by numerical integration of the equation of motion (7.5.6) with
m ¼ 10kNm1 s2 , T ¼ 0:5s, F ¼ 23:685kN, u0 ¼ 0:55m, u_ 0 ¼ using the pro-
gram aem_nlin.m developed in Chapter 5. The numerical results coincide with
those obtained by the above-presented analytical solution. The motion is a
vibration with a period T ¼ 2p=w, which means that the Coulomb friction does
not affect the frequency or the period of vibration. The amplitude of vibration is
reduced in each cycle by 4F=k. A consequence of this is that the envelopes of
the curve are straight lines, unlike in the cases of viscous or hysteretic damping
where the envelopes are exponential functions. The motion of the system con-
tinues until the elastic force ku becomes smaller than the force F of the friction.
Until now, no difference was made between static friction Fs ¼ ms N and
dynamic friction Fd ¼ md N . The first occurs when the body is stationary and
the second when the body moves. Generally, it is md < ms , hence the dynamic
friction coefficient md will be used in the equation of motion while the static
friction coefficient ms is used for the control of the motion. In viscous or hys-
teretic damping, theoretically, the body does not stop moving because the
amplitude of the vibration reduces exponentially. Nevertheless, real structural
systems stop after a finite time. This is due to the fact that the Coulomb friction
coexists with other forms of damping and forces the moving systems to stop.
7.5.2 Forced vibrations with Coulomb damping

In this case, the equation of motion becomes
m u€ + F signðu_ Þ + ku ¼ pðt Þ (7.5.17)
An analytical solution can be achieved by splitting the above equation into
two equations. Thus the equation describing the motion to the right is
m u€ + ku ¼ F + pðt Þ (7.5.18)
Z
F 1 t
u ðt Þ ¼ Acos wt + B sin wt + pðτÞ sin wðt τÞdτ (7.5.19)
k mw 0
On the other hand, the equation describing the motion to the left is
m u€ + ku ¼ F + pðt Þ (7.5.20)
Z
F 1 t
u ðt Þ ¼ A cos wt + B sin wt + + pðτÞsin wðt τÞdτ (7.5.21)
k mw 0
@CivilMethod
t the general solution is (see Eq. 3.2.8)

For a harmonic load pðt Þ ¼ p0 sin w
F p0 1
w
u ðt Þ ¼ A cos wt + B sin wt + t, b ¼
sin w (7.5.22)
k k 1 b2 w
Apparently, the derivation of the solution for forced vibrations becomes

quite complicated if we follow the procedure applied for the free oscillations,
that is, by splitting the equation of motion into two equations. Nevertheless, for
small damping, we can approximate the solution in the phase of the steady-state
response. Thus, the solution is given by Eq. (7.2.2).
The graph of the friction force versus displacement in one cycle takes the
form of the rectangle of Fig. 7.5.3. Hence, the work produced by the friction
force in a complete oscillation is
WD ¼ 4Fr (7.5.23)
We can determine an equivalent coefficient of viscous damping by equating
the loss of energy (Eq. 7.5.23) with that of the viscous damping given by
Eq. (7.3.1). Namely,
FIG. 7.5.3 Hysteresis loop in a system with Coulomb damping.
wr2 ¼ 4Fr
ceq p (7.5.24)
which gives
4F
ceq ¼ (7.5.25)
p
wr
and an equivalent damping ratio
ceq 2F
xeq ¼ ¼ (7.5.26)
2mw pkrb
@CivilMethod
Substituting this value of the damping ratio in Eq. (7.2.3) yields

p0 1
r¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi (7.5.27)
k 2
2 2
1 b + ð4F=pkrÞ
which is solved for r to give

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
p0 u1 ð4F=pp0 Þ2
r¼ t 2 (7.5.28)
k 1 b2
Eq. (7.5.28) holds if 1 ð4F=pp0 Þ2 > 0, that is, when F=p0 < p=4. Obvi-
ously, for F=p0 > p=4 r becomes imaginary and this method for determining
an equivalent damping coefficient does not apply.
The phase angle results from Eq. (7.2.4) by setting x ¼ xeq and taking the
value of r from Eq. (7.5.28). Thus, we have
4F=pp0
tan q ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7.5.29)
1 ð4F=pp0 Þ2
The plus sign is valid when b < 1 while the minus sign is valid when b > 1.
7.6 Damping modeling via fractional derivatives

7.6.1 Introduction
In the modeling of the linear elastic behavior of structures, damping has been
classically modeled as being linearly dependent on velocity. This damping
model is adequate for the dynamic analysis of lightly damped structures. How-
ever, this model, while mathematically straightforward, fails to describe the
damping response of heavily damped structures. This drawback is more pro-
nounced in materials used in devices aimed at increasing the damping of struc-
tures, such as seismic isolators.
Historically, the need for more refined models to control damping has
pushed researchers to use viscoelastic models in the dynamic analysis of struc-
tures. Applicable viscoelastic models are expressed as a series of ordinary time
derivatives relating time-dependent stress to strain fields. These models yield
acceptable plots of material properties, but they have drawbacks. Normally,
these models contain many terms, making them mathematically cumbersome
and increasing the order of the differential equations describing the response
of the system. An alternative representation of these models is realized via
fractional derivatives, that is, derivatives of noninteger order [6]. The real
strength of this approach is that the fractional derivative models describe
@CivilMethod
damping effects using fewer material parameters than the integer order differ-
ential models but with equal precision. This approach involves fractional dif-
ferential equations, which needs acquaintance with fractional calculus.
7.6.2 The fractional derivative

The fractional (noninteger order) derivatives are as old as calculus. The theory
of derivatives of noninteger order goes back to G. W. Leibnitz. After Leibnitz
defined the derivative of integer order, d n y=dx n with n being an integer,
L’H^ opital asked: “What if n is a fraction, say n ¼ 1=2?” Leibnitz gave an
answer to this question (Sept. 30, 1695) concluding “It will be a paradox”
and added prophetically “From this apparent paradox, one day useful conse-
quences will be drawn” [7]. For three centuries, the fractional derivative
inspired pure theoretical mathematical developments useful only for mathema-
ticians. The integer order derivative allows giving geometrical interpretations to
the proposed physical models resulting from Newton’s law of motion. Appar-
ently, this made the then-revolutionary concepts accessible to contemporary
scientists, who were well experienced in geometry. In a sense, the long delay
to apply fractional calculus may be attributed to this fact. We recall that frac-
tional derivatives lack the straightforward geometrical interpretation of their
integer counterparts. However, research carried out in recent years has pointed
out that fractional order derivatives provide an effective tool to reliably model
many complex physical and engineering systems. In recent years, this fact gave
a great boost to the study of fractional differential equations. In this context,
many books on fractional calculus as well as numerous publications on the
study of physical and engineering systems via fractional derivatives have been
written; see for example [8] and the references therein.
There are several definitions of the fractional derivative [8]. The Riemann-
Liouville and Caputo fractional derivatives are among the most widely used.
Nevertheless, as the scope of this section is not to give a detailed account of
this subject, we will restrict our discussion to the Caputo derivative, defined as
Z t
1 u ðmÞ ðτÞ
DCa u ðt Þ ¼ dτ, m 1 < a < m (7.6.1)
Gðm aÞ 0 ðt τÞa + 1m
where Gðz Þ is the gamma function with argument z, a is the order of the frac-
tional derivative, and m an integer. It can be shown that
lim DCa u ðt Þ ¼ u ðmÞ ðt Þ lim DCa u ðt Þ ¼ u ðm1Þ ðt Þ u ðm1Þ ð0Þ (7.6.2)
a!m a!m1
Obviously, for m ¼ 1, the fractional derivative interpolates the first-order

derivative [9]. Fig. 7.6.1 shows the fractional derivative of the function
u ¼ t t 3 =6 + t 5 =120 for various orders of the fractional derivative.
@CivilMethod
1.2
1 D 1u
c
0.8 D 0.99999u
c
D 0.9u
0.6 c
D 0.5u
0.4 c
D 0.1u
c
0.2
D 0.00001u
c
0 u
−0.2
−0.4
−0.6
0 0.5 1 1.5 2 2.5 3
FIG. 7.6.1 The fractional derivative of the function u ¼ t t 3 =6 + t 5 =120 for various orders.
If damping is modeled via the fractional derivative, the equation of motion

of a SDOF system is written as
m u€ + cD aC u + ku ¼ pðt Þ (7.6.3)
Obviously, for a ! 1, we obtain the equation of motion with viscous

damping
m u€ + cu_ + ku ¼ pðt Þ (7.6.4)
while for a ! 0 the equation of motion becomes
m u€ + ðc + k Þu ¼ pðt Þ + cu 0 (7.6.5)
that is, it yields the elastic response with stiffness k ∗ ¼ c + k and excitation
force p∗ ¼ pðt Þ + cu 0 . The Caputo derivative is employed because, contrary
to other types of fractional derivatives, it allows the application of initial con-
ditions having a direct physical significance.
Eq. (7.6.3) represents a fractional differential equation. Analytical solutions
of such equations are difficult or impossible to obtain. This reason has recently
boosted the development of efficient numerical methods for solving fractional
differential equations [10, 11].
Fig. 7.6.2 shows the free vibration response of an DOF system for various
values of the order a
@CivilMethod
FIG. 7.6.2 Free vibration response of a SDOF system for various values of the order a.
The solution of Eq. (7.6.3) has been obtained using the method presented
in [11]. The developed MATLAB program has been given the name
three_term_FD.m and is included on this book’s companion website. Note
that for a 1 and a 0 the solutions are identical to the corresponding
analytical ones.
The fractional calculus has allowed the definition of any order of fractional
derivative, real or imaginary. This fact enables us to consider the fractional
derivative to be a function of time (explicit variable-order fractional derivative)
or of some other time-dependent variable (implicit variable-order fractional
derivative). Thus, the variable-order Caputo derivative for m ¼ 1 reads
Z t
aðt Þ 1 u_ ðτÞ
DC u ð t Þ ¼ dτ ðExplicitÞ (7.6.6a)
Gð1 aðt ÞÞ 0 ðt τÞaðt Þ
Z t
aðu Þ 1 u_ ðτÞ
DC u ð t Þ ¼ dτ ðImplicitÞ (7.6.6b)
Gð1 aðu ÞÞ 0 ðt τÞaðu Þ
Z t
aðu_ Þ 1 u_ ðτÞ
DC u ð t Þ ¼ dτ ðImplicitÞ (7.6.6c)
Gð1 aðu_ ÞÞ 0 ðt τÞaðu_ Þ
The concept of a variable-order fractional derivative exhibits notable advan-
tages over the constant order derivative and it has been recently used to model
the dynamic response of actual structures [12, 13].
7.7 Measurement of damping

As stated in Chapter 1, the mass and stiffness of a dynamic system can be deter-
mined from its physical characteristics. The mass can be determined from the
geometry and the mass density of the structural elements. The stiffness is also
@CivilMethod
determined if the geometry and material properties of the structural elements

are known. However, it is difficult or at least impractical to relate damping
to known or measurable physical characteristics of the system. Therefore, the
damping of a given structure cannot be measured precisely during the design
phase, but only experimentally after its construction. There are several exper-
imental techniques to measure the damping of a structure. Most of them are
based on the assumption that the damping is viscous. When damping is not vis-
cous in nature, an equivalent viscous damping is usually determined.
7.7.1 Free vibration decay method

This is the simplest and most frequently used method of determining the viscous
damping ratio x through experimental measurements. This method was already
presented in Chapter 2 when we discussed the free damped vibrations. The sys-
tem is excited by means of an appropriate experimental instrumentation and is
left to perform free oscillations. Subsequently, the peak amplitudes ui and ui + n
over n consecutive cycles are measured. As shown in Section 2.3.2, the damp-
ing ratio can be calculated using Eq. (2.3.24), namely

2npx ui
pffiffiffiffiffiffiffiffiffiffiffiffi ¼ ln (7.7.1)
1 x2 ui + n
which yields
dn
x ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7.7.2)
4p n 2 + d n 2
2
where

ui
dn ¼ ln (7.7.3)
ui + n
This method was illustrated in Example E6.2.1.
When the damping is hysteretic, the damping force is given by Eq. (7.4.2). In
w, provided that the damping is small. Hence,
free vibrations, it can be set as w
the equivalent damping coefficient ch and the damping ratio are obtained from
the relations
k
ch ¼ (7.7.4a)
w

xh ¼ (7.7.4b)
2
The hysteretic damping coefficient is obtained by using Eq. (7.7.2), if it is
set x ¼ xh . This yields
@CivilMethod
2dn
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7.7.5)
4p n 2 + dn 2
2
Example 7.7.1 Evaluation of damping in a SDOF system

In a test, the silo in Example 2.3.1 is subjected to harmonic excitation at a fre-
quency 15s1 and the force-displacement relation is obtained at the steady state.
The amplitude of the displacement is measured as 10 cm, and the energy loss per
cycle 50kNm. If damping is considered to be viscous, determine c and x.
Solution
The mass of the system is m ¼ 100kNm1 s2 , the stiffness k ¼ 2721:6kN=m,
and frequency w ¼ 5:2169s1 , taken from Example 2.3.1.
(i) For viscous damping, the energy loss per cycle is given by Eq. (7.2.6),
namely
wr2
WD ¼ pc (1)
1
¼ 15s , r ¼ 0:1m, WD ¼ 50kNm gives
which for w
WD 50
c¼ ¼ ¼ 106:10 (2)
wr2 p 15 0:12
p
c 106:10
x¼ ¼ ¼ 0:102 (3)
2mw 2 100 5:2169s1
7.7.2 Resonance amplitude method

t. In the phase of
The system is subjected to harmonic excitation pðt Þ ¼ p0 sin w
the steady-state response, the amplitude and the phase angle are given by
Eq. (3.2.27), namely
p0 h i 1
2 2
r¼ 1 b2 + ð2xb Þ2 (7.7.6a)
k

2xb
q ¼ tan 1 (7.7.6b)
1 b2
For b ¼ 1 Eq. (7.7.6b) yields q ¼ 90° regardless of the value of x. If the
employed experimental instrumentation allows the measurement of the phase
angle q, then we adjust the excitation frequency so that q ¼ 90° and we measure
the amplitude of vibration r. Besides, setting b ¼ 1 in Eq. (3.2.28) we obtain
r 1
D ¼ max jRðt Þj ¼ ¼ (7.7.7)
p0 =k 2x
which yields
@CivilMethod
p0 =k
x¼ (7.7.8)
2r
This previous method requires knowledge of the stiffness k of the structure,
which is determined either from the physical characteristics of the structure or
experimentally, for example, by imposing a load and measuring the resulting
displacement.
If the measurement of the phase difference is not easy, then we measure
experimentally the amplitude of the vibration in the range of resonance. Sub-
sequently, we plot the curve Dðb Þ ¼ rðb Þ=ðp0 =k Þ (see Fig. 7.7.1) and determine
its maximum. Then Eq. (3.2.31), namely
1
Dmax ¼ pffiffiffiffiffiffiffiffiffiffiffiffi (7.7.9)
2x 1 x 2
FIG. 7.7.1 Graphical representation of the curve D ðbÞ in the range of resonance.
allows the evaluation of x. For small values of x, it is x 2 0, hence

x 1=2Dmax . This method also requires knowledge of the stiffness k of the
structure.
7.7.3 Width of response curve method

The width of the response curve DðbÞ in the range of resonance can be used to
determine the damping of the structure. In this method, the frequencies corre-
sponding to the phase angles 45° are measured. One of these frequencies is
@CivilMethod
below the resonance frequency while the other is above it (see Fig. 7.7.2). The
respective values of b are obtained from Eq. (7.7.6b). Thus, we have
2xb1
¼1 (7.7.10a)
1 b21
2xb2
¼ 1 (7.7.10b)
1 b22
which can be written as
1 b21 2xb1 ¼ 0 (7.7.11a)
1 b22 2xb2 ¼0 (7.7.11b)
FIG. 7.7.2 Response curve DðbÞ in the neighborhood of resonance to determine bI and b II .
Subtracting the previous equations gives

1 1w 1
2 w
x ¼ ðb 2 b 1 Þ ¼ (7.7.12)
2 2 w
1 and w
Eq. (7.7.12) can be used to calculate x when the frequencies w 2 have
been measured and the natural frequency w is known or can be determined.
The previously described method is based on the capability of measuring the
phase angle, which may not be simple because it requires complex instrumen-
tation. However, another property of the response curve may be used to deter-
mine the damping ratio. This method is not based on the measurement of the
phase angle, but on experimental measurements. The response curve D ðb Þ is
plotted in the range of resonance and the values b corresponding to
@CivilMethod
pffiffiffi
1= 2Dmax are determined. It is apparent from Fig. 7.7.2 that these values are
two, which are denoted by bI and bII . For small values of the damping ratio, it is
Dmax 1=2x. Besides, using Eq. (3.2.28) we can write
1 1 1
pffiffiffi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(7.7.13)
2 2x 2
1 b2 + ð2xb Þ2
The solution of the above equation gives two values of b

2
b I ¼ 1 2x 2x 1 + x2
2
(7.7.14a)
b 2II ¼ 1 2x2 + 2x 1 + x2 (7.7.14b)
Using the binomial theorem to expand 1 + x 2 and neglecting terms of
order higher than the second, the above relations reduce to
b2I 1 2x 2x 2 (7.7.15a)
b2II 1 + 2x 2x 2 (7.7.15b)
or after expanding in Taylor series

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3
bI 1 2x 2x2 1 x x 2 (7.7.16a)
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3
b II 1 + 2x 2x2 1 + x x 2 (7.7.16b)
2
which give
1
x ¼ ðbII b I Þ (7.7.17)
2
Eq. (7.7.17) is similar to Eq. (7.7.12), but avoids the measurement of the
phase angles.
7.8 Problems
Problem 7.1 The damping force applied to a system moving in a fluid is given
by the relation fD ¼ lu_ a , where l and a are real constants. Give:
(i) The graphical representation of the displacement u ðt Þ when the system
performs free vibrations with initial conditions u ð0Þ ¼ 0:02m, u_ ð0Þ ¼ 0.
(ii) The graphical representation of the displacement u ðt Þ when the system is
t with zero initial conditions.
subjected to the harmonic load pðt Þ ¼ p0 sin w
Data: m ¼ 10kNm1 s2 , k ¼ 1500kN=m, l ¼ 100, a ¼ 3, p0 ¼ 300kN, and
¼ 2s1 .
w
@CivilMethod
Problem 7.2 The system of Fig. P7.2 is subjected to Coulomb damping. Give:
(i) The graphical representation of the displacement u ðt Þ when the system
performs free vibrations with initial conditions (a) u ð0Þ ¼ 0:40m,
u_ ð0Þ ¼ 0 and (b) u ð0Þ ¼ 0, u_ ð0Þ ¼ 2:8ms1 .
(ii) The graphical representation of the displacement u ðt Þ when the system is
t with zero initial conditions.
subjected to the harmonic load pðt Þ ¼ p0 sin w
(iii) Calculate the equivalent damping ratio xeq of the equivalent viscous
damping.
Data: m ¼ 10:132kNm1 s2 , k ¼ 1600kN=m, b ¼ w
=w ¼ 0:4, N ¼ 70kN,
m ¼ 0:32, and p0 ¼ 1:57F.
FIG. P7.2 System subjected to Coulomb damping.
Problem P7.3 Two bodies B1 and B2 with masses m1 and m2 , respectively, are
placed on two inclined planes whose angles are f1 and f2 , as shown in
Fig. P7.3. The bodies are connected by a massless cable of length L and axial
stiffness k ¼ EA=L. The friction coefficient between the bodies and the inclined
planes is m while between the cable and the pulley it is zero. Determine the
equation of motion of the system.
Problem P7.4 The damping of a SDOF system is expressed by the Caputo frac-
tional derivative of order a ¼ 0:5. Compare the response of the system with that
of viscous damping. Data: m ¼ 10kNm1 s2 , x ¼ 0:1, k ¼ 500kN=m,
p ¼ p0 sin 5t, and p0 ¼ 2kN. Hint: Use the program three_term_FD.m avail-
able on this book’s companion website.
@CivilMethod

[1] L. Rayleigh, Theory of Sound, second ed., Dover Publications, New York, 1877; reissued
1945.
[2] R.W. Clough, J. Penzien, Dynamics of Structures, second ed., McGraw-Hill, New York, 1993.
[5] S. Adhikari, Damping Models for Structural Vibration, Dissertation submitted to the Univer-
sity of Cambridge for the Degree of Doctor of Philosophy, Trinity College, Cambridge, 2000.
[6] R.L. Bagley, P.J. Torvik, Fractional calculus—a different approach to the analysis of viscoe-
lastically damped structures, AIAA J. 27 (1998) 1412–1417.
[7] G.W. Leibniz, S€amtliche Schriften und Briefe, 3te Reihe, Mathematischer, Naturwis-
senschaftlicher und Technischer Briefwechsel, Band VI, Letter 163, 2004, p. 510. Berlin-
Brandenburgischen Akademie der Wissenschaften und Akademie der Wissenschaften in
G€ottingen.
[8] I. Podlubny, Fractional Differential Equations, Academic Press, New York, 1999.
[9] J.T. Katsikadelis, Generalized fractional derivatives and their applications to mechanical sys-
tems. Arch. Appl. Mech. 85 (2015) 1307–1320, https://doi.org/10.1007/s00419-014-0969-0.
[10] K. Diethelm, N.J. Ford, A.D. Freed, Y. Luchko, Algorithms for fractional calculus: a selection
for numerical Methods, Comput. Methods Appl. Mech. Eng. 194 (2005) 743–773.
[11] T. Katsikadelis, Numerical Solution of multi-term fractional differential equations, ZAMM,
Z. Angew. Math. Mech. 89 (7) (2009) 593–608.
[12] C.F.M. Coimbra, Mechanics with variable-order differential operators, Ann. Phys. (Leipzig)
12 (11–12) (2003) 692–703, https://doi.org/10.1002/andp.200310032.
[13] J.T. Katsikadelis, Numerical solution of variable order fractional differential equations,
arXiv:1802.00519 [math.NA] (2018).
@CivilMethod
Chapter 8
Generalized single-degree-of-
freedom systems—Continuous
systems
Chapter outline
8.1 Introduction 269 8.3.3 Free vibrations of beams 286
8.2 Generalized single-degree-of- 8.3.4 Orthogonality of the
freedom systems 275 free-vibration modes 291
8.3 Continuous systems 284 8.3.5 Forced vibrations of
8.3.1 Introduction 284 beams 293
8.3.2 Solution of the beam 8.4 Problems 295
equation of motion 285 References and further reading 297
8.1 Introduction
In this chapter, the method of global shape functions is employed to approx-
imate the response of continuous systems by SDOF systems, which we call
generalized SDOF systems. The example that follows helps in understanding
the basic ingredients of the method as well as the error introduced by the
lumped mass assumption.
In order to study the dynamic response of the frame shown in Fig. 8.1.1a, we
approximate it by the model of Fig. 8.1.1b. As mentioned, in formulating this
model, it was assumed that the mass of the columns is concentrated at their ends,
that is, half at the top and half at the foot of the column. The consequence of this
assumption is that the elastic curve of the columns has the form of an unloaded
beam fixed at both ends, whose end cross-sections undergo a relative displace-
ment u ðt Þ. This assumption is, however, not entirely correct because the mass of
the columns is actually distributed along their length. This fact, apparently, pro-
u€
duces inertial forces fI ðx, t Þ ¼ m ðx, t Þ, where uðx, t Þ is the actual elastic curve
of the columns (see Fig. 8.1.1c). Thus, the problem would be correctly
addressed if the columns were treated as systems of infinite degrees of freedom,
that is, continuous systems whose top cross-sections are connected by the rigid
beam of mass m. Such an approach, however, would be quite difficult because
the analysis leads to partial differential equations (see Section 1.1). Neverthe-
less, it is possible to approximate the system by another model, which is closer
Dynamic Analysis of Structures. https://doi.org/10.1016/B978-0-12-818643-5.00008-X

@CivilMethod
to the actual system than that of Fig. 8.1.1b. In this model, we consider that the
mass is distributed along the length of columns, but their deformed shape is cho-
sen so that it is close to the actual one. We observe that the end cross-sections of
the columns in Fig. 8.1.1a do not rotate during the motion. Hence the shape func-
tion ðx Þ ( see Eq. (8.1.1)) representing the shape of the elastic curve should be
chosen so that it satisfies the geometrical boundary conditions at the ends of the
columns, which demand ð0Þ ¼ 0, 0 ð0Þ ¼ 0, ðh Þ 6¼ 0, and 0 ðh Þ ¼ 0. The func-
tions ðx Þ are not unique. They constitute an infinite set of functions called
geometrically admissible functions.
(a) (b)
(c)
FIG. 8.1.1 Model of a two-column frame with rigid beam.
After this consideration, we may set

uðx, t Þ ¼ ðx Þu ðt Þ (8.1.1)
where u ðt Þ is a time-dependent function denoting the displacement of a
certain cross-section of the column, say at x ¼ a, 0 < a h. This yields
uða, t Þ ¼ ða Þu ðt Þ ¼ u ðt Þ, which implies that ða Þ ¼ 1. If it is taken a ¼ h, then
u ðt Þ ¼ uðh, t Þ represents the displacement at the top of the column.
Eq. (8.1.1) states that the shape of the elastic curve remains the same
during motion while its values are multiplied by the common factor u ðt Þ.
The function ðx Þ, which approximates the elastic curve, is referred to as shape
function.
In the example of Fig. 8.1.1, a shape function for the columns is the elastic
curve of a beam with constant stiffness EI , whose ends are subjected to a rel-
ative displacement equal to one. Therefore, the shape function is obtained from
the solution of the differential equation of the elastic curve of the beam in the
absence of loading (Fig. 8.1.2a)
d 4 ðx Þ
EI ¼0 (8.1.2)
dx 4
@CivilMethod
Generalized single-degree-of-freedom systems—Continuous systems Chapter 8 271
(a) (b)
FIG. 8.1.2 Shape function (a) and and deflection curve (b) in Example 8.1.1
with boundary conditions

0 0
ð0Þ ¼ 0, ð0Þ ¼ 0, ðh Þ ¼ 1, ðh Þ ¼ 0 (8.1.3)
Integrating Eq. (8.1.2) yields
x3 x2
ðx Þ ¼ c 1 + c2 + c3 x + c4 (8.1.4)
6 2
where ci (i ¼ 1, 2, 3, 4) are arbitrary constants.
Applying now the boundary conditions (8.1.3), we obtain
12 6
c1 ¼ , c2 ¼ , c3 ¼ 0, c4 ¼ 0 (8.1.5)
h3 h2
Then substituting these values for the arbitrary constants into Eq. (8.1.4)
yields the shape function
x 2 x 3
ðx Þ ¼ 3 2 ¼ 3x2 2x3 , x ¼ x=h (8.1.6)
h h
After the selection of the shape function, the motion of the system is deter-
mined by the time-dependent function u ðt Þ. The equation of motion can be
derived using different methods such as Hamilton’s principle, Lagrange equa-
tions, or the principle of virtual work. To implement these methods, it is nec-
essary to determine the elastic energy U and the kinetic energy T of the system
as well as the virtual work of the external force pðt Þ.
(a) Evaluation of the elastic energy U . The elastic energy is due to the strain
energy of the two columns because of bending. The strain energy per unit
volume is
1
U0 ¼ s x e x (8.1.7)
2
@CivilMethod
where sx is the normal stress on the cross-section of the column and ex the cor-
responding strain. For a linearly elastic material with modulus of elasticity E,
it is ex ¼ sx =E. Taking into account that the bending stress in a beam is
sx ¼ Mx y=I ðx Þ, we obtain

1 Mx y 2
U0 ¼ (8.1.8)
2E I ðx Þ
where I ðx Þ is the moment of inertia of the, in general, variable cross-section
and Mx the bending moment. Hence, the total elastic energy of the one
column is
Z
U
¼ U0 dV
2 V
Z (8.1.9)
1 Mx y 2
¼ dV
V 2E I ðx Þ
where V is the volume of the column. It is known from the beam theory
that
Mx ¼ EI ðx Þu00 ðx, t Þ (8.1.10)
which is introduced into Eq. (8.1.9) to yield
Z
U 1 h
EI ðx Þ½u00 ðx, t Þ dx
2
¼ (8.1.11)
2 2 0
For I ðx Þ ¼ I ¼ constant and uðx, t Þ ¼ ðx Þu ðt Þ, the previous equation
becomes
Z h
U EI
½ 00 ðx Þ dx
2 2
¼ ½u ðt Þ (8.1.12)
2 2 0
Differentiating Eq. (8.1.6) twice with respect to x gives

00 6
ðx Þ ¼ ð1 2xÞ (8.1.13)
h2
Substituting Eq. (8.1.13) into Eq. (8.1.12) and integrating give
U 1 12EI 2
¼ u (8.1.14)
2 2 h3
Therefore, the elastic energy of the two columns is
12EI 2
U¼ u (8.1.15)
h3
and its variation
12EI
dU ¼ 2 udu (8.1.16)
h3
@CivilMethod
(b) Evaluation of the kinetic energy T . The kinetic energy is due to the velocity
u_ ðt Þ of the mass m of the horizontal beam as well as to the velocity u_ ðx, t Þ
of the mass m at points x of the column axis. Thus, we have
Z h
1 1
½u_ ðx, t Þ dx
2
T ¼ m u_ + 2
2
m (8.1.17)
2 0 2
By virtue of Eq. (8.1.1), the above relation is written as
Z h
1
T ¼ m u_ 2 + m u_ 2 ½ ðx Þ2 dx (8.1.18)
2 0
Then using Eq. (8.1.6) and integrating give

1
T ¼ m u_ 2 + 0:371mh
u_ 2 (8.1.19)
2
and
Þud
dT ¼ ðm + 0:742h m _ u_ (8.1.20)
(c) Evaluation of the virtual work dWnc . This is due to the nonconservative
external force pðt Þ. This is
dWnc ¼ pðt Þdu (8.1.21)
Now we proceed to the derivation of the equation of motion using:
1. Hamilton’s principle. Substituting Eqs. (8.1.16), (8.1.20), (8.1.21) into
Eq. (1.7.13) and taking into account that A ¼ 0, we obtain
Z t2
EI
24 3 udu ðm + 0:742h m Þud
_ u_ pðt Þdu dt ¼ 0 (8.1.22)
t1 h
which after elimination of du_ using integration by parts yields the equa-
tion of motion
m ∗ u€ + k ∗ u ¼ p∗ ðt Þ (8.1.23)
where
24EI

m ∗ ¼ m + 0:741mh, k∗ ¼ , p∗ ðt Þ ¼ pðt Þ (8.1.24)
h3
2. Lagrange equations. The equation of motion will result from

Eq. (1.8.11) by setting qi ¼ u, A ¼ 0 and Qi ¼ pðt Þ. This gives

d ∂T ∂T ∂U
+ ¼ pðt Þ (8.1.25)
dt ∂u_ ∂u ∂u
which by virtue of Eqs. (8.1.15), (8.1.19) yields Eq. (8.1.23).
@CivilMethod
3. The principle of virtual work. The principle of virtual work stated in

Section 1.6 for rigid bodies can also be extended to deformable bodies,
but now we must take account of the work done by the internal forces in
riding through the virtual deformation. Thus, for a deformable body, the
principle of virtual work becomes
dWex ¼ dWin (8.1.26)
where dWex and dWin denote the virtual work of the external and inter-
nal forces, respectively.
For the system of Fig. 8.1.1, the external forces are the excitation
force pðt Þ and the inertial forces. Hence
Z h
€ 2
dWex ¼ m udu u€
m ðx, t Þd uðx, t Þdx + pðt Þdu
0
€ 0:742mhdu
dWex ¼ m udu + pðt Þdu (8.1.27)
The virtual work of the internal forces for the two columns is
obtained from the relationa
Z
dWin ¼ 2 sx dex dV (8.1.28)
V
where V is the volume of the column.

Taking into account that sx ¼ Mx y=I , sx ¼ Eex , Mx ¼ EI u00 ðx, t Þ,
we obtain from Eq. (8.1.28)
12EI
dWin ¼ 2 udu (8.1.29)
h3
Substituting Eqs. (8.1.27), (8.1.29) into Eq. (8.1.26) yields the equation
of motion (8.1.23).
a
The expression for the strain energy of a beam due to bending is derived from the strain energy
of the beam by taking its variation ([1], Chap. 1):
Z Z
1 1
Win ¼ sx ex dV ¼ Es2 dV (a)
2 V 2 V x
which by taking the variation gives
Z Z
dWin ¼ Eex dex dV ¼ sx dex dV (b)
V V
@CivilMethod
FIG. 8.1.3 Dependence of T*/T on the ratio of the column mass over total beam mass in the frame
of Fig. 8.1.1
If the mass of the column is assumed concentrated at the ends of the col-
umns, the coefficients of the corresponding equation of motion result as
24EI

m ∗ ¼ m + mh, k∗ ¼ , p∗ ðt Þ ¼ pðt Þ (8.1.30)
h3
We observe that the generalized mass is less by 0:259mh. Fig. 8.1.3

shows the variation of the ratio T ∗ =T versus the ratio mh=m, where T ∗
is the natural period of the generalized single-degree-of-freedom system
and T the the period of the model in Fig. 8.1.1b. We observe that the lumped
mass assumption has a small influence on the natural period when the mass
of the columns with respect to the mass of the horizontal beam is small.
Illustrative examples facilitating the comprehension of all concepts are pre-
sented and the pertinent bibliography with recommended references for fur-
ther study is also included. The chapter is enriched with problems to be
solved.
8.2 Generalized single-degree-of-freedom systems

After the introductory example in the previous section, we can extend the dis-
cussed method to approximate the response of a more complex system with a gen-
eralized SDOF system. When the members of the system are undeformed, the
substitute SDOF system represents the actual response of the system. Such exam-
ples were discussed in Chapter 1. Example 1.7.3 is a representative case. The
shape function of that example is shown in Fig. E1.11. However, when the mem-
bers connecting the lumped masses are deformable, then the generalized SDOF
system simulating the actual one can be used to approximate its response. The
accuracy of the approximation depends on the selection of the shape function.
@CivilMethod
Nevertheless, the method allows us to approximate easily the dynamic character-

istics of a complex system, thus circumventing exact solution methods.
To illustrate the application of the method to complex systems, we consider
the beam of Fig. 8.2.1a, which is fixed at point 1 and simply supported at point 2.
Moreover, the beam rests on the nonhomogeneous Winkler’s type elastic foun-
dation with variable reaction modulus k ðx Þ. The foundation reacts also with dis-
tributed linear damping having coefficient cðx Þ. The cross-section of the beam
is variable, hence I ¼ I ðx Þ and m ¼ m ðx Þ. The beam caries the lumped masses
mi with rotational inertia Ii at points x ¼ xi , i ¼ 1, 2, …,N . The beam is loaded
by the distributed load pðx, t Þ and the axial load P at the end cross-section 2.
The beam may also be subjected to concentrated loads and/or concentrated and
distributed moments. However, for the convenience of the illustration of the
method, these loads are not considered.
To treat the structure as a generalized SDOF, we will seek the transverse
deflection in the form of Eq. (8.1.1), that is, uðx, t Þ. The shape function ðx Þ
will be selected so that
0 0
ð0Þ ¼ 0, ð0Þ ¼ 0, ðLÞ ¼ 0, ðLÞ 6¼ 0 (8.2.1)
The deflection curve of the beam fixed at one end and simply supported at
the other is a geometrically admissible function and thus can be used as a shape
function. Taking ðL=2Þ ¼ 1, we may write
uðx, t Þ ¼ ðx Þu ðt Þ (8.2.2)
Obviously, it is uðL=2, t Þ ¼ u ðt Þ, hence u ðt Þ represents the displacement at
the middle of the beam.
We will use Hamilton’s principle to derive the equation of motion of
the generalized SDOF system. For this purpose, we evaluate the quantities
U , K , dWnc , and A.
(a)
(b)
(c)
(d)
FIG. 8.2.1 Beam resting on foundation with nonhomogeneous elastic and damping reaction.
@CivilMethod
(a) Evaluation of the elastic energy U . This energy consists of the strain energy
of the beam due to bending and the elastic energy of the springs of Winkler’s
model. That is,
Z Z
1 L 00 2 1 L
U¼ EI ðx Þ½u ðx, t Þ dx + k ðx Þ½uðx, t Þ2 dx (8.2.3)
2 0 2 0
which yields
Z L Z L
dU ¼ EI ðx Þu00 ðx, t Þd u00 ðx, t Þdx + k ðx Þuðx, t Þd uðx, t Þdx (8.2.4)
0 0
Eq. (8.1.1) gives duðx, t Þ ¼ ðx Þdu, du00 ðx, t Þ ¼ 00

ðx Þdu. Thus,
Eq. (8.2.4) becomes
dU ¼ k ∗ udu (8.2.5)
where
Z L Z L
00 2
k∗ ¼ EI ðx Þ½ ðx Þ dx + k ðx Þ½ ðx Þ2 dx (8.2.6)
0 0
(b) Evaluation of the kinetic energy T . This energy consists of the kinetic
energy of the distributed mass m ðx Þ and the kinetic energy of the lumped
masses mi .That is
Z h i2
1 L 1X N
1X N
0
m ðx Þ½u_ ðx, t Þ dx + mi ½u_ ðxi , t Þ + Ii u_ ðxi , t Þ
2 2
T¼
2 0 2 i¼1 2 i¼1
(8.2.7)
0
where u_ ðx, t Þ and u_ ðx, t Þ are the transverse velocity and the angular veloc-
ity, respectively, at the cross-section x. Eq. (8.2.7) gives
Z L X
N
dT ¼ m ðx Þu_ ðx, t Þdu_ ðx, t Þdx + mi u_ ðxi , t Þdu_ ðxi , t Þ
0 i¼1
(8.2.8)
X
N
0 0
+ Ii u_ ðxi , t Þdu_ ðxi , t Þ
i¼1
which by virtue of Eq. (8.1.1) results in

_ u_
dT ¼ m ∗ ud (8.2.9)
where
Z L X
N X
N
Ii ½ 0 ðxi Þ
2
m∗ ¼ m ðx Þ½ ðx Þ2 dx + mi ½ ðxi Þ2 + (8.2.10)
0 i¼1 i¼1
@CivilMethod
(c) Evaluation of the virtual work dWnc of the nonconservative forces. This is
due to the damping force fD ðx, t Þ ¼ cðx Þu_ ðx, t Þ and the external force
pðx, t Þ. Hence we have
Z L Z L
dWnc ¼ cðx Þu_ ðx, t Þd uðx, t Þdx + pðx, t Þd uðx, t Þdx (8.2.11)
0 0

dWnc ¼ ½c∗ u_ + p∗ ðt Þdu (8.2.12)
where
Z L
c∗ ¼ cðx Þ ðx Þdx (8.2.13a)
0
Z L
p∗ ðt Þ ¼ pðx, t Þ ðx Þdx (8.2.13b)
0
(d) The potential A of the conservative forces. This is due to the work of the
axial force P. Obviously, if the axial deformation is neglected, it is
A ¼ 0. But if we consider shortening of the beam due to bending, that is,
if we adopt large displacements, then the shortening is expressed by the
nonlinear term of the strain-displacement relation. Thus, according to the
nonlinear theory of elasticity, we have [1, 2]

1 ∂u 2
ex ¼ (8.2.14)
2 ∂x
and the total shortening is

Z L
1
½u0 ðx, t Þ dx
2
DL ¼ (8.2.15)
2 0
Consequently
Z L
1
½u0 ðx, t Þ dx
2
A ¼ PDL ¼ P (8.2.16)
2 0
and
Z L
dA ¼ P u0 ðx, t Þd u0 ðx, t Þdx (8.2.17)
0

dA ¼ k∗ udu (8.2.18)
@CivilMethod
where
Z L
k∗ ¼ P ½ 0 ðx Þ dx
2
(8.2.19)
0
Substituting Eqs. (8.2.5), (8.2.9), (8.2.12), (8.2.18) into Hamilton’s prin-

ciple, Eq. (1.7.13), results in
Z t2 Z t2

_ u_ k∗ udu dt
k ∗ udu m ∗ ud ½c∗ u_ + p∗ ðt Þdudt ¼ 0 (8.2.20)
t1 t1
or after integration by part to eliminate the derivative from du_

Z t2

m ∗ u€ + c∗ u_ + k ∗ k∗ u p∗ ðt Þ dudt ¼ 0 (8.2.21)
t1
from which we obtain the equation of motion of the generalized SDOF

system

m ∗ u€ + c∗ u_ + k ∗ k∗ u ¼ p∗ ðt Þ (8.2.22)
Looking at the equation of motion (8.2.22), we draw a useful conclusion
regarding the stability of the structure. It is evident that the stiffness of the
system vanishes when the axial load P takes the critical value
Z h
½ 0 ðx Þ dx
2
Pcr ¼ k ∗ = (8.2.23)
0
This value of the axial force is the buckling load of the structure.
A consequence of this is the vanishing of the natural frequency,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

w¼ k ∗ k∗ =m ∗ ¼ 0. Therefore, a method of determining the buckling
load is to find the value of the compressive axial load, which annuls the nat-
ural frequency. This method is known as the dynamic criterion for buckling.
The method of approximating the continuous systems by a generalized
SDOF system can be successful with regard to the displacement. However,
we should be careful when we need to determine the stress resultants from
the obtained deflection curve using the known relations
M ðx, t Þ ¼ EI u00 ðx, t Þ (8.2.24)
000
Q ðx, t Þ ¼ EI u ðx, t Þ (8.2.25)
The stress resultants resulting from the above relations may deviate con-
siderably from the actual ones. This is illustrated by the following example.
We consider the cantilever beam of Fig. 8.2.2. The function
px
ðx Þ ¼ 1 cos (8.2.26)
2L
@CivilMethod
is a geometrically admissible function, that is,

0 0
ð0Þ ¼ 0, ð0Þ ¼ 0, ðLÞ ¼ 1, ðLÞ 6¼ 0 (8.2.27)
FIG. 8.2.2 Cantilever beam as generalized SDOF system
Therefore, it can be used as a shape function for the cantilever beam and
the resulting displacement is
px
uðx, t Þ ¼ 1 cos u ðt Þ (8.2.28)
2L
Eq. (8.2.25) gives
p 2 px
Q ðx, t Þ ¼ EI sin u ðt Þ (8.2.29)
2L 2L
which results in
Q ð0, t Þ ¼ 0 (8.2.30)
This result is absurd. Nevertheless, this problem can be circumvented
if the stress resultant is evaluated using the procedure described in
Example 8.2.1.
Example 8.2.1 The industrial chimney of length L ¼ 75m shown in Fig. E8.1
consists of the outer reinforced concrete shell, which supports the linings. The
thickness of the thermal insulation layer is ti ¼ 0:10m and that of the refractory
layer tr ¼ 0:10m. The chimney is fixed on the ground.
1. Determine the natural frequency of the structure using two different shape
functions: (i) the elastic curve of a cantilever with constant cross-section
under a uniformly distributed load, and (ii) the first vibration mode of a
cantilever with constant cross-section.
2. Study the dynamic response of the chimney subjected to the impulsive wind
pressure shown in Fig. E8.1b. The analysis will be done using the shape
function that produces more accurate results.
3. Compute the bending moment and shear force at the base of the chimney and
give their expressions as a function of time.
4. Compute the dynamic magnification factor D ¼ max jRðt Þj for the
displacement.
Data:
Specific weight of reinforced concrete: g b ¼ 24kN=m3
@CivilMethod
Specific weight of thermal insulation: g i ¼ 0:7kN=m3

Specific weight of refractory bricks: g r ¼ 20kN=m3
Peak value of wind pressure: pw ¼ 1kN=m2
Time duration of wind pressure: t1 ¼ 2s
(a) (b)
FIG. E8.1 Industrial chimney and loading.
Modulus of elasticity of reinforced concrete: E ¼ 2:1 107 kN=m2

Solution
Computation of m ∗ ,k ∗ ,p∗ ðt Þ.
Mean radius of reinforced concrete shell: r ðx Þ ¼ 3:05 0:0243x
Mean radius of thermal insulation: ri ðx Þ ¼ 2:85 0:0233x
Mean radius of the refractory lining: rr ðx Þ ¼ 2:75 0:0233x
Thickness of reinforced concrete: t ðx Þ ¼ 0:30 0:002x
The mass and the moment of inertia
m ðx Þ ¼ ½2pr ðx Þt ðx Þg b + 2pri ðx Þti g i + 2prr ðx Þtr g r =g (1)
I ðx Þ pr ðx Þ3 t ðx Þ (2)
1. The natural frequency of the chimney

(i) The shape function is the elastic curve of a cantilever with constant
cross-section under a uniformly distributed load p ¼ 8EI =L4 , that is,
@CivilMethod
1 2
ðx Þ ¼ 6x 4x3 + x4 , x ¼ x=L (3)
3
Computation of the integrals using MATLAB gives
Z L
EI ðx Þ½ 00 ðx Þ dx ¼ 2977:6218
2
k ¼
∗ (4)
0
Z L
m∗ ¼ m ðx Þ½ ðx Þ2 dx ¼ 120:9955 (5)
0
Hence
pffiffiffiffiffiffiffiffiffiffiffiffiffi
w¼ k ∗ =m ∗ ¼ 4:9608 (6)
(ii) The shape function is the first vibration mode of a cantilever with a
uniform cross-section (see Section 8.3.3.2)
1
ðx Þ ¼ ½ cosh lx cos lx 0:7341ð sinh lx sinh lx Þ, l ¼ 1:8751=L (7)
3
Z L
EI ðx Þ½ 00 ðx Þ dx ¼ 2715:2000
2
k∗ ¼ (8)
0
Z L
m∗ ¼ m ðx Þ½ ðx Þ2 dx ¼ 116:8332 (9)
0
pffiffiffiffiffiffiffiffiffiffiffiffiffi
w¼ k ∗ =m ∗ ¼ 4:8208 (10
Obviously, case (ii) is more accurate because it yields a smaller
natural frequency (see Chapter 12).
2. The dynamic response of the chimney
The outer diameter of the chimney is
d ðx Þ ¼ 6:40 0:0507x (11)
hence the wind load per unit length is given by
pðx, t Þ ¼ pw d ðx Þf ðt Þ (12)
where
8
> t
>
> 2 if 0 t t1 =2
>
< t1
f ðt Þ ¼ t (13)
>
> 2 1 if t1 =2 t t1
>
> t
: 1
0 if t1 t
hence the peak load of the generalized SDOF system is
Z L
p∗w ¼ pw d ðx Þ ðx Þdx ¼ 106:81 (14)
0
@CivilMethod
The dynamic response will be obtained from the solution of the equation
of motion
m ∗ u€ + k ∗ u ¼ p∗w f ðt Þ (15)
The solution of Eq. (15) is obtained analytically in three phases.
Phase Ι. 0 t 1. The solution is given by Eq. (3.4.15)

2p∗w sin wt
uI ¼ t ¼ 0:039337t 0:008160 sin 4:8208t (16a)
k ∗ t1 w
Phase ΙΙ. 1 t 2, et ¼ t 1. The solution is given by Eqs. (3.3.14),
(3.5.10)

u_ I ð1Þ p∗w sin wet e
uII ðt Þ ¼ sin wet + uI ð1Þcos wet + 1 cos wet + t (16b)
w k∗ w
where
uI ð1Þ ¼ 0:04745m, u_ I ð1Þ ¼ 0:03508m=s
Phase IΙΙ. 2 t, ^t ¼ t 2. The solution is given by Eq. (2.2.13)
u_ II ð2Þ
uIII ðt Þ ¼ sin w^t + uII ð2Þcos w^t (16c)
w
where
uII ð2Þ ¼ 0:01447m, u_ II ð2Þ ¼ 0:00759m=s
Fig. E8.2 shows the graphical representation of the response ratio Rðt Þ,
from which we conclude that max jRðt Þj occurs in phase ΙI. The numerical
solution gives D ¼ 1:2673 occurring at t ¼ 1:12s.
FIG. E8.2 Graphical representation of the response ratio in Example 8.2.1
@CivilMethod
The stress resultants at the base of the chimney are obtained by considering
the equilibrium of all external forces, that is, Q ð0, t Þ, M ð0, t Þ, the wind pres-
sure pðx, t Þ, and the inertia force fI ðx, t Þ. Thus, referring to Fig. E8.3, we have
FIG. E8.3 External forces in Example 8.2.1
Z L Z L
Q ð0, t Þ + fI ðx, t Þdx pðx, t Þdx ¼ 0
0 0
Z L Z L
M ð0, t Þ xf I ðx, t Þdx + xpðx, t Þdx ¼ 0
0 0
By setting in the previous equilibrium equations

fI ðx, t Þ ¼ m ðx Þu€
ðx, t Þ ¼ m ðx Þ ðx Þu€ðt Þ and pðx, t Þ ¼ pw d ðx Þf ðt Þ
we obtain after the evaluation of the integrals
Z L Z L
Q ð0, t Þ ¼ u€ðt Þ m ðx Þ ðx Þdx + pw f ðt Þ d ðx Þdx
0 0
Z L Z L
M ð0, t Þ ¼ u€ðt Þ m ðx Þ ðx Þxdx pw f ðt Þ d ðx Þxdx
0 0
which yield
Q ð0, t Þ ¼ 211:6454u€ðt Þ + 337:4063f ðt Þ (17)
M ð0, t Þ ¼ 10421:7751u€ðt Þ 10870:3125f ðt Þ (18)
8.3 Continuous systems

8.3.1 Introduction
The modeling of a structure as a continuous system can accurately express its
real response, provided that the employed constitutive relations represent
the actual physical law that relates the stresses and strains. As we saw in
@CivilMethod
Section 1.1, this modeling leads to partial differential equations of the

hyperbolic type, namely, for the vibrations of the beam [3]
∂4 u ∂2 u
EI +
m ¼ pðx, t Þ (8.3.1)
∂x 4 ∂t 2
where u ¼ u ðx, t Þ represents the transverse displacement, EI the bending stiff-
ness of the beam, m the mass per unit length, and pðx, t Þ the transverse load
density.
Similarly, for the vibrations of the thin plate [1]
∂4 u ∂4 u ∂4 u r ∂2 u pðx, y, t Þ
+ 2 + + ¼ (8.3.2)
∂x 4 ∂x 2 ∂y 2 ∂y 4 D ∂t 2 D
where u ¼ u ðx, y, t Þ represents the transverse displacement, D the bending
stiffness of the plate, r the surface mass density, and pðx, y, t Þ the transverse
load density. The above equations are more complicated for the shell and
become even more complicated if the thickness of the structural element is var-
iable or the response is nonlinear [1, 4]. The effort to obtain analytical solutions
of the equations governing the response of continuous systems has inspired
mathematicians and applied physicists to develop the theory of partial differen-
tial equations [5, 6]. This, however, did not satisfy people who wanted to use the
derived solutions in engineering praxis because these solutions treat a very lim-
ited number of problems. The development of approximate and especially of
modern numerical solutions (e.g., FEM, BEM, MM) has provided us with
efficient tools to solve problems of mathematical physics and engineering
described by complicated ordinary and partial differential equations. The ques-
tion that arises nowadays and demands investigation is how reliably these equa-
tions describe the actual response of physical systems. The system identification
based on experimental data can answer this question. Though the available pro-
fessional computer codes efficiently analyze engineering problems, they cannot
give adequate insight into the dynamic response of structures. Therefore, the use
of analytical methods is necessary to understand the response of structures and
to develop simple and desired-accuracy solutions, which are very useful to
check new numerical methods. Because the scope of this book is the analysis
of structures consisting of straight-line structural elements, the discussion will
be limited to the solution of the equation of the vibrating beam. Results of this
analysis are used in SDOF generalized systems.
8.3.2 Solution of the beam equation of motion

The transverse flexural vibrations of the beam with constant stiffness are gov-
erned by Eq. (8.3.1) subjected to the specified boundary and initial conditions.
The solution u ¼ u ðx, t Þ can be obtained in the form
u ¼ uh + up (8.3.3)
@CivilMethod
where uh is the solution of the homogeneous equation

∂4 uh ∂2 uh
EI 2 ¼0
+m (8.3.4)
∂x 4 ∂t
and up a particular solution of the nonhomogeneous equation
∂4 up ∂2 up
EI +
m ¼ pðx, t Þ (8.3.5)
∂x 4 ∂t 2
8.3.3 Free vibrations of beams

For pðx, t Þ ¼ 0, Eq. (8.3.1) yields the homogeneous equation, which represents
the equation of free vibrations of the beam. Assuming constant mass,
m ¼ constant, and dropping the subscript for convenience in Eq. (8.3.4),
we write this equation as
∂4 u ∂2 u
EI + m ¼0 (8.3.6)
∂x 4 ∂t 2
Eq. (8.3.6) is solved using the method of separation of variables for partial
differential equations [6], that is, the solution is sought in the form
u ðx, t Þ ¼ fðx ÞY ðt Þ (8.3.7)
This equation indicates that the free vibration motion is the product of a
specific deflection shape fðx Þ and an amplitude Y ðt Þ depending on time.
Introducing the above expression for the deflection into Eq. (8.3.6) gives
ðx ÞY€ ðt Þ ¼ 0
EI fðivÞ ðx ÞY ðt Þ + mf (8.3.8)
which is written as
EiI fðivÞ ðx Þ Y€ ðt Þ
¼ (8.3.9)
m fðx Þ Y ðt Þ
For the obvious problem associated with the possibility of the vanishing of
the product fðx ÞY ðt Þ, we refer to [7, 8]. Here, we just state that the roots of the
denominator in Eq. (8.3.9) coincide with those of the nominator, an assumption
that permits this division.
Because the left side of Eq. (8.3.9) is independent of t and the right side
independent of x, this equation is valid only if both sides are equal to a
constant l. Hence
EI fðivÞ ðx Þ Y€ ðt Þ
¼ ¼l
m fðx Þ Y ðt Þ
or
Y€ ðt Þ + lY ðt Þ ¼ 0 (8.3.10a)
@CivilMethod

m
fðivÞ ðx Þ l
f ðx Þ ¼ 0 (8.3.10b)
EI
If l 0, the solution of Eq. (8.3.10a) does not represent an oscillatory
motion. Therefore, l must be a positive constant, l ¼ w2 . Thus, the solution
of Eq. (8.3.10a) is (see Section 2.2)
Y ðt Þ ¼ A cos wt + B sin wt (8.3.11)
where A,B are arbitrary constants depending on the initial conditions Y ð0Þ and
Y_ ð0Þ. Thus we have
Y_ ð0Þ
Y ðt Þ ¼ Y ð0Þcos wt + sin wt (8.3.12)
w
Apparently, w is the natural frequency of the vibration, unknown in the first
instance.
Similarly, Eq. (8.3.10b) is written as
fðivÞ ðx Þ b 4 fðx Þ ¼ 0 (8.3.13)

where
m
b 4 ¼ w2 (8.3.14)
EI
The solution of Eq. (8.3.13) is sought in the form
fðx Þ ¼ C 0 ekx (8.3.15)

which is introduced into Eq. (8.3.13) to give the characteristic equation
k 4 b4 ¼ 0 (8.3.16)
whose roots are
k1, 2 ¼ ib, k3, 4 ¼ b (8.3.17)
Using each of these roots in Eq. (8.3.15) yields four terms, which are added
to give the general solution
fðx Þ ¼ C10 eibx + C20 eibx + C30 ebx + C40 ebx (8.3.18)
where C 0 i ði ¼ 1, 2, 3, 4Þ are arbitrary complex constants.
Using Euler’s formula (2.2.8) and the expressions of the hyperbolic sine and
cosine, Eq. (8.3.18) becomes
fðx Þ ¼ C1 cos bx + C2 sin bx + C3 cosh bx + C4 sinh bx (8.3.19)
in which Ci ði ¼ 1, 2, 3, 4Þ are new arbitrary constants related to
C 0 i ði ¼ 1, 2, 3, 4Þ, and can be determined from the boundary (support) condi-
tions of the one-span beam.
@CivilMethod
In the following, the above-described procedure for free-vibration analysis

is illustrated by analyzing the simply supported and the cantilever beams.
For further reading, the reader is advised to look in Refs. [9–13].
8.3.3.1 The simply supported beam

For a simply supported beam of length L (Fig. 8.3.1) the support conditions are
realized as:
u ð0, t Þ ¼ 0, M ð0, t Þ ¼ EI u 00 ð0, t Þ ¼ 0 (8.3.20a)
u ðL, t Þ ¼ 0, M ðL, t Þ ¼ EI u 00 ðL, t Þ ¼ 0 (8.3.20b)
which by virtue of Eq. (8.3.7) become
fð0ÞY ðt Þ ¼ 0, f00 ð0ÞY ðt Þ ¼ 0 (8.3.21a)
fðLÞY ðt Þ ¼ 0, f00 ðLÞY ðt Þ ¼ 0 (8.3.21b)
FIG. 8.3.1 Vibration modes and natural frequencies of a uniform simply supported beam.
Inasmuch as Eqs. (8.3.21a), (8.3.21b) are valid for all values of t, they are
satisfied only if
fð0Þ ¼ 0, f00 ð0Þ ¼ 0 (8.3.22a)
fðLÞ ¼ 0, f00 ðLÞ ¼ 0 (8.3.22b)
Introducing Eq. (8.3.19) into Eq. (8.3.22a) yields
C1 + C3 ¼ 0 (8.3.23a)
@CivilMethod
C1 + C3 ¼ 0 (8.3.23b)
C1 ¼ C3 ¼ 0 (8.3.24)
Further, introducing Eq. (8.3.19) into Eq. (8.3.22b), we obtain
C2 sin bL + C4 sinh bL ¼ 0 (8.3.25a)
b2 ½C2 sin bL + C4 sinh bL ¼ 0 (8.3.25b)
Eqs. (8.3.25a), (8.3.25b) provide the system of two homogeneous algebraic

equations for the evaluation of C2 ,C4 , that is,

sin bL sinh bL C2 0
¼ (8.3.26)
sin bL sinh bL C4 0
The system has a nontrivial solution if its determinant vanishes, that is,
sin bL sinh bL ¼ 0 (8.3.27)
Nevertheless, because sinh bL 6¼ 0, it must be
sin bL ¼ 0 (8.3.28)
which is satisfied if
np
bn ¼ , n ¼ 1, 2, … (8.3.29)
L
Introducing this value of b n into Eq. (8.3.14) gives the corresponding natural
frequencies
rffiffiffiffiffiffiffiffiffiffi
EI
wn ¼ n p
2 2
(8.3.30)
4
mL
Substituting Eq. (8.3.28) into Eq. (8.3.26), we obtain
" #( ) ( )
0 sinh bL C2 0
¼ (8.3.31)
0 sinh bL C4 0
which yields C4 ¼ 0 and C2 ¼ arbitrary, therefore

fðx Þ ¼ C2 sin bx (8.3.32)
On the basis of Eq. (8.3.29), we obtain the mode shapes
np
fn ðx Þ ¼ C2 sin x , n ¼ 1, 2, … (8.3.33)
L
The first three of these mode shapes are shown in Fig. 8.3.1 along with their
natural frequencies
@CivilMethod
8.3.3.2 The cantilever beam

For a cantilever beam of length L the support conditions are realized as:
u ð0, t Þ ¼ 0, u 0 ð0, t Þ ¼ 0 (8.3.34a)
00 000
M ðL, t Þ ¼ EI u ðL, t Þ ¼ 0, Q ðL, t Þ ¼ EI u ðL, t Þ ¼ 0 (8.3.34b)
which by virtue of Eq. (8.3.7) become
fð0ÞY ðt Þ ¼ 0, f0 ð0ÞY ðt Þ ¼ 0 (8.3.35a)
f00 ðLÞY ðt Þ ¼ 0, f000 ðLÞY ðt Þ ¼ 0 (8.3.35b)
Because Eqs. (8.3.35a), (8.3.35b) are valid for all values of t, they are
satisfied only if
fð0Þ ¼ 0, f0 ð0Þ ¼ 0 (8.3.36a)
f00 ðLÞ ¼ 0, f000 ðLÞ ¼ 0 (8.3.36b)
Introducing Eq. (8.3.19) into Eqs. (8.3.36a), (8.3.36b) yields the system of
equations for the evaluation of the coefficients Ci
2 38 9 8 9
1 0 1 0 > C1 >
> > > >0> >
6 0 1 0 1 7 < C2 = < 0 =
6 7 ¼ (8.3.37)
4 cos bL sin bL coshbL sinh bL 5> C3 > > 0 >
: >
> ; > : > ;
sin bL cos bL sinh bL cosh bL C4 0
Eq. (8.3.37) has a nontrivial solution if the determinant of the coefficient
matrix vanishes, that is,

1 0 1 0

0 1 0 1
¼0 (8.3.38)
cos bL sin bL cosh bL sinh bL

sin bL cos bL sinh bL cosh bL
or after evaluation of the determinant
cos bL cosh bL + 1 ¼ 0 (8.3.39)
The roots bn , n ¼ 1, 2, … of Eq. (8.3.39) are used in Eq. (8.3.14) to obtain
the natural frequencies of the vibrating cantilever beam.
Substituting Eq. (8.3.39) into Eq. (8.3.37) and solving the resulting
homogeneous system of equations, we obtain
cos b n L + cosh bn L
C3 ¼ C1 , C4 ¼ C2 , C2 ¼ C1 (8.3.40)
sin b n L + sinh bn L
which are introduced into Eq. (8.3.19) to give the mode shapes

cos b n L + cosh bn L
fn ðx Þ ¼ C1 cos bn x cosh b n x ð sin bn x sinh bn x Þ
sin b n L + sinh bn L
(8.3.41)
@CivilMethod
Table 8.3.1 gives the first five roots of bn L of Eq. (8.3.39). They have been
obtained numerically using the function fsolve of MATLAB. Note that for n > 3
they can be obtained from the relation
p
bn L ð2n 1Þ (8.3.42)
2
The first three of these mode shapes are shown in Fig. 8.3.2 along with their
natural frequencies
TABLE 8.3.1 First five roots of Eq. (8.3.39).
n bn L
1 1.8751040688
2 4.6940911329
3 7.8547574382
4 10.995540734
5 14.137168391
FIG. 8.3.2 Vibration modes and natural frequencies of a uniform cantilever beam.
8.3.4 Orthogonality of the free-vibration modes

The infinite set of free-vibration mode shapes F : ffn ðx Þg, n ¼ 1, 2, … has a
nice property. They are orthogonal in the interval ½0, L, that is, they satisfy
the orthogonality condition (see Section 3.6.2)
@CivilMethod
Z L
0 if r 6¼ n
fr ðt Þfn ðt Þdt ¼ (8.3.43)
0 cn if r ¼ n
for any two functions fn ,fr F, m, nN . In the language of partial differential
equations, the free-vibration modes are called the eigenfunctions of the eigenvalue
problem described by the differential equation (8.3.13) and its boundary condi-
tions. The orthogonality condition is readily proved by proceeding as follows.
The mode shapes fn ,fr satisfy Eq. (8.3.13), that is,
fðnivÞ ðx Þ b4n fn ðx Þ ¼ 0 (8.3.44a)
fðrivÞ ðx Þ b 4r fr ðx Þ ¼ 0 (8.3.44b)
Multiplication of Eq. (8.3.44a) by fr ðx Þ and integrating over the interval
½0, L gives
Z L Z L
b4n fn ðx Þfr ðx Þdx ¼ fðnivÞ ðx Þfr ðx Þdx (8.3.45)
0 0
Further, integrating the right side of the above equation twice by parts gives
Z Z L
L
00 00

00 0
L
000 L
fn ðx Þfr ðx Þdx fn ðx Þfr ðx Þ 0 + fn ðx Þfr ðx Þ 0 ¼ bn
4
fn ðx Þfr ðx Þdx
0 0
(8.3.46)
Obviously, the terms in square brackets in the above equation vanish if
either end of the beam is simply supported, fixed, or free. Thus, we have
Z L Z L
4
bn fn ðx Þfr ðx Þdx ¼ f00n ðx Þf00r ðx Þdx (8.3.47a)
0 0
Similarly, multiplying Eq. (8.3.44b) by fn ðx Þ, integrating over the interval

½0, L, and performing the integrations by parts yields the symmetric relation
Z L Z L
b4r fn ðx Þfr ðx Þdx ¼ f00n ðx Þf00r ðx Þdx (8.3.47b)
0 0
Subtracting Eq. (8.3.47b) from Eq. (8.3.47a) gives

Z
L
0 ¼ b 4n b4r fn ðx Þfr ðx Þdx (8.3.48)
0
which for b 4n 6¼ b4r results in the orthogonality condition for the mode shapes
Z L
fn ðx Þfr ðx Þdx ¼ 0 (8.3.49)
0
It can also be shown that the set F : ffn ðx Þg is complete, that is there is no
other function outside the set F, which satisfies the condition (8.3.49).
@CivilMethod
8.3.5 Forced vibrations of beams

The method of separation of variables can be employed to study the forced
vibrations of one-span beams by solving Eq. (8.3.1). The beam may be
subjected to initial conditions
u ðx, 0Þ ¼ f ðx Þ, u_ ðx, 0Þ ¼ g ðx Þ
The total flexural displacement of the beam is obtained as the superposition
of all modal contributions, namely
X
∞
u ðx, t Þ ¼ fn ðx ÞYn ðt Þ (8.3.50)
n¼1
Substituting the previous equation into Eq. (8.3.1), multiplying by fr ðx Þ,

and integrating over the interval ½0, L, we obtain
X∞ Z L X ∞ Z L
Yn ð t Þ ðivÞ
EI fn ðx Þfr ðx Þdx + €
Y n ðt Þ n ðx Þfr ðx Þdx
mf
n¼1 0 n¼1 0
Z L
¼ fr ðx Þpðx, t Þ (8.3.51)
0
Using now the orthogonality condition (8.3.49) and taking into account
Eqs. (8.3.45), (8.3.14), we obtain
Mn Y€ n ðt Þ + Kn Yn ðt Þ ¼ Pn ðt Þ (8.3.52)
where
Z L Z L

Mn ¼ m f2n ðx Þdx, Kn ¼ w2n Mn , Pn ¼ fn ðx Þpðx, t Þdx (8.3.53)
0 0
denote the modal mass, the modal stiffness, and the modal force. These quan-
tities are also referred to as the generalized mass, the generalized stiffness, and
the generalized force, respectively.
The solution of Eq. (8.3.52) is given by Eq. (3.3.14), that is,
Z t
Y_ n ð0Þ 1
Y n ðt Þ ¼ sin wn t + Yn ð0Þcos wn t + Pn ðτÞsin ½wn ðt τÞdτ
wn M n wn 0
(8.3.54)
The initial conditions Yn ð0Þ, Y_ n ð0Þ for the time function result from
Eq. (8.3.50). This yields
X
∞
u ðx, 0Þ ¼ fn ðx ÞYn ð0Þ ¼ f ðx Þ (8.3.55)
n¼1
Multiplying the previous equations by fn ðx Þ, integrating over the interval

½0, L, and using the orthogonality condition (8.3.49), we obtain
@CivilMethod
Z L
f ðx Þfn ðx Þdx
Yn ð 0Þ ¼ 0
Z L
(8.3.56a)
f2n ðx Þdx
0
Similarly, we obtain
Z L
g ðx Þfn ðx Þdx
Y_ n ð0Þ ¼ 0
Z L
(8.3.56b)
f2n ðx Þdx
0
Example 8.3.1 A simply supported beam of length L is subjected to a suddenly

applied uniform p0 under zero initial conditions, f ðx Þ ¼ g ðx Þ ¼ 0. Determine the
expressions of the displacement, bending moment, and shear force.
Solution
1. Determine the natural frequencies and mode shapes of the vibration.
They are obtained from Eqs. (8.3.30), (8.3.33):
rffiffiffiffiffiffiffiffiffiffi np
EI
wn ¼ n p
2 2
, f ð x Þ ¼ sin x , n ¼ 1, 2, … (1)
4
mL n
L
2. Determine the modal mass, modal force, and initial conditions. They are
obtained from Eqs. (8.3.53), (8.3.56a), (8.3.56b):
Z L np mL
Mn ¼ m sin 2 x dx ¼ (2)
0 L 2
Z L np L
Pn ¼ p 0 sin x dx ¼ p0 ½1 ð1Þn (3)
0 L np
Yn ð0Þ ¼ 0, Y_ n ð0Þ ¼ 0 (4)
3. Determine the time-varying amplitudes Yn ðt Þ. Substituting Eqs. (2)–(4)

into Eq. (8.3.54) and using Eq. (3.4.3) give
4p0 L4
Yn ð t Þ ¼ ð1 cos wn t Þ, n ¼ 1, 3, 5, … (5)
n 4 p4 EI
Hence
X∞
4p0 L4 X
∞
1 np
u ðx, t Þ ¼ fn ðx ÞYn ðt Þ ¼ 4 4
sin x ð1 cos wn t Þ,
n¼1
p EI n¼1 n L (6)
n ¼ 1, 3, 5, …
@CivilMethod
The bending moment and the shear force are evaluated from the
expressions
M ðx, t Þ ¼ EI u 00 ðx, t Þ ¼ EI f00 ðx ÞY ðt Þ (7)
000 000
Q ðx, t Þ ¼ EI u ðx, t Þ ¼ EI f ðx ÞY ðt Þ (8)
8.4 Problems
Problem 8.1 The television tower of Fig. P8.1 is subjected to seismic ground
motion ug ðt Þ. Derive the equation of motion if the structure is approximated by
a SDOF system. The reaction moment
of the elastic ground is represented by the
nonlinear expression MR ¼ CR f + 14 f2 , where CR ¼ KI f ; If is the moment of
inertia of the planform of the fundament and K ¼ E=10h the foundation mod-
ulus with E being the modulus of elasticity of the material of the structure. The
cross-section of the flexible column, the planform of the fundament, and the
body B are circular with diameters D, Df ¼ 8D, and DB ¼ 5D, respectively.
The density of the material is r. The fundament and the body B are
assumed rigid.
FIG. P8.1 Television tower in problem P8.1
Problem P8.2 The continuous beam of Fig. P8.2 rests on Winkler’s elastic
foundation with variable modulus k ðx Þ. The beam is axially subjected to the
load P. Determine the value of the load that produces buckling.
FIG. P8.2 Continous beam on nonhomogeneous Winkler’s elastic foundation in problem P8.2
@CivilMethod
Problem P8.3 A vehicle of weight W ¼ 30kN is traveling on a simply sup-

ported bridge with velocity v ¼ 80km=h as shown in Fig. P8.3a. The vehicle
is simulated by a single wheel of negligible mass. Determine the dynamic mag-
nification factor for the bridge when the beam is approximated by a generalized
SDOF system. The material of the bridge is prestressed concrete with modulus
of elasticity E ¼ 2:1 107 kN=m2 and specific weight g ¼ 25kN=m3 . The
cross-section of the bridge is shown in Fig. P8.3b.
(a)
(b)
FIG. P8.3 Simple supported bridge in Problem P8.3
Problem P8.4 Use the method of separation of variables to solve the equation
of free flexural vibrations of the:
(a) Fixed-fixed beam.
(b) Fixed-simply supported beam
In both cases, determine the natural mode shapes and the frequency equation.
Compute the first three natural frequencies.
Problem P8.5 Analyze the free flexural vibrations of the two-span continuous
beam of Fig. P8.5 by solving the equation of motion of the beam. Hint: Consider
the continuity condition at support 2, that is, u 0 I ðL, t Þ ¼ u 0 II ð0, t Þ.
FIG. P8.5 Two-span continuous beam in problem P8.5
@CivilMethod

[1] J.T. Katsikadelis, The Boundary Element Method for Plate Analysis, Academic Press,
[2] D.O. Brush, B.O. Almroth, Buckling of Bars, Plates and Shells, McGraw-Hill, New York,
1975.
[3] S. Timoshenko, D.E. Young, W. Weaver Jr., Vibration Problems in Engineering, fourth ed.,
John Wiley & Sons, New York, 1974.
[4] J.T. Katsikadelis, G.C. Tsiatas, Nonlinear dynamic analysis of beams with variable stiffness,
J. Sound Vib. 270 (2004) 847–863.
[5] E. Zauderer, Partial Differential Equations of Applied Mathematics, second ed., John Wiley &
Sons, Singapore, 1989.
[7] E.C. Titchmarsh, Eigenfunction Expansions, Part I (1958) & Part II (1962), Oxford University
Press, Oxford, UK, 1958.
[8] G.L. Cain, G.H. Meyer, Separation of Variables for Partial Differential Equations: An Eigen-
function Approach, in: Studies in Advanced Mathematics, CRC Press, London, 2006.
[9] R.W. Clough, J. Penzien, Dynamics of Structures, McGraw-Hill, New York, 1993.
[12] S.S. Rao, Vibration of Continuous Systems, John Wiley & Sons, New Jersey, 2007.
[13] L. Meirovitch, Analytical Methods in Vibration, Macmillan, London, 1967.
@CivilMethod
Chapter 9
Analysis in the frequency

domain
Chapter outline
9.1 Introduction 299 9.6 Discrete Fourier transform 310
9.2 Complex form of the Fourier 9.7 Application of the discrete
series 301 Fourier transform
9.3 Complex dynamic response to dynamic analysis 313
to periodic load 303 9.8 Fast Fourier transform 314
9.4 Fourier integral 9.8.1 The Sande-Tukey
representation of a algorithm 315
nonperiodic load 304 9.9 Problems 320
9.5 Response to a nonperiodic References and further reading 322
load 307
9.1 Introduction
The methods we discussed in the previous chapters for solving the equation of
motion of a SDOF system were accomplished using time as an independent var-
iable or, said differently, the solution was obtained in the time domain. These
methods are either analytical, which in the general case lead to the evaluation of
the Duhamel integral, or numerical, implemented by the step-by-step integra-
tion methods. Occasionally, simpler or more convenient analytical or numerical
solutions may be possible for certain types of dynamics problems, using integral
transforms such as the Laplace transform or the Fourier transform. The integral
converts the linear differential equation into a linear algebraic equation, from
which the integral transform of the unknown function is obtained. Then the
inverse transform results in the solution in the time domain.
We have already discussed the Laplace transform in Section 3.3, where it
was employed to solve the equation of motion of a SDOF system under an arbi-
trary external excitation. The Laplace transform uses a parameter that does not
have a direct physical meaning. Instead, the parameter in the Fourier transform
has the physical meaning of frequency [1–3]. The method of analyzing dynam-
ical systems using the Fourier transform is known as the analysis in the fre-
quency domain. It plays an important role in studying the dynamic response
of linear systems, that is, systems described by linear differential equations.

@CivilMethod
The analysis in the frequency domain is employed in many fields of engineering

science and physics. Therefore, the available related literature is extensive and
reference to it is made with regard to the specific scientific discipline under
treatment.
As we have seen in Section 3.4, the drawback of time-domain analysis is the
evaluation of the Duhamel integral, which can be achieved analytically only for
very simple loading functions. Instead, the Fourier transform converts the con-
volution integral, such as the integral of Duhamel, into the product of the trans-
forms of the functions involved in the convolution. Then the inverse transform
of the product gives the response of the system in the time domain.
The efficiency of the method depends on the possibility of finding the Fou-
rier transform of the functions of the convolution as well as the inverse of their
product. There are tables [4] that give the Fourier transform and its inverse for a
variety of functions. Unfortunately, these functions are rather simple while the
use of tables requires tedious analytical work with the risk of a possible error.
Certainly, the symbolic languages such as MAPLE, MATHEMATICA, and
WOLFRAM ALPHA offer new possibilities for the application of the method.
Nevertheless, the Fourier transform method cannot always meet the practi-
cal needs, especially when the load function is specified by a set of values, as for
example the accelerogram of an earthquake. For this reason, the Fourier trans-
form for a long time has not been equally applied to the dynamic analysis of
structures as it has to other areas of engineering science. The use of numerical
methods of computing the Fourier transform has brought to light this method in
its discrete form, referred to as the discrete Fourier transform. Of course, the
discrete Fourier transform is not simpler or more functioning than the numerical
methods of evaluating the Duhamel integral. Therefore, its use would have been
limited if a specific algorithm for the numerical computation of the Fourier
transform had not been developed, the so-called Fast Fourier transform, which
limits the numerical computations by several orders and makes dynamic anal-
ysis in the frequency domain quite effective.
Besides its effectiveness, the analysis in the frequency domain has other
important advantages. It shows clearly the frequencies of the loading, providing
thus the possibility of detecting those frequencies that unfavorably excite the
structure. It is also suitable for the study of systems involving large or infinite
regions, such as the interaction of dams in large reservoirs with waves, soil
structure interaction in earthquake engineering, and, in general, problems where
the physical characteristics of the system such as stiffness or damping depend
on the frequency of the oscillation. For these reasons, it was considered appro-
priate to include a brief discussion of the analysis in the frequency domain in
this book. Apparently, this would enable the engineers involved in earthquake
analysis to understand the capabilities of the method.
The method could be presented starting from the mathematical development
of the Fourier transform [4], but in order to understand its physical significance,
the complex Fourier series and the response of the SDOF system to periodic
@CivilMethod
Analysis in the frequency domain Chapter 9 301
loading are preceded. Then, the method extends to nonperiodic loads by intro-
ducing the Fourier integral and the Fourier transform. The chapter closes with
the presentation of the discrete Fourier transform and the fast Fourier transform.
9.2 Complex form of the Fourier series

In Section 3.6, the dynamic response of a structure under a general periodic load
was obtained in the time domain by expanding the periodic load in Fourier
series, that is, by representing it through an infinite sum of harmonic loads. This
procedure was discussed in detail there. However, the study of the response of a
system in the frequency domain is facilitated by representing the Fourier series
(3.6.2) in exponential form. We recall that
X
1
pðt Þ ¼ a0 + t + bn sin n w
ðan cos n w t Þ (9.2.1)
n¼1
Z T =2 Z
1 1 T
a0 ¼ pðt Þdt ¼ pðt Þdt (9.2.2a)
T T =2 T 0
Z T =2
2
an ¼ tdt, n ¼ 1, 2, 3…
pðt Þcos n w (9.2.2b)
T T =2
Z T =2
2
bn ¼ tdt,
pðt Þ sin n w n ¼ 1, 2, 3… (9.2.2c)
T T =2
From Euler’s formula (2.2.8), we have

einwt + einwt
t ¼
cos n w (9.2.3a)
2
einwt einwt
t ¼
sin n w (9.2.3b)
2i
Substituting Eq. (9.2.3a) into Eq. (9.2.2b) gives
Z
1 T =2
an ¼ pðt Þ einwt + einwt dt (9.2.4a)
T T =2
Replacing n with n in the previous equation gives

Z
1 T =2
an ¼ pðt Þ einwt + einwt dt ¼ an (9.2.4b)
T T =2
Similarly, substituting Eq. (9.2.3b) into Eq. (9.2.2c) yields

Z
1 T =T ein wt einwt
bn ¼ pðt Þ dt: (9.2.5a)
T T =2 i
@CivilMethod
Hence
Z T =2
1 einwt einwt
bn ¼ pðt Þ dt ¼ bn (9.2.5b)
T T =2 i
The Fourier series (9.2.1) by virtue of Eqs. (9.2.3a), (9.2.3b) becomes
X1
einwt + einwt X 1
ein wt einwt
pðt Þ ¼ a0 + an + bn
n¼1
2 n¼1
2i
(9.2.6)
1X 1
1X 1
¼ ao + ðan ibn Þeinwt + ðan + ibn Þeinwt
2 n¼1 2 n¼1
Further, using Eqs. (9.2.4a), (9.2.5a), the previous equation becomes

1X 1
1 X1
pðt Þ ¼ ao + ðan ibn Þeinwt + ðan ibn Þein wt
2 n¼1 2 n¼1
(9.2.7)
X
1
t
¼ cn e in w
n¼1
in which the series coefficients are given as

1
cn ¼ ðan ibn Þ
2
Z T =2
1
¼ pðt Þ einwt + einwt ein wt einwt dt (9.2.8a)
2T T =2
Z
1 T =2
¼ pðt Þeinwt dt
T T =2
and
1
cn ¼ ðan + ibn Þ
Z2 (9.2.8b)
1 T =2
¼ pðt Þeinwt dt
T T =2
Z
1 T =2
c0 ¼ pðt Þdt ¼ a0 (9.2.8c)
T T =2
Eq. (9.2.7) represents the complex form or exponential form of the Fourier
series.
From Eq. (9.2.7), the reasonable question arises as to how it is possible for a
real function to be expressed as a sum of complex terms. It is easy, however, to
prove that the right side of this equation is a real function as long as we think that
to each term cn einwt the term cn ein wt corresponds, whose sum yields a real
function.
@CivilMethod
9.3 Complex dynamic response to periodic load

In Section 3.6.3, the method of determining the steady-state response of a SDOF
system subjected to a periodic load was presented. In this method, the periodic
load was analyzed into harmonic terms, both sine and cosine, by expanding it in
a Fourier series. After establishing the steady-state response for each term, the
total response was obtained as the sum of all responses. Following a similar pro-
cedure, it is easy to determine the steady-state response to a periodic load when
it is has been expanded in a complex Fourier series. To this end, we must first
determine the steady-state response of the system when subjected to the load
pðt Þ ¼ p0 eiwt . In this case, the equation of the SDOF system is written
m u€ + cu_ + ku ¼ p0 eiwt (9.3.1)
The solution will be obtained as a sum of the homogeneous solution uh and a
particular solution up of the nonhomogeneous equation. The homogeneous
solution is given by Eq. (3.2.22), namely
uh ¼ exwt ðA cos wD t + B sin wD t Þ (9.3.2)
The particular solution is sought in the form
up ¼ Ceiwt (9.3.3)
Substituting the previous expression into Eq. (9.3.1) gives
p0
C¼ (9.3.4)
ðm w + icw
2 + kÞ
which is inserted into Eq. (9.3.3) to yield
p0
up ¼ eiwt (9.3.5)
m w 2 + icw
+k
Hence, the general solution of Eq. (9.3.1) is
p0
u ðt Þ ¼ exwt ðA cos wD t + B sin wD t Þ + eiwt (9.3.6)
2
m w + icw+k
The first term in Eq. (9.3.6) becomes negligible with increasing time and it
represents the transient response of the system. The second term expresses the
steady-state response and can be written as
Þp0 eiwt
u ðt Þ ¼ H ðw (9.3.7)
where it was set
1
Þ ¼
H ðw
m w2 + icw+k
1 (9.3.8)
¼ , b ¼ w
=w
k 1 b 2 + 2ixb
@CivilMethod
Þ is called the complex frequency response function, also

The function H ðw
known as the transfer function. In summary, setting p0 ¼ cn we may write the
steady-state response to a periodic load as
X
1
u ðt Þ ¼ Þeinwt
c n H ðw (9.3.9)
n¼1
where
Z T =2 Z
1 1 T
cn ¼ pðt Þeinwt dt ¼ pðt Þein wt dt (9.3.10)
T T =2 T 0
9.4 Fourier integral representation of a nonperiodic load

When the load is nonperiodic, it cannot be represented as a Fourier series. But it
is possible to express the nonperiodic function in the form of an integral over the
interval ð1, 1Þ, which one might regard as a Fourier series whose period is
infinitely large. In Section 3.6.2, we saw that a function periodic in the interval
½0, 1Þ or ð1, 1Þ is expanded in a Fourier series, Eq. (9.2.1). The coefficients
in this equation are written as
Z
1 T =2
a0 ¼ pðτÞdτ (9.4.1a)
T T =2
Z
2 T =2
an ¼ pðτÞcos n wτdτ, n ¼ 1, 2, 3… (9.4.1b)
T T =2
Z
2 T =2
bn ¼ τdτ, n ¼ 1, 2, 3…
pðτÞ sin n w (9.4.2)
T T =2
The previous equations resulted from Eq. (9.2.2a), Eq. (9.2.2b), Eq. (9.2.2c)
by substituting t with τ in the integrand. Note that it is allowed because τ is a
dummy variable.
Substituting Eqs. (9.4.1a), (9.4.1b), (9.4.2) in Eq. (9.2.1) yields
Z " #
1 Z T =2
1 T =2 2X
pðt Þ ¼ pðτÞdτ + τ cosn w
pðτÞ cos n w tdτ
T T =2 T n¼1 T =2
" # (9.4.3)
1 Z T =2
2X
+ τ sin n w
pðτÞ sin n w tdτ
T n¼1 T =2
or
Z "Z #
1 T =2
2X 1 T=2
pðt Þ ¼ pðτÞdτ + ðτ t Þdτ
pðτÞcos n w (9.4.4)
T T =2 T n¼1 T =2
@CivilMethod
If the function pðt Þ is not periodic, we may set

( Z " #)
1 Z T =2
1 T=2 2X
pðt Þ ¼ lim pðτÞdτ + ðτ t Þdτ
pðτÞcos n w
T !1 T T=2 T n¼1 T =2
(9.4.5)
in which t does not change in passing to the limit. It is assumed that the function
pðt Þ satisfies the Dirichlet conditions (see Section 3.6.2). Therefore, it is abso-
lutely integrable over the interval ½T =2, T =2, that is,
Z
1 T =2 1 Z T =2
k
j a0 j ¼ pðτÞdτ jpðτÞjdτ < (9.4.6)
T T =2 T T =2 T
which vanishes for T ! 1. Hence

( " #)
1 Z T =2
2X
pðt Þ ¼ lim ðτ t Þdτ
pðτÞ cos n w (9.4.7)
T !1 T T =2
n¼1
If we set now
2p
¼w
nw n , D n + 1 w
w¼w n ¼ (9.4.8)
T
we may write Eq. (9.4.7) as
X
1
pðt Þ ¼ lim n ÞD
P ðw w (9.4.9)
T !1
n¼1
n Þ is the value of the function

where P ðw
Z
1 T =2
Þ ¼
P ðw ðτ t Þdτ
pðτÞcos w (9.4.10)
p T =2
¼w
for w is understood as a continuous variable (Fig. 9.4.1)
n , where w
FIG. 9.4.1 Fourier transform of p(t).
@CivilMethod
Þ, we readily deduce that the sum

From the graphical representation of P ðw
X
1
n ÞD
P ðw w (9.4.11)
n¼1
Þ. When T ! 1, then D

approaches the area under the curve y ¼ P ðw w ! 0 and
the sum (9.4.11) becomes a definite integral. Consequently, we may write
Eq. (9.4.9) as
Z 1
pðt Þ ¼ Þd w
P ðw (9.4.12)
0
or by virtue of Eq. (9.4.10)

Z 1 Z 1
1
pðt Þ ¼ ðτ t Þdτ d w
pðτÞcos w (9.4.13)
0 p 1
The integral (9.4.13) is known as the Fourier integral. The Fourier integral con-
verges to the function if it is piecewise continuous in every finite interval and
absolutely integrable over ð1, 1Þ. At points of discontinuity, the Fourier
integral converges to the mean value.
The Fourier integral can be expressed in a complex form using Euler’s for-
mula, Thus, writing
1h i
cos wðτ t Þ ¼ eiwðτt Þ + eiwðτt Þ (9.4.14)
2
and inserting into Eq. (9.4.13) splits it into two integrals, that is,
Z Z 1 Z Z 1
1 1 ðτt Þ 1 1 ðτt Þ
i w
pðt Þ ¼ pðτÞe iw
+
dτ d w pðτÞe
dτ d w
2p 0 1 2p 0 1
Changing the integration variable from w to w in the second integral, we

obtain
Z Z 1
1 1
pðt Þ ¼ pðτÞeiwτ dτ eiwt d w
(9.4.15)
2p 1 1
The forgoing relation allows writing
Z 1
Þ ¼
P ðw pðτÞeiwτ dτ
1
or returning to the variable t from τ

Z 1
P ðwÞ ¼ pðt Þeiwt dt (9.4.16)
1
and inserting it into Eq. (9.4.15), gives

Z
1 1
pðt Þ ¼ Þeiwt d w
P ðw (9.4.17)
2p 1
@CivilMethod
The function P ðwÞ defined by Eq. (9.4.16) is called the (direct) Fourier
transform of pðt Þ while the function pðt Þ resulting from Eq. (9.4.17) is called
the inverse Fourier transform of P ðwÞ.
In the time domain, a function will be denoted by a small letter while its
Fourier transform is by the same capital letter. The relationship between them
will be symbolized by
Þ
pðt Þ , P ðw (9.4.18)
Usually, we denote the Fourier transform of a function pðt Þ by F ½pðt Þ while
its inverse is by F 1 ½pðt Þ, namely
Þ ¼ F ½pðt Þ
P ðw (9.4.19)
pðt Þ ¼ F 1 ½P ðw
Þ (9.4.20)
The Fourier transform of the derivative of a function pðt Þ is readily estab-
lished by applying integration by parts to Eq. (9.4.16). Generally, for a function
of which the ðn 1Þ order derivatives are continuous and the nth order deriv-
ative is piecewise continuous, it can be shown that
Z 1
pðnÞ ðt Þeiwt dt ¼ ðiw
Þn P ðw
Þ (9.4.21)
1
Example 9.4.1 The Fourier transform of a function

Find the Fourier transform of the function pðt Þ ¼ ejt j .
Solution Z 1
Þ ¼
P ðw ejt j eiwt dt
1
Z 0 Z 1
¼ eð1iwÞt dt + eð1 + iwÞt dt
1 0
1 1
¼ +
1 iw 1 + iw

2
¼
1+w 2
9.5 Response to a nonperiodic load

In Section 3.7.3, we have seen that the response of a SDOF system to an arbi-
trary load is given by the convolution integral
Z t
u ðt Þ ¼ pðt Þ∗ h ðt Þ ¼ pðτÞh ðt τÞdτ (9.5.1)
0
where h ðt τÞ is the response to the unit impulse given by

exwðtτÞ
h ðt τ Þ ¼ sin wD ðt τÞ, t > τ (9.5.2)
mwD
@CivilMethod
for x 6¼ 0 and
1
h ðt τ Þ ¼ sin wðt τÞ, t > τ (9.5.3)
mw
for x ¼ 0.
The integral (9.5.1) can be also written as
Z 1
u ðt Þ ¼ pðt Þ∗ h ðt Þ ¼ pðτÞh ðt τÞdτ (9.5.4)
1
because pðτÞ ¼ 0, when τ < 0 and h ðt τÞ ¼ 0, when τ > t.

The establishment of the response to an arbitrary load in the frequency
domain is achieved by taking the Fourier transform of the convolution
(9.5.4). Thus, we have
Z 1 Z 1
Þ ¼
U ðw pðt Þh ðt τÞdτ eiwt dt (9.5.5)
1 1
If we set t τ ¼ s, then t ¼ τ + s and Eq. (9.5.5) gives

Z 1 Z 1
Þ ¼
U ðw h ðs Þeiws ds pðτÞeiwτ dτ
Z1
1
1
¼ ÞpðτÞeiwτ dτ
H ðw
1 (9.5.6)
Z 1
¼ H ðwÞ pðτÞeiwτ dτ
1
ÞH ðw
¼ P ðw Þ
From Eq. (9.5.6) we deduce that the Fourier transform of the response to an
arbitrary load, namely of the convolution integral, is equal to the product of the
Fourier transforms of the functions in the convolution. Hence, we may write
symbolically
Þ
pðt Þ , P ðw (9.5.7)
Þ
h ðt Þ , H ðw (9.5.8)
Þ
u ðt Þ ¼ pðt Þ ∗h ðt Þ , U ðw (9.5.9)
The Fourier transform of the function u ðt t0 Þ is obtained as
Z 1
F½uðt t0 ¼ u ðt t0 Þeiwt dt
1
Z 1
¼ u ðxÞeiwðx + t0 Þ dt
1 (9.5.10)
Z 1
¼ eiwt0 u ðx Þeiwx dt
1
¼ eiwt0 F ½u ðt Þ
@CivilMethod
The forgoing equation represents the shifting property of the Fourier

transform.
The Fourier transform H ðw Þ of the response h ðt τÞ to the unit impulse can
be obtained as follows.
The equation of motion of the response h ðt τÞ to the unit impulse results by
setting pðt Þ ¼ d ðt τÞ in the equation of the SDOF system. That is,
m h€ + ch_ + kh ¼ dðt τÞ (9.5.11)
By virtue of Eqs. (9.5.10), (9.4.21), (3.7.6b), Eq. (9.5.11) gives

m w2 + ci w
+ k H ðwÞ ¼ 1 (9.5.12)
1
Þ ¼
H ðw (9.5.13)
2
ðm w + kÞ
+ ci w
or
1
Þ ¼ 2
H ðw (9.5.14)
k b 1 + 2ixb
Eq. (9.5.14) is identical to Eq. (9.3.8), that is, the complex response function
is the Fourier transform of the response of the SDOF system to the unit impulse.
Example 9.5.1 The Fourier transform method for the dynamic response of a
SDOF system
Determine the response of a SDOF (k,w,x) system subjected to the load
pðt Þ ¼ p0 , 0 < t using the frequency domain analysis.
Solution
The Fourier transform of the load is
Z Z
1 1 t 1 1 p0
Þ ¼
P ðw p0 e i w
dt ¼ p0 eiwt dt ¼ (1)
2p 1 2p 0
2pi w
The Fourier transform of the response to the unit impulse is
1
Þ ¼ 2
H ðw , b ¼ w
=w (2)
k b 1 + 2ixb
Consequently, by virtue of Eq. (9.5.9), we obtain
p0
U ðw Þ ¼ P ðw
ÞH ðw
Þ ¼ 2 (3)
k b 1 + 2ixb
2pi w
The response in the time domain results as the inverse Fourier transform of
Þ, that is,
U ðw
Z 1
p0 eiwt
u ðt Þ ¼ 2 dw
(4)
2pik 1 w
b + 2ixb 1
@CivilMethod
which is further written as

Z 1
p0 eiwbt
u ðt Þ ¼ db (5)
2pikw 1 b ðb b 1 Þðb b 2 Þ
where b 1 , b2 are the roots of the polynomial b2 + 2ixb 1, namely

qffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffi
b 1 ¼ ix + 1 x2 , b 2 ¼ ix 1 x 2 (6)
The integral (5) is evaluated using the method of closed line integrals in the
complex domain of b. This method yields
u ðt Þ ¼ 0, t 0
" !#
p0 x
u ðt Þ ¼ 1e xwt
cos wD t + pffiffiffiffiffiffiffiffiffiffiffiffi sin wD t , 0 < t
k 1 x2
From the last example, we observe that the dynamic analysis in the fre-
quency domain requires the evaluation of complicated integrals, even for the
simplest load cases. This problem is circumvented by applying numerical
methods for the evaluation of the Fourier transform, such as the discrete Fourier
transform (DFT), and the fast Fourier transform (FFT). These methods are dis-
cussed in the next sections.
9.6 Discrete Fourier transform

In the previous paragraph, we developed the method of dynamic analysis in the
frequency domain. The steps we followed are summarized as follows:
1. We take the Fourier transforms of the loading function, P ðw Þ ¼ F ½pðt Þ and
the response function to the unit pulse H ðwÞ ¼ F ½h ðt Þ.
2. We evaluate the product of the two transforms U ðw Þ ¼ P ðw
ÞH ðwÞ obtained
in the previous step.
3. We evaluate the inverse Fourier transform, which gives the response of the
system u ðt Þ ¼ F 1 ½U ðw
Þ in the time domain.
The success of the method is based on the capability of finding the Fourier trans-
form of a given function and its inverse. However, this is not always easy. With
the exception of simple functions whose Fourier transforms are obtained from
tables [4], their establishment requires the computation of complicated inte-
grals, a task that is difficult, tedious, or even impossible.
Very often, the loading function is specified by a set of values at distinct time
instants, for example, the accelerogram of an earthquake. As we mentioned, the
previous difficulties can be overcome by developing methods for the numerical
computation of the Fourier transform (direct and inverse). The discrete
Fourier transform (DFT) is among them. Below, we present the numerical
@CivilMethod
implementation of the DFT and its application for dynamic analysis in the fre-
quency domain.
We consider the function pðt Þ of Fig. 9.6.1, which is defined in the interval
0 t ttot .
tot
FIG. 9.6.1 Nonperiodic function expanded perodically.
Then we assume that the function is periodically extended from 1 to +1

with a period T0 ttot to include a number of zero values of pðt Þ in T0 .
If this function is expanded in Fourier series, we obtain by virtue of
Eqs. (9.2.7), (9.2.8a), (9.2.8b), (9.2.8c)
X
1
pðt Þ ¼ cn ein wt (9.6.1)
n¼1
Z T0 =2
1
cn ¼ pðt Þeinwt dt
T0 T0 =2
Z T0 , n ¼ 0, 1, 2, … (9.6.2)
1 t
in w
¼ pðt Þe dt
T0 0
The integral (9.6.2) can be evaluated numerically. For this purpose, we

divide the interval T0 into N equal subintervals of length Dt ¼ T0 =N and
approximate the integral with a sum of N rectangles.
If we set
s ðt Þ ¼ pðt Þeinwt (9.6.3)
then we have
1 NX1
cn ¼ s ðtk ÞDt (9.6.4)
T0 k¼0
where s ðtk Þ is the value of the integrand at instants tk ¼ kDt. That is,
s ðtk Þ ¼ pðtk ÞeinwkDt (9.6.5)
@CivilMethod

T0 ¼ N Dt, ¼ 2p=T0 ¼ 2p=N Dt
w (9.6.6)
we obtain
s ðtk Þ ¼ pðtk Þe2pikn=N (9.6.7)
1 NX
1
cn ¼ pðtk Þe2pikn=N (9.6.8)
N k¼0
As was shown in Section 3.6, the Fourier series can be approximated by a

sum of finite terms, Eq. (3.6.9), with very good accuracy even in cases of a dis-
continuous function, as demonstrated by Example 3.6.2. Therefore, on the basis
of Eq. (9.6.6) we can write Eq. (9.6.1) as
X
M
pðtk Þ ’ cn e2pikn=N (9.6.9)
n¼M
In the foregoing equation, we split the sum into two sums

X
1 X
M
pðtk Þ ¼ cn e2pikn=N + cn e2pikn=N (9.6.10)
n¼M n¼0
then taking into account that the function pðt Þ is periodic with a period
T0 ¼ N Dt, we can write the first sum as
X
1 X
1
cn e2pikn=N ¼ cn + N e2pik ðn + N Þ=N (9.6.11)
n¼M n¼M
Subsequently, setting M ¼ ðN 1Þ=2, n~ ¼ n + N and taking into account

the periodicity of pðt Þ, we can write the right side of the previous equation as
X
2M X
M
cn~ e2pik n~ =N ¼ cn e2pikn=N (9.6.12)
n~ ¼M + 1 n¼1
Substituting Eq. (9.6.11) into Eq. (9.6.10), and taking into account
X
2M
pðtk Þ ¼ cn e2pikn=N (9.6.13)
n¼0
or because 2M ¼ N 1, it follows:
X
N 1
pðtk Þ ¼ cn e2pikn=N (9.6.14)
n¼0
@CivilMethod
In order to match the expressions between the continuous Fourier transform

and the discrete Fourier transform, we set cn ¼ Pn and Eqs. (9.6.8), (9.6.14) are
written
1 NX
1
Pn ¼ pk e2pikn=N , k ¼ 0, 1, 2, …,N 1 (9.6.15)
N k¼0
X
N 1
pk ¼ Pn e2pikn=N , k ¼ 0, 1, 2, …,N 1 (9.6.16)
n¼0
The foregoing relations express the discrete Fourier transform (DFT), direct
and inverse, respectively.
The DFT approximates numerically the continuous Fourier transform,
defined by Eqs. (9.4.16), (9.4.17). The accuracy of the DFT is very good if
Dt is selected small. However, there is a fundamental difference between the
continuous Fourier transform and the discrete Fourier transform. The first pro-
vides the exact transform of the actual function while the second assumes a peri-
odic extension of the function. This means that the discrete transform is
applicable when the interval T0 is finite. It holds only within the period. Outside
it, the two transforms are completely different unless the function happens to be
periodic.
9.7 Application of the discrete Fourier transform

to dynamic analysis
As we mentioned in Section 9.5, the response in the time domain of a SDOF
system subjected to an arbitrary load is given by the convolution integral of
the load function pðt Þ and the response function to the unit impulse h ðt Þ.
Namely,
u ðt Þ ¼ pðt Þ ∗h ðt Þ (9.7.1)
Eq. (9.7.1) can be used as the basis to obtain the response in the frequency
domain by the DFT adhering to the following steps:
1. We compute the DFT of the function pðt Þ, 0 t ttot , which is assumed
extended periodically with a period T0 ttot so that pðtN Þ ¼ 0.
1 NX
1
Pn ¼ pðtk Þe2pikn=N (9.7.2)
N k¼0
2. We compute the DFT of the response function h ðt Þ. This requires the con-
finement of h ðt Þ in an interval equal or smaller than T0 .
1 NX
1
Hn ¼ h ðtk Þe2pikn=N (9.7.3)
N k¼0
@CivilMethod
3. We compute the DFT of the product

U n ¼ Pn H n (9.7.4)
4. We compute the inverse DFT of the product
X
N 1
u ðtk Þ ¼ Un e2pikn=N , k ¼ 0, 1, 2, …,N 1 (9.7.5)
n¼0
which yields the response in the time domain.

Details about the application of the DFT method to dynamic analysis can be
found in the relevant literature, for example, Refs. [3, 5].
9.8 Fast Fourier transform

The discrete convolution of two functions pðt Þ, h ðt Þ is defined as
X
N 1
u ðtk Þ ¼ pðtm Þh ðtkm ÞDt, tk ¼ kDt, tkm ¼ ðk m ÞDt (9.8.1)
m¼0
where both functions pðt Þ and h ðt Þ are periodic, that is,

pðtm + rN Þ ¼ pðtm Þ (9.8.2a)
h ðtm + rN Þ ¼ h ðtm Þ, r ¼ 0, 1, 2, … (9.8.2b)
Eq. (9.8.1) gives the dynamic response u ðtk Þ directly in the time domain at
time tk if pðt Þ represents the load and h ðt Þ the response to the unit
impulsive load.
We observe that the time-domain analysis based on Eq. (9.8.1) requires N 2
multiplications between real numbers. Instead, for the dynamic analysis in the
frequency domain, the required multiplications are N 2 between real and com-
plex numbers as dictated by each of Eqs. (9.7.2), (9.7.3), N of complex numbers
as dictated by Eq. (9.7.4), and N 2 multiplications of complex numbers as dic-
tated by Eq. (9.7.5).
It is clear that the dynamic analysis in the frequency domain using the DFT
requires significantly more computations, a fact that does not encourage its use.
However, it is possible to reduce the number of computations drastically by tak-
ing advantage of the harmonic properties of the involved functions. The algo-
rithm that reduces the calculations is referred to as the fast Fourier transform
(FFT). The appearance of the FFT gave an impetus to dynamic analysis in the
frequency domain. The number of operations (multiplications) required by the
FFT decreases from N 2 to N log 2 N . Fig. 9.8.1 illustrates the advantage of FFT
@CivilMethod
over DFT. The first FFT algorithm was developed by Gauss in the early 19th
century [6]. Also, the contributions of Runge, Danielson, Lanczos, and others
in the early 20th century were significant. However, its use did not attract the
interest of many researchers because the calculations had to be performed by
hand. It was only with the advent of computers that the FFT came to the fore-
ground. In 1965, J. W. Cooley and J. W. Tukey published an algorithm for cal-
culating the FFT [7, 8]. This algorithm is similar to that of Gauss and others and
is named after them as the Cooley-Tukey algorithm. Today, there are several
algorithms for FFT based on this algorithm. Below we present the Sande-Tukey
algorithm that is a variation of the Cooley-Tukey.
FIG. 9.8.1 Number of operations in DFT and FFT.
9.8.1 The Sande-Tukey algorithm

In this algorithm, we assume that N is a power of 2, that is,
N ¼ 2M (9.8.3)
where M is an integer. This constraint is introduced to simplify the algorithm.
In general, the DFT can be represented as
X
N 1
Pn ¼ ~k e2pikn=N , p
p ~k ¼ pk =N , n ¼ 0, 1, 2, …,N 1 (9.8.4)
k¼0
@CivilMethod
Eq. (9.8.4) can also be written in the form
X
N 1
Pn ¼ ~k W nk
p (9.8.5)
k¼0
where W is the complex weight function defined as
W ¼ e2pi=N (9.8.6)
We divide now the interval into two subintervals, and we express

Eq. (9.8.4) as
ðNX
=2Þ1 X
N 1
Pn ¼ pk e2pikn=N +
~ ~k e2pikn=N , n ¼ 0, 1, 2, …,N 1 (9.8.7)
p
k¼0 k¼N =2
Then we introduce a new variable m ¼ k N =2, so that the total number

of the indices is the same in both sums. Thus, we may write the foregoing
equation as
ðNX
=2Þ1 ðNX
=2Þ1
Pn ¼ ~k e2pikn=N +
p ~m + N =2 e2pinðm + N =2Þ=N
p (9.8.8)
k¼0 m¼0
or
=2Þ1
ðNX
Pn ¼ ~k + epin p
p ~k + N =2 e2pikn=N (9.8.9)
k¼0
n
We observe that eipn ¼ ðeip Þ ¼ ð1Þn . Consequently, for points with
even n this factor is equal to one while with odd n it is equal to 1. The next
step is to separate the terms of Eq. (9.8.9) into two sums corresponding to the
even and odd values of n. Hence, for even values, we have
=2Þ1
ðNX
P2n ¼ p ~k + N =2 e2pik ð2nÞ=N
~k + p
k¼0
(9.8.10)
=2Þ1
ðNX
¼ p ~k + N =2 e2pikn=ðN =2Þ
~k + p
k¼0
@CivilMethod
while for odd values
=2Þ1
ðNX
P2n + 1 ¼ p ~k + N =2 e2pik ð2n + 1Þ=N
~k p
k¼0
(9.8.11)
=2Þ1
ðNX
¼ p ~k + N =2 e2pik=N e2pikn=ðN =2Þ
~k p
k¼0
where n ¼ 0, 1, 2, …, ðN =2Þ 1.
By virtue of Eq. (9.8.6), Eqs. (9.8.10), (9.8.11) may be written
=21
NX
P2n ¼ p ~k + N =2 W 2kn
~k + p (9.8.12)
k¼0
=21
NX
P2n + 1 ¼ p pk + N =2 W k W 2kn
~k ~ (9.8.13)
k¼0
We can now make an important observation, which is the key to the method.
The even and the odd expressions can be considered as two DFTs of N =2 points
each. We further set
~k + p
gk ¼ p ~k + N =2 (9.8.14)

hk ¼ p~k p
~k + N =2 W k , k ¼ 0, 1, 2, …, ðN =2Þ 1 (9.8.15)
Hence
)
P2n ¼ Gn
, n ¼ 0, 1, 2, …, ðN =2Þ 1 (9.8.16)
P2n + 1 ¼ Hn
In other words, a computation at N points has been replaced by two com-

putations at N =2 points each. Because each of the latter computations requires
ðN =2Þ2 complex calculations against N 2 required by the initial computation,
the number of multiplications is reduced to 2ðN =2Þ2 ¼ N 2 =2. It is obvious,
therefore, that the process of bisecting the transforms can be repeated in a sec-
ond stage. We can thus compute 4 DFT with N =4 points each using the first and
last point of each sequence of N =4 points. This technique continues and ends in
@CivilMethod
DFT with two points (see Fig. 9.8.2). The total number of complex multiplica-
tion for a given DFT reduces to N log 2 N . The importance of FFT over DFT is
demonstrated in Fig. 9.8.1.
FIG. 9.8.2 Flow chart of the first substitution stage of the DFF with N points by two DFF of N =2
points each when N ¼ 8.
On the basis of the previous analysis, a computer program has been

written in MATLAB that evaluates the dynamic response of a SDOF. The pro-
gram uses the MATLAB functions fft.m and ifft.m. This program, which is
given the name FFT_dymamic.m, is available on this book’s companion
website.
Example 9.8.1 The Fourier transform for the dynamic response of a SDOF
system
Determine the response of SDOF system subjected to the Athens earthquake
in 1999 using the FFT method. Data: m ¼ 1, x ¼ 0:1, k ¼ 25, and pðt Þ ¼ u€g ðt Þ
where u€g ðt Þ is the accelerogram of the earthquake.
Solution
The solution is obtained using the program FFT_dymamic.m. The response
of the system is shown in Fig. E9.1 as compared with the numerical solution
in the time domain using the program aem.lin.m given in Section 4.4. It
is obvious that the computed responses by both methods are graphically
identical.
@CivilMethod
u(t) (m)
t
u,t(t) (m/s)
t
u,tt(t) (m/s2)
t
FIG. E9.1 Dynamic response of the SDOF system using FFT in Example 9.8.1.
@CivilMethod
Example 9.8.2 Amplitude spectrum of an accelerogram

Compute and plot the amplitude spectrum of the 1999 Athens earthquake
using the FFT method.
Solution
The amplitude spectrum is established by computing the magnitude of the
values of the Fourier transform of the accelerogram and it is plotted versus
the frequency (Hz) in the interval of the duration of the excitation. Fig. E9.2
shows the amplitude spectrum of the Athens 1999 earthquake as computed
using the program FFT_Ampl_Spectrum.m, which utilizes the function fft.m
of MATLAB. The program FFT_Ampl_Spectrum.m is available on this
book’s companion website.
FIG. E9.2 Amplitude spectrum of the 1999 Athens earthquake in Example 9.8.2.
9.9 Problems
Problem P9.1 Write a computer program for the evaluation of DFT and com-
pute the DFT of the function shown in Fig. P9.1. Then compute the inverse DFT
and compare the results with the exact function.
FIG. P9.1 Function p(t) in problem P9.1
@CivilMethod
Problem P9.2 The SDOF system m, x,k is subjected to a ground motion ug ðt Þ.

Study its dynamic response using:
(a) The exact analytic method in the time domain.
(b) The numerical computation of the convolution integral in the time domain.
(c) The DFT in the frequency domain.
h i
Data: x ¼ 0:1, u€gi ¼ 20 1 + ð1Þi ði + 5Þ=ði + 1Þ , ti ¼ 0:05i, i ¼ 1, 2, …,20,
u ð0Þ ¼ u_ ð0Þ ¼ 0, and w ¼ 4ps1 .
Problem P9.3 Give the graph of the Fourier transform Aðw Þ versus the fre-
quency (Hz) of the accelerogram of the Mexico City earthquake using the
FFT. The file Mexico_Earthquake.txt including the values of the accelerogram
is available on this book’s companion website.
Problem P9.4 The water tower of Fig. P9.4a is subjected to the load
pðt Þ ¼ 150sin pt=t1 , t1 ¼ 0:5 (Fig. P9.4b), which acts at the center of mass of
the tank. Approximating the structure by a generalized SDOF system, study
the dynamic response in the frequency domain for ttot ¼ 1s. The material of
the structure is reinforced concrete with a specific weight g ¼ 25kN=m3 . The col-
umn and the tank have a circular cross-section. The interior of the tank is divided
into small compartments to eliminate the liquid-structure interaction phenomena.
Use the first mode shape of the cantilever with a constant cross-section as a shape
function, that is, ψðx Þ ¼ ½ coshlx cos lx 0:7341ð sinh lx sinhlx Þ=3,
l ¼ 1:8751=L, where L is the length of the cantilever.
(b)
(a)
FIG. P9.4 Water tower (a) and load (b) in problem P9.4
@CivilMethod

[1] R. Bracewell, The Fourier Transform and Its Applications, McGraw-Hill, New York, 1978.
[4] F. Oberhettinger, Tabellen zur Fourier Transformation, 1957 Sringer-Verlag, Berlin, 1957.
[5] R.R. Craig Jr., A.J. Kurdila, Fundamentals of Structural Dynamics, second ed., John Wiley,
New Jersey, 2006.
[6] M.T. Heideman, D.H. Johnson, C.S. Burrus, Gauss and the history of the fast Fourier transform,
IEEE ASSP Mag. 1 (4) (1984) 14–21.
[7] J.W. Cooley, J.W. Tukey, An algorithm for the machine calculation of complex Fourier series,
Math. Comput. 19 (1965) 297–301.
[8] J.F. Hall, A FFT algorithm for structural dynamics, Earthq. Eng. Struct. Dyn. 10 (1982)
797–811.
[9] W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterlin, Numerical Recipes in FORTRAN,
second ed., Cambridge University Press, New York, 1992.
@CivilMethod
Chapter 10
Multi-degree-of-freedom
systems: Models and equations
of motion
Chapter outline
10.1 Introduction 325 10.5 Systems with distributed
10.2 Systems with localized mass mass and distributed stiffness 341
and localized stiffness 327 10.5.2 The method of global
10.3 Systems with distributed shape functions 342
mass and localized stiffness 328 10.6 Mixed systems 347
10.4 Systems with localized mass 10.7 Transformations of the
and distributed stiffness 330 equations of motion 351
10.4.1 The method of 10.8 Problems 354
influence coefficients 334 References and further reading 358
10.1 Introduction
So far, we have studied the dynamic response of SDOF systems. We have also
shown how a system with infinite degrees of freedom can be approximated by
a SDOF system. The trustworthiness of this approximation depends on various
issues. If the actual distribution of the physical properties of the structure, that
is, mass and stiffness, and that of the external force produce deformation during
the motion similar to the assumed, then the approximation with a SDOF system
gives acceptable results. A key shortcoming of this approximation is the diffi-
culty in determining the degree of reliability of the obtained results. In general,
however, the study of the dynamic response of structures requires their model-
ing with MDOF systems, especially when the deformation shapes are compli-
cated. In engineering structures, the mass, though distributed to all its members,
is usually lumped at certain points or regions. For example, in buildings the
mass is lumped at the levels of the stories or in a water tower at the top of
the column that supports the tank. This fact allows describing the motion of
a structure with that of a MDOF system with deformation parameters the dis-
placements of the points where the dynamic characteristics (mass and moment
of inertia) are concentrated.
Fig. 10.1.1a shows a three-story frame whose horizontal beams are virtually
rigid. In this structure, the mass of the columns is negligible compared to that of

@CivilMethod
326 PART II Multi-degree-of-freedom systems
the beams. Hence, the mass is lumped at the level of the beams and the structure
can be approximated by the model of Fig. 10.1.1b. The deformation shape dur-
ing motion is shown in Fig. 10.1.1c. Obviously, its motion can be determined by
establishing the displacements u1 ðt Þ, u2 ðt Þ, and u3 ðt Þ, that is, the system has
three degrees of motion.
(a) (b) (c)

FIG. 10.1.1 Three-story frame (a), dynamic model (b) and deformation shape (c).
(a) (b)
FIG. 10.1.2 Water tower (a) and its deformed dynamic model (b).
Fig. 10.1.2a shows a water tower. With the assumption that the mass of the
column is negligible compared with that of the tank, we can model the water
tower with the system of Fig. 10.1.2b. That is, the water tower is simulated
by a flexible column, which is fixed at the ground and has a mass m at its
top with a moment of inertia Io . The mass m can move horizontally and rotate
within the plane. The determination of the motion requires the establishment of
the displacement u ðt Þ and the rotation fðt Þ of the top cross-section of the col-
umn, that is, the system has two degrees of freedom.
The MDOF systems can be categorized as follows:
1. Systems with localized mass and localized stiffness.
2. Systems with distributed mass and localized stiffness.
@CivilMethod
MDOF systems: Models and equations of motion Chapter 10 327
3. Systems with localized mass and distributed stiffness.

4. Systems with distributed mass and distributed stiffness.
5. Mixed systems resulting from a combination of the systems of previous
categories.
In what it follows, the term localized mass denotes a lumped mass that may have
a moment of inertia while the term concentrated mass denotes a mass that is
concentrated at a point and has a zero moment of inertia. The above classifica-
tion is useful as it dictates a convenient method to formulate the equations of
motion of the MDOF system. Its effectiveness is explained by illustrative exam-
ples. The pertinent bibliography with recommended references for further study
is also included. The chapter is enriched with problems to be solved.
10.2 Systems with localized mass and localized stiffness

The system of Fig. 10.2.1a consists of two rigid bodies with masses m1 and m2 ,
whose horizontal motion is elastically constrained by two massless springs k1 and
k2 . The system has localized masses and stiffnesses. The degrees of freedom are
two, the displacements u1 ðt Þ and u2 ðt Þ. The equations of motion can be derived
using the method of the equilibrium of the forces (D’Alembert’s principle) pre-
sented in Section 1.5. The forces applied to the masses are shown in Fig. 10.2.1b.
The equilibrium of mass m1 gives
fI 1 + fD1 + fS1 ¼ p1 (10.2.1)
or
m1 u€ 1 + c1 ðu_ 1 u_ 2 Þ + k1 ðu1 u2 Þ ¼ p1 ðt Þ (10.2.2)
Similarly, the equilibrium of mass m2 gives
fI 2 + fD2 fD1 + fS2 fS1 ¼ p2 (10.2.3)
or
m2 u€2 c1 u_ 1 + ðc2 + c1 Þu_ 2 k1 u1 + ðk2 + k1 Þu2 ¼ p2 ðt Þ (10.2.4)
Eqs. (10.2.2), (10.2.4) are written in matrix form

m1 0 u€1 c1 c1 u_ 1 k1 k1 u1
+ +
0 m2 u€2 c1 c2 + c1 u_ 2 k1 k2 + k1 u2

p1
¼ (10.2.5)
p2
or
u + Cu_ + Ku ¼pðt Þ
M€ (10.2.6)
where

m1 0 c1 c1 k1 k1
M¼ , C¼ , K¼ (10.2.7)
0 m2 c1 c1 + c2 k1 k1 + k2
@CivilMethod
are the mass, the damping, and stiffness matrices of the system, respec-
tively, and

u1 p1 ð t Þ
u¼ , pðt Þ ¼ (10.2.8)
u2 p2 ðt Þ
the displacement and load vectors.
(a)
(b)
FIG. 10.2.1 System with localized mass and localized stiffness (a). Forces applied to the masses
m1 and m2 (b).
10.3 Systems with distributed mass and localized stiffness

The system of Fig. 10.3.1a consists of the rigid bar AC , whose mass is uni-
formly distributed along its length with line mass density m, and the circular
rigid body of total mass m ¼ mL, also uniformly distributed. The motion of
the system is elastically constrained by the two massless springs, k1 ¼ k and
k2 ¼ 2k. A damper with damping coefficient c acts at point C . The system is
loaded by the moment M ðt Þ and the force pðt Þ. This structure, which repre-
sents a system with distributed mass and localized stiffness, has two degrees
of freedom. We choose the vertical displacement u ðt Þ of the end C of the bar
and the rotation fðt Þ of the circular body to determine the motion of the sys-
tem. Referring to Fig. 10.3.1b, the equilibrium of the rigid bar with respect to
point A gives:
@CivilMethod
(a)
(b)
FIG. 10.3.1 System with distributed mass and localized stiffness (a). Forces acting on the two
bodies (b).
MIA + 2Lf D + Lf S1 + 2Lf S2 ¼ pðt ÞL (10.3.1)

Obviously, we have
u€ ð2LÞ3 u€ 4
MIA ¼ IA
¼m ¼ mL _ fS1 ¼ 0:5ku
2 u€ , fD ¼ cu,
2L 3 2L 3
fS2 ¼ 2k ðu 0:5LfÞ
which are substituted into Eq. (10.3.1) to yield
2 9 1
u€ + cu_ + ku kLf ¼ pðt Þ
mL (10.3.2)
3 4 2
Similarly, the equilibrium of the circular body with respect to point O gives
MIO 0:5Lf S2 ¼ M ðt Þ (10.3.3)

mL 1 3€
mL
MIO ¼ IO f€ ¼ p ð 0:5L Þ 4€
f ¼ f, fS2 ¼ 2k ðu 0:5LfÞ
pð0:5LÞ2 2 8
and substituting into Eq. (10.3.3), we obtain
3€
mL
f kLu + 0:5L2 kf ¼ M ðt Þ (10.3.4)
8
@CivilMethod
Eqs. (10.3.2), (10.3.4) are the equations of motion, which in matrix form
become
M€ (10.3.5)
where
2 3
2 " #
0
mL 9
63 7 c 0 k kL
M¼4 , C¼ , K¼
35 4 (10.3.6)
mL 0 0
0 kL 0:5L2 k
8
represent the mass, damping, and stiffness matrices of the system and
8 9
< 1 pðt Þ =
u
u¼ , pðt Þ ¼ 2 (10.3.7)
f : ;
M ðt Þ
the displacement and load vectors.
10.4 Systems with localized mass and distributed stiffness

The system of Fig. 10.4.1 a represents an example of this case. It consists of the
flexible column AO, whose mass is assumed negligible, and the plane square
body of side length a and mass m uniformly distributed.
(a) (b)
FIG. 10.4.1 System with localized mass and distributed stiffness (a). Deformed dynamic model (b).
The equations of motion will results from the motion of the plane rigid body in
its plane. We examine the motion with respect to the point O, which does not coin-
cide with the center of mass of the body. In general, the system has three degrees of
freedom, namely the horizontal displacement, the vertical displacement, and the
rotation about O. Because the column is flexible, the horizontal displacement and
the rotation are due to the bending deformation. The vertical displacement is
caused by (i) the axial deformation of the column, which is very small and thus
@CivilMethod
neglected, and (ii) the shortening of the chord of the deflection curve, which is
also neglected in the linear theory. Therefore, the parameters of the motion are
the horizontal displacement u ðt Þ and the rotation fðt Þ, Fig. 10.4.1b. In the follow-
ing, the equations of motion are derived using two different methods.
1. The method of the Lagrange equations
(i) Elastic energy: This is due to the bending deformation of the column. The
deflection curve can be set in the form
uðx, t Þ ¼ 1 ðx Þu ðt Þ + 2 ðx Þfðt Þ (10.4.1)
where 1 ðx Þ and 2 ðx Þ are the elastic curves of the column for u ðt Þ ¼ 1,

fðt Þ ¼ 0 and fðt Þ ¼ 1, u ðt Þ ¼ 0, respectively. They can be obtained from
the solution of the following two boundary value problems
d4 1 0 0
¼ 0, 1 ð0Þ ¼ 0, 1 ð0Þ ¼ 0, 1 ðh Þ ¼ 1, 1 ðh Þ ¼ 0 (10.4.2a)
dx 4
and
d4 2 0 0
¼ 0, 2 ð0Þ ¼ 0, 2 ð0Þ ¼ 0, 2 ðh Þ ¼ 0, 2 ðh Þ ¼ 1 (10.4.2b)
dx 4
Integrating of the differential equation (10.4.2a) gives
1 3 1 2
1 ðx Þ ¼ c1 x + c2 x + c3 x + c4 (10.4.3)
6 2
After evaluation of the arbitrary constants by applying the boundary
conditions, we obtain
x 2 x 3
1 ðx Þ ¼ 3 +2 ¼ 3x2 + 2x3 , x ¼ x=h (10.4.4)
h h
x 2 x
2 ð x Þ ¼ h 1 ¼ hx 2 ðx 1Þ, x ¼ x=h (10.4.5)
h h
The elastic energy is given (see Eq. 8.1.11)
Z
1 h
EI ½u00 ðx, t Þ dx
2
U¼
2 0
Z
1 h 00 2
U¼ EI 1 ðx Þu ðt Þ + 002 ðx Þfðt Þ dx (10.4.6)
2 0
Differentiating Eq. (10.4.6) with respect to u ðt Þ and fðt Þ gives
@CivilMethod
Z h
∂U 00 00
00
¼ EI 1 ðx Þu ðt Þ + 2 ðx Þfðt Þ 1 ðx Þdx
∂u 0
Z h Z
00
2 h
00 00
¼ EIu ðt Þ (10.4.7a)
1 ðx Þ dx + EI fðt Þ 2 ðx Þ 1 ðx Þdx
0 0
12EI 6EI
¼ u+ 2 f
h3 h
Z h
∂U 00 00
00
¼ EI 1 ðx Þu ðt Þ + 2 ðx Þfðt Þ 2 ðx Þdx
∂f 0
Z h Z h
00 00
00
2
¼ EIu ðt Þ (10.4.7b)
1 ðx Þ 2 ðx Þdx + EI fðt Þ 2 ðx Þ dx
0 0
6EI 4EI
¼ u+ f
h2 h
(ii) Kinetic energy: The kinetic energy with respect to point O is evaluated
from Eq. (1.5.8). Taking the origin of the coordinates at point O we have:
XP ¼ u, YP ¼ 0, w ¼ f, _ xc ¼ 0, yc ¼ a=2, IP ¼ IO and Eq. (1.5.8)
becomes
1 1 a
T ¼ m u_ 2 + IO f_ 2 m u_ f_ (10.4.8)
2 2 2
Differentiating Eq. (10.4.8) with respect to u_ and f_ gives
∂T a
¼ m u_ m f_ (10.4.9a)
∂u_ 2
∂T a
¼ IO f_ m u_ (10.4.9b)
∂f_ 2
(iii) Generalized forces: They result from Eq. (1.8.23) by setting q1 ¼ u,

q2 ¼ f. Thus, we obtain
Q1 ¼ pðt Þ, Q2 ¼ pðt Þa=2 (10.4.10a,b)
(iv) The potential of the external conservative forces: Because there are no
conservative forces, it is A ¼ 0.
Substituting Eqs. (10.4.7a), (10.4.7b), (10.4.9a) (10.4.9b), (10.4.10a,b)
into the Lagrange equation (1.8.11) gives
a 12EI 6EI
m u€ m f€ + 3 u + 2 f ¼ pðt Þ (10.4.11a)
2 h h
a 6EI 4EI a
IO f€ m u€ + 2 u + f ¼ pðt Þ (10.4.11b)
2 h h 2
@CivilMethod
or in matrix form
M€
u + Ku ¼pðt Þ (10.4.12)
where
2 3
2 ma 3 12EI 6EI
m 6 h3 h2 7
M ¼ 4 ma 2 5, K¼6 4 6EI 4EI 5
7 (10.4.13a,b)
IO
2 h2 h
( )
u pðt Þ
u¼ , pðt Þ ¼ a (10.4.13c,d)
f pðt Þ
2
2. The method of equilibrium of forces

The forces acting on the square rigid body during the motion are shown in
Fig. 10.4.2. The equilibrium of forces may be considered either with respect
to point O or to the center of mass C .
FIG. 10.4.2 Free body diagram.
a. Equilibrium with respect to point O.

Because point O does not coincide with the mass center, the equations
of motion will be obtained from Eqs. (1.5.14a), (1.5.14b), (1.5.14c) by
taking the origin of axes at the point O and setting: P O, uP ¼ u,
vP ¼ 0, xc ¼ 0, yc ¼ a=2, Fx ¼ pðt Þ fS , Fy ¼ 0, MO ¼ pðt Þa=2 MS .
Thus, we obtain
a
m€u m f€ + fS ¼ pðt Þ (10.4.14a)
2
a a
m u€ + IO f€ + MS ¼ pðt Þ (10.4.14b)
2 2
The elastic forces fS and MS are the shear force and the bending
moment of a beam fixed at end A and subjected to the displacement u
and rotation f at end O. Thus, we have
12EI 6EI
fS ¼ 3
u+ 2 f (10.4.15a)
h h
@CivilMethod
6EI 4EI
MS ¼ 2
u+ f (10.4.15b)
h h
Obviously, they are identical to those given by Eqs. (10.4.7a),
(10.4.7b).
Substituting the previous expressions for fS and MS in
Eqs. (10.4.14a), (10.4.14b) yields
a 12EI 6EI
u m f€ + 3 u + 2 f ¼ pðt Þ
m€ (10.4.16a)
2 h h
a 6EI 4EI a
IO f€ m u€ + 2 u + f ¼ pðt Þ (10.4.16b)
2 h h 2
which are identical to Eqs. (10.4.11a), (10.4.11b).
b. Equilibrium with respect to the mass center C .
The equations of motion will be obtained from Eqs. (1.5.11a),
(1.5.11b), (1.5.11c) by setting uc ¼ u af=2, Fx ¼ fS + pðt Þ,
MC ¼ MS fS a=2. Thus, we obtain
a
m€u m f€ + fS ¼ pðt Þ (10.4.17a)
2
a
Ic f€ + MS + fS ¼ 0 (10.4.17b)
2
The first of the above equations is identical to (10.4.14a). The second
equation, however, looks different from (10.4.14b). Nevertheless, multi-
plying Eq. (10.4.17a) by a=2 and adding it to (10.4.17b) gives
a 2
a a
m€ u + IC + m f€ + MS ¼ pðt Þ (10.4.18a)
2 2 2
Because IC + m ða=2Þ2 ¼ IO (Steiner’s formula), the previous equa-
tion becomes
a a
m€ u + IO f€ + MS ¼ pðt Þ (10.4.18b)
2 2
which is identical to (10.4.14b).
The above transformation of the equations of motion from the center
of mass to point O is rather occasional. A formal method to transform the
equations of motion when we change the point of reference is presented in
Section 10.7.
10.4.1 The method of influence coefficients

The equations of motion of linear systems with localized masses and distributed
stiffnesses can be derived using the method of influence coefficients, which is
based on the superposition principle. A frame structure can be idealized as an
assemblage of beam elements interconnected at nodal points. We illustrate this
@CivilMethod
method with the plane frame of Fig. 10.4.3. The masses are localized at the
nodes, where the external loads are applied. In the general case in which the
axial deformation of the beams and columns is not neglected, each node i
has three degrees of freedom with respect to the global system of axes X Y ,
Y axes, respectively, and a rota-
two translations ui , vi in the directions of the X,

tion fi about the Z axes. Hence, the frame of Fig. 10.4.3 with n ¼ 6 free nodes
has in total N ¼ 3n ¼ 18 degrees of freedom.
FIG. 10.4.3 Frame with n ¼ 6 nodes.
Referring to Fig. 10.4.4, we represent the displacement vector by

T ¼ u1 v1 f1 u2 v2 f2 ⋯ ui vi fi ⋯ un vn fn
u (10.4.19)
or if they are labeled from 1 to N ¼ 3n, we may write
uT ¼ f u1 u2 u3 u4 u5 u6 ⋯ u3i2 u3i1 u3i ⋯ uN g (10.4.20)
where
u3i2 ¼ ui , u3i1 ¼ vi , u3i ¼ fi , i ¼ 1, 2, …,n (10.4.21)
FIG. 10.4.4 Nodal displacements of the frame. N ¼ 3n ¼ 18.
@CivilMethod
Similarly, we formulate the vector of the external nodal loads (Fig. 10.4.5)
FIG. 10.4.5 External loads acting on the nodes of the frame.
pðt ÞT ¼ f px1 py1 M1 px2 py2 M2 ⋯ pxi pyi Mi ⋯ MN g

(10.4.22)
¼ f p1 p2 p3 p4 p5 p6 ⋯ p3i2 p3i1 p3i ⋯ pN g
where
p3i2 ¼ pxi , p3i1 ¼ pyi , p3i ¼ Mi , i ¼ 1, 2, …,n (10.4.23)
The number of equations will be equal to the number of degrees of freedom.
The equations of motion can be formulated using the method of equilibrium of
forces. If fIi , fDi , fSi represent the inertia, damping, and elastic forces, respec-
tively, and pi ðt Þ the external load in the direction of the displacement ui , then
it should be (see Fig. 10.4.6)
fIi + fDi + fSi ¼ pi ðt Þ, i ¼ 1, 2, …,N (10.4.24)
(a) (b)
FIG. 10.4.6 Equilibrium of forces (a) or moments (b) in the direction of the displacement ui .
Eq. (10.4.24) holds in the direction of all displacement components

u1 , u2 ,…,uN , N ¼ 3n. The expressions of the forces fIi , fDi , fSi can be deter-
mined using the following method, which is referred to as the method of influ-
ence coefficients.
@CivilMethod
10.4.1.1 Elastic forces

In a SDOF system, the elastic force depends only on one displacement. Appar-
ently, in a MDOF system this force will depend on all the displacements. Thus,
for a linear system, the superposition principle allows writing
fSi ¼ ki1 u1 + ki2 u2 + … + kij uj + … + kiN uN (10.4.25)
FIG. 10.4.7 Stiffness influence coefficients ki4 .
where kij are constants expressing the stiffness influence coefficients or simply
the stiffness coefficients. They relate the elastic force fSi to the displacement uj .
The physical meaning of the coefficient kij results by setting uj ¼ 1 and uk6¼j ¼ 0
in Eq. (10.4.25), that is, by applying a unit displacement along the degree of
freedom j, holding all other displacements zero, as shown in Fig. 10.4.7. This
yields
fSi ¼ kij (10.4.26)
Eq. (10.4.26) states that the stiffness coefficient kij expresses the elastic
force applied along the degree of freedom i for a unit displacement along the
degree of freedom j, that is, the displacement uj , while all other displacements
are zero.
Fig. 10.4.7 shows the deformation of the frame when the displacement u4 ¼ 1
is applied at node 4 while all other displacements are zero, that is, uj ¼ 0
( j ¼ 1, 2, …, N , j 6¼ 4). Obviously, the stiffness influence coefficients are equal
to the forces required to maintain the deformed shape of the frame when sub-
jected to the displacement u4 ¼ 1.
For i ¼ 1, 2, …, N , Eq. (10.4.25) yields N equations, which we write in
matrix form
8 9 2 38 9
> f > k k ⋯ k1N > > u1 >
< S1 >
> = 6 11 12 7< u2 =
>
fS2 k k ⋯ k
¼6 21 22
4 ⋯ ⋯ ⋯ ⋯ 5> ⋮ >
2N 7 (10.4.27)
> ⋮ >
>
>
: ; : >
> ;
fSN kn1 kn2 ⋯ kNN uN
@CivilMethod
or
f S ¼ ku (10.4.28)
where
8 9 8 9 2 3
> fS1 > > u1 > k11 k12 ⋯ k1N
>
> >
> > >
> >
< fS1 = < u2 = 6k k ⋯ k 7
6 21 22 2N 7
fS ¼ , u¼ , k¼6 7 (10.4.29)
> ⋮ >
> > > ⋮ >
> > 4⋯ ⋯ ⋯ ⋯ 5
>
: >
; >
: ; >
fSN uN kn1 kn2 ⋯ kNN
The vector f S with dimensions N 1 represents the vector of the elastic
forces, the vector u with dimensions N 1 represents the vector of the displace-
ments, and the matrix k with dimensions N N represents the stiffness matrix
of the structure. Obviously, this method, which explains the physical meaning
of the stiffness coefficients, is by no means suitable for their evaluation because
it requires N static analyses of the fixed structure successively for u1 ¼ 1,
u2 ¼ 1, …, uN ¼ 1. The stiffness matrix, however, can be established using
other methods of structural analysis, for example, the flexibility method or
the direct stiffness method. Anyhow, the establishment of the stiffness matrix
of a structure is a subject of the static structural analysis and it is evaluated using
the most suitable method for a particular structure.
10.4.1.2 Damping forces

The concept of the influence coefficients can be employed to express the damp-
ing forces fDi . Thus, assuming viscous damping we may write
fDi ¼ ci1 u_ 1 + ci2 u_ 2 + … + cij u_ j + … + ciN u_ N (10.4.30)
Applying Eq. (10.4.30) for i ¼ 1, 2, …, N yields N equations, which we
write in matrix form
8 9 2 38 9
> fD1 > c11 c12 ⋯ c1N > u_ 1 >
> >
> > >
>
< f = 6 c c ⋯ c 7> < u_ > =
D2 6 21 22 2N 7 2
¼6 7 (10.4.31)
>
>
> ⋮ >>
>
4 ⋯ ⋯ ⋯ ⋯ 5> >
> ⋮ >>
>
: ; : ;
fDN cn1 cn2 ⋯ cNN u_ N
or
f D ¼ cu_ (10.4.32)
where
8 9 8 9 2 3
> fD1 > > u_ 1 > c11 c12 ⋯ c1N
>
> >
> > >
> >
< fD2 = < u_ 2 = 6c c ⋯ c
6 21 22 2N
7
7
fD ¼ , u_ ¼ , c¼6 7 (10.4.33)
> ⋮ >
> > > ⋮ >
> > 4⋯ ⋯ ⋯ ⋯ 5
>
: >
; >
: ; >
fDN u_ N cn1 cn2 ⋯ cNN
@CivilMethod
FIG. 10.4.8 Damping influence coefficients ci4 .
The vector f D with dimensions N 1 represents the vector of the damping

forces, the vector u_ with dimensions N 1 represents the vector of the veloc-
ities, and the matrix c with dimensions N N represents the damping matrix of
the structure. The elements cij of the damping matrix are the damping influence
coefficients. Their physical meaning is analogous to that of the stiffness influ-
ence coefficients, namely cij expresses the damping force applied along the
degree of freedom i for unit velocity along the degree of freedom j, that is,
u_ j ¼ 1, while all other velocities are zero. Fig. 10.4.8 show the damping influ-
ence coefficients at the nodes of the frame, when u_ 4 ¼ 1, u_ j ¼ 0 (j ¼ 1, 2, …, N ,
j 6¼ 4). Note that the damping coefficients, contrary to the stiffness coefficients,
cannot be established from the geometrical data of the structure and the physical
properties of its elements. In Section 12.11, we will show how the damping
matrix of the structure can be established by assuming known damping ratios
for each mode shape estimated from experimental data obtained from similar
structures.
10.4.1.3 Inertial forces

Similarly, we can relate the inertial force fIi to the acceleration u€ j . Thus, we may
write
fIi ¼ mi1 u€ 1 + mi2 u€ 2 + … + mij u€ j + … + miN u€ N (10.4.34)
For i ¼ 1, 2, …, N , Eq. (10.4.34) yields N equations, which we write in
matrix form
8 9 2 38 9
> fI 1 >
> m11 m12 ⋯ m1N > u€ 1 >
>
< = 6 7 < >
> =
fI 2 m m ⋯ m u€2
¼6 21 22 2N 7
(10.4.35)
> ⋮ > 4 ⋯ ⋯ ⋯ ⋯ 5> ⋮ >
: >
> ; : >
> ;
fIN mn1 mn2 ⋯ mNN u€ N
or
f I ¼ m€
u (10.4.36)
@CivilMethod
where
8 9 8 9 2 3
> fI 1 > > u€ 1 > m11 m12 ⋯ m1N
< >
> = < >
> = 6 m21 m22
fI 2 u€ 2 ⋯ m2N 7
fI ¼ , €¼
u , m¼6
4 ⋯ ⋯
7 (10.4.37)
> ⋮ > > ⋮ >
> ⋯ ⋯ 5
: >
> ; : >
;
fIN u€ N mn1 mn2 ⋯ mNN
The vector f I with dimensions N 1 represents the vector of the inertial
forces, the vector u € with dimensions N 1 represents the vector of the accel-
erations, and the matrix m with dimensions N N represents the mass matrix
or inertial matrix of the structure. The elements mij of the mass matrix are the
mass influence coefficients. Their physical meaning is analogous to that of the
stiffness and damping influence coefficients, namely mij expresses the inertial
force applied along the degree of freedom i for unit acceleration along the
degree of freedom j, that is, u€ j ¼ 1, while all other accelerations are zero.
Fig. 10.4.9 shows the mass influence coefficients at the nodes of the frame,
when u€ 4 ¼ 1, u€ j ¼ 0 (j ¼ 1, 2, …, N , j 6¼ 4).
FIG. 10.4.9 Mass influence coefficients mi4 .
In actual structures, the mass is distributed. The model that considers the
mass lumped at certain points of the structure, for example, at the nodes of a
frame, approximates adequately the dynamic response of the structure. When
it is assumed that the lumped mass has no geometrical dimensions, that is, it
is simulated by a material particle, then its rotational inertia is zero and the
respective influence coefficients vanish.
Writing now Eq. (10.4.24) for all directions i ¼ 1, 2, …, N , we obtain
fI 1 + fD1 + fS1 ¼ p1 ðt Þ
fI 2 + fD2 + fS2 ¼ p2 ðt Þ
(10.4.38)
… … … …
fIN + fDN + fSN ¼ pN ðt Þ
or
f I + f D + f S ¼ p ðt Þ (10.4.39)
@CivilMethod
which by virtue of Eqs. (10.4.28), (10.4.32), (10.4.36) gives the equation of

motion of the structure
u + cu_ + ku ¼ pðt Þ
m€ (10.4.40)
10.5 Systems with distributed mass and distributed stiffness

Systems with distributed mass and distributed stiffness are also referred to as
continuous systems or distributed parameter systems. Theoretically, they have
infinite degrees of freedom and their motion is described by partial differential
equations. The systems of Fig. 10.5.1 belong to this category.
(a) (b)
FIG. 10.5.1 Continuous systems: (a) Cantilever beam, (b) Chimney fixed on the ground.
Continuous systems can be approximated by MDOF systems, even if the

mass distribution does not necessarily have a particular concentration at some
points. This approximation can be realized using two methods. The first method
analyzes the structure by expressing its deformed shape as a superposition of a
series of global shape functions of the spatial coordinates, each multiplied by its
own generalized coordinate. This method is known as the method of global
shape functions. On the contrary, the second method treats the structure as a
discrete MDOF system in which the mass and stiffness are concentrated at cer-
tain points, but with interacting displacements. The discretization can be per-
formed either by the flexibility method or the stiffness method. The latter is
realized by the finite element method (FEM) [1], in which the structure is
approximated by a set of discrete elements with known deformation and inertial
properties. Both methods, that is, the method of the global shape functions and
the FEM, convert the governing partial differential equation into a system of
ordinary differential equations of motion, which can be solved by well-
established numerical methods. Consequently, the dynamic analysis of contin-
uous systems through a direct analytical solution of partial differential
equations, a very difficult and in most cases insurmountable mathematical
@CivilMethod
problem, is circumvented. Nevertheless, the research is ongoing and new effi-

cient methods have been developed for solving time-dependent structural prob-
lems, for example, the boundary element method (BEM) [2] or the meshless
methods beyond the element methods [3]. The method of global shape functions
is presented below while the FEM is presented in this Chapter 11.
10.5.1 The method of global shape functions

In this method, it is assumed that the continuous function u ðx, t Þ, which repre-
sents the displacement of the points of the structure, can be approximated by the
finite superposition series
X
N
uðx, t Þ ¼ 1 ðx Þu1 ðt Þ + 2 ðx Þu2 ðt Þ + … + N ðx ÞuN ðt Þ ¼ i ðx Þui ðt Þ
i¼1
(10.5.1)
The functions i ðx Þ express shapes of deformation of the structure and are
referred to as global shape functions while ui ðt Þ are functions of time that
express the generalized coordinates, for example, displacements or rotations
of certain points of the structure. The minimum requirements for the
expression (10.5.1) to approximate the actual deformation u ðx, t Þ are: (i) the
shape functions must be at least geometrically admissible, that is, satisfy
the geometric (essential) boundary (support) conditions of the structure, and
(ii) must be linearly independent, that is, any of them cannot result as a linear
combination of two or more of the others.
The generalized coordinates are the unknown time functions, which must be
determined by the solution of equations of motion. The number of generalized
coordinates is equal to the degrees of freedom of the substitute MDOF system,
which approximates the actual continuous system. The equations of motion are
usually derived using the principle of virtual work, the Lagrange equations, or
the Hamilton principle. Here, the latter approach, as presented for the general-
ized SDOF system in Section 8.2, is employed to illustrate the method of global
shape functions.
FIG. 10.5.2 Cantilever beam.
@CivilMethod
Suppose we are going to apply the method of global shape functions to

approximate the dynamic response of the cantilever beam of Fig. 10.5.2. The
cantilever has variable stiffness I ðx Þ and mass m ðx Þ. It is loaded by the trans-
verse load pðx, t Þ and the axial force P at the free end.
(i) Elastic energy: The elastic energy is equal to the strain energy of the can-
tilever and it is given (see Eq. 8.1.11)
Z
1 L
EI ðx Þ½u00 ðx, t Þ dx
2
U¼ (10.5.2)
2 0
Its variation is
Z L
dU ¼ EI ðx Þu00 ðx, t Þd u00 ðx, t Þdx (10.5.3)
0
or using Eq. (10.5.1)

Z L ! !
X
N X
N
00 00
dU ¼ EI ðx Þ i ui j duj dx
0 i¼1 j¼1
Z !
L N X
X N
00 00
¼ EI ðx Þ i j ui duj dx (10.5.4)
0 i¼1 j¼1
X
N X
N
¼ kij ui duj
i¼1 j¼1
where
Z L
00 00
kij ¼ EI ðx Þ i j dx (10.5.5)
0
(ii) Kinetic energy: The kinetic energy is due to the transverse displacements
and rotations of the mass elements. Thus, we have
Z Z h i2
1 L 1 L 0
m ðx Þ½u_ ðx, t Þ dx + I ðx Þ u_ ðx, t Þ dx
2
T¼ (10.5.6)
2 0 2 0
The second term in the above expression is due to the rotation of the
cross-sections and its contribution is small. In the following development,
without limiting the generality, we omit this term for the sake of simplicity.
The variation of the kinetic energy is
Z L
dT ¼ m ðx Þu_ ðx, t Þd u_ ðx, t Þdx (10.5.7)
0
@CivilMethod
Z ! !
L X
N X
N
dT ¼ m ðx Þ i u_ i j d u_ j dx
0 i¼1 j¼1
Z !
L X
N X
N
¼ m ðx Þ i j u_ i d u_ j dx (10.5.8)
0 i¼1 j¼1
X
N X
N
¼ mij u_ i d u_ j
i¼1 j¼1
where
Z L
mij ¼ m ðx Þ i j dx (10.5.9)
0
(iii) Virtual work of the nonconservative forces:

The nonconservative forces are the load pðx, t Þ and the damping force.
The damping may be external or internal. As in the case of generalized
SDOF systems (see Fig. 8.2.1), we assume that the external damping force
is viscous and is distributed along the length of the beam, that is,
fD ðx, t Þ ¼ cðx Þu_ ðx, t Þ, where cðx Þ is the damping coefficient. The internal
damping is due to the deformation of the element of the beam, resists it, and
depends on the velocity of the strain. Hence, if the produced stress is denoted
by sD , we may write
∂ex
sD ¼ cs (10.5.10)
∂t
where cs is the coefficient of the internal damping and ex the stain. Then,
the virtual work of the internal nonconservative forces is
Z
dWnc ¼ sD dex dV
in
(10.5.11)
V
sx M
ex ¼ , sx ¼ y, M ¼ EI ðx Þu00 ðx, t Þ (10.5.12)
E I ðx Þ
Eq. (10.5.11) becomes
Z L
00
dWncin
¼ cs I ðx Þu_ ðx, t Þd u00 ðx, t Þdx (10.5.13)
0
Z L ! !
X
N X
N
00 00
¼ cs I ðx Þ i u_ i
in
dWnc j duj dx
0 i¼1 j¼1
Z !
L X
N X
N
00 00
¼ cs I ðx Þ i j u_ i duj dx (10.5.14)
0 i¼1 j¼1
N X
X N
¼ cijin u_ i duj
i¼1 j¼1
@CivilMethod
where
Z L
00 00
cijin ¼ c s I ðx Þ i j dx (10.5.15)
0
The virtual work of the external damping force is

Z L
dWncex
¼ cðx Þu_ ðx, t Þd uðx, t Þdx (10.5.16)
0

Z L ! !
X
N X
N
dWnc ¼ c ðx Þ i u_ i
ex
j duj dx
0 i¼1 j¼1
Z !
L N X
X N
¼ c ðx Þ i j u_ i duj dx (10.5.17)
0 i¼1 j¼1
X
N X
N
¼ cijex u_ i duj
i¼1 j¼1
where
Z L
cijex ¼ c ðx Þ i j dx (10.5.18)
0
Finally, the virtual work due the external nonconservative load pðx, t Þ is
Z L
dWnc ¼
p
pðx, t Þd uðx, t Þdx (10.5.19)
0

Z L X
N
p
dWnc ¼ pðx, t Þ j duj dx
0 j¼1
(10.5.20)
X
N
¼ pj ðt Þduj
j¼1
where
Z L
pj ðt Þ ¼ pðx, t Þ j dx (10.5.21)
0
Consequently, the virtual work due to all nonconservative forces is

X
N X
N X
N
dWnc ¼ cij u_ i duj + pj ðt Þduj (10.5.22)
i¼1 j¼1 j¼1
@CivilMethod
where it was set

cij¼ cijin + cijex (10.5.23)
(iv) The potential energy of the conservative forces: The potential energy of the
conservative forces is due to the constant axial force P. Thus, we have
A ¼ Pe (10.5.24)
where e is the shorting of the elastic curve due to bending and is given as
Z L Z
1 L 0 2
e¼ ex dx ¼ ½u ðx, t Þ dx (10.5.25)
0 2 0
The variation of the potential energy is

dA ¼ Pde
Z L (10.5.26)
¼ P u0 ðx, t Þd u0 ðx, t Þdx
0

Z ! !
L X
N X
N
0 0
dA ¼ P i ui j duj dx
0 i¼1 j¼1
Z !
L X
N X
N
0 0
¼ P i j ui duj dx (10.5.27)
0 i¼1 j¼1
X
N X
N
¼ P kGij ui duj
i¼1 j¼1
where
Z L
0 0
kGij ¼ i j dx (10.5.28)
0
Substituting the previous expressions for dU ,dK , dWnc , and dA in

Hamilton’s principle, Eq. (1.7.13), produces the following equations of
motion
X
N X
N X
N
mij u€ j + cij u_ j + kij Pk Gij uj ¼ pi , i ¼ 1, 2…, N (10.5.29)
j¼1 j¼1 j¼1
or in matrix form
u + Cu_ + ðK PKG Þu ¼pðt Þ
M€ (10.5.30)
The matrices M, C, K represent the mass, damping, and stiffness matri-
ces, respectively. Their elements are evaluated from Eqs. (10.5.9),
@CivilMethod
(10.5.23), (10.5.5). The matrix KG , whose elements are evaluated from

Eq. (10.5.28), is referred to as the geometric stiffness matrix. It expresses
here the influence of the axial force on the dynamic response of the struc-
ture. It is obvious that the compressive axial force reduces the stiffness of the
system while the tensile force increases it. The vanishing of the determinant
of the total stiffness matrix K PKG , namely
det ðK PKG Þ ¼ 0 (10.5.31)
yields the N values of the buckling load.
The method of global shape functions is also known as the Ritz method. The
shape functions are, as we have already mentioned, geometrically admissible
and linearly independent functions representing deformation patterns of the
entire structure. They are chosen arbitrarily but appropriately, based on the
experience of the user of the method. The success of the method depends on
the choice of shape functions, but this is still a difficult problem [4]. The use
of the Ritz method has been substantially reduced with the emergence of mod-
ern computational methods, such as the finite element method (FEM) [1], the
boundary element method (BEM) [2], and the meshless methods [3]. However,
it is useful either as the theoretical background for developing new computa-
tional methods or for testing them.
10.6 Mixed systems

Mixed systems are those resulting from the combination of the previously dis-
cussed categories. As an example of this case, we refer to the structure of
Fig. 10.6.1a. Indeed, this system has localized masses (m1 ,m2 ), distributed
masses (mass of bars), localized stiffness (stiffness kR of the rotational spring),
and distributed stiffnesses (stiffness of bars). We assume: m1 ¼ m2 ¼ m,
m ¼ m=a, kR ¼ EI =4a.
(a)
(b)
FIG. 10.6.1 Mixed system (a) and motion parameters (b).
@CivilMethod
For the determination of motion, we take as degrees of freedom the displace-

ments of the masses, namely the rotation f1 of the mass m1 , the displacement u,

and the rotation f2 of the mass m2 ; see Fig. 10.6.1b. The equations of motion
will be derived using the Lagrange equations.
(i) Elastic energy: If u1 , u2 , u3 , u4 denote the displacements and the rotations at
the ends of the beam element as in Fig. 10.6.2, its elastic curve can be set in
the form
uðx, t Þ ¼ 1 ðx Þu1 + 2 ðx Þu2 + 3 ðx Þu3 + 4 ðx Þu4 (10.6.1)
The shape functions i express the elastic curves of the beam for ui ¼ 1,
when uj ¼ 0, j 6¼ i. Following the procedure presented in Section 10.4, we
obtain.
FIG. 10.6.2 Degrees of freedom of the beam element.
x 2 x 3
1 ðx Þ ¼ 1 3 +2 ¼ 1 3x 2 + 2x 3 (10.6.2a)
L L
x x 2
2 ðx Þ ¼ L 1 ¼ Lx ðx 1Þ2 (10.6.2b)
L L
x 2 x 3
3 ðx Þ ¼ 3 2 ¼ 3x2 2x3 (10.6.2c)
L L
x 2 x
4 ð x Þ ¼ L 1 ¼ Lx2 ðx 1Þ (10.6.2d)
L L
where
x ¼ x=h (10.6.3)
For the element 2–3 holds: u1 ¼ af1 , u2 ¼ f1 , u3 ¼ u a f2 =2, u4 ¼ f2 ,
L ¼ 3a.
For the element 4–5 holds: u1 ¼ u + a f2 =2, u2 ¼ f2 , u3 ¼ 0, u4 ¼ 0,
L ¼ 3a.
The elastic energy is given by the expression
Z Z
1 3 1 5 1
EI ½u00 ðx, t Þ dx + EI ½u00 ðx, t Þ dx + kR f1
2 2 2
U¼ (10.6.4)
2 2 2 4 2
@CivilMethod
Z
1 3a
f1 , f2 ¼
U u, EI 00
1 ðx Þa f1 +
00
2 ðx Þf1 +
00
3 ðx Þ u a f2 =2 + 00 2
4 ðx Þf2 dx
2 0
Z
1 3a 1
00
u + a f2 =2 + 00 2 + kR f1
2
+ EI 1 ðx Þ 2 ðx Þf2 dx
2 0 2
(10.6.5)
Differentiating the above relation yields after evaluation of the integrals
∂U
¼ k11 u + k12 f1 + k13 f2 (10.6.6a)
∂u
∂U
¼ k21 u + k22 f1 + k23 f2 (10.6.6b)
∂f1
∂U
¼ k31 u + k32 f1 + k33 f2 (10.6.6c)
∂f2
where
8 EI 10 EI
k11 ¼ , k12 ¼ k21 ¼ , k13 ¼ k31 ¼ 0
9 a3 9 a2
121 EI 17 EI 38 EI
k22 ¼ , k23 ¼ k32 ¼ , k33 ¼
36 a 9 a 9 a
Therefore the stiffness matrix of the system is
2 3
32 40a 0
EI 4
k¼ 40a 121a2 17a 2 5 (10.6.7)
36a 3
0 17a 2 152a 2
(ii) Kinetic energy: The kinetic energy results from the expression
Z Z
1 3 1 5
½u_ ðx, t Þ dx + ½u_ ðx, t Þ dx
2 2
T¼ m m
2 2 2 4
(10.6.8)
1 1 1
+ I1 f_ 12 + m u_ 2 + IC f_ 22
2 2 2
1 Z 3a h i2
_ f_ 1 , f_ 2 ¼
T u, 1 ðx Þa f_ 1 + 2 ðx Þf_ 1 + 3 ðx Þ u_ a f_ 2 =2 +
m _
4 ð x Þf 2 dx
2 0
Z
1 3a h i2
+ 1 ðx Þ u_ + a f_ 2 =2 + 2 ðx Þf_ 2 dx
m
2 0
2 2
1 1 1
+ I1 f_ + m u_ + IC f_
2
2 1 2 2 2
(10.6.9)
@CivilMethod
Differentiating the above relation yields after evaluation of the integrals

d ∂T
+ m12 f€1 + m13 f€2
¼ m11 u€ (10.6.10a)
dt ∂u_
!
d ∂T
+ m22 f€1 + m23 f€2
¼ m21 u€ (10.6.10b)
dt ∂f_
1
!
d ∂T
+ m32 f€1 + m33 f€2
¼ m31 u€ (10.6.10c)
dt ∂f_
2
where
2712 558
m11 ¼ m, m12 ¼ m21 ¼ ma, m13 ¼ m31 ¼ 0
840 840
2294 2 675 2 1832 2
m22 ¼ ma , m23 ¼ m32 ¼ ma , m33 ¼ ma
840 840 840
Therefore, the mass matrix of the system is
2 3
2712 558a 0
m 4
m¼ 558a 2294a 2 675a 2 5 (10.6.11)
840
0 675a 2 1832a 2
(iii) Generalized forces: The virtual work of the nonconservative forces is due
to the load pðt Þ. Thus, we have

Q1 d u + Q2 df1 + Q3 df2 ¼ pðt Þd u + a f2 =2
(10.6.12)
¼ pðt Þd u pðt Þadf2 =2
Q1 ¼ pðt Þ
Q2 ¼ 0
Q3 ¼ pðt Þa=2
Hence the equation of motion of the mixed system is
2 38 9
€
m6
2712 558a 0 
> =
2 7 f €

4 558a 2294a 675a 5
2
1
840 : € >
> ;
0 675a 2 1832a 2 f2
2 38 9 8 9 (10.6.13)
32 40a 0 >
= >
< pðt Þ >
=
EI 6 2 7
+ 4 40a 121a 2
17a 5 f ¼ 0
36a3 >
: >
1
; > : >
;
0 17a 2 152a 2 f2 pðt Þa=2
@CivilMethod
10.7 Transformations of the equations of motion

In the previous sections, we presented various methods for the formulation of
the equations of motions of a MDOF system with N degrees of freedom. In each
case, a set of independent displacements is selected, which, as stated in
Section 1.8, expresses the generalized coordinates or simply the coordinates,
as we will now be referring to, of the system and are denoted by the vector
of the displacements u in the equation of motions. It is often necessary or con-
venient to express the equations in a system of also independent coordinates u

other than that originally selected. The transformation of the coordinates from u
to u is a linear relation of the form
u ¼R
u (10.7.1)
where R is the N N square matrix referred to as the transformation matrix. If
M, C, and K are the mass, damping, and stiffness matrices, respectively, and
pðt Þ the vector of the external loads in the original system of coordinates u,
we need to determine the matrices M, C, K and the vector pðt Þ in the trans-
formed system of coordinates u . This can be achieved using the principle of vir-
tual work as follows.
The elastic forces in the coordinates u are given by the relation
f S ¼Ku (10.7.2)
by the relation
while in the coordinates u
f S ¼K
u (10.7.3)
We give u a virtual displacement du. Then the corresponding virtual dis-
placement du results from the relation (10.7.1)
du ¼Rd
u (10.7.4)
The virtual work done by the elastic forces f S due to the virtual displacement
du is
dWS ¼duT f S
(10.7.5)
¼ duT Ku
or using Eqs. (10.7.1), (10.7.4)
dWS ¼d
uT RT KR
u (10.7.6)
Similarly, the virtual work of the elastic forces in the system of the coordi-
is
nates u
S ¼d
dW uT f S
(10.7.7)
¼ d
uT Ku
@CivilMethod
S ¼ dWS , which gives

Obviously, it must be d W
uT RT KR
d u ¼ d u
uT K
or

uT K
d RT KR u¼0 (10.7.8)
and d
Because the vectors u u are nonzero, it must be
¼ RT KR
K (10.7.9)
Similarly, by considering the virtual work of the inertial forces f I ¼ M€
u and
f I ¼ Mu
€ in the two systems of coordinates u and u , we obtain
¼ RT MR
M (10.7.10)
Further, by considering the virtual work of the damping forces f D ¼ Cu_ kai
f D ¼ C
u_ in the two systems of coordinates, we prove that
¼ RT CR
C (10.7.11)
Finally, by expressing the virtual work of the external forces in the two sys-
tems of coordinates, we have
uÞT pðt Þ
dWp ¼ duT pðt Þ ¼ ðRd
(10.7.12)
uT RT pðt Þ
¼ d
and
p ¼ d
dW ðt Þ
uT p (10.7.13)
p , we obtain
Because these two virtual works are equal, dWp ¼ d W
ðt Þ ¼ RT pðt Þ
p (10.7.14)
Example 10.7.1 The equation of motion of the system shown in Fig. 10.4.1 with
respect to point O is (see Eqs. 10.4.11a, 10.4.11b):
2 ma 3
m ( pðt Þ )
6 2 7 u€ + k11 k12 u
¼
4 ma 5 € a (1)
IO f k21 k22 f pðt Þ
2 2
where
12EI 6EI 4EI
k11 ¼ , k12 ¼ k21 ¼ , k22 ¼ (2)
h3 h2 h
Transform Eq. (1) with respect to the mass center C of the body.
@CivilMethod
Solution
If u and f denote the displacement of the mass center and its rotation about it,
then the transformation relations result by geometrical consideration as
a
u ¼ u + f (3)
2
f ¼ f (4)
or
" a #
u 1 u
¼ 2 (5)
f 0 1 f
Hence
"
a# " #
1 1 0
R¼ 2 , RT ¼ a (6)
1
0 1 2
Then Eq. (10.7.10) gives
2 a 32 a 3
" # m m
1 0
¼ a 6 2 74 1 2 5
M 4 5
1 a
2 m IO 0 1
2 (7)
" #
m 0
¼
0 IC
where
a2
IC ¼ m (8)
6
is the moment of inertia with respect to the center of mass C .
Further, Eq. (10.7.9) gives
" #" #2 3
1 0 k11 k12 1 a
¼ a
K 4 25
1 k21 k22
2 0 1
" # (9)
k11 k12
¼
k21 k22
where
12EI
k11 ¼ k11 ¼ 3
h
a 6EI
k 12 ¼ k 21 ¼ k11 + k12 ¼ 3 ða + h Þ (10)
a 2
2 h
12EI a 2 ah h 2
k 22 ¼ k11 + ak 12 + k22 ¼ 3 + +
2 h 4 2 3
@CivilMethod
Finally, Eq. (10.7.14) gives

2 38 9
1 0 < pðt Þ =
ðt Þ ¼ 4 a 5
p a
1 : pðt Þ ;
2 2 (11)
( )
pðt Þ
¼
0
Therefore, the transformed equation of motion is
" #( ) 2 3( ) ( )
2 a+h
m 0 u€ 6EI 4 u pðt Þ
+ 3 3a 2 + 6ah + 2h 2 5 ¼ (12)
0 IC f€ h a+h f 0
6
Note that we obtain the same equations if the equations are derived directly
with respect to center of mass C .
10.8 Problems
Problem P10.1 Formulate the equations of motion of the system shown in
Fig. P10.1. Data: k1 ¼ 3k, k2 ¼ 2k, k3 ¼ k, c1 ¼ c3 ¼ c, c2 ¼ 2c, m1 ¼ m2 ¼ m,
and m3 ¼ 2m.
Problem P10.2 A square plate with a side length a, constant thickness

d ¼ a=10, and mass density g is supported by three columns of height
h ¼ a=2 and square cross-section of a side length b ¼ a=20 (Fig. P10.2).
Neglecting the axial deformation and the mass of the columns, formulate the
equations of motion of the structure using the Lagrange equations. The module
of elasticity is E and the shear modulus G ¼ 0:4E.
@CivilMethod
FIG. P10.2 Structure in problem P10.2.
Problem P10.3 A rigid bar of total mass m is supported by the three springs
k1 , k2 , k3 and the damper c as shown in Fig. P10.3. The circular disc of mass
0:5m at the end D of the bar has a diameter 0:2L. Formulate the equation of
motion. Data: k2 ¼ 1:5k1 and k3 ¼ 2k1 .
Problem P10.4 Formulate the equations of motion of the plane frame of

Fig. P10.4 when the supports 1 and 2 are subjected to the horizontal displace-
ments ug1 ðt Þ and ug2 ðt Þ, respectively. Give the expressions for the evaluation of
the support reactions. The beams are assumed rigid (I ¼ 1) while the columns
have the same cross-section and modulus of elasticity E.
FIG. P10.4 Frame in problem P10.4.
@CivilMethod
Problem P10.5 The beam of Fig. P10.5 consists of the flexible and massless
part AB and the rigid part BC of mass m per unit length. The beam is clamped
at A while the hinged support at C is elastically restrained by the rotational
spring CR ¼ EI =10L. The beam is loaded by a concentrated moment M ðt Þ
applied at B. Formulate the equation of motion of the structure and give the
expressions for the evaluation of the support reactions.
FIG. P10.5 Beam in problem P10.5.
Problem P.10.6 The structure of Fig. P10.6 consists of the two bars AB and
BC . The bar AB is massless and flexible and is simply supported at A while
per unit length and is elastically restrained at
the bar BC is rigid with mass m
the hinged end C by the rotational spring CR ¼ EI =10L. The structure is
loaded by a concentrated moment M ðt Þ applied at B. Formulate the equation
of motion of the structure and give the expressions for the evaluation of the
support reactions.
FIG. P10.6 Structure in problem P10.6.
Problem P10.7 Formulate the equation of motion of the plane frame of

Fig. P10.7. The massless cables FB and GC have cross-sectional area A and
are prestressed so that they can undertake compressive forces. Data:
¼ m=a (kg=m), and CR ¼ EI =10L.
I =A ¼ a2 =25, m
@CivilMethod
FIG. P10.7 Plane frame in problem P10.7.
Problem P10.8 Formulate the equations of motion of the system shown in

per unit length while the cable
Fig. P10.8. The bar AB is rigid and has mass m pffiffiffi
and the pulley are assumed massless. The total length of the cable is 1:5a 2.
The hinged support at A is elastically restrained by the rotational spring
CR ¼ EAa=250. The hanging body of mass m ¼ ma is moving downward with
a velocity v ¼ v0 sin wt.
Problem P10.9 The thin spherical tank of Fig. P10.9 has a diameter R ¼ a=3
and is full with a liquid of density g. It is supported on the ground through nine,
hinged at both ends, massless rods of cross-sectional area A and modulus of
elasticity E. The points 10 ,20 ,30 lie on a horizontal circular ring of the spherical
tank at a height a from the ground and at a distance R=3 from the center of the
sphere. The mass of the tank is neglected.
@CivilMethod
Formulate the equation of motion when

1. The structure performs free vibrations.
2. The structure is subjected to ground motion ug ðt Þ in the direction b with
respect to the x axis.
FIG. P10.9 Spherical tank in problem P10.9.
Problem P10.10 A vehicle traveling along a bridge is idealized as shown in

Fig. P10.10. The unsprung mass mc is connected to the sprung mass through
the suspension system (k,c). The bridge has a span L and is simply supported.
Using the Ritz method, formulate the equation of motion of the system bridge-
vehicle when the vehicle is traveling with a constant velocity v. The bridge is
simulated by a two-degree-of-freedom system with global shape functions
n ¼ sin ðnpx=LÞ, n ¼ 1, 2. The mass mc maintains its contact to the bridge dur-

ing the motion. Data: mc ¼ m0 =10, m0 ¼ mL=20, and k ¼ EI =100L.
FIG. P10.10 Vehicle traveling on a bridge in problem P10.10.

[1] O. Zienkiewicz, R. Taylor, The finite element method, seventh ed., Butterworth-Heinemann,
Oxford, UK, 2013.
[2] J.T. Katsikadelis, The boundary element method for engineers and scientists, second ed.,
Academic press, Elsevier, Oxford, UK, 2016.
[3] G.R. Liu, Meshfree methods: moving beyond the finite element method, second ed., CRC Press,
Boca Raton, FL, 2010.
[4] J.T. Katsikadelis, A generalized ritz method for partial differential equations in domains of arbi-
trary geometry using global shape functions. Eng. Anal. Bound. Elem. 32 (5) (2008) 353–367,
https://doi.org/10.1016/j.enganabound.2007.001.
@CivilMethod
Chapter 11
The finite element method

Chapter outline
11.1 Introduction 359 11.8 The finite element method
11.2 The finite element method for the plane grid 453
for the plane truss 360 11.8.1 Properties of the
11.2.1 Properties of the plane plane grid element 453
truss element 360 11.8.2 Transformation of the
11.2.2 Transformation of the nodal coordinates
nodal coordinates of of the plane grid
the truss element 374 element 463
11.2.3 Equation of motion of 11.9 The finite element method
the plane truss 377 for the space frame 476
11.2.4 Steps to formulate the 11.9.1 Properties of the
equations of motion space frame element 476
for a plane truss by the 11.9.2 Transformation of the
finite element method 381 nodal coordinates of
11.2.5 Modification of the the space frame
equations of motion element 482
due to the supports 11.10 The finite element method
of the structure 382 for the space truss 494
11.3 The finite element method 11.10.1 Properties of the
for the plane frame 392 space truss element 494
11.3.1 Properties of the 11.10.2 Transformation of
plane frame element 392 the nodal coordinates
11.3.2 Transformation of the of the space truss
nodal coordinates of the element 498
plane frame element 410 11.11 Rigid bodies within flexible
11.4 Static condensation: Guyan’s skeletal structures 503
reduction 424 11.11.1 Rigid bodies in
11.5 Flexural vibrations of a spaces frames 503
plane frame 429 11.11.2 Rigid bodies in
11.6 Reduction of the degrees spaces trusses, plane
of freedom due to grids, plane frames,
constraints 439 and plane trusses 510
11.7 Axial constraints in the 11.12 Problems 517
plane frame 442 References and further reading 521
Dynamic Analysis of Structures. https://doi.org/10.1016/B978-0-12-818643-5.00011-X

@CivilMethod
11.1 Introduction
The distributed mass is the usual case we encounter in actual structures, that is,
the actual structures should be modeled as continuous systems. This modeling
leads to partial differential equations with prescribed initial and boundary con-
ditions. The solution of such equations is a difficult mathematical problem, even
for individual components of structures (e.g., beams, plates, walls, etc.). On the
other hand, the method of global shape functions, which can provide an accept-
able solution, has limited capabilities due to the difficulty in choosing the global
shape functions [1]. In practice, this method cannot cope with conventional
structures such as a multistory frame or a grid. Besides, it cannot be pro-
grammed for automatic use on a computer. Therefore, neither method is
suitable for the dynamic analysis of realistic structures in engineering. At this
dead end, the finite element method (FEM) method gives a way out.
The FEM represents a major breakthrough in the field of computational
mechanics. Extensive literature on the general formulation of the FEM [2] as
well as on its application to dynamic problems is now available [3].
In the finite element approach, the structure being analyzed (e.g., frame,
wall, plate, shell, three-dimensional deformable body, or a combination of
them) is divided into a finite number of small subregions, elements, which
are interconnected at discrete points, the nodes, where the compatibility
condition for the displacements and the equilibrium of equivalent nodal forces
are ensured. After the discretization, we assume that the field function (the dis-
placement in this case) within each element varies according to a known law,
which is expressed as a superposition of shape patterns, the shape functions,
corresponding to unit values of the element nodal displacements. After that,
the elastic and kinetic energy as well as the virtual work of the nonconservative
nodal forces in terms of the nodal displacements, viewed as generalized coor-
dinates, are established. The equivalent nodal forces (elastic, inertial, damping,
and external forces) may result by the use of the Lagrange equations or the prin-
ciple of virtual work. Finally, the compatibility of the nodal displacements and
the equilibrium of the nodal forces give the differential equations of motion of
the structure. The FEM is presented below only for skeletal structures, that is,
structures consisting of straight-line elements (trusses, frames, grids) because
surface structures (walls, plates, shells) as well as three-dimensional (3D)
deformable bodies do not fall within the scope of this book.
11.2 The finite element method for the plane truss

11.2.1 Properties of the plane truss element
We consider a plane truss element with a variable cross-section Aðx Þ, mass per
unit length m ðx Þ, and modulus of elasticity E, whose local axis in the unde-
formed state coincides with the x axis. The ends j, k of the element are called
nodes of the element, thus any quantity (force, displacement) associated with
them is referred to as nodal quantity (e.g., nodal force, nodal displacement).
@CivilMethod
The finite element method Chapter 11 361
Fig. 11.2.1 Degrees of freedom of a plane truss element.
In the deformed state at time t, the ends of the elements are displaced to points
j 0 , k 0 and the element occupies the position j 0 k 0 (Fig. 11.2.1).
The element has four degrees of freedom, the displacements u1 ðt Þ, u2 ðt Þ of
node j and the displacements u3 ðt Þ, u4 ðt Þ of node k. These quantities constitute
the nodal coordinates of the plane truss element. A typical point x of the ele-
ment axis undergoes two displacements, the axial displacement u ðx, t Þ and the
transverse one v ðx, t Þ. We consider the element as a generalized system and
apply the Ritz method presented in Section 10.5.1. Thus, we set the axial
displacement in the form
u ðx, t Þ ¼ u1 ðt Þ 1 ðx Þ + u 3 ðt Þ 3 ðx Þ (11.2.1)
where 1 ðx Þ and 3 ðx Þ are shape functions expressing the axial deformation for
u1 ¼ 1, u3 ¼ 0 and u1 ¼ 0, u3 ¼ 1, respectively. The functions 1 ðx Þ and 3 ðx Þ
can be established as follows.
The equilibrium of the element dx (see Fig. 11.2.2) yields
dN
¼0 (11.2.2)
dx
N ¼ Aðx Þsx
¼ EAðx Þex (11.2.3)
d i
¼ EAðx Þ
dx
we write Eq. (11.2.2) as

d d i
EAðx Þ ¼0 (11.2.4)
dx dx
and after integration
Z
dx
i ¼ c1 + c2 (11.2.5)
Aðx Þ
Fig. 11.2.2 Beam element dx.

@CivilMethod
The arbitrary constants c1 and c2 are evaluated from the boundary condi-
tions. It is obvious that for a variable cross-section, the determination of the
shape functions requires the evaluation of the integral in Eq. (11.2.5). This
relation, although it gives shape functions expressing the exact static axial
deformation, is not suitable for the automation of the method because different
shape functions have to be determined for elements with a different law of
variation of the cross-section. This difficulty is surpassed if we accept the
same shape functions for all elements, regardless of the variation law of the
cross-section and, indeed, those resulting from Eq. (11.2.4) for a constant
cross-section. In that case, we obtain
i ¼ c1 x + c2
which for 1 ð0Þ ¼ 1 and 1 ð LÞ ¼ 0 gives
1 ðx Þ ¼ 1 x, x ¼ x=L (11.2.6a)
and for 3 ð 0Þ ¼ 0 and 3 ð LÞ ¼ 1
3 ðx Þ ¼ x, x ¼ x=L (11.2.6b)
The shape functions (11.2.6a), (11.26b) can also result if we arbitrarily
accept a linear law of variation of axial displacement within the element.
The assumption of linear variation of the displacement is at the expense of accu-
racy. However, the error is acceptable when the element is small and the cross-
sectional variation is not intense. This becomes clear from Example 11.2.1.
Taking into account that the element is a straight line in the deformed state,
we can write the transverse displacement asa
v ðx, t Þ ¼ u2 ð1 x Þ + u4 x
(11.2.7)
¼ u2 2 ðx Þ + u4 4 ðx Þ
a. Due to the axial deformartion we should write

v ðx, t Þ ¼ u2 ð1 x 0 Þ + u4 x 0 , x0 ¼ x 0 =L0 (a)
But it is
x0 x + u ðx, t Þ x=L + u ðx, t Þ=L u ðx, t Þ
x0 ¼ ¼ ¼ x+ (b)
L0 L + ðu3 u1 Þ 1 + ðu3 u1 Þ=L L
because ðu3 u1 Þ=L≪1:

Eq. (a) by virtue of Eq. (b) becomes

u ðx, t Þ u ðx, t Þ
v ðx, t Þ u2 1 x + u4 x +
L L
u4 u2
u2 ð1 x Þ + u4 x u ðx, t Þ
L
u2 ð1 x Þ + u4 x
because the nonlinear term u4 u

L u ðx, t Þ is much smaller compared to the linear ones and thus it can
2
be neglected.
@CivilMethod
Obviously, it is 2 ðx Þ ¼ 1 ðx Þ and 4 ðx Þ ¼ 3 ðx Þ.
The equivalent nodal forces in the directions of the degrees of freedom can
be established using the method of the Lagrange equations or the principle of
virtual work. Both methods are presented in the following.
11.2.1.1 The method of the Lagrange equations

We refer to Eq. (1.8.11), which we write for the truss element

d ∂T ∂T ∂U ∂A
+ + ¼ Qi , i ¼ 1, 2, 3, 4 (11.2.8)
dt ∂u_ i ∂ui ∂ui ∂ui
The forces ∂A=∂ui , derived from the potential A, are gravitational forces. As
shown in Section 1.9, they are ignored in the linear dynamic analysis but the
deformation produced by them should be superimposed to that resulting from
the dynamic analysis in order to obtain the total deformation of the structure.
Obviously, we deduce from Eq. (11.2.8) that the terms

d ∂T ∂T
fIi ¼ (11.2.9)
dt ∂u_ i ∂ui
∂U
fSi ¼ (11.2.10)
∂ui
express the nodal inertial force and the nodal elastic force, respectively, in the
direction of the generalized coordinate ui . Finally, the force Qi in Eq. (11.2.8)
includes the nonconservative generalized forces. In our case, these are the
damping forces fDi and the external forces pi . Regarding the sign convention
for the nodal forces, they are positive when their direction coincides with the
positive direction of the axes they refer.
(i) Nodal elastic forces and stiffness matrix of the truss element
Eq. (11.2.7) expresses the displacement of the element as a rigid body.
Therefore, the elastic energy is due only to the axial deformation of the element
given by
Z
1
U¼ sx ex dV (11.2.11)
2 V
Taking into account that sx ¼ Eex , ex ¼ ∂u=∂x, the previous relation
becomes
Z Z 2
1 L ∂u
U¼ E dxdydz
2 0 A ∂x
Z L 2 (11.2.12)
1 ∂u
¼ EAðx Þ dx
2 0 ∂x
@CivilMethod
Z
1 L 0 0
2
U ðu1 , u3 Þ ¼ EAðx Þ u1 1 ðx Þ + u3 3 ðx Þ dx (11.2.13)
2 0
The nodal elastic forces result from Eq. (11.2.10) for i ¼ 1, 2, 3, 4. Thus,
we obtain
∂U
fS1 ¼ ¼ k11 u1 + k13 u3 (11.2.14a)
∂u1
∂U
fS2 ¼ ¼0 (11.2.14b)
∂u2
∂U
fS3 ¼ ¼ k31 u1 + k33 u3 (11.2.14c)
∂u3
∂U
fS4 ¼ ¼0 (11.2.14d)
∂u4
where
Z L
kij ¼ EAðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 3 (11.2.15)
0
Eqs. (11.2.14a)–(11.2.14d) are written in matrix form as

8 9 2 38 9
> fS1 > k11 0 k13 0 > u1 >
> >
>
> > 6
> 7>> >
> >
>
> >
< fS2 = 6 0 0 0 07 < >
> =
6 7 u2
¼6 7 (11.2.16)
> > 6 07 > u3 >
> fS3 >
>
> > 4 k31 0 k33
> 5>> >
> >
>
>
: ; > : >
> ;
fS4 0 0 0 0 u4
or
f eS ¼ ke ue (11.2.17)
where u are the vectors of the nodal elastic forces and the nodal displace-
f eS , e
ments, respectively, and ke the stiffness matrix of the e truss element. Hence, the
element matrix ke for the plane truss element is defined as
2 3
k11 0 k13 0
6 7
60 0 0 07
6 7
k ¼6
e
7 (11.2.18)
6 k31 0 k33 0 7
4 5
0 0 0 0
Obviously, we deduce from Eq. (11.2.15) that kij ¼ kji . Hence, the stiffness
matrix is symmetric.
For an element with a constant cross-section Aðx Þ ¼ Ae and a length Le ,
Eq. (11.2.15) is integrated analytically and Eq. (11.2.18) yields
@CivilMethod
2 3
1 0 1 0
e6
EA 0 0 0 07
ke ¼ e 6 7 (11.2.19)
L 4 1 0 1 05
0 0 0 0
It should be noted that Eq. (11.2.10) expresses the Castigliano theorem.
Therefore, the previous method for establishing the stiffness matrix is identical
to the so-called energy method.
(ii) Nodal inertial forces and mass matrix of the truss element
In the FEM, the equivalent inertial forces are obtained by two different assump-
tions of the mass distribution on the element: the consistent mass assumption,
which assumes a continuous distribution of the mass on the element, and the
lumped mass assumption, which lumps the mass at its nodes. The inertial mass
matrices resulting from both assumptions are derived below.
(a) Consistent mass matrix
During the motion, the infinitesimal mass m ðx Þdx undergoes the two displace-
ments u ðx, t Þ and v ðx, t Þ. Therefore, the kinetic energy of the truss element will
be given by the expression
Z n o
1 L
T¼ m ðx Þ ½u_ ðx, t Þ2 + ½v_ ðx, t Þ2 dx (11.2.20)
2 0
or using Eqs. (11.2.1), (11.2.7)
Z n o
1 L 2 2
T ðu_ 1 , …, u_ 4 Þ ¼ m ðx Þ ½u_ 1 1 ðx Þ + u_ 3 3 ðx Þ + ½u_ 2 1 ðx Þ + u_ 4 3 ðx Þ dx
2 0
(11.2.21)
The inertial forces result from Eq. (11.2.9) for i ¼ 1, 2, 3, 4. Thus, after
performing the differentiations we obtain

d ∂T ∂T
fI 1 ¼ ¼ m11 u€1 + m13 u€3 (11.2.22a)
dt ∂u_ 1 ∂u1

d ∂T ∂T
fI 2 ¼ ¼ m22 u€2 + m24 u€4 (11.2.22b)
dt ∂u_ 2 ∂u2

d ∂T ∂T
fI 3 ¼ ¼ m31 u€1 + m33 u€3 (11.2.22c)
dt ∂u_ 3 ∂u3

d ∂T ∂T
fI 4 ¼ ¼ m42 u€2 + m44 u€4 (11.2.22d)
dt ∂u_ 4 ∂u4
where
Z L
mij ¼ mi + 1, j + 1 ¼ m ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 1, 3 (11.2.23)
0
@CivilMethod
Eqs. (11.2.22a)–(11.2.22d) are written in matrix form

8 9 2 38 9
fI 1 > u€
> > 1>
>
m11 0 m13 0
>
> > >
> >
>
>
< fI 2 = 6 7>> >
60 m22 0 m24 7 u€2 =
<
6
¼6 7 (11.2.24)
>
> fI 3 >
>
> 4 m31 0 m33 0 7 5>> u€3 >
>
>
> ;
: > > >
>
: >
;
fI 4 0 m42 0 m44 u€4
or
€e
f eI ¼ me u (11.2.25)
€e are the vectors of the nodal inertial forces and the nodal acceler-
where f eI , u
ations, respectively, and me the mass matrix of the e truss element. Therefore,
the element mass matrix me is defined as
2 3
m11 0 m13 0
60 m22 0 m24 7
me ¼ 64 m31 0
7 (11.2.26)
m33 0 5
0 m42 0 m44
Obviously, we deduce from Eq. (11.2.23) that mij ¼ mji . Hence, the mass
Eq. (11.2.23) is integrated
For an element with a constant mass, m ðx Þ ¼ m,
analytically and Eq. (11.2.26) gives
2 3
2 0 1 0
e6
m 60 2 0 17 7
me ¼ (11.2.27)
6 41 0 2 05
0 1 0 2
e Le is the total mass of the e element.
where m e ¼ m
(b) Lumped mass matrix
According to this assumption, the mass of the element is concentrated at its
nodes by static considerations, that is, they are obtained as the reactions of a
simply supported beam under the load m ðx Þ (see Fig. 11.2.3). Thus, we have
Z L
m1 ¼ m ðx Þð1 x Þdx (11.2.28a)
0
Z L
m2 ¼ m ðx Þxdx (11.2.28b)
0
Therefore, the kinetic energy of the truss element is given by the expression
1 1
T ¼ m1 u_ 1 2 + u_ 2 2 + m2 u_ 3 2 + u_ 4 2 (11.2.29)
2 2
@CivilMethod
Fig. 11.2.3 Lumped mass assumption.
The inertial forces result from Eq. (11.2.9) for i ¼ 1,2, 3, 4. Hence, after
performing the differentiations, we obtain

d ∂T ∂T
fI 1 ¼ ¼ m11 u€1 (11.2.30a)
dt ∂u_ 1 ∂u1

d ∂T ∂T
fI 2 ¼ ¼ m22 u€2 (11.2.30b)
dt ∂u_ 2 ∂u2

d ∂T ∂T
fI 3 ¼ ¼ m33 u€3 (11.2.30c)
dt ∂u_ 3 ∂u3

d ∂T ∂T
fI 4 ¼ ¼ m44 u€4 (11.2.30d)
dt ∂u_ 4 ∂u4
where
m11 ¼ m22 ¼ m1 , m33 ¼ m44 ¼ m2 (11.2.31)
8 9 2 38 9
fI 1 > u€
> > 1>
>
m11 0 0 0
>
> > >
> >
>
>
< fI 2 = 6 7>> >
60 m22 0 0 7 u€2 =
<
6
¼6 7 (11.2.32)
>
>
> fI 3 >
>
> 40 0 m33 0 7 5>>
> u€3 >
>
>
> ;
: > : >
> ;
fI 4 0 0 0 m44 u€4
Therefore, the element mass matrix me is defined as
2 3
m11 0 0 0
60 0 7
6 m22 0 7
me ¼ 6 7 (11.2.33)
40 0 m33 0 5
0 0 0 m44
We observe that the lumped mass assumption results in a diagonal mass matrix.
For an element with constant mass, m ðx Þ ¼ m Eqs. (11.2.28a),
(11.2.28b) give
e Le =2
m11 ¼ m22 ¼ m33 ¼ m44 ¼ m (11.2.34)
@CivilMethod
and the mass matrix, Eq. (11.2.33), becomes

2 3
1 0 0 0
1 e6 0 1 0 07
m ¼ m 6
e 7 (11.2.35)
2 40 0 1 05
0 0 0 1
e Le is the total mass of the truss element e.
where m e ¼ m
(iii) Nodal damping forces and damping matrix of the truss element
We consider only internal damping. It is therefore due to the deformation of the
truss element; it resists it and depends on the velocity of the strain. If we denote
with sD the axial stress due to damping, we can write
∂ex
sD ¼ cs (11.2.36)
∂t
where cs is the coefficient of the internal damping and ex the axial strain.
The virtual work of the internal damping force in the truss element is
Z
dWnc ¼ sD dex dV
in
(11.2.37)
V
Taking into account that ex ¼ ∂u=∂x, the previous relation becomes

Z L
∂2 u ðx, t Þ ∂u ðx, t Þ
dWncin
¼ c s Að x Þ d dx (11.2.38)
0 ∂x∂t ∂x
Z L
0 0
0 0

dWncin
¼ cs Aðx Þ u_ 1 1 ðx Þ + u_ 3 3 ðx Þd u1 1 ðx Þ + u 3 3 ðx Þ dx
0
¼ ðc11 u_ 1 + c13 u_ 3 Þdu1 ðc31 u_ 1 + c33 u_ 3 Þdu3
(11.2.39)
where
Z L
cij ¼ cs Aðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 3 (11.2.40)
0
The damping forces are obtained as generalized forces from Eqs. (11.2.8),
(1.8.8) for i ¼ 1, 2, 3, 4. Thus we have
fD1 ¼ Q1 ¼ c11 u_ 1 + c13 u_ 3 (11.2.41a)
fD2 ¼ Q2 ¼ 0 (11.2.41b)
fD3 ¼ Q3 ¼ c31 u_ 1 + c33 u_ 3 (11.2.41c)
fD4 ¼ Q4 ¼ 0 (11.2.41d)
@CivilMethod
The minus sign in the above relations is due to the fact that they are shifted
to the left side of Eq. (11.2.8) in order that their positive direction is the same
with that of fIi and fSi .
8 9 2 38 9
> fD1 > c11 0 c13 0 > u_ 1 >
>
> > >
> > >
> >
>
< fD2 = 6 7> >
6 0 0 0 0 7< u_ 2 =
¼66 7 (11.2.42)
> fD3 > 7 u_ >
>
> >
> 4 c31 0 c33 0 5> >
> 3>>
> ;
: > > ;
: >
fD4 0 0 0 0 u_ 4
or
f eD ¼ ce u_ e (11.2.43)
where f eD , u_ e are the vectors of the nodal damping forces and nodal velocities,
respectively, and ce is the damping matrix of the e truss element. Therefore, the
element damping matrix ce is defined as
2 3
c11 0 c13 0
60 0 0 07
6 7
ce ¼ 6 7 (11.2.44)
4 c31 0 c33 0 5
0 0 0 0
Obviously, we deduce from Eq. (11.2.40) that cij ¼ cji . Hence, the damping
For an element with a constant cross-section, Aðx Þ ¼ Ae , Eq. (11.2.40) are
integrated analytically and Eq. (11.2.44) becomes
2 3
1 0 1 0
6 7
cs A e 6 0 0 0 0 7
ce ¼ e 6 7 (11.2.45)
L 6 4 1 0 1 0 5
7
0 0 0 0
The evaluation of the matrix ce requires knowledge of the coefficient cs , the
determination of which is practically not feasible. The damping matrix in engi-
neering praxis, as we will see in Section 12.11.3, is constructed by assuming
known values of the damping ratio for each vibration mode, which are estimated
by experiments on similar structures. Nevertheless, we presented this approach
for the completeness of FEM, which is structured through the element approach.
(iv) Equivalent nodal loads of the truss element
The truss element can also be subjected to axial external loading, which may
consist of a distributed load pðx, t Þ and a finite number of concentrated loads
Pk ðt Þ, k ¼ 1, 2, …, K , acting at the points xk of the element, Fig. 11.2.4.
@CivilMethod
Fig. 11.2.4 Loading and equivalent nodal loads of a plane truss element.
Obviously, equivalent nodal loads will act only in directions of u1 and u3 .

The loads pðx, t Þ and Pk ðt Þ are considered as nonconservative, thus the equiv-
alent loads can be established as generalized forces in the directions of u1 and
u3 . The virtual work due to axial displacement is
Z L X
K
dWnc ¼
p
pðx, t Þdu ðx, t Þdx + Pk du ðxk , t Þ
0 k¼1
Z L X
K
¼ pðx, t Þ½du1 1 ðx Þ + du3 3 ðx Þdx + Pk ðt Þ½du1 1 ðxk Þ + du3 3 ðxk Þ
0 k¼1
¼ p1 ðt Þdu1 + p3 ðt Þdu3
(11.2.46)
where it was set
Z L X
K
pi ðt Þ ¼ pðx, t Þ i ðx Þdx + Pk ðt Þ i ðxk Þ, i ¼ 1, 3 (11.2.47)
0 k¼1
The nodal load forces are obtained from Eq. (11.2.8) for i ¼ 1, 2, 3, 4, that is,
Q1 ¼ p1 , Q2 ¼ 0, Q3 ¼ p3 , Q4 ¼ 0. Hence, the vector of the equivalent nodal load
forces for the e element is
p e ð t Þ ¼ f p 1 ð t Þ 0 p 3 ð t Þ 0g T (11.2.48)
Example 11.2.1 Compute the stiffness, mass, and damping matrices as well as
the vector of the nodal load forces of an element with variable cross-section and
compare them to those of the element with a constant cross-section equal to the
mean value of the variable cross-section.
Data: Aðx Þ ¼ A0 1 + x x2 ,
x ¼ x=L, pðx, t Þ ¼ p0 ðt Þ 1 + x x . Modulus of elasticity E, material density
2
r, and damping coefficient cs .

Solution
The mean cross-sectional area is
Z L Z L

¼1
A Aðx Þdx ¼ A0 1 + x x 2 dx ¼
7A0
(1)
L 0 0 6
@CivilMethod
Moreover, from Eq. (11.2.6a), we obtain

0 1 0 1
1 ðx Þ ¼ , 3 ðx Þ ¼ (2)
L L
Stiffness matrix. The elements of the stiffness matrix are computed from
Eq. (11.2.15), which gives
Z L
kij ¼ EAðx Þ 0i ðx Þ 0j ðx Þdx
0 (3)
EA
¼
L
Obviously, the stiffness matrix of the element with a variable cross-section
is the same as that of an element with a constant cross-section equal to the mean
cross-section. Hence, we may write
2 3
1 0 1 0
EA 6 0 0 0 07
ke ¼ 6 7 (4)
L 4 1 0 1 0 5
0 0 0 0
Mass matrix. The elements of the consistent mass matrix are computed
from Eq. (11.2.23), which gives
2 3
23 0 12 0
me 6
60 23 0 12 7
7
m ¼
e
(5)
70 4 12 0 23 0 5
0 12 0 23
where m e ¼ 7rA0 L=6 is the total mass of the element. Comparison with the
mass matrix of the element of constant cross-section, Eq. (11.2.27), shows that
the deviation is negligible, ð23=70Þ=ð2=6Þ 0:99, ð12=70Þ=ð1=6Þ 1:03.
The elements of the mass matrix for lumped mass assumption are computed
from Eq. (11.2.28), which give
Z 1
7
m11 ¼ m22 ¼ rA0 L 1 + x x 2 xdx ¼ rLA0 (6)
0 12
Z 1
7
m33 ¼ m44 ¼ rA0 L 1 + x x 2 ð1 x Þdx ¼ rLA0 (7)
0 12
2 3
1 0 0 0
1 60 1 0 07
me ¼ m e 6 7 (8)
2 40 0 1 05
0 0 0 1
where m e ¼ 7rA0 L=6 is the total mass of the element.
@CivilMethod
Damping matrix. The elements of the damping matrix are computed from
the Eq. (11.2.40), which gives
2 3
1 0 1 0
6 0 0 0 07
cs A
ce ¼ 6 7 (9)
L 4 1 0 1 0 5
0 0 0 0
Nodal Load forces. They are computed from Eq. (11.2.47), which gives
pe ðt Þ
pe ðt Þ ¼ f 1 0 1 0 gT (10)
2
where pe ðt Þ ¼ 7p0 ðt ÞL=6 is the total load of the element.
11.2.1.2 The method of virtual work

The method of virtual work is the usual method employed in FEM to derive the
pertinent nodal forces and matrices of the element. This principle states that if a
deformable system being in equilibrium under a set of forces is given virtual
displacements that are consistent with the constraints on the system, the virtual
work dWex done by the external (applied) forces riding the virtual displace-
ments equals the virtual work done by the internal forces, dWin . We illustrate
this method by deriving the equivalent elastic nodal and inertial forces as well as
the stiffness and mass matrices of the truss element. For the other element
equivalent nodal forces matrices, we refer to the method of the Lagrange equa-
tions, where the virtual work of the damping and external forces was evaluated,
cases (iii) and (iv).
(i) Nodal elastic forces and stiffness matrix of the truss element
The virtual work due to the axial elastic deformation of the element when it is
given a virtual displacement du is equal to the virtual strain energy
Z Z L
dWin ¼ sx dex dx ¼ Aðx ÞEex dex dx (11.2.49)
V 0
where V is the volume of the element.

Taking into account Eq. (11.2.1), we obtain

∂u u
ex ¼ ¼ u1 01 + u3 03 ¼ ½ 01 03 1 ¼ N0 u (11.2.50)
∂x u3

∂u du1
dex ¼ d ¼ du1 01 + du3 03 ¼ ½ 01 03 ¼ duT N0
T
(11.2.51)
∂x du3
where

0 0 0 u1 du1
N ¼½ 1 3 , u¼ , du ¼ (11.2.52)
u3 du3
@CivilMethod

Z L
0T 0
dWin ¼ du T
EAðx ÞN N dx u (11.2.53)
0
Besides, the virtual work of the equivalent elastic nodal forces

f S ¼ ½ fS1 fS3 T is
dWex ¼ duT f (11.2.54)
Inasmuch as dWin ¼ dWex we obtain

fS1 k k u1
¼ 11 13 (11.2.55)
fS3 k31 k33 u3
where
Z L
kij ¼ EAðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 3 (11.2.56)
0
Further, taking into account that fS2 ¼ fS4 ¼ 0, we can write Eq. (11.2.55)
8 9 2 38 9
> fS1 > k11 0 k13 0 > u1 >
> >
> 6 > >
> =
< 7>< > =
fS2 6 0 0 0 0 7 u2
¼6 7 (11.2.57)
>
>
> fS3 >
>
>
4 k31 0 k33 0 5> >
> u3 >
>
>
: ; : ;
fS4 0 0 0 0 u4
which is identical to Eq. (11.2.16).
The internal inertial force of mass element m ðx Þdx is
fI ðx, t Þ ¼ m ðx Þ½u€ðx, t Þ + v€ððx, t Þdx (11.2.58)
and the virtual work produced when the end displacements are given a virtual
displacement ðdu1 , du2 , du3 , du4 Þ is
Z L
dWin ¼ m ðx Þ½u€ðx, t Þdu + v€ðx, t Þdv dx (11.2.59)
0
which by virtue of Eqs. (11.2.1), (11.2.7) is written as

Z L
dWin ¼ duT m ðx ÞNT Ndx u € (11.2.60)
0
where
1
N¼½ 1 2 3 4¼ ½ ð1 xÞ ð1 x Þ x x (11.2.61)
L
The integral in Eq. (11.2.60) gives the mass matrix of the truss element
@CivilMethod
2 3
m11 0 m13 0
Z 6 7
L 60 m22 0 m24 7
me ¼ m ðx ÞNT Ndx ¼ 6
6m 0
7 (11.2.62)
0 4 31 m33 0 7 5
0 m42 0 m44
where
Z L
mij ¼ mi + 1, j + 1 ¼ m ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 1, 3 (11.2.63)
0
Consequently, Eq. (11.2.60) becomes

dWin ¼ duT m€
u (11.2.64)
Besides, the virtual work of the equivalent nodal inertial forces is
dWex ¼ duT f I (11.2.65)
Inasmuch as dWin ¼ dWex we obtain
8 9 2 38 9
fI 1 > u€
> > 1>
>
m11 0 m13 0
> > >
> >
> > 6
= > >
7> >
< fI 2 60 m22 0 m24 7 u€2 =
<
¼6
6
7 (11.2.66)
>
>
> fI 3 >
>
> 4 m31 0 m33 0 7 5>
>
> u€3 >
>
>
: >
> ; : >
> ;
fI 4 0 m42 0 m44 u€4
which is identical to (11.2.32).
From the previous discussion, it becomes clear that the method of the
Lagrange equations is more convenient to derive the equivalent element
quantities (nodal forces and matrices), provided that the reader is familiar with
this method.
11.2.2 Transformation of the nodal coordinates of the truss element

In order to formulate the equilibrium equations at the nodes of the truss, all ele-
ment nodal quantities, forces, and displacements should be referred to a single
system of axes. The key to achieving this is the transformation matrix of the
components of a vector with respect to a rotated system of axes. Thus, we con-
sider two systems of axes in the plane, Ox y and Oxy. The second system results
by rotating the first through an angle f about the z axis (Fig. 11.2.5) in the pos-
itive sense, that is, counterclockwise. If ða1 , a2 Þ and ða1 , a2 Þ denote the com-
ponents of a vector a with respect to these systems of axes, the following
transformation relation is valid
( ) " #( )
a1 cos f sin f a1
¼ (11.2.67)
a2 sin f cos f a2
@CivilMethod
Fig. 11.2.5 Transformation of the coordinates of a vector.
The previous relation is readily obtained by projecting the broken line OAP
successively onto the axes Ox and Oy and taking into account that OA ¼ a1 ,
AP ¼ a2 , OC ¼ a1 , and CP ¼ a2 (see Fig. 11.2.5).
a ¼L
a (11.2.68)
where

a1 a1 cos f sin f
a¼ , a ¼ , L¼ (11.2.69)
a2 a2 sin f cos f
The matrix L is referred to as the transformation matrix of the components
of a plane vector.
It can be readily shown that
L1 ¼LT (11.2.70)
that is, the matrix L is orthonormal. Thus Eq. (11.2.68) is readily inverted to
a ¼LT a (11.2.71)
!
We now consider the element e, whose longitudinal axis j k defines the
right-handed system of axes jxy. Let O xy be another arbitrarily chosen right-
handed system of axes and let f be the angle between x and x axes, that is,
the angle that the axis x sweeps when it is rotated counterclockwise to coincide
with the x axis, Fig. 11.2.6. The axes x, y are called the global axes while the
axes x, y defined by the truss element are called the local axes. Any quantity,
force, or displacement referred to the global axes will be henceforth designated
by an overbar.
Fig. 11.2.6 Nodal displacements of the truss element in global and local axes.
@CivilMethod
Empoying Eq. (11.2.67) for the nodal displacements, we obtain

u1 cos f sin f u1
¼ (11.2.72a)
u2 sin f cos f u2

u3 cos f sin f u3
¼ (11.2.72b)
u4 sin f cos f u4
The previous two relations can be combined as
8 9 2 38 9
>
> u1 >
< = 6> cos f sin f 0 0 > u1 >
> >
u2 sin f cos f 0 0 7< u2 =
¼ 6 7 (11.2.73)
> u > 4 0 0 cos f sin f 5> u >
: 3>
> ; : 3>
> ;
u4 0 0 sin f cos f u4
or
ue ¼Re ue (11.2.74a)
and after inverting
ue ¼ðRe ÞT ue (11.2.74b)

where
2 3
cos fe sin fe 0 0
6 sin fe cos f 0 0 7
Re ¼ 6
4 0
7
e5 (11.2.75)
0 cos fe sin f
0 0 sin fe cos fe
represents the transformation matrix of the plane truss element e.
The vectors of the equivalent nodal forces are defined with respect to
the global axes as shown in Fig. 11.2.7. It is obvious that they also obey the
transformation law (11.2.74b). Thus, we have
f e ¼ðRe ÞT f e (11.2.76)
S S
f e ¼ðRe ÞT f e (11.2.77)
I I
f e ¼ðRe ÞT f e (11.2.78)
D D
pe ðt Þ ¼ðR Þ p ðt Þ
e T e
(11.2.79)
@CivilMethod
Fig. 11.2.7 Nodal forces of the truss element in global and local axes.
Eq. (11.2.76) is further written as

f e ¼ðRe ÞT f e
S S
¼ ðRe ÞT ke ue (11.2.80)
e T e
¼ ðR Þ k Re ue
or
f e ¼ke ue (11.2.81)
S
where
k ¼ ðRe ÞT ke Re :
e
(11.2.82)
is the element stiffness matrix with respect to the global axes.
f e ¼m
e ue (11.2.83)
I
f D ¼
ce ue (11.2.84)
where
e ¼ ðRe ÞT me Re
m (11.2.85)

ce ¼ ðRe ÞT ce Re (11.2.86)
11.2.3 Equation of motion of the plane truss

The formulation of the equations of motion in FEM can be viewed as a process
of assembling of the element stiffness, mass, and damping matrices as well as of
the external nodal forces into the corresponding global matrices and global load
nodal vector of the structure. This process involves first the formulation of the
global displacement vector u. This vector contains all nodal displacements of
the truss referred to the global axes including those of the supports. If the truss
has n nodes, then the vector u will have N ¼ 2n elements. The displacements of
the i node will take the places 2i 1 and 2i in the global vector. Thus, if j, k
denote the numbers of origin and the end of the e element, the displacements
ue will take the places u2j1 , u2j , u2k1 , u2k in u. This can be realized compu-
tationally by means of the linear transformation
ue ¼ae u (11.2.87)
where ae is an 4 N Boolean matrix, that is, a matrix consisting of zeros and ones.
ð11:2:88Þ
@CivilMethod
The above transformation is inverted asb
u ¼ðae ÞT ue (11.2.89)

If ^f e
denotes the enlarged vector with dimension N , then the assemblage
S
of the elements of f S can be realized according to Eq. (11.2.89)
e
^f e ¼ ðae ÞT f e (11.2.90)
S S
The next step is to write the relation f S

¼k u as a relation between ^f S and u.
e e
e e
Starting with Eq. (11.2.90) and using the transformations (11.2.87), we obtain
^f e ¼ ðae ÞT f e
S S
e T e e
¼ ða Þ k u
(11.2.91)
¼ ðae ÞT k ae u
e
¼K^ e u
where
^ e ¼ ðae ÞT ke ae
K (11.2.92)
denotes the enlarged stiffness matrix with dimensions N N . Obviously, the
places of the elements of k in K
^ e are:
e
k^2j1,2j1 ¼ k11 , k^2j1,2j ¼ k12 , k^2j1, 2k1 ¼ k13 , k^2j1, 2k ¼ k14
k^2j, 2j1 ¼ k21 , k^2j,2j ¼ k22 , k^2j,2k1 ¼ k23 , k^2j, 2k ¼ k24
k^2k1,2j1 ¼ k31 , k^2k1,2j ¼ k32 , k^2k1, 2k1 ¼ k33 , k^2k1, 2k ¼ k34
k^2k, 2j1 ¼ k41 , k^2k,2j ¼ k42 , k^2k,2k1 ¼ k43 , k^2k, 2k ¼ k44

Following the same process, we obtain the relations
^f e ¼M
^ e u€ (11.2.93)
I
^f e ^ e u_
¼C (11.2.94)
D
b. The matrix a is not square, therefore it cannot be inverted. The inverse transformation (11.2.89)
can be proved either by inspection or by considering the equality of the work produced by the nodal
forces in the two systems of diplacements ue and u. Indeed, according to Eq. (11.2.87) we have
f e ¼ae^f e
where f and ^f e are element force vectors with a dimension 4 and N , respectively.
e
Equating the works gives

T
^f e u ¼ f e ue
T
¼ ae ^f e ue
T
¼ ^f e ðae ÞT ue
@CivilMethod
which results in Eq. (11.2.89).
ê ðt Þ ¼ ðae ÞT pe ðt Þ
p (11.2.95)
where ^f e , ^f e ,
and p ê ðt Þ denote the enlarged nodal inertial, damping matrices,
I D
and load vector, respectively, while
^ e ¼ ðae ÞT m
M e ae (11.2.96)
ê
C ¼ ða Þ e T e e
ca (11.2.97)
ðt Þ with a dimen-
Finally, all nodal loads are assembled into a load vector P
sion N , corresponding to u. Hence, the loads applied to node i will take the
places P2i1 and P2i in P
ðt Þ.
The equations of motion can be derived using the method of the Lagrange
equations with generalized coordinates the components of the vector u. Thus,
we have:
Elastic energy. The elastic energy of the e element is
1
Ue ¼ uT ^f S
e
2 (11.2.98)
1 ê
¼ uT K u
2
and the total elastic energy is
1X Ne
^ e u
U¼ uT K
2 e¼1
1
¼ uT K u (11.2.99)
2
1X N X N
ij ui uj
¼ K
2 i¼1 j¼1
where
X
Ne
¼
K ê
K (11.2.100)
e¼1
represents the stiffness matrix of the structure with Ne denoting the number of
all elements.
Kinetic energy. Similarly, the kinetic energy of the structure results as the
sum of the kinetic energies of all elements
1X Ne
T ê
T¼ u_ M u_
2 e¼1
1 T
¼ u_ M u_
2
1X N X N
ij u_ i u_ j
¼ M (11.2.101)
2 i¼1 j¼1
@CivilMethod
where
X
Ne
¼
M ê
M (11.2.102)
e¼1
represents the mass matrix of the structure.

Virtual work of the damping forces. For the e element, we have
uT ^f D
e
dWDe ¼ d
(11.2.103)
¼ d ^ e u_
uT C
and the total virtual work of the damping forces is
X
Ne
dWD ¼ ^ e u_
uT C
d
e¼1
¼ d u_
uT C
X
N X
N
¼ ij u_ j d
C ui (11.2.104)
i¼1 j¼1
where
X
Ne
¼
C ê
C (11.2.105)
e¼1
represents the damping matrix of the structure.

Virtual work of the external loads. It results as the sum of all external loads,
namely of the equivalent element nodal loads p ðt Þ applied
ê ðt Þ and the loads P
directly to the nodes of the truss. Therefore, we have
X
Ne
dWp ¼ d
uT Pðt Þ + ^ ðt Þe
uT p
d
e¼1
!
X
Ne (11.2.106)
¼ d
u T ðt Þ +
P ^ e ðt Þ
p
e¼1
u pðt Þ
¼ d T
where
X
Ne
ðt Þ +
pðt Þ ¼ P ^ e ðt Þ
p (11.2.107)
e¼1
represents the vector of all external loads.

Applying Eq. (1.8.11), we obtain
∂T
¼0 (11.2.108a)
∂
ui
@CivilMethod
X N
d ∂T
¼ Mij u€j (11.2.108b)
dt ∂u_ i j¼1
∂U X N
¼ Kij uj (11.2.108c)
∂
u i j¼1
ij u_ j + pi ðt Þ
Qi ¼ C (11.2.108d)
and the equation of motion becomes
X
N X
N X
N
ij u€j +
M ij u_ j +
C ij uj ¼ pi ðt Þ
K (11.2.109)
j¼1 j¼1 j¼1
or
M u_ + K
u€ + C u ¼
pðt Þ (11.2.110)
Obviously, Eq. (11.2.110) has the same form as that obtained using the
methods discussed in Chapter 10.
11.2.4 Steps to formulate the equations of motion for a plane truss

by the finite element method
The formulation of the equations of motion for a structure (here the plane truss)
by the FEM based on the analytical procedure described in the previous section
can be summarized as a sequence of the following steps:
Fig. 11.2.8 Plane truss (n ¼ 5, N ¼ 2n ¼ 10, Ne ¼ 8).
(i) Idealization of the truss by a set of finite elements interconnected at its

nodes. This step includes (Fig. 11.2.8):
a. Selection of the system of global axes x y.
b. Numbering the nodes of the truss ð1, 2, …, n Þ and determining their
coordinates with respect to the global axes.
c. Determination of the vector of the N ¼ 2n degrees of freedom u.
d. Formulation of the vector P ðt Þ of the external loads directly applied to
the nodes of the truss.
@CivilMethod
e. Numbering of the elements ð1, 2, …, Ne Þ and determination of their

orientation by selecting the positive direction of the local x axis.
(ii) For each element ðe ¼ 1, 2, …, Ne Þ
a. Compute the matrices:
– ke using Eq. (11.2.18)
– me using Eq. (11.2.26) or (11.2.33) depending on the mass
assumption
– ce using Eq. (11.2.44), if the coefficient cs is given
– pe ðt Þ using Eq. (11.2.48)
– Re using Eq. (11.2.75)
– ae using Eq. (11.2.88)
b. Compute the matrices and load vector:
k ¼ ðRe ÞT ke Re
e
e ¼ ð Re Þ T m e Re
m
ce ¼ ðRe ÞT ce Re
pe ðt Þ ¼ ðRe ÞT pe ðt Þ
c. Using Eqs. (11.2.92), (11.2.96), (11.2.97), (11.2.95), formulate the

enlarged matrices and load vector
– K ^ e, C
^ e, M ^ e , and p
^ e ðt Þ
(iii) Compute the matrices and the load vector of the structure
X
Ne X
Ne X
Ne X
Ne
¼
K K ¼
^ e, M M ¼
^ e, C ^ e , pðt Þ ¼ P
C ðt Þ + ^ e ðt Þ
p
e¼1 e¼1 e¼1 e¼1
11.2.5 Modification of the equations of motion due to the supports

of the structure
Eq. (11.2.110) describes the motion of the free structure. However, the truss
may be supported at some of its nodes. Supporting the structure means con-
straining certain degrees of freedom or, in other words, specifying certain nodal
displacements. The minimum number of degrees of freedom that should be con-
strained is three in order to prevent the movement of the structure as a rigid
body. These degrees of freedom are selected so that an infinitesimal movement
is excluded.
Let s be the number of the specified degrees of freedom, hence s elements of
the vector u are known while f ¼ N s are unknown. We reorder the elements
of u to separate the known displacements from the unknown ones and we denote
by us the vector of the known (support) displacements and by uf the vector of
the unknown (free) displacements. The elements of us are equal to zero for
@CivilMethod
nonyielding supports. In general, the elements of us may depend on time,

us ¼ us ðt Þ, as in the case of support excitation. The nodal quantities ui and
pi ðt Þ are referred to as dual quantities. For a well-posed problem, the dual quan-
tities cannot be specified simultaneously. Either ui is specified and pi ðt Þ is
unknown or pi ðt Þ is specified and ui is unknown.
Consequently, the vector ps ðt Þ, which corresponds to us , contains the
unknown nodal forces, namely the support reactions plus the element nodal
contributions, while the vector pf ðt Þ, which corresponds to uf , contains the
known nodal forces.
The N equations of motion (11.2.110) will be solved to determine the
unknown vectors uf and ps ðt Þ. The solution procedure requires first the rearran-
gement of the equations of motion, which is done as follows.
We rearrange the elements of u so that the known elements occupy the last
positions. This rearrangement can be done computationally by the transformation
u ¼Ve
u (11.2.111)
where e u is the rearranged vector and V is a square Boolean matrix. As an
example, we consider the truss of Fig. 11.2.8. The specified displacements
are u1 ¼ u2 ¼ u4 ¼ 0. Therefore the rearranged displacement vector is
e e1 u
uT ¼ f u e2 ue3 ue4 ue5 ue6 ue7 ue8 ue9 ue10 g
(11.2.112)
¼ f u3 u5 u6 u7 u8 u9 u10 u1 u2 u4 g
and the matrix V is defined from the relation
8 9 2 38 9
>
> u1 >> 0 0 0 0 0 0 0 1 0 0 > > u3 >>
> >
> > >
>
>
> u2 >> 6 1 07 > u5 >
> >
>
>
> >
> 60 0 0 0 0 0 0 0 7>>
> >
>
>
> >
> 6 7> >
>
>
> u >
3 > 61 0 0 0 0 0 0 0 0 0 7> >
> u 6 >
>
>
>
> > 6
>
> 60 0 0 0 0 0 0 0
7>
7 >
>
>
>
>
>
> u4 >> 6 0 1 7> > u 7 >
>
>
> >
> > >
>

0 0 7< u8 =
¼6 7
5
60 0 1 0 0 0 0 0 or u ¼ Ve
u (11.2.113)
>
>
> u6 >>
> 6 0 07 7>>
> u9 >>
>
> >
> 6 7> >
>
>
> >
7 > 60 0 0 1 0 0 0 0 0 07 > u10 >
> >
>
>
>
u >
> 6 7>>
> >
>
>
> u8 >> 60 0 0 0 1 0 0 0 7
0 0 7> > >
>
>
> >
> 6 >
> u 1 >
>
>
> >
> 6 7>> >
>
> > > >
>
>
>
u 9 > 40 0 0 0 0 1 0 0
>
>
0 0 >5
>
>
u 2 >
>
>
: ; : ;
u10 0 0 0 0 0 0 1 0 0 0 u4
We can readily show that the matrices M, K
C, and the vector pðt Þ
are rearanged by virtue of matrix V as
e ¼ VT MV
M (11.2.114a)
C
e ¼ VT CV (11.2.114b)
K
e ¼VT KV (11.2.114c)
pðtÞ ¼VT pðtÞ
e (11.2.114d)
@CivilMethod
Thus, the rearranged equations of motion become
e€ + C
eu
M eeu_ + Ke
e u ¼e
pðt Þ
which are partitioned after separating the known from the unknown
displacements
ð11:2:115Þ
or
" #
" #
" #

e ff M
M e fs e€f
u e ff C
C e fs e_ f
u e ff K
K e fs e
uf e
p f ðt Þ
+ + ¼
e e
Msf Mss u€s
e e e
Csf Css u_ s
e e e
Ksf Kss e
us e
p s ðt Þ
(11.2.116)
Performing the matrix multiplications in the previous equation yields
M u€f + C
e ff e u_ f + K
e ff e e ff e pf∗ ðt Þ
uf ¼ e (11.2.117)
M u€f + M
e sf e u€s + C
e ss e u_ f + C
e sf e u_ s + K
e ss e e sf e e ss e
uf + K us ¼ e
p s ðt Þ (11.2.118)
where it was set
pf∗ ðt Þ ¼ e
e pf ðt Þ M u€s C
e fs e u_ s K
e fs e e fs e
us (11.2.119)
Eq. (11.2.117) is the equation of motion of the supported truss. Its solution
gives the vector of the unknown displacements e uf , which is then inserted in
Eq. (11.2.118) to obtain the vector of nodal forces e
ps ðt Þ. Note that this vector con-
tains the equivalent element nodal loads contributing to the supported node. Obvi-
ously, when the support displacements do not depend on time, Eq. (11.2.119)
becomes
pf∗ ðt Þ ¼ e
e e fs e
pf ðt Þ K us (11.2.120)
@CivilMethod
and Eq. (11.2.118) is written
M u€f + C
e sf e u_ f + K
e sf e e sf e e ss e
uf + K us ¼ e
ps ðt Þ (11.2.121)
Finally, when the supports do not yield, the foregoing equations are further
simplified and become
pf∗ ðt Þ ¼ e
e p f ðt Þ (11.2.122)
M u€f + C
e sf e u_ f + K
e sf e e sf e
uf ¼ e
p s ðt Þ (11.2.123)
Example 11.2.2 Formulate the equations of motion of the truss in Fig. E11.1a
when (i) It is subjected to the external loading. (ii) The support 1 is
subjected to the vertical motion ug ðt Þ. Assumed data: Nodal coordinates:
1ð0, 0Þ, 2ð3, 3:5Þ, 3ð6, 0Þ; cross-sectional areas of the bars: A1 ¼ 1:5A,
A2 ¼ A3 ¼ A; distributed load on the element 1: pðx, t Þ ¼ P ðt Þ=L1 ; damping
coefficient: cs ¼ 0, modulus of elasticity E; material density: r, lumped mass
assumption.
(a)
Fig. E11.1a Plane truss in Example 11.2.2.
Solution
(i) Equation of motion under the external loading.
(b)
Fig. E11.1b Nodal displacements of the plane truss in Example 11.2.2.
@CivilMethod
The truss has n ¼ 3 nodes, hence the free structure has N ¼ 2n ¼ 6 degrees of
freedom (see Fig. E11.2b). The numbering of the elements and the positive
direction of the local axes are shown in Fig. E11.1b.
TABLE E11.1 Geometrical data of the elements in Example 11.2.2.
e xj xk D
x yj yk D
y Le cos fe sinfe
1 0 6 6 0 0 0 6 1 0
2 0 3 3 0 3.5 3.5 4.61 0.651 þ0.759
3 3 6 3 3.5 0 3.5 4.61 0.651 0.759
1. Computation of ke , me , ce , pe ðt Þ, Re , e ¼ 1, 2, 3
Matrices ke . The elements have constant cross-section. Thus Eq. (11.2.19) gives
2 3 2 3
0:25 0 0:25 0 0:217 0 0:217 0
6 0 0 0 07 6 0 0 0 07
k1 ¼ EA6 7 2 6
4 0:25 0 0:25 0 5, k ¼ k ¼ EA4 0:217 0 0:217 0 5
3 7
0 0 0 0 0 0 0 0
Matrices me . The elements have constant cross section. Thus, we obtain
from Eq. (11.2.35)
2 3 2 3
4:5 0 0 0 2:305 0 0 0
6 0 4:5 0 0 7 60 2:305 0 0 7
m1 ¼ rA6 7
4 0 0 4:5 0 5, m ¼ m ¼ rA4 0
2 3 6 7
5
0 2:305 0
0 0 0 4:5 0 0 0 2:305
Matrices ce . It is given cs ¼ 0. Thus, Eq. (11.2.45) gives ce ¼ 0.

Vectors pe ðt Þ. They are obtained from Eq. (11.2.47), which gives
8 9 8 9
> 0:5P ðt Þ >
> > 0>
>
< = < >
> =
0 0
p1 ðt Þ ¼ , p2 ðt Þ ¼ p3 ðt Þ ¼
> 0:5P ðt Þ >
> > 0>
>
: ; : >
> ;
0 0
Matrices Re . The direction cosines cos fe and sin fe of the element axes are
calculated from the global coordinates of the element nodes and are given in
Table E11.1. Thus, we have
2 3 2 3
1 0 0 0 0:651 0:759 0 0
60 1 0 07 6 0:759 0:651 0 0 7
R1 ¼ 6 7
4 0 0 1 0 5, R ¼ 4 0
2 6 7
0 0:651 0:759 5
0 0 0 1 0 0 0:759 0:651
@CivilMethod
2 3
0:651 0:759 0 0
6 0:759 0:651 0 0 7
R3 ¼ 6
40
7
0 0:651 0:759 5
0 0 0:759 0:651
2. Computation of k , m
e
e , pe ðt Þ, e ¼ 1, 2, 3
Matrices k . Eq. (11.2.82) gives
e
2 3
0:25 0 0:25 0
6 0 0 0 07
k1 ¼ EA6
4 0:25 0 0:25
7,
05
0 0 0 0
2 3
0:092 0:107 0:092 0:107
6 0:107 0:125 0:107 0:125 7

k ¼ EA4
2 6 7
0:092 0:107 0:092 0:107 5
0:107 0:125 0:107 0:125
2 3
0:092 0:107 0:092 0:107
6 0:107 0:125 0:107 0:125 7
k3 ¼ EA64 0:092 0:107
7
0:092 0:107 5
0:107 0:125 0:107 0:125
e . Eq. (11.2.85) gives
Matrices m
2 3 2 3
4:5 0 0 0 2:305 0 0 0
6 0 4:5 0 0 7 60 2:305 0 0 7
m ¼ rA4
1 6 7 6
, m ¼ m ¼ rA4
2 3 7
0 0 4:5 0 5 0 0 2:305 0 5
0 0 0 4:5 0 0 0 2:305
Vectors pe ðt Þ. Eq. (11.2.79) gives

8 9 8 9
> 0:5P ðt Þ >
> > 0>
>
< = < >
> =
0 0
p ðt Þ ¼
1
, p ðt Þ ¼ p ðt Þ ¼
2 3
> 0:5P ðt Þ >
> > 0>
>
: ; : >
> ;
0 0
^ e, M
3. Computation K ^ e, p
ê ðt Þ, e ¼ 1, 2, 3
The element assembly matrices ae are
2 3 2 3 2 3
1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0
6 0 1 0 0 0 07 6 07 6 07
a1 ¼ 6 7, a2 ¼ 6 0 1 0 0 0 7, a3 ¼ 6 0 0 0 1 0 7
40 0 0 0 1 05 40 0 1 0 0 05 40 0 0 0 1 05
0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 1
@CivilMethod
They yield the enlarged element matrices:

^ e . Eq. (11.2.92) gives
Matrices K
2 3
0:25 0 0 0 0:25 0
6 0 0 0 0 0 07
6 7
6 07
K^ 1 ¼ EA6 0 0 0 0 0 7,
6 0 0 0 0 0 07
6 7
4 0:25 0 0 0 0:25 0 5
0 0 0 0 0 0
2 3
0:092 0:107 0:092 0:107 0 0
6 0:107 0:125 0:107 0:125 0 0 7
6 7
6 7
K^ 2 ¼ EA6 0:092 0:107 0:092 0:107 0 0 7
6 0:107 0:125 0:107 0:125 0 0 7
6 7
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
0 0 0 0 0 0
60 0 0 0 0 0 7
6 7
6 0 0 0:092 0:107 0:092 0:107 7
^
K ¼ EA6
3 6 7
0 0 0:107 0:125 0:107 0:125 7
6 7
4 0 0 0:092 0:107 0:092 0:107 5
0 0 0:107 0:125 0:107 0:125
^ e . Eq. (11.2.96) gives
Matrices M
2 3
4:5 0 0 0 0 0
6 0 4:5 0 0 0 0 7
6 7
6 7
^ 1 ¼ rA6 0 0 0 0 0 0 7
M 60 0 0 0 0 0 7
6 7
4 0 0 0 0 4:5 0 5
0 0 0 0 0 4:5
2 3
2:305 0 0 0 0 0
60 2:305 0 0 0 07
6 7
6 0 0 2:305 0 0 07
^
M ¼ rA6
2 6 7
60 0 0 2:305 0 0 77
40 0 0 0 0 05
0 0 0 0 0 0
2 3
0 0 0 0 0 0
60 0 0 0 0 0 7
6 7
6 7
M^ 3 ¼ rA6 0 0 2:305 0 0 0 7
60 0 0 2:305 0 0 7
6 7
40 0 0 0 2:305 0 5
0 0 0 0 0 2:305
@CivilMethod
ê ðt Þ. Eq. (11.2.95) gives

Vectors p
8 9 8 9
>
> 0:5P ðt Þ >
> >
> 0>>
> >
> > >
>
> 0 >
>
>
> 0>>
>
< = < >
> =
0 0
^ ðt Þ ¼
p 1
^ ðt Þ ¼ p
, p2
^ ðt Þ ¼
3
>
> 0 >
> >
> 0>>
> > > >
>
> 0:5P ðt Þ >
>
>
>
> 0>>
>
: ; : >
> ;
0 0
M
4. Computation of the total matrices K, and total load vector pðt Þ of
the truss
2 3
0:342 0:107 0:092 0:107 0:25 0
6 7
6 0:107 0:125 0:107 0:125 0 0 7
6 7
X e 6 7
3 6 0:092 0:107 0:184 0 0:092 0:107 7
¼
K ^ ¼ EA6
K 7
6 0:107 0:125 0 0:107 0:125 7
6 0:25 7
e¼1
6 7
6 0:25 0:092 0:107 0:342 0:107 7
4 0 5
0 0 0:107 0:125 0:107 0:125
2 3
6:805 0 0 0 0 0
6 7
60 6:805 0 0 0 0 7
6 7
X e
3 60 0 4:610 0 0 0 7
¼
M ^ ¼ rA6
M 6
7
7,
60 0 0 4:610 0 0 7
e¼1 6 7
6 7
40 0 0 0 6:805 0 5
0 0 0 0 0 6:805
8 9 8 9 8 9
>
> P1 > > >
> 0:5P ðt Þ >
> >
> P1 + 0:5P ðt Þ >
>
> > > >
> > >
>
>
> P2 > >
>
>
>
> 0 >
>
>
>
> P2 >
>
>
X3 < = < = < =
ðt Þ + P ð t Þ 0 P ð t Þ
pðt Þ ¼ P ^ e ðt Þ ¼
p + ¼
> 3P ðt Þ >
> > >
> 0 >
> > 3P ðt Þ >
>
> > > > > > >
> > >
> 0:5P ðt Þ > > >
e¼1
> 0
> > >
> >
> >
> 0:5P ðt Þ >
>
: ; : ; : ;
P6 0 P6
Note that P1 , P2 , P6 denote the unknown support reactions

e M
5. Computation of the modified matrices K, e and load vector e
pðt Þ
Due to the supports, the displacement vector should be modified as
e
uT ¼ f u e2 u
e1 u e3 u
e4 u
e5 u
e6 g ¼ f u3 u4 u5 u1 u2 u6 g
@CivilMethod
Therefore, the matrix V defined by Eq. (11.2.111) is

2 3
0 0 0 1 0 0
60 0 0 0 1 07
6 7
61 0 0 0 0 07
V¼6 60 1 0 0 0 07
7
6 7
40 0 1 0 0 05
0 0 0 0 0 1
Then, applying Eqs. (11.2.114a), (11.2.114c), (11.2.114d) gives
– – –
– – –
– – –
– – –
– –
– –
e ff , M
6. Formulation of the matrices M e fs , M
e sf , M
e ss , K
e ff , K
e fs , K
e sf , K
e ss and
pf ðt Þ, e
vectors e ps ðt Þ
These matrices are obtained after partitioning M, e K,
e e pðt Þ on the basis of the
previously indicated partition. Thus, we have
2 3 2 3
4:610 0 0 0 0 0
Me ff ¼ rA4 0 4:610 0 5, M e fs ¼ M
e sf ¼ rA4 0 0 05
0 0 6:805 0 0 0
2 3
6:805 0 0
Me ss ¼ rA4 0 6:805 0 5
0 0 6:805
@CivilMethod
2 3 2 3
0:184 0 0:092 0:092 0:107 0:107
e ff ¼ EA6
K 4 0 0:25
7 e
0:107 5, K
6 7
fs ¼ EA4 0:107 0:125 0:125 5
0:092 0:107 0:342 0:25 0 0:107
2 3 2 3
0:092 0:107 0:25 0:342 0:107 0
e sf ¼ EA4 0:107 0:125 0
K 5, K e ss ¼ EA4 0:107 0:125 0 5
0:107 0:125 0:107 0 0 0:125
8 9 8 9
< P ðt Þ = < P1 + 0:5P ðt Þ =
e
pf ðt Þ ¼ 3P ðt Þ , e
ps ðt Þ ¼ P2
: ; : ;
0:5P ðt Þ P6
7. Equation of motion of the truss

Because eus ¼ 0, it is also eu€s ¼ 0 and Eq. (11.2.119) gives epf∗ ðt Þ ¼ e
pf ðt Þ.
Hence, Eq. (11.2.117) results in the following equation of motion
2 38 € 9 2 38 9
4:610 0 0 >
> e1 >> 0:184 0 0:092 > u e
>

= > 1> >
6 7 6 7< =
rA6 7 u € 6 7 e
4 0 4:610 0 5> e 2 > + EA 4 0 0:250 0:107 u
5> 2 >
>
> >
> : >
> ;
: ;
0 0 6:805 e€3
u 0:092 0:107 0:342 e3
u
8 9
> P ðt Þ >
>
< >
=
¼ 3P ðt Þ
>
> >
>
: ;
0:5P ðt Þ
The solution of the equation of motion yields the vector euf and
Eq. (11.2.118) the nodal forces at the supports
8 9 2 38 9
<e ps1 = 0:125 0 e1 =
0:107 < u
ps2 ¼ EA4 0
e 0:092 0:092 5 ue2
: ; : ;
e
ps3 0:125 0:107 0 e3
u
Hence, the reactions at the supports are
8 9 8 9 8 9
< P1 = < e
ps1 = < 0:5P ðt Þ =
P ¼ ep 0
: 2 ; : s2 ; : ;
P6 e
ps3 0
(ii) Equation of motion due to the vertical motion ug ðt Þ of the support 1.
In this case it is
8 9 8 9
< P1 = <0 =
e
pf ðt Þ ¼ 0, e
ps ðt Þ ¼ P2 , e
us ¼ ug ðt Þ
: ; : ;
P6 0
@CivilMethod
and Eq. (11.2.119) gives
pf∗ ðt Þ ¼ M
e u€s C
e fs e u_ s K
e fs e e fs e
us
or 8 9
< 0:107 =
pf∗ ðt Þ ¼ EA 0:125 ug ðt Þ
e
: ;
0
Hence, the equation of motion becomes
2 38 € 9 2 38 u 9
4:610 0 0 >
= E 0:184 0 0:092 >< e1 >
=
40 4:610 0 5 u e€2 + 4 0 0:250 0:107 5 ue2
> > ; r 0:092 0:107 0:342 > : >
0 0 6:805 : u e€3 e3
u
;
8 9
0:107 =
E<
¼ 0:125 ug ðt Þ
r: ;
0
and Eq. (11.2.123) gives the reactions
8 9 2 38 9 8 9
< P1 = e1 =
0:092 0:107 0:250 < u <0 =
P2 ¼ EA4 0:107 0:125 0 5 u
e2 + rA 6:805 u€g
: ; : ; : ;
P6 0:107 0:125 0:107 e3
u 0
8 9
< 0:107 =
+ EA 0:125 ug
: ;
0
Remark: The displacement vector u ef ¼ f u
e1 u
e2 u e3 gT can be converted to
the original vector of the free nodal displacements, that is, u ¼ f u3 u4 u5 gT
by inverting the transformations that have created uef .
11.3 The finite element method for the plane frame

11.3.1 Properties of the plane frame element
We consider a plane frame element with variable cross-section Aðx Þ, cross-
sectional moment of inertia I ðx Þ, mass per unit length m ðx Þ, and modulus of
elasticity E. The element local axis in the undeformed state coincides with the
x axis. The frame element is a beam, therefore it is also called a beam element.
The ends j,k are the nodes of the element, thus any quantity (force, displace-
ment) associated with them is referred to as nodal quantity (i.e., nodal force,
nodal displacement). In the deformed state at time t, the ends of the element
are displaced to the points j 0 , k 0 and the element occupies the position j 0 k 0
(Fig. 11.3.1). The element, besides the axial deformation, undergoes a flexural
deformation due to the end transverse displacements and rotations. The element
has six degrees of freedom: the displacements u1 ðt Þ, u2 ðt Þ and the rotation
u3 ðt Þ ¼ q1 of node j as well as the displacements u3 ðt Þ, u4 ðt Þ and rotation
@CivilMethod
u6 ðt Þ ¼ q2 of node k. These six quantities represent the nodal coordinates of the

plane frame element. A typical point x of the element axis undergoes two dis-
placements: the axial displacement u ðx, t Þ and the transverse displacement
v ðx, t Þ. The rotation of the cross-section v 0 ðx, t Þ has a small influence on the
dynamic properties of the beam so it is ignored in the presentation that follows.
We consider the element as a generalized system and apply the Ritz method
presented in Section 10.5.1. Thus, we set the axial displacement in the form
u ðx, t Þ ¼ u1 1 ðx Þ + u 4 4 ðx Þ (11.3.1)
where 1 ðx Þ and 4 ðx Þ are shape functions expressing the axial deformation for
u1 ¼ 1, u4 ¼ 0 and u1 ¼ 0, u4 ¼ 1, respectively. These functions for an element
with constant cross-section were obtained in Section 11.2.1 as
1 ðx Þ ¼ 1 x (11.3.2a)
4 ðx Þ ¼ x (11.3.2b)
where x ¼ x=L.
Fig. 11.3.1 Degrees of freedom of a plane frame element.
The transverse displacement can be set in the form

v ðx, t Þ ¼ u2 2 ðx Þ + u3 3 ðx Þ + u5 5 ðx Þ + u6 6 ðx Þ (11.3.3)
where i ðx Þ are shape functions expressing the elastic curve of the element for
ui ¼ 1, ði ¼ 2, 3, 5, 6Þ and uj ¼ 0, j 6¼ i.
These functions can be determined from the solution of the beam equation
under bending. Without restricting the generality, we assume that the Bernoulli
assumption is valid for the bending of the element. Therefore, the transverse
deformation results only from bending. In this case, the elastic curve is obtained
by successive integration of the equation

d2 d 2 ðx Þ
EI ðx Þ ¼0 (11.3.4)
dx 2 dx 2
The solution of this equation, although it gives shape functions expressing
the exact static transverse deformation of the beam element, is not suitable for
the automation of the method because different shape functions have to be
determined for elements with a different law of variation of the moment of
@CivilMethod
inertia. As we pointed out in Section 11.2.1, this difficulty is surpassed if we

accept the same shape functions for all elements, regardless of the variation
law of the cross-section. Apparently, the shape functions resulting from
Eq. (11.3.4) for a constant cross-section are suitable. In this case, the solution
of Eq. (11.3.4) is given by the cubic polynomial
x3 x2
ðx Þ ¼ c 1 + c2 + c3 x + c4 (11.3.5)
6 2
and its first derivative by
0 x2
ðx Þ ¼ c1 + c2 x + c3 (11.3.6)
2
The arbitrary constants are evaluated from the boundary conditions.
Thus, the shape function
0 0
2 ðx Þ results for 2 ð0Þ ¼ 1, 2 ð0Þ ¼ 0, 2 ðLÞ ¼ 0, 2 ðLÞ ¼ 0
0 0
0 0
0 0
This procedure yields

2 ðx Þ ¼ 1 3x
2
+ 2x 3 (11.3.7a)

3 ðx Þ ¼ L x 2x + x
2 3
(11.3.7b)
5 ðx Þ ¼ 3x 2x3
2
(11.3.7c)

6 ðx Þ ¼ L x + x
2 3
(11.3.7d)
where x ¼ x=L.
Fig. 11.3.2 Shape functions of the transverse deformation of the beam element.
@CivilMethod
The cubic polynomials given by Eqs. (11.3.7a)–(11.3.7d) are known as

Hermitian polynomials. Their graphic representation is shown in Fig. 11.3.2.
Obviously, they satisfy the differential equation of the elastic curve of a beam
with a constant cross-section. However, as mentioned, they can be used as shape
functions of elements with a variable cross-section, provided that the elements
are small. The accuracy of the solution is improved by subdividing the element
into smaller elements. This technique increases the number of elements.
The equivalent nodal forces of the element will be obtained using the
methods employed for the truss element. It should be recalled here that, unlike
the classic notation of statics, the nodal forces are positive when they act in
the direction of the axis they refer. This is shown in Fig. 11.3.3. The actions
in the directions of the coordinates u1 , u2 , u4 , and u5 are forces while in the
directions of u3 and u6 are moments. In the following, the nodal elastic forces
are established using both methods, the method of the Lagrange equations and
that of the virtual work.
Fig. 11.3.3 Positive direction of the nodal forces af the plane frame element.
11.3.1.1 The method of the Lagrange equations

(i) Nodal elastic forces and stiffness matrix of the beam element
The elastic energy is due to the axial deformation of the element given by the
Eq. (11.2.12), that is,
Z
1 L
EAðx Þ½u 0 ðx, t Þ dx
2
Ua ¼ (11.3.8)
2 0
as well as to the bending deformation of the element given by Eq. (10.5.2),
that is,
Z
1 L
EI ðx Þ½v 00 ðx, t Þ dx
2
Ub ¼ (11.3.9)
2 0
Hence, the total elastic energy is
U ¼ Ua + U b
Z Z
1 L 1 L (11.3.10)
EAðx Þ½u 0 ðx, t Þ dx + EI ðx Þ½v 00 ðx, t Þ dx
2 2
¼
2 0 2 0
@CivilMethod

Z
1 L 2
U ðu1 , u2 , …, u6 Þ ¼ EAðx Þ u1 01 ðx Þ + u4 04 ðx Þ dx
2 0
Z
1 L 2
+ EI ðx Þ u2 002 ðx Þ + u3 003 ðx Þ + u5 00
5 ðx Þ + u6
00
6 ðx Þ dx
2 0
(11.3.11)
The nodal elastic forces result from Eq. (11.2.10) for i ¼ 1, 2, …,6.
Thus, after performing the differentiation, we obtain
∂U
fS1 ¼ ¼ k11 u1 + k14 u4 (11.3.12a)
∂u1
∂U
fS2 ¼ ¼ k22 u2 + k23 u3 + k25 u5 + k26 u6 (11.3.12b)
∂u2
∂U
fS3 ¼ ¼ k32 u2 + k33 u3 + k35 u5 + k36 u6 (11.3.12c)
∂u3
∂U
fS4 ¼ ¼ k41 u1 + k44 u4 (11.3.12d)
∂u4
∂U
fS5 ¼ ¼ k52 u2 + k53 u3 + k55 u5 + k56 u6 (11.3.12e)
∂u5
∂U
fS6 ¼ ¼ k62 u2 + k63 u3 + k65 u5 + k66 u6 (11.3.12f)
∂u6
where
Z L
kij ¼ EAðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 4 (11.3.13a)
0
Z L
00 00
kij ¼ EI ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 2, 3, 5, 6 (11.3.13b)
0
Eqs. (11.3.12a)–(11.3.12f) are written in matrix form as

8 9 2k 38 9
> fS1 > 11 0 0 k14 0 0 > u1 >
>
> > 6 > u2 >
>
>
> f > > 6 0 k22 k23 0 k25 k26 7 7>
>
>
< S2 > = 6 7 < >
> =
fS3 0 k k 0 k k u
¼66
32 33 35 36 7 3
7 u4 > (11.3.14)
> fS4 >
> > 6 k41 0 0 k44 0 0 7> >
>
> >
> >u > >
: fS5 >
> ;
4 0 k52 k53 0 k55 k56 5> : 5>
> >
;
fS6 0 k62 k63 0 k65 k66 u6
or
f eS ¼ ke ue (11.3.15)
where f eS ,
ue are the vectors of the nodal elastic forces and the nodal displace-
ments, respectively, and ke is the stiffness matrix of the e beam element.
Therefore, the matrix ke for the plane beam is defined as
@CivilMethod
2 3
k11 0 0 k14 0 0
60 k22 k23 0 k25 k26 7
6 7
60 k32 k33 0 k35 k36 7
k ¼6
e
6 k41
7 (11.3.16)
6 0 0 k44 0 0 7 7
40 k52 k53 0 k55 k56 5
0 k62 k63 0 k65 k66
Obviously, we deduce from Eqs. (11.3.13a), (11.3.13b) that kij ¼ kji . Hence,
the stiffness matrix is symmetric.
For an element with a constant cross-section Aðx Þ ¼ Ae , I ðx Þ ¼ I e and
length Le , Eqs. (11.3.13a), (11.3.13b) are integrated analytically and
2 3
EA EA
6 L 0 0 0 0 7
6 L 7
6 12EI 6EI 12EI 6EI 7
6 0 0 7
6 L3 L2 L3 L2 7
6 2EI 7
6 0 6EI 4EI

6EI 7
6 0 7
k ¼6
e 6 L 2 L L 2 L 7 (11.3.17)
EA EA 7
6 0 0 0 0 7
6 L L 7
6 7
6 12EI 6EI 12EI 6EI 7
6 0 3 2 0 7
6 L L L3 L2 7
4 6EI 2EI 6EI 4EI 5
0 0
L2 L L2 L
Note that the superscript e in Ae , I e , and Le has been dropped from the ele-
ments of the matrix in Eq. (11.3.17) for the sake of simplicity of the expressions.
(ii) Nodal inertial forces and mass matrix of the beam element
As in the case of the truss, the equivalent inertial nodal forces for the beam ele-
ment are obtained by two different assumptions of the mass distribution on
the element: the consistent mass assumption, which assumes a continuous dis-
tribution of the mass on the element, and the lumped mass assumption, which
lumps the mass at its nodes. The inertial mass matrices resulting from both
assumptions are derived below.
During the motion, the infinitesimal mass m ðx Þdx undergoes the two displace-
ments u ðx, t Þ and v ðx, t Þ. Therefore, the kinetic energy of the beam element will
be given by the expression.c
c. Due to bending, the cross-sections of the beam rotate by an angle v 0 ðx, t Þ. Under this rotation the
mass element m ðx Þdx exhibits a kinetic energy rðx ÞI ðx Þv_ 0 ðx, t Þdx=2, where rðx Þ is the mass
density of the material. Therefore, Eq. (11.3.18) should also include the term
Z Z
1 L 1 L
rðx ÞI ðx Þv_ 0 ðx, t Þdx ¼ rðx ÞI ðx Þ u_ 2 02 ðx Þ + u_ 3 03 ðx Þ + u_ 5 05 ðx Þ + u_ 6 06 ðx Þ dx
2 0 2 0
@CivilMethod
However, it can be shown that the contribution of this term is small and it can be neglected.
Z L n o
1
T¼ m ðx Þ ½u_ ðx, t Þ2 + ½v_ ðx, t Þ2 dx (11.3.18)
2 0
or using Eqs. (11.3.1), (11.3.3), the previous equation becomes

Z h i
1 L
T ðu_ 1 , u_ 2 , …, u_ 6 Þ ¼ m ðx Þ ½u_ 1 1 ðx Þ + u_ 4 4 ðx Þ2 dx
2 0
Z
1 L 2
+ m ðx Þ½u_ 2 2 ðx Þ + u_ 3 3 ðx Þ + u_ 5 5 ðx Þ + u_ 6 6 ðx Þ dx
2 0
(11.3.19)
The nodal inertial forces result from Eq. (11.2.9) for i ¼ 1, 2, …, 6. Thus,
after performing the differentiations, we obtain

d ∂T ∂T
fI 1 ¼ ¼ m11 u€1 + m14 u€4 (11.3.20a)
dt ∂u_ 1 ∂u1

d ∂T ∂T
fI 2 ¼ ¼ m22 u€2 + m23 u€3 + m25 u€5 + m26 u€6 (11.3.20b)
dt ∂u_ 2 ∂u2

d ∂T ∂T
fI 3 ¼ ¼ m32 u€2 + m33 u€3 + m35 u€5 + m36 u€6 (11.3.20c)
dt ∂u_ 3 ∂u3

d ∂T ∂T
fI 4 ¼ ¼ m41 u€1 + m44 u€4 (11.3.20d)
dt ∂u_ 4 ∂u4

d ∂T ∂T
fI 5 ¼ ¼ m52 u€2 + m53 u€3 + m55 u€5 + m56 u€6 (11.3.20e)
dt ∂u_ 5 ∂u5

d ∂T ∂T
fI 6 ¼ ¼ m62 u€2 + m63 u€3 + m65 u€5 + m66 u€6 (11.3.20f)
dt ∂u_ 6 ∂u6
where
Z L
mij ¼ m ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 1, 4 and i, j ¼ 2, 3, 5, 6 (11.3.21)
0

8 9 2m 38 9
> fI 1 > 11 0 0 m14 0 0 > u€1 >
>
> >
> >
> >
>
> f > 60 m22 m23 0 m25 m26 7 > u€2 >
>
< > I 2
= 660
7>< >
=
fI 3 m m33 0 m35 m36 7 u€
¼6
6 m41 0
32 7 3
(11.3.22)
>
> >f I 4 >
> 6 0 m44 0 0 7 7>> u€4 >
>
>
> > >
> fI 5 >
: > 40
; m52 m53 0 m55 m56 >5 > >
: u€5 >
;
fI 6 0 m62 m63 0 m65 m66 u€6
or
€e
f eI ¼ me u (11.3.23)
@CivilMethod
€e are the vectors of the nodal inertial forces and the nodal acceler-
where f eI , u
ations, respectively, and me the mass matrix of the e beam element. Therefore,
the mass matrix me is defined as
2 3
m11 0 0 m14 0 0
60 m22 m23 0 m25 m26 7
6 7
60 m35 m36 7
6 m32 m33 0 7
m ¼6
e
7 (11.3.24)
6 m41 0 0 m44 0 0 7
6 7
40 m52 m53 0 m55 m56 5
0 m62 m63 0 m65 m66
Obviously, we deduce from Eq. (11.3.21) that mij ¼ mji . Hence, the mass
Eq. (11.3.21) are integrated
For an element with constant mass, m ðx Þ ¼ m,
analytically and Eq. (11.3.24) becomes
2 3
140 0 0 70 0 0
60 156 22L 0 54 13L 7
6 7
e 60 2
3L 27
m 6 22L 4L 0 13L 7
me ¼ 6 7 (11.3.25)
420 6 70 0 0 140 0 0 7
6 7
40 54 13L 0 156 22L 5
0 13L 3L 0 2
22L 4L2
where m e ¼ m e Le is the total mass of the e element. Note that the superscript
e
e in L has been dropped from the elements of the matrix in Eq. (11.3.25) for the
sake of simplicity of the expressions.
The shape functions employed to derive the mass matrix are the same as
those employed to derive the stiffness matrix. The mass matrix resulting in this
way is referred to as the consistent mass matrix.
Z L
m1 ¼ m ðx Þð1 x Þdx (11.3.26a)
0
Z L
m2 ¼ m ðx Þxdx (11.3.26b)
0
Therefore, the kinetic energy of the beam element will be given by the
expression
1 1
T ¼ m1 u_ 1 2 + u_ 2 2 + m2 u_ 4 2 + u_ 5 2 (11.3.27)
2 2
@CivilMethod
The nodal forces result from Eq. (11.2.9) for i ¼ 1, 2, 3, 4. Thus, after
performing the differentiations, we obtain

d ∂T ∂T
fI 1 ¼ ¼ m11 u€1 (11.3.28a)
dt ∂u_ 1 ∂u1

d ∂T ∂T
fI 2 ¼ ¼ m22 u€2 (11.3.28b)
dt ∂u_ 2 ∂u2

d ∂T ∂T
fI 3 ¼ ¼0 (11.3.28c)
dt ∂u_ 3 ∂u3

d ∂T ∂T
fI 4 ¼ ¼ m44 u€4 (11.3.28d)
dt ∂u_ 4 ∂u4

d ∂T ∂T
fI 5 ¼ ¼ m55 u€5 (11.3.28e)
dt ∂u_ 5 ∂u5

d ∂T ∂T
fI 6 ¼ ¼0 (11.3.28f)
dt ∂u_ 6 ∂u6
where
m11 ¼ m22 ¼ m1 , m44 ¼ m55 ¼ m2 (11.3.29)
8 9 2 38 9
> fI 1 >
>
> >
>
> 6
m11 0 0 0 0 0 > > u€1 >
>
> fI 2 >
>
> > 7>> u€2 >
>
>
<f >
> = 6
0 m 22 0 0 0 0 7>>
< >
=
6 0 0 0 0 0 0 7 u€
¼6 7
I3 3
60 (11.3.30)
>
> fI 4 >
> 6 0 0 m44 0 077>> u€4 >
>
> >
> 5> >
>
>
> >
fI 5 >
4 0 0 0 0 m 0 >
>
> u€5 >
>
>
>
: ; > 55
: ;
0 0 0 0 0 0 u€6
fI 6
Eqs. (11.3.26a), (11.3.26b)
For an element with constant mass, m ðx Þ ¼ m,
are integrated analytically to give
e Le =2
m11 ¼ m22 ¼ m33 ¼ m44 ¼ m (11.3.31)
and the mass matrix (11.3.30) becomes
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
e 60 0 07
m 6 0 0 0 7
me ¼ 6 7 (11.3.32)
2 60 0 0 1 0 07
6 7
40 0 0 0 1 05
0 0 0 0 0 0
e Le is the total mass of the e element.
where m e ¼ m
We observe that the lumped mass assumption results in a diagonal mass
matrix.
@CivilMethod
(iii) Nodal damping forces and damping matrix of the beam element
As in the case of the truss element, we assume only internal damping. It is there-
fore due to the deformation of the beam element, it resists it, and depends on the
velocity of the strain. If we denote by sD the axial stress due to damping, we
may write
∂ex
sD ¼ cs (11.3.33)
∂t
where cs is the coefficient of the internal damping and ex the axial strain.
The virtual work of the internal damping force in the beam element is
Z
dWnc in
¼ sD dex dV (11.3.34)
V
sx N ðx Þ M ðx Þ
ex ¼ , sx ¼ + y, N ðx Þ ¼ EAðx Þu 0 , M ðx Þ ¼ EI ðx Þv 00 ðx, t Þ
E Aðx Þ I ðx Þ
the previous relation gives

Z L
dWnc ¼ cs Aðx Þu_ 0 ðx, t Þdu 0 ðx, t Þdx
Z0 L (11.3.35)
cs I ðx Þv_ 00 ðx, t Þdv 00 ðx, t Þdx
0
or using Eqs. (11.3.1), (11.3.3)

Z L

dWnc ¼
in
cs Aðx Þ u_ 1 01 ðx Þ + u_ 4 0
4 ðx Þ d u1 0 0
1 ðx Þ + u 4 4 ðx Þ dx
0
Z L
cs I ðx Þ u_ 2 002 ðx Þ + u_ 3 003 ðx Þ + u_ 5 005 ðx Þ + u_ 6 006 ðx Þ
0

d u2 02 ðx Þ + u3 03 ðx Þ + u5 05 ðx Þ + u6 06 ðx Þ dx
¼ ðc11 u_ 1 + c14 u_ 4 Þdu1 ðc41 u_ 1 + c44 u_ 4 Þdu4
ðc22 u_ 2 + c23 u_ 3 + c25 u_ 5 + c26 u_ 6 Þdu2
ðc32 u_ 2 + c33 u_ 3 + c35 u_ 5 + c36 u_ 6 Þdu3
ðc52 u_ 2 + c53 u_ 3 + c55 u_ 5 + c56 u_ 6 Þdu5
ðc62 u_ 2 + c63 u_ 3 + c65 u_ 5 + c66 u_ 6 Þdu6
(11.3.36)
where
Z L
cij ¼ cs Aðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 4 (11.3.37a)
0
@CivilMethod
Z L
00 00
cij ¼ c s I ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 2, 3,5, 6 (11.3.37b)
0
Writing Eq. (1.8.11) for the beam element, we obtain the nodal damping
forces as generalized forces for i ¼ 1, 2, …,6. Thus we have
fD1 ¼ Q1 ¼ c11 u_ 1 + c14 u_ 4 (11.3.38a)
fD2 ¼ Q2 ¼ c22 u_ 2 + c23 u_ 3 + c25 u_ 5 + c26 u_ 6 (11.3.38b)
fD3 ¼ Q3 ¼ c32 u_ 2 + c33 u_ 3 + c35 u_ 5 + c36 u_ 6 (11.3.38c)
fD4 ¼ Q4 ¼ c41 u_ 1 + c44 u_ 4 (11.3.38d)
fD5 ¼ Q5 ¼ c52 u_ 2 + c53 u_ 3 + c55 u_ 5 + c56 u_ 6 (11.3.38e)
fD6 ¼ Q6 ¼ c62 u_ 2 + c63 u_ 3 + c65 u_ 5 + c66 u_ (11.3.38f)
8 9 2 38 9
>
> fD1 >
> c11 0 0 c14 0 0 > > u_ 1 >
>
> >
>
> > 6
> 7>
>
>
>
>
>
>
> f >
D2 > 6 0 c c 0 c c 7
26 > 2 >
> u_ >
> >
> > 6 22 23 25
7>
> >
>
< fD3 >
> = 6 6 0 c32 c33 0 c35 c36 7 u_ 3 =
7>
< >
¼66 7 (11.3.39)
> fD4 > 7 u_ >
>
> >
> 6 c41 0 0 c44 0 0 7> > 4> >
>
> >
> 6 7>
>
> >
>
>
> >
> 6 7>
> _ >
>
>
> fD5 >
> 4 0 c52 c 53 0 c 55 c 56 5>
> u 5 >
>
>
: ; > : >
> ;
fD5 0 c62 c63 0 c65 c66 u_ 6
or
f eD ¼ ce u_ e (11.3.40)
where f eD , u_ e are the vectors of the nodal damping forces and nodal velocities,
respectively, and ce is the damping matrix of the beam element e. Hence, the
damping matrix ce is defined as
2 3
c11 0 0 c14 0 0
6 7
6 0 c22 c23 0 c25 c26 7
6 7
6 7
6 0 c 32 c 33 0 c 35 c36 7
c ¼6
e 6 7 (11.3.41)
7
6 c41 0 0 c44 0 0 7
6 7
60 c c 0 c c 7
4 52 53 55 56 5
0 c62 c63 0 c65 c66
For an element with constant cross-section Aðx Þ ¼ Ae , I ðx Þ ¼ I e and length

e
L , Eqs. (11.3.37a), (11.3.37b) are integrated analytically and Eq. (11.3.41)
becomes.
@CivilMethod
2 3
cs A cs A
6 L 0 0 0 0 7
6 L 7
6 12c I 6c I 12cs I 6cs I 7
6 0 s s
3 7
6 0
L2 7
6 L3 L2 L 7
6 2cs I 7
6 6cs I 4cs I 6cs I 7
6 0 0 2 7
6 L 2 L L L 7
ce ¼ 6 7 (11.3.42)
6 cs A cs A 7
6 0 0 0 0 7
6 L L 7
6 7
6 12c I 6c I 12cs I 6cs I 7
6 0 s
3 2
s
0 2 7
6 L3 L 7
6 L L 7
4 6cs I 2cs I 6cs I 4cs I 5
0 2
0 2
L L L L
Note that, the superscript e in Ae , I e , and Le has been dropped from the
elements of the matrix in Eq. (11.3.42) for the sake of simplicity of the
expressions.
(iv) Equivalent nodal loads of the beam element
The equivalent nodal forces can be evaluated in two ways: (a) as generalized
forces in the direction of the nodal displacements ue and (b) as static equivalent
forces. In the first case, the shape functions utilized to derive elastic forces are
employed too. The so-resulting vector of the equivalent nodal forces is referred
to as the consistent load vector.
Fig. 11.3.4 Loading and consistent equivalent nodal loads of a plane beam element.
(a) Consistent nodal load vector

Referring to Fig. 11.3.4, the virtual work produced by the displacements of the
beam element dx is
Z L Z L
dWnc ¼
p
px ðx, t Þdu ðx, t Þdx + py ðx, t Þdv ðx, t Þdx (11.3.43)
0 0

p
dWnc ¼ p1 ðt Þdu1 + p2 ðt Þdu2 + p3 ðt Þdu3 + p4 ðt Þdu4 + p5 ðt Þdu5 + p6 ðt Þdu6
(11.3.44)
@CivilMethod
where it was set

Z L
pi ðt Þ ¼ px ðx, t Þ i ðx Þdx, i ¼ 1, 4 (11.3.45a)
0
Z L
pi ðt Þ ¼ py ðx, t Þ i ðx Þdx, i ¼ 2, 3, 5, 6 (11.3.45b)
0
The nodal load vector results from Eq. (11.3.44)
pe ðt Þ ¼ f p1 ðt Þ p2 ðt Þ p3 ðt Þ p4 ðt Þ p5 ðt Þ p6 ðt Þ gT (11.3.46)
(b) Statically equivalent nodal load vector

In this case, the bending deformation of the beam element is ignored and its
transverse deformation is assumed linear, hence it is given by the relation
v ðx, t Þ ¼ u2 2 ðx Þ + u 5 5 ðx Þ
in which 2 ðx Þ ¼ 1 ðx Þ ¼ 1 x, 5 ðx Þ ¼ 4 ðx Þ ¼ x and Eq. (11.3.44) becomes

p
dWnc ¼ p1 ðt Þdu1 + p2 ðt Þdu2 + p4 ðt Þdu4 + p5 ðt Þdu5 (11.3.47)
where it was set
Z L
0
Z L
pi ðt Þ ¼ py ðx, t Þ i ðx Þdx, i ¼ 2, 5 (11.3.48b)
0
Consequently
pe ðt Þ ¼ f p1 ðt Þ p2 ðt Þ 0 p4 ðt Þ p5 ðt Þ 0 gT (11.3.49)
Inserting the expressions of the linear shape function in Eqs. (11.3.48a),
(11.3.48b) yields
Z L Z L
p1 ðt Þ ¼ px ðx, t Þð1 xÞdx, p4 ðt Þ ¼ px ðx, t Þxdx (11.3.50)
0 0
Z L Z L
p 2 ðt Þ ¼ py ðx, t Þð1 xÞdx, p5 ðt Þ ¼ py ðx, t Þxdx (11.3.51)
0 0
namely, the equivalent nodal loads are obtained as the reactions of a simply sup-
ported beam under the loads px ðx, t Þ and py ðx, t Þ. Hence, their name statically
equivalent nodal loads.
@CivilMethod
The forces resulting in this way are less accurate because the transverse
displacement is approximated by a linear polynomial instead of a cubic one.
The linear shape functions are not geometrically admissible, which means that
the conditions of the Ritz method are violated. Nevertheless, the accuracy of the
results is acceptable, especially when the elements are small. The statically
equivalent nodal load vector is particularly useful when inertial forces due
to the rotations of the cross-sections are ignored, as in the lumped mass
assumption.
(v) Geometric stiffness matrix of the beam element
When the produced rotation of the cross-sections is significant, there may be
shortening of the length of the element even if the axial deformation is ignored
(∂u=∂x ¼ 0). This shortening is due to the nonlinear term of the strain-
displacement relation [4,5], that is,

1 ∂v ðx, t Þ 2
ex ¼ (11.3.52)
2 ∂x
In the presence of an axial force N ðx Þ, the elastic energy due to its
shortening is
Z
1 L
UG ¼ N ðx Þex dx
2 0
Z (11.3.53)
1 L ∂v ðx, t Þ 2
N ðx Þ dx
2 0 ∂x
Z
1 L 2
UG ¼ N ðx Þ u2 02 ðx Þ + u3 03 ðx Þ + u5 0 0
5 ðx Þ + u6 6 ðx Þ dx
2 0
Differentiation of the previous expression for UG with respect to ui ,
i ¼ 1, 2, …,6, results in the additional elastic nodal forces
∂UG
fG1 ¼ ¼0 (11.3.54a)
∂u1
∂UG
fG2 ¼ ¼ kG22 u2 + kG23 u3 + kG25 u5 + kG26 u6 (11.3.54b)
∂u2
∂UG
fG3 ¼ ¼ kG32 u2 + kG33 u3 + kG35 u5 + kG36 u6 (11.3.54c)
∂u3
∂UG
fG4 ¼ ¼0 (11.3.54d)
∂u4
@CivilMethod
∂UG
fG5 ¼ ¼ kG52 u2 + kG53 u3 + kG55 u5 + kG56 u6 (11.3.54e)
∂u5
∂UG
fG6 ¼ ¼ kG62 u2 + kG63 u3 + kG65 u5 + kG66 u6 (11.3.54f)
∂u6
where
Z L
kGij ¼ N ðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 2, 3, 5, 6 (11.3.55)
0

8 9 2 38 9
> fG1 > 0 0 0 0 0 0 > u1 >
> >
>
> >
> 6 0 kG22 > >
>
> >
>
> f > kG23 0 kG25 kG26 7 > u 2>
< >
> = 6 7> >
G2
6 0 kG32 7 < =
fG3 kG33 0 kG35 kG36 7 u3
¼66 (11.3.56)
> fG4 >
> > 60 0 0 0 0 0 7 7>> u4 >
>
>
> >
> > >
>
> fG5 >
> >
4 0 kG52 kG53 0 kG55 kG56 5> > u5 >
> >
: ; : ;
fG6 0 kG62 kG63 0 kG65 kG66 u6
or
f eG ¼ keG ue (11.3.57)
where ue are the vectors of the elastic nodal forces and the nodal displace-
f eG ,
ments, respectively, and keG is the geometric stiffness matrix of the e beam
element. Therefore, the matrix keG for the plane frame is defined as
2 3
0 0 0 0 0 0
6 0 kG22 kG23 0 kG25 kG26 7
6 7
6 0 kG32 kG33 0 kG35 kG36 7
keG ¼ 6
60 0
7 (11.3.58)
6 0 0 0 0 7 7
4 0 kG52 kG53 0 kG55 kG56 5
0 kG62 kG63 0 kG65 kG66
When the axial force is constant, N ðx Þ ¼ N , Eq. (11.3.55) are integrated
analytically and Eq. (11.3.58) becomes
2 3
0 0 0 0 0 0
6 12 12 7
60 1 0 1 7
6 L L 7
6 7
6 4L L 7
60 1 0 1 7
N 6 3 7
keG ¼ 6 6
3 7 (11.3.59)
10 6 0 0 0 0 0 0 77
6 7
6 0 12 1 0 12 1 7
6 7
6 L L 7
4 5
L 4L
0 1 0 1
3 3
The axial force may be a priori known, as it happens when the structure is
under stress before the dynamic loads are applied. The total stiffness of the
@CivilMethod
element is the sum ke + keG . Thus, we conclude from Eq. (11.3.59) that the stiff-
ness increases if the axial force is tensile while it decreases if it is compressive.
The vanishing of the stiffness leads to the buckling of the structure.
The axial force is given by the relationd
∂u
N ðx Þ ¼ EA
∂x (11.3.60)
EA
¼ ðu1 + u4 Þ
L
namely, it depends on the displacements. Therefore, if the geometric stiffness is
taken into account in the analysis, the problem becomes nonlinear.
11.3.1.2 The method of virtual work

The application of this method is illustrated with the derivation of the stiffness
matrix of the beam element, under both axial and bending deformation.
The virtual work due to the axial deformation is written as
Z Z L
dWina ¼ sax deax dx ¼ AEeax deax dx (11.3.61)
V 0
Using Eq. (11.3.1) and taking into account that

8 9
>
> u1 >
>
> >
>
>
> u2 >
>
>
∂u < =
0 0 u3
0 0
ex ¼ ¼ u 1 1 + u 4 4 ¼ ½ 1 0 0 4 0 0
a
¼ N0a u (11.3.62)
∂x >
> u4 >
>
> >
>
> u >>
: 5>
> ;
u6
deax ¼ N0a du (11.3.63)
where
Na ¼ ½ 1 0 0 4 0 0 (11.3.64a)
u ¼ f u1 u2 u3 u4 u5 u6 gT (11.3.64b)
Substituting now Eqs. (11.3.62), (11.3.63) into Eq. (11.3.61) yields
Z L
EAðx ÞN0 a N0a dx u
T
dWina ¼ duT (11.3.65)
0
The virtual work due to the bending deformation is

Z
dWin ¼ sbx debx dx
b
(11.3.66)
V
" #
∂u 1 ∂v 2
d. The exact expression for the axial force is N ðx Þ ¼ EA + [4].
∂x 2 ∂x
@CivilMethod
Using Eq. (11.3.3), we obtain

8 9
>
> u1 >
>
>
> u >>
M < 2>
> =
u3
sbx ¼ y ¼ Eyv 00 ¼ Ey ½ 0 00 00
0 00 00
6 u ¼ EyN00b u
I 2 3 5
>
> >4 >
>
>
> >
: u5 >
> ;
u6
debx ¼ yN00b du
where
N00b ¼ ½ 0 00
2
00
3 0 00
5
00
6 (11.3.67)
Substituting the foregoing expressions for sbx and debx into Eq. (11.3.66)
and integrating over the cross-section, we obtain
Z L
EI ðx ÞN00 b N00b dx u
T
dWinb ¼ duT (11.3.68)
0
The total virtual work of the internal forces is

Z L Z L
EAðx ÞN0 a N0a dx + EI ðx ÞN00 b N00b dx u
T T
dWin ¼ dWina + dWinb ¼ duT
0 0
(11.3.69)
which is set equal to the virtual work of the equivalent nodal forces, that is,
dWex ¼ duT f S (11.3.69)
to yield
8 9 2k 38 9
> fS1 > 11 0 0 k14 0 0 > u1 >
> > 6 > u2 >
>
> fS2 >
> > 60
> k22 k23 0 k25 k26 7
7>
>
>
>
< = 6 7 < >
> =
fS3 0 k32 k33 0 k35 k36 7 u3
¼6
6 k41 (11.3.70)
>
> fS4 >
> 6 0 0 k44 0 0 7 7>> u4 >
>
>
> >
> 40 5 >
> >
>
: fS5 >
> ; k52 k53 0 k55 k56 > : u5 >
;
fS6 0 k62 k63 0 k65 k66 u6
where
Z L
kij ¼ EAðx Þ 0i ðx Þ 0j ðx Þdx, i, j ¼ 1, 4 (11.3.71a)
0
Z L
00 00
kij ¼ EI ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 2, 3, 5, 6 (11.3.71b)
0
Obviously, Eq. (11.3.70) is identical to Eq. (11.3.14). A similar procedure

will give the remaining element matrices.
@CivilMethod
Example 11.3.1 Compute the stiffness and mass matrices and the nodal load
vector
of a beam element, which is subjected to the loads px ðx, t Þ ¼ pðt Þ
1 + x x 2 and py ðx, t Þ ¼ 3pðt Þð1 x Þ, x ¼ x=L. The height of the cross-
section varies linearly, h ðx Þ ¼ h ð1 + 0:5x Þ, while its width b is constant.
The material density is r. Consider the lumped mass assumption.
Solution
i. Stiffness matrix: We set A ¼ bh, I ¼ bh 3 =12. Hence Aðx Þ=A ¼ ð1 + 0:5x Þ,
I ðx Þ=I ¼ ð1 + 0:5x Þ3 . Then we obtain from Eq. (11.3.13)
Z L
0 0
k11 ¼ EAðx Þ 1 ðx Þ 1 ðx Þdx ¼ 1:25k11
c
0
Z L
0 0
k14 ¼ EAðx Þ 1 ðx Þ 4 ðx Þdx ¼ 1:25k14
c
0
Z L
00 00
k22 ¼ EI ðx Þ 2 ðx Þ 2 ðx Þdx ¼ 2:09375k22
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
Z L
00 00
c
0
The superscript c refers to the element of constant cross-section A ¼ bh.
@CivilMethod
ii. Mass matrix: Substituting m ðx Þ ¼ rAðx Þ ¼ rAð1 + 0:5x Þ into

Z L
7
m11 ¼ m22 ¼ m ðx Þð1 xÞdx ¼ rAL
0 12
Z L
8
m44 ¼ m55 ¼ m ðx Þxdx ¼ rAL
0 12
iii. Nodal load vector: Using Eqs. (11.3.50), (11.3.51), we obtain
Z L
7
p1 ðt Þ ¼ px ðx, t Þð1 x Þdx ¼ pðt ÞL
0 12
Z L
7
p4 ðt Þ ¼ px ðx, t Þxdx ¼ pðt ÞL
0 12
Z L
p2 ðt Þ ¼ py ðx, t Þð1 x Þdx ¼ pðt ÞL
0
Z L
1
p5 ðt Þ ¼ py ðx, t Þxdx ¼ pðt ÞL
0 2
11.3.2 Transformation of the nodal coordinates

of the plane frame element
The translational components u1 and u2 are transformed from the local to the
global system of axes according to Eq. (11.2.67), while the component u3 ,
which expresses a rotation about the z axis, remains unaltered in the rotated sys-
tem because the local axis z and the global axis z are identical, Fig. 11.3.5.
Hence, we can write

u1 cos f sin f u1
¼ (11.3.72a)
u2 sin f cosf u2
u3 ¼ u3 (11.3.72b)
which may be combined as
( ) 2 3( )
u1 cos f sin f 0 u1
u2 ¼ 4 sin f cos f 0 5 u2 (11.3.73a)
u3 0 0 1 u3
@CivilMethod
Fig. 11.3.5 Nodal displacements of the beam element in global and local axes.
( ) 2 3( )
u4 cosf sin f 0 u4
u5 ¼ 4 sin f cos f 0 5 u5 (11.3.73b)
u6 0 0 1 u6
Eqs. (11.3.73a), (11.3.73b) are further combined as
8 9 2 38 9
>
> u1 >>
> 6
cos f sin f 0 0 0 0 >> u1 >>
> >
>
> > > u2 >
7> >
>
< >
> u 2
= 6
sin f cos f 0 0 0 0 7>
>
< >
=
u3 6 0 0 1 0 0 0 7 u
¼6
6
7 3
(11.3.74)
> u4 >
> > 6 0 0 0 cos f sin f 0 7> u4 >
7> >
> >
> 5> >
>
> u >
>
4 0 0 0 sin f cos f 0 >
> u > >
> ;
: 5
: 5>
> ;
u6 0 0 0 0 0 1 u6
or
ue ¼ Re ue (11.3.75a)
hence
ue ¼ ðRe ÞT ue (11.3.75b)

where
2 3
cos fe sin fe 0 0 0 0
6 sin fe cos fe 0 0 0 07
6 7
6 0 0 1 0 0 07
R ¼6
e
6 0
7 (11.3.76)
6 0 0 cos fe sin fe 077
4 0 0 0 sin fe cos fe 05
0 0 0 0 0 1
represents the transformation matrix of the e plane frame element.
Eq. (11.3.75b) holds because the matrix Re is orthonormal.
The global vectors of the nodal forces f S , f I , f D and pe ðt Þ are defined in
e e e
relation to ue . Their transformation obeys the same law, Eq. (11.3.75b). Thus,
we have
@CivilMethod
f e ¼ðRe ÞT f e (11.3.77)
S S
f e ¼ðRe ÞT f e (11.3.78)
I I
f e ¼ðRe ÞT f eD (11.3.79)
D
pe ðt Þ ¼ðR Þ p ðt Þ
e T e
(11.3.80)
Using Eqs. (11.3.15), (11.3.23), (11.3.40), (11.3.75a), the first three of the
foregoing equations are transformed in the global axes as
f e ¼ ke ue (11.3.81)
S
f e e €e
u
¼m (11.3.82)
I
f D ¼ c u
e _e
(11.3.83)
where
k ¼ ðRe ÞT ke Re
e
(11.3.84)
m (11.3.85)
ce ¼ ðRe ÞT ce Re (11.3.86)

e e , and ce representing the stiffness, mass, and damping matrices of the
with k , m
e beam element in global axes, respectively. Note that the lumped mass matrix
remains unaltered under this transformation.
Applying the procedure presented in Section 11.2.3 for the plane truss,
we obtain the equation of motion of the plane frame
M u_ + K
u€ + C u ¼
pð t Þ (11.3.87)
which gives after applying the support conditions and partitioning
u€f + C
e ff e
M u_ f + K
uf ¼ e (11.3.88)
M u€f + M
e sf e u€s + C
e ss e u_ f + C
e sf e u_ s + K
uf + K us ¼ e
p s ðt Þ (11.3.89)
where
pf∗ ðt Þ ¼ e
e fs e u_ s K
e fs e e fs e
us (11.3.90)
Apparently, Eqs. (11.3.88), (11.3.89), (11.3.90) are the counterpart of
Eqs. (11.2.117), (11.2.118), (11.2.119), which hold for the plane truss.
For the sake of convenience, we write Eq. (11.3.88) as
M€
u + C€
u + Ku ¼ pðt Þ (11.3.91)
@CivilMethod
Example 11.3.2 Formulate the equation of motion of the frame shown in

Fig. E11.2.
Fig. E11.2 Frame in Example 11.3.2
Assumed data:
Coordinates of frame nodes: 1ð0, 0Þ, 2ð0, 5Þ, 3ð4, 8Þ, 4ð10, 0Þ.
Properties of the elements: A1 ¼ A, A2 ¼ A3 ¼ 2A, I1 ¼ I , I2 ¼ I3 ¼ 8I ,
I =A ¼ ðL1 =50Þ2 . Load pðx, t Þ ¼ P ðt Þ=L2 . Damping coefficient cs ¼ 0,
modulus of elasticity E, material density r. Lumped mass assumption.
Solution
The system has n ¼ 4 nodes, hence the free structure has N ¼ 3n ¼ 12 degrees
of freedom, (u3i2 , u3i1 , u3i , i ¼ 1, 2, 3, 4). The numbering of the elements and
the positive direction of the local axes are chosen as in Fig. E11.2.
TABLE E11.2 Geometrical data of the elements in Example 11.3.2.

e xj xk D
x yj yk D
y Le cos fe sinfe
1 0 0 0 0 5 5 5 1 0
2 0 4 4 5 8 3 5 0.8 0.6
3 4 10 6 8 0 8 10 0.6 0.8
1. Computation of ke , me , ce , pe ðt Þ, Re for e ¼ 1, 2, 3
Matrices ke . The elements have constant cross-section. Therefore, Eq. (11.3.17)
is employed, which gives
@CivilMethod
2 3
2500 0 0 2500 0 0
6 0 12 30 0 12 30 7
6 7
EI 66 0 30 50 7
0 30 100 7
k ¼
1
6 7
125 6 2500 0 0 2500 0 07
6 7
4 0 12 30 0 12 30 5
0 30 50 0 30 100
2 3
5000 0 0 5000 0 0
6 0 96 240 0 96 240 7
6 7
EI 66 0 240 400 7
0 240 800 7
k ¼
2
6 7
125 6 5000 0 0 5000 0 07
6 7
4 0 96 240 0 96 240 5
0 240 400 0 240 800
2 3
2500 0 0 2500 0 0
6 0 12 60 0 12 60 7
6 7
EI 6 0 60 400 0 60 200 7
k ¼
3 6 7
125 6
6 2500 0 0 2500 0 0 7
7
4 0 12 60 0 12 60 5
0 60 200 0 60 400
Matrices me . The elements have constant cross-section. Therefore,
Eq. (11.3.32) is employed for lumped mass assumption, which gives
2 3 2 3
2:5 0 0 0 0 0 5:0 0 0 0 0 0
6 0 2:5 0 0 0 0 7 6 0 5:0 0 0 0 0 7
6 7 6 7
6 0 0 0 0 0 07 7 60 0 0 0 0 07
m1 ¼ rA6 6 0 0 0 2:5 0 0 7 , m 2
¼ rA 6
6 0 0 0 5:0 0 0 7
7
6 7 6 7
4 0 0 0 0 2:5 0 5 4 0 0 0 0 5:0 0 5
0 0 0 0 0 0 0 0 0 0 0 0
2 3
10:0 0 0 0 0 0
6 0 10:0 0 0 0 07
6 7
6 0 0 0 0 0 07
m ¼ rA6
3 6 7
6 0 0 0 10:0 0 0 7 7
4 0 0 0 0 10:0 0 5
0 0 0 0 0 0
Matrices ce . It is cs ¼ 0 and Eq. (11.3.42) gives ce ¼ 0.
Vectors pe ðt Þ. To be consistent to the lumped mass assumption, the statically
equivalent nodal load vector will be employed. Therefore, Eqs. (11.3.50), (11.3.51)
give for the load-free elements 1 and 3, p1 ðt Þ ¼ p2 ðt Þ ¼ p3 ðt Þ ¼ p4 ðt Þ ¼ 0 while
for element 2, we obtain
p1 ðx, t Þ ¼ PLð2t Þ cos fð2Þ sin fð2Þ , p2 ðx, t Þ ¼ PLð2t Þ cos 2 fð2Þ
@CivilMethod
where cos fð2Þ ¼ 0:8, sin fð2Þ ¼ 0:6. Thus, we have

8 9 8 9
>
> 0:048P ðt Þ > > >
> 0>>
> >
> > >
>
> 0:064P ð t Þ >
>
>
> 0>>
>
< = < >
> =
0 0
p ðt Þ ¼
2
1 3
>
> 0:048P ðt Þ > > >
> 0>>
> > > >
>
> 0:064P ðt Þ > >
>
>
> 0>>
>
: ; : >
> ;
0 0
Matrices Re . The direction cosines cos fe and sin fe are computed from the
Cartesian coordinates of the element nodes and are given in Table E11.2. Thus
employing Eq. (11.3.76), we obtain
2 3 2 3
0 1 0 0 0 0 0:8 0:6 0 0 0 0
6 1 0 0 0 0 0 7 6 0:6 0:8 0 0 0 0 7
6 7 6 7
6 0 0 1 0 0 07 7 6 0 0 1 0 0 07
R1 ¼ 6
6 0 0 0 0 1 07 , R 2
¼ 6
6 0 0 0 0:8 0:6 0 7
7
6 7 6 7
4 0 0 0 1 0 0 5 4 0 0 0 0:6 0:8 0 5
0 0 0 0 0 1 0 0 0 0 0 1
2 3
0:6 0:8 0 0 0 0
6 0:8 0:6 0 0 0 07
6 7
6 0 0 1 0 0 07
R ¼6
3 6 7
60 0 0 0:6 0:8 0 7 7
40 0 0 0:8 0:6 0 5
0 0 0 0 0 1
e
e , pe ðt Þ for e ¼ 1, 2, 3
Matrices k . They are computed using Eq. (11.3.84)
e
2 3
12 0 30 12 0 30
6 0 2500 0 0 2500 07
6 7
EI 6 30 0 100 30 0 50 7

k ¼
1 6 7
125 6
6 12 0 30 12 0 30 7
7
4 0 2500 0 0 2500 05
30 0 50 30 0 100
2 3
3234:6 2353:9 144:0 3234:6 2353:9 144:0
6 2353:9 1861:4 192:0 2353:9 1861:4 192:0 7
6 7
EI 6 144:0 192:0 800:0 144:0 192:0 400:0 7

k ¼
2 6 7
125 66 3234:6 2353:9 144:00 3234:6 2353:9 144:0 7 7
4 2353:9 1861:4 192:0 2353:9 1861:4 192:0 5
144:0 192:0 400 144:0 192:0 800:0
@CivilMethod
2 3
907:7 1194:2 48:0 907:6 1194:2 48:0
6 1194:2 1604:3 36:0 1194:2 1604:3 36:0 7
6 7
EI 6 48:0 36:0 400:0 48:0 36:0 200:0 7
k3 ¼ 6 7
125 6
6 907:6 1194:2 48:0 907:7 1194:2 48:0 7
7
4 1194:2 1604:3 36:0 1194:2 1604:3 36:0 5
48:0 36:0 200:0 48:0 36:0 400:0
e . They are computed using Eq. (11.3.85)
Matrices m
2 3 2 3
2:5 0 0 0 0 0 5:0 0 0 0 0 0
60 2:5 0 0 0 0 7 60 5:0 0 0 0 07
6 7 6 7
60 0 0 0 0 07 60 0 0 0 0 07
1 ¼ rA6
m 60
7, m
7 2
¼ rA 6
60
7
6 0 0 2:5 0 0 7 6 0 0 5:0 0 077
40 0 0 0 2:5 0 5 40 0 0 0 5:0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
10:0 0 0 0 0 0
6 0 10:0 0 0 0 07
6 7
6 0 0 0 0 0 07
3 ¼ rA6
m 6 0
7
6 0 0 10:0 0 077
4 0 0 0 0 10:0 05
0 0 0 0 0 0
Vectors pe ðt Þ. They are computed using Eq. (11.3.80).
8 9 8 9
>
> 0 >
> >
> 0>>
> >
> > >
>
> 0:08P ð t Þ >
>
>
> 0>>
>
< = < >
> =
0 0
p ðt Þ ¼
2
1 3
>
> 0 >
> >
> 0>>
> >
> > >
>
> 0:08P ð t Þ >
>
>
> 0>>
>
: ; : >
> ;
0 0
^ e, M
3. Computation of the enlarged matrices K ^ e, p
ê ðt Þ for e ¼ 1, 2, 3
Assembly matrices ae . By inspection, we obtain
2 3 2 3
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
60 1 0 0 0 0 0 0 0 0 0 07 60 0 0 0 1 0 0 0 0 0 0 07
6 7 6 7
6 7 6 7
60 0 1 0 0 0 0 0 0 0 0 07 60 0 0 0 0 1 0 0 0 0 0 07
6 7 6 7
a ¼6
1
7, a ¼ 6
2
7
60 0 0 1 0 0 0 0 0 0 0 07 60 0 0 0 0 0 1 0 0 0 0 07
6 7 6 7
6 7 6 7
40 0 0 0 1 0 0 0 0 0 0 05 40 0 0 0 0 0 0 1 0 0 0 05
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
@CivilMethod
2 3
0 0 0 0 0 0 1 0 0 0 0 0
60 0 0 0 0 0 0 1 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 1 0 0 07
6 7
a ¼6
3
7
60 0 0 0 0 0 0 0 0 1 0 07
6 7
6 7
40 0 0 0 0 0 0 0 0 0 1 05
0 0 0 0 0 0 0 0 0 0 0 1
^ e ¼ ðae ÞT ke ae
^ e . They are computed using the relation K
Matrices K
2 3
12 0 30 12 0 30 0 0 0 0 0 0
6 0 2500 0 0 2500 0 0 0 0 0 0 07
6 7
6 7
6 30 0 100 30 0 50 0 0 0 0 0 0 7
6 7
6 12 30 0 0 0 0 0 0 7
6 0 30 12 0 7
6 7
6 0 2500 0 0 2500 0 0 0 0 0 0 07
6 7
6 30 100 0 0 0 0 0 0 7
^1 ¼ EI 6 0 50 30 0 7
K 6 7
125 6 0 0 0 0 0 0 0 0 0 0 0 07
6 7
6 0 0 0 0 0 0 0 07
6 0 0 0 0 7
6 7
6 0 0 0 0 0 0 0 0 0 0 0 07
6 7
6 0 0 0 0 0 0 0 07
6 0 0 0 0 7
6 7
4 0 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
6 7
60 0 0 07
6 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 0 3234:6 2353:9 144:0 3234:6 2353:9 144:0 0 0 07
6 7
6 7
60 0 0 2353:9 1861:4 192:0 2353:9 1861:4 192:0 0 0 07
6 7
6 7
60 0 0 144:0 192:0 800:0 144:0 192:0 400:0 0 0 07
^ 2 ¼ EI 6
K 6
7
7
125 6 0 0 0 3234:6 2353:9 144:0 3234:6 2353:9 144:0 0 0 07
6 7
6 7
60 0 0 2353:9 1861:4 192:0 2353:9 1861:4 192:0 0 0 07
6 7
6 7
60 0 0 144:0 192:0 400:0 144:0 192:0 800:0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
4 5
0 0 0 0 0 0 0 0 0 0 0 0
@CivilMethod
2 3
0 0 0 0 0 0 0 0 0 0 0 0
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 07
6 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
EI 60 0 0 0 0 0 0 0 0 0 0 07
^3 ¼
K 6 7
125 6
60 0 0 0 0 0 907:6 1194:2 48 907:6 1194:2
7
48 7
6 7
60 0 0 0 0 0 1194:2 1194:2 1604:3 36 7
6 1604:3 36 7
6 7
60 0 0 0 0 0 48 36 400 48 36 200 7
6 7
6 7
60 0 0 0 0 0 907:6 1194:2 48 907:6 1194:2 48 7
6 7
6 7
40 0 0 0 0 0 1194:2 1604:3 36 1194:2 1604:3 36 5
0 0 0 0 0 0 48 36 200 48 36 400
Matrices M ^ e . They are computed using the relation M

^ e ¼ ðae ÞT m
e ae
2 3
2:5 0 0 0 0 0 0 0 0 0 0 0
60 2:5 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 07
6 0 0 0 7
6 7
60 0 0 2:5 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 2:5 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 07
^ 6 0 0 0 7
M ¼ rA6
1
7,
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 07
6 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
40 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
6 0 0 0 5: 0 0 0 0 0 0 0 0 7
6 7
6 0 0 0 0 5: 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
^ ¼ rA6 6 7
M 2
7
6 0 0 0 0 0 0 5: 0 0 0 0 0 7
6 7
6 0 0 0 0 0 0 0 5: 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
40 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
@CivilMethod
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 07
6 0 0 0 0 0 0 0 0 0 0 7
60 07
6 0 0 0 0 0 0 0 0 0 0 7
6 7
^ 60 0 0 0 0 0 0 0 0 0 0 07
M ¼ rA6
3
7
60 0 0 0 0 0 10: 0 0 0 0 07
6 7
60 0 0 0 0 0 0 10: 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 10: 0 07
6 7
40 0 0 0 0 0 0 0 0 0 10: 05
0 0 0 0 0 0 0 0 0 0 0 0
ê ðt Þ. They are computed using the relation p
Vectors p ^ e ðt Þ ¼ ðae ÞT pe ðt Þ
8 9 8 9
>
> 0 >
> >
> 0> >
>
> 0 >
> >
> 0> >
>
> >
> >
> >
>
>
> 0 >
> >
> >
>
>
> >
> >
> 0 >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
>
> >
> >
> >
>
> 0:08P ð t Þ >
> > 0 >
>
< = < >
> =
0 0
^ ðt Þ ¼
p 2
^ ðt Þ ¼ p
, p1
^ ðt Þ ¼
3
>
> 0 >
> >
> 0> >
> >
> > >
>
>
> 0:08P ð t Þ >
>
>
>
> 0> >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
> >
> > >
>
> 0
> >
> >0>
>
> >
>
>
> >
> >
> >
>
> >
> > >
>
: 0 ; : >
> 0 ;
0 0
K,
4. Computation of the total matrices M, pðt Þ of the frame
2 3
2:5 0 0 0 0 0 0 0 0 0 0 0
60 2:5 0 0 0 0 0 0 0 0 0 07
6 7
60 07
6 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 7:5 0 0 0 0 0 0 0 07
6 7
60 0 0 0 7:5 0 0 0 0 0 0 07
X 6 7
3 6 07
¼
M ^ e ¼ rA6 0
M
0 0 0 0 0 0 0 0 0 0 7
60 0 0 0 0 0 15: 0 0 0 0 07
e¼1 6 7
60 07
6 0 0 0 0 0 0 15: 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 10: 0 07
6 7
40 0 0 0 0 0 0 0 0 0 10: 05
0 0 0 0 0 0 0 0 0 0 0 0
@CivilMethod
X
3
¼
K ê
K
e¼1
2 3
12 0 30 12 0 30 0 0 0 0 0 0
6 0 2500 0 0 2500 0 0 0 0 0 0 07
6 7
6 7
6 30 0 100 30 0 50 0 0 0 0 0 07
6 7
6 12 2353:9 114 3234:6 2353:9 144 07
6 0 30 3246:6 0 0 7
6 7
6 0 2500 0 2353:9 4361:4 192 2353:9 1861:4 192 0 0 07
6 7
EI 66 30 0 50 114 192 900 144 192 400 0 0 077
¼ 6 7
125 6 0 0 0 3234:6 2353:9 144 4142:28 1159:7 192 907:68 1194:2 48 7
6 7
6 0 0 2353:9 1861:4 192 156 1194:2 1604:3 36 7
6 0 1159:7 3465:7 7
6 7
6 0 0 0 144 192 400 192 156 1200 48 36 200 7
6 7
6 0 907:68 1194:2 48 907:68 1194:2 48 7
6 0 0 0 0 0 7
6 7
4 0 0 0 0 0 0 1194:2 1604:3 36 1194:2 1604:3 36 5
0 0 0 0 0 0 48 36 200 48 36 400
8 9
>
>
0 >
>
>
> >
>
>
> 0 >
>
>
> >
>
>
> 0 >
>
>
> ð Þ >
>
>
> 2P t >
>
>
> >
>
>
> 0:08P ðt Þ >
>
X3 < =
ðt Þ + 0
pðt Þ ¼ P ^ e ðt Þ ¼
p
>
> 0 >
>
> >
> 1:08P ðt Þ >
e¼1
>
> >
>
>
> >
>
> 0
> >
>
>
> >
>
> 0
> >
>
>
> >
>
>
> 0 >
>
: ;
0
K,
5. Modification of the matrices M, pðt Þ due to support conditions of
the frame
Referring to Fig. E11.2, the displacement vector should be rearranged as
e e1 u
uT ¼ f u e2 ue3 ue4 ue5 ue6 ue7 ue8 ue9 ue10 ue11 ue12 g
¼ f u4 u5 u6 u7 u8 u9 u12 u1 u2 u3 u10 u11 g
Hence, the matrix V is defined from the relation
8 9 2 38 9
> e1 >
u 0 0 0 0 0 0 0 1 0 0 0 0 > u4 >
> >
> > >
>u
>
> e >
> 60 07 > u5 >
> >
>
> 2 > 6 0 0 0 0 0 0 0 1 0 0 7>
> >
>
>
>
> e3 >
u >
>
>
60
6 0 0 0 0 0 0 0 0 1 0 07 >
>
>
7> u 6
>
>
>
>
> > > >
>
>
> e4 >
u > 6
>
61 0 0 0 0 0 0 0 0 0 0 07 >
7>
> u 7 >
>
>
> >
> > >
>
>
> e
u5 > >
60 1 0 0 0 0 0 0 0 0 0 07 > u8 >
>
7> >
>
< = 660 < =
e6
u 0 1 0 0 0 0 0 0 0 0 077 u9
¼6
60
>
> e
u 7 >> 6 0 0 1 0 0 0 0 0 0 0 07 > u12 >
7> >
>
>u >
> 60 > u1 >
> >
>
> e 8 >> 6 0 0 0 1 0 0 0 0 0 0 077>
> >
>
>
>u >
> 60 >
> >
>
>
> e 9 >> 6 0 0 0 0 1 0 0 0 0 0 07 >
7> u 2 >>
>
> >
> 60 > u >
> >
>
> e
u >
10 > 6 0 0 0 0 0 0 0 0 0 1 07 >
7> 3 >>
>
> >
> > >
>
>
>
: e
u 11 >
>
;
40 0 0 0 0 0 0 0 0 0 0 1 5>
>
>
: u 10 >
>
;
e12
u 0 0 0 0 0 0 1 0 0 0 0 0 u11
Applying Eqs. (11.2.114a), (11.2.114c), (11.2.114d), we obtain
@CivilMethod
e ff , M
6. Formulation of the matrices M e fs , M
e sf , M
e ss , K
e ff , K
e fs , K
e sf , K
e ss ,
e
pf ðt Þ, e
p s ðt Þ
e M,
These matrices result directly from K, e e pðt Þ taking into account the partition-
ing dictated by the separation of the free displacements from the specified
(support) ones. Thus, we have
2 3 2 3
7:5 0 0 0 0 0 0 0 0 0 0 0
60 7:5 0 0 0 0 07 60 0 0 0 07
6 7 6 7
60 0 0 0 0 0 07 7 60 0 0 0 07
6 6 7
e ff ¼ rA6 0 0 0 15: 0 0 0 7 e 6 7
M 6 7, Mfs ¼ rA6 0 0 0 0 0 7
60 0 0 0 15: 0 0 7 7 6 7
6 60 0 0 0 07
40 0 0 0 0 0 0 5 4 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3 2 3
0 0 0 0 0 0 0 2:5 0 0 0 0
60 0 0 0 0 0 07 60 2:5 0 0 0 7
6 7 6 7
Me sf ¼ rA6 0 0 0 0 0 0 0 7, M e ss ¼ rA6 0 0 0 0 0 7
6 7 6 7
40 0 0 0 0 0 05 40 0 0 10: 0 5
0 0 0 0 0 0 0 0 0 0 0 10:
2 3
3246:6 2353:9 114 3234:6 2353:9 144 0
6 2353:9 4361:4 192 2353:9 1861:4 192 07
6 7
6 114 192 900 144 192 400 0 7
EI 6 7
e
Kff ¼ 6 3234:6 2353:9 144 4142:2 1159:7 192 48 7
6
125 6 7
7
6 2353:9 1861:4 192 1159:7 3465:7 156 36 7
4 144 192 400 192 156 1200 200 5
0 0 0 48 36 200 400
2 3
12 0 30 0 0
6 0 2500 0 0 0 7
6 7
6 30 0 50 0 0 7
EI 6 7
Ke fs ¼ 6 0 0 0 907:6 1194:2 7
125 6
6 0
7
6 0 0 1194:2 1604:3 7 7
4 0 0 0 48 36 5
0 0 0 48 36
@CivilMethod
2 3
12 0 30 0 0 0 0
6 0 2500 0 0 0 0 07
6 7
e sf ¼ EI 6 30
K 0 50 0 0 0 07
125 46 7
0 0 0 907:6 1194:2 48 48 5
0 0 0 1194:2 1604:3 36 36
2 3
12 0 30 0 0
6 0 2500 0 0 0 7
e EI 66
7
Kss ¼ 6 30 0 100 0 0 77
125 4
0 0 0 907:6 1194:2 5
0 0 0 1194:2 1604:3
8 9
>
> 2:0P > > 8 9
>
>
>
> 0:08P > >
>
> >
> P1 >>
> >
> > >
>
< 0 = < P2 >
> =
e
pf ðt Þ ¼ 0 , e
ps ðt Þ ¼ P3
> > > >
> 1:08P >
>
> >
> > P10 >
>
> >
>
>
> >
> : ;
>
> 0 >
> P11
: ;
0
7. Equation of motion
In as much as e u€s ¼ 0 and Eq. (11.3.90) yields e
us ¼ 0, it is also e pf∗ ðt Þ ¼ e
pf ðt Þ.
The equation of motion results from Eq. (11.3.88). Thus we obtain
2 38 9
7:5 0 0 0 0 0 0 > > e€1 >
u >
60 > >
6 7:5 0 0 0 0 077>
> u >
> e€2 >
> >
60 7 > >
>
6 0 0 0 0 0 0 7><ue€3 >
=
6 7 €
rA6 0 0 0 15:0 0 0 07 u e4
6 7> € >
60 0 0 0 15:0 0 0 7>>
>ue5 >>
>
6 7>> >
40 0 0 0 0 0 0 5>> e
u€6 >
>
>
>
:€ > ;
0 0 0 0 0 0 0 e7
u
2 38 9
3246:6 2353:9 114:0 3234:6 2353:9 144:0 0 >
>ue1 >>
6 2353:9 4361:4 192:0 2353:9 1861:4 192:0 0 7>>
> e2 >
u
>
>
6 7>> >
>
6 114:0 144:0 192:0 400:0 0 7>7 > e3 >>
EI 66 192:0 900:0 >
6 2353:9 1861:4 192:0 1159:7 3465:7 156:0 36:0 7> >ue5 >>
6 7>>
>
>
>
>
4 144:0 5
192:0 156:0 1200:0 200:0 > > e >
192:0 400:0 > >
:
u 6 >
;
0 0 0 48:0 36:0 200:0 400:0 e7
u
8 9
>
> 2:0P >>
>
> 0:08P > >
>
> >
>
>
> >
>
>
< 0 >
=
¼ 0
>
> >
>
>
> 1:08P >>
>
>
>
> >
>
>
> 0 >
>
: ;
0
@CivilMethod
The solution of the equation of motion gives the vector e

uf . Then Eq. (11.3.89)
yields the support reactions because there are no nodal element loads in the
direction of the supports, hence
e e sf e
p s ðt Þ ¼ K uf
or
8 9
2 3>> e1 >
u >
8 9 >
> >
> P 1 > 12 0 30 0 0 0 0 >
> e2 >
u >
>
>
> >
> 6 7 >
> >
>
> P >
< 2 = EI 6 0 2500 0 0 0 0 0 >
7< 3 >
e
u =
6 7
P3 ¼ 6 30 0 50 0 0 0 0 7 u e4
>
> >
> 125 6 7> >
>
> P10 >
: >
;
4 0 0 0 907:68 1194:2 48 48 5>>
>
> e5 >
u >
>
>
> >
P11 0 0 0 1194:2 1604:3 36 36 >>
> e
u >
6>
: > ;
e7
u
11.4 Static condensation: Guyan’s reduction

Often the inertial forces in the directions of some coordinates are small and can
be omitted. This allows reducing significantly the number of equations of
motion to be solved. Such a case is common in the FEM, where the kinetic
energy of the rotational degrees of freedom is neglected because it is small
compared to that of the translational degrees.
If the elements of the displacement vector are reordered so that the vector ut
includes the translational degrees of freedom and the vector uq the rotational
ones, then for C ¼ 0 and M diagonal Eq. (11.3.91) is written as

Mtt 0 €t
u Ktt Ktq ut pt
+ ¼ (11.4.1)
0 0 u €q Kqt Kqq uq pq
where the vector pq ðt Þ includes the actions (moment) applied directly to
the nodes.
Obviously, Eq. (11.4.1) yields after performing the multiplications
€t + Ktt ut + Ktq uq ¼ pt ðt Þ
Mtt u (11.4.2a)
Kqt ut + Kqq uq ¼ pq ðt Þ (11.4.2b)
Eq. (11.4.2b) is quasistatic and can be used to eliminate uq from
Eq. (11.4.2a). Τhe matrix Kqq is square and can be inverted. Thus solving
Eq. (11.4.2b) for uq gives
uq ¼ K1
qq ðpq ðt Þ Kqt ut Þ
which is substituted into Eq. (11.4.2a) to yield

€t + Ktt∗ ut ¼ pt∗ ðt Þ
Mtt u (11.4.3)
@CivilMethod
where
Ktt∗ ¼ Ktt Ktq K1
qq Kqt (11.4.4a)
pt∗ ¼ pt Ktq K1
qq pq (11.4.4b)
The matrix Ktt∗ is called the statically condensed stiffness matrix while the
procedure producing it is referred to as the static condensation. Obviously,
when pq ¼ 0, pt∗ ¼ pt .
The reduction of the degrees of freedom can also be achieved when the
mass matrix is not diagonal or the inertial forces in the rotational degrees
of freedom are not negligible. In the employed technique, a subset of the coor-
dinates is selected arbitrarily as the set of active (or “master”) coordinates, say
ut , and the remaining coordinates are dependent (or “slave”) coordinates, say
uq . In this case, the mass matrix of Eq. (11.4.1) is written in the partitioned
form

Mtt Mtq
M¼ (11.4.5)
Mqt Mqq
where now the matrices Mtq , Mqt , Mqq may not vanish.
The starting point is Eq. (11.4.2b), which for pq ðt Þ ¼ 0 becomes
uq ¼ K1
qq Kqt ut (11.4.6)
In the foregoing equation ut represents the independent coordinates and
uq the dependent ones. Thus, we can write

ut I
¼ ut (11.4.7)
uq K1qq Kqt
or
u ¼ Tut (11.4.8)
where

I
T¼ (11.4.9)
K1
qq Kqt
represents the transformation matrix of the independent coordinates ut to u.

By virtue of the transformation (11.4.9), the elastic energy is written
1
U ¼ uT Ku
2
1
¼ uTt TT KTut (11.4.10)
2
1
¼ uTt K ut
2
@CivilMethod
where
K ¼TT KT
(11.4.11)
¼Ktt Ktq K1
qq Kqt
Similarly, the kinetic energy is written

1
K ¼ u_ T Mu_
2
1
¼ u_ Tt TT MTu_ t (11.4.12)
2
1
¼ u_ Tt M u_ t
2
where
M ¼TT MT
" #" #
1
Mtt Mtq I
¼ I Kqt Kqq (11.4.13)
Mqt Mqq K1qq Kqt
¼Mtt Mtq K1 1 1 1

qq Kqt + Kqt Kqq Mqq Kqq Kqt Kqt Kqq Mqt
Obviously, if M is diagonal then Mtq ¼ Mqt ¼ 0 and the foregoing equation

becomes
M ¼ Mtt + Kqt K1 1
qq Mqq Kqq Kqt (11.4.14)
Finally, the virtual work of the external nodal loads is written
dWp ¼duT pðt Þ
¼duTt TT pðt Þ (11.4.15)
¼duTt p ∗ ðt Þ
where
p ∗ ðt Þ ¼TT pðt Þ
(11.4.16)
¼pt ðt Þ K1
qq Kqt pq ðt Þ
This reduction of the degrees of freedom is known as the Guyan reduc-

tion [6]. Obviously, in this reduction, the stiffness of the structure is not
affected, as all elements of the original stiffness matrix contribute to the reduced
stiffness matrix. On the contrary, the reduced mass matrix is affected as it
involves combinations of elements of the mass and stiffness matrix. This tech-
nique reduces the number of equations. Nevertheless, it introduces approxima-
tions, and the reliability of the results depends on the selection of the eliminated
degrees of freedom.
@CivilMethod
Example 11.4.1 Formulate the equation of motion of the frame in Example

11.3.2 after the static condensation of the rotational degrees of freedom.
Solution
The rotational degrees of freedom are u6 , u9 , u12 , will take the places u
e3 , u
e6 , u
e7
in the modified displacement vector euf . Therefore, the elements of the vector
e
uf must be reordered as follows:
8 9 2 38 9
> u1 >> 1 0 0 0 0 0 0 > e1 >
u
>
> >
> > 6 > >
>
7> >
>
>
> u 2>> 6 0 1 0 0 0 0 0 >
7> e
u 2>>
>
> >
> 6 7>
> >
>
> > 6 0 0 0 0 1 0 0 7> e4 >
< u3 >
> = 6 7
>

=
u4 ¼ 6 6 0 0 1 0 0 0 0 7 u
7 e 5
>u >
> > 6 0 0 0 1 0 0 0 7>
> >u >
>
>
> >
> 6 7>
> e 3>>
> > >
>
5
> >
> > 6 7> >
>
>
> u 6>> 4 0 0 0 0 0 1 0 5>
> e
u 6>>
: >
> ; : >
> ;
u7 0 0 0 0 0 0 1 e7
u
which defines the transformation matrix

2 3
1 0 0 0 0 0 0
60 1 0 0 0 0 07
6 7
6 7
60 0 0 0 1 0 07
6 7
6 7
VC ¼ 6 0 0 1 0 0 0 07
6 7
60 0 0 1 0 0 07
6 7
6 7
40 0 0 0 0 1 05
0 0 0 0 0 0 1
and modifies the matrices M e ff and the vector e

e ff , K pf as
2 3
7:5 0 0 0 0 0 0
6 0 0 0 07
6 7:5 0 0 7
6 7
6 0 0 15:0 0 0 0 07
6 7
T e 6 7
M ¼ VC Mff VC ¼ 6 0 0 0 15:0 0 0 07
6 7
6 0 0 0 0 0 0 07
6 7
6 7
4 0 0 0 0 0 0 05
0 0 0 0 0 0 0
@CivilMethod
The implied partitioning of the matrices M, K, and vector pðt Þ in the foregoing
equations yields
2 3
7:5 0 0 0
60 7:5 0 0 7
Mtt ¼ rA6
40
7
0 15 0 5
0 0 0 15
2 3
3246:6 2353:9 3234:6 2353:9
EI 66 2353:9 4361:4 2353:9 1861:4 7,
7
Ktt ¼ 4
125 3234:6 2353:9 4142:2 1159:7 5
2353:9 1861:4 1159:7 3465:7
2 3
114 144 0
EI 6
6 192 192 0 7
7
Ktq ¼ 4 5
125 144 192 48
192 156 36
2 3 2 3
114 192 144 192 900 400 0
EI 4 EI
Kqt ¼ 144 192 192 156 5, Kqq ¼ 4 400 1200 200 5
125 125
0 0 48 36 0 200 400
2 3
2P ðt Þ 2 3
6 0:08P ðt Þ 7 0
pt ðt Þ ¼ 6
4 0
7, pq ðt Þ ¼ 4 0 5
5
0
1:08P ðt Þ
Applying Eqs. (11.4.4a), (11.4.4b) gives (11.4.4a)
2 3
3222:71 2389:02 3205:83 2387:19
EI 66 2389:02 4308:10 2396:65 1810:18 7,
7
Ktt∗ ¼ 4 5
125 3205:83 2396:65 4101:75 1196:10
2387:19 1810:18 1196:10 3412:98
8 9
> 2:0P ðt Þ >
>
> >
>
∗
< 0:08P ðt Þ =
pt ðt Þ ¼
>
> 0 >
>
>
: >
;
1:08P ðt Þ
@CivilMethod
Hence, the equation of motion after the static condensation is

2 38 9
7:5 0 0 > u€1 >
0
6 7>> > >
60 7:5 0 0 7< u€2 =
rA6
60
7
4 0 15 0 7
5>> u€4 >
>
: >
> ;
0 0 0 15 u€5
2 38 9 8 9
2389:02 3205:83 2387:19 > u1 > > 2:0 >
3222:71
6 >
7> > >> > >
>
EI 6 2389:02 4308:10 2396:65 1810:18 7< u2 = < 0:08 =
6 7
+
125 6 7 u > > 0 >P ð t Þ
¼
4 3205:83 2396:65 4101:75 1196:10 5>> 4> > >
>
: >
> ; > : ;
2387:19 1810:18 1196:10 3412:98 u5 1:08
11.5 Flexural vibrations of a plane frame

In a frame structure, the work due to the axial deformation of its elements is
small compared to that due to bending deformation. Thus, in the usual cases,
the axial deformation is omitted in order to reduce the number of equations
of motion to be solved, which is an important issue in the dynamic analysis.e
This omission is particularly convenient in small plane frames with elements
parallel to the global axes (e.g., rectangular frames), where the equations of
motion can be easily derived by hand without recourse to the FEM. This is illus-
trated with the examples below. However, the general problem of neglecting the
axial deformations in frames is discussed in Section 11.7.
In flexural vibrations, the displacement vector results by omitting the
degrees of freedom u1 and u4 in the direction of the element local x axis.
The remaining degrees of freedom are shown in Fig. 11.5.1 after the renumber-
ing. Thus, the displacement vector is written now
u ¼ f u1 u2 u3 u4 gT (11.5.1)
where u1 , u3 are the translational degrees of freedom and u2 , u4 the rotational
ones. Then the reduced stiffness, mass, and damping matrices and the vectors
of the nodal forces of the element will result from the corresponding original
ones by omitting the nodal forces in these directions. Thus, for an element
with a constant cross-section, the stiffness matrix results from Eq. (11.3.17)
by omitting the first and fourth line as well as the first and fourth column. This
yields
e. The omission of the axial deformation is not always permitted because it may lead to considerable
mistakes in certain structures such as high-rise buildings, whose overall deformation pattern resem-
bles that of a cantilever when subjected to horizontal loads. Obviously, this deformation cannot be
realized when the axial deformation of the columns is neglected.
@CivilMethod
Fig. 11.5.1 Degrees of freedom of a beam element with bending deformation.
2 3
12EI 6EI 12EI 6EI

6 L 3 L2 L3 L2 7
6 7
6 6EI 4EI 6EI 2EI 7
6 2 7
6 2 L 7
ke ¼ 6 L L L 7 (11.5.2)
6 12EI 6EI 12EI 6EI 7
6 3 2 7
6 L L L3 L2 7
4 6EI 2EI 6EI 4EI 5
2
L2 L L L
Similarly, the consistent mass matrix is obtained from Eq. (11.3.25) as

2 3
156 22L 54 13L
me 66 22L 4L2 13L 3L2 7 7
me ¼ 6 7 (11.5.3)
420 4 54 13L 156 22L 5
13L 3L2 22L 4L2
and the lumped mass matrix from Eq. (11.3.32) as
2 3
1 0 0 0
me 6
60 0 0 07
7
me ¼ 6 7 (11.5.4)
2 40 0 1 05
0 0 0 0
The damping matrix is obtained from Eq. (11.3.42) as
2 3
12cs I 6cs I 12cs I 6cs I
6 3
6 L3 L2 L L2 7 7
6 6cs I 4c I 6c I 2c sI 7
6 s
2
s 7
6 L2 7
c ¼6
e 6 L L L 7 (11.5.5)
6 2 7
6 L3 L2 L3 L 7 7
2
L2 L L L
@CivilMethod
while the geometric stiffness matrix for constant axial force results from
Eq. (11.3.59) 2 3
12 12
1 1
6 L L 7
6 4L L 7
6 1 1 7
N66 3 3 7
7
kG ¼ 6 12
e
12 7 (11.5.6)
10 6 1 7
6 L 1 L 7
4 L 4L 5
1 1
3 3
Finally, the consistent nodal load vector is obtained from Eq. (11.3.46),
which for a constant load py ðx, t Þ ¼ pe becomes

T
p e Le L L
pe ðt Þ ¼ 1 1 (11.5.7)
2 6 6
and the static equivalent nodal load vector is obtained from Eq. (11.3.49) as
p e Le
f 1 0 1 0 gT
pe ðt Þ ¼ (11.5.8)
2
In the following examples, the formulation of the equation of motion of
certain frames with elements parallel to the global axes is presented. The method
of influence coefficients is employed to establish the matrices of the structure.
Example 11.5.1 Formulate the equation of motion of the structure shown in
Fig. E11.3a. Consider the lumped mass assumption for the column. Examine
the case I0 ¼ 0. Assumed material density r ¼ m=10AL.
Solution
The structure may be viewed as a plane frame consisting of a single element, the
beam column. Obviously, the system has two free degrees of freedom in global
axes: the nodal displacement u1 and the rotation u2 of the top cross-section
(see Fig. E11.3b). Hence, the stiffness matrix of the structure is
(a) (b) (c) (d) (e)

Fig. E11.3 Structure in Example 11.5.1 (a), parameters of motion (b), deformation patterns (c), (d),
and beam element degrees of freedom (e)

k12
¼ k 11
K (1)

k 21 k22
@CivilMethod
The elements of the stiffness matrix K are evaluated by considering the

deformation of the system in global axes. The matrix element kij expresses
the elastic force in the direction of ui , when uj ¼ 1 while the remaining displace-
ments are zero (i, j, ¼ 1, 2). The deformation patterns are shown in Fig. E11.3c
and d. The nodal elastic forces can be evaluated using the stiffness matrix of the
beam element with a constant cross-section, Eq. (11.5.2), when matching
u1 ¼ u1 , u2 ¼ u2 , u3 ¼ u4 ¼ 0 (see Fig. E11.3e). Consequently, we obtain:
With reference to Fig. E11.3c
12EI 6EI
k11 ¼ k11 ¼ 3 , k21 ¼ k21 ¼ 2
ð1Þ ð1Þ
L L
With reference to Fig. E11.3d
6EI 4EI
k12 ¼ k12 ¼ 2 , k22 ¼ k22 ¼
ð1Þ ð1Þ
L L
Obviously, it is not necessary to evaluate the element k21 because the stiff-
ness matrix is symmetric. Thus, we have
2 12EI 6EI 3
¼6
K 4
L3 L2 7
5 (2)
6EI 4EI
L2 L
The mass matrix is

m11 0
M¼ (3)
0 m22
where
ð1Þ
11 ¼ m11 + m ¼ rAL=2 + m ¼ 1:05m and m22 ¼ I0
m
Hence, the mass matrix of the structure is

M ¼ 1:05m 0 (4)
0 I0
The vector of the nodal loads is

pðt Þ
pðt Þ ¼ (5)
M ðt Þ
Therefore, the equation of motion of the structure is

m 11 0 u€1 k11 k12 u1 pðt Þ
+ ¼ (6)
0 22
m u€2 k 21 k 22 u2 M ðt Þ
or after performing the operations
11 u€1 + k11 u1 + k12 u2 ¼ pðt Þ
m (7a)
22 u€2 + k21 u1 + k22 u2 ¼ M ðt Þ
m (7b)
@CivilMethod
22 ¼ I0 ¼ 0 static condensation is applied and Eq. (7b) becomes

When m
k21 u1 + k22 u2 ¼ M ðt Þ (8)
1 1
u2 ¼ k22 M ðt Þ k22 k21 u1 (9)
Then substituting into Eq. (7a) gives
11 u€1 + k11
m ∗
u1 ¼ p ∗ ðt Þ (10)
where
1 3EI
k11
∗
¼ k11 k12 k22 k21 ¼ 3 (11)
L
1 3
p ∗ ðt Þ ¼ pðt Þ k12 k22 M ðt Þ ¼ pðt Þ M ðt Þ (12)
2L
Inserting the values for k11
∗
and p ∗ ðt Þ into Eq. (10) yields the condensed
equation of motion
3EI M ðt Þ
1:2m u€1 + 3
u1 ¼ pðt Þ 1:5 (13)
L L
We observe that the moment M ðt Þ in the direction of u2 adds a force in the
direction of u1 .
Example 11.5.2 Formulate the equation of motion of the plane frame shown
Fig. E11.4 by neglecting the axial deformation and the damping of the elements.
Column 1 is elastically supported on the ground, whose stiffness is simulated by
the rotational spring kR ¼ EI =2L. All elements have a constant cross-section.
Their properties are shown in the figure. The mass matrix will be obtained
using the lumped mass assumption. The rotational degrees of freedom will
be condensed. Assume M ðt Þ ¼ 2P ðt ÞL.
Fig. E11.4 Frame in Example 11.5.2.
@CivilMethod
Solution
Inasmuch as the axial deformation of the elements is neglected, the degrees of
freedom of the structure are reduced to seven, two translational, u1 , u2 , and five
rotational, u3 , u4 , u5 , u6 , u7 . The degrees of freedom are shown in Fig. E11.5.
Fig. E11.5 Degrees of freedom of the frame in Example 11.5.2.
Stiffness matrix. The stiffness matrix will have the form
(a) (b)
(c) (d)
(e) (f)
(g)
Fig. E11.6 Deformation patterns of the frame in Example 11.5.2 due to unit nodal displacements.
@CivilMethod
2 3
k11 k12 k13 k14 k15 k16 k17
6 7
6 k 21 k22 k23 k24 k25 k26 k27 7
6 7
6 k31 k32 k33 k34 k35 k36 k37 7
6 7
¼6
K 6 k41 k42 k43 k44 k45 k46
7
k47 7 (1)
6 7
6 k51 k52 k53 k54 k55 k56 k57 7
6 7
6 7
4 k 61 k62 k63 k64 k65
k66 k67 5
k 71 k 72 k 73 k 74 k 75 k76
k77
Its elements will be computed as the elastic forces at the nodes by giving a
unit value to each displacement and keeping the other displacements equal to
zero, following the method of the influence coefficients discussed in
Section 10.4.1. For the economy of the computation, we will take advantage
of the symmetry of the matrix and restrict the computations to the elements
of the main diagonal and those below it.
The elements of the 1st column of K are computed by referring to
Fig. E11.6a. The displacement u1 ¼ 1 “activates” (deforms) the elements
1 and 3. Their matching with the beam element is shown in Fig. E11.5, namely:
Element 1: u1 ¼ u1 , u2 ¼ 0, u3 ¼ 0, u4 ¼ 0
Element 3: u1 ¼ u1 , u2 ¼ 0, u3 ¼ 0, u4 ¼ 0
Then, with reference to Eq. (11.5.2), we obtain
12EI 12EI 27 EI 12EI
k11 ¼ k11 + k11 ¼
ð1Þ ð3Þ ð3Þ
+ 3 ¼ , k 21 ¼ k31 ¼ 3
ð2LÞ3 L 2 L3 L
6EI 3 EI 6EI 3 EI
k31 ¼ k41 ¼
ð1Þ ð1Þ
¼ , k 41 ¼ k21 ¼ ¼
ð2LÞ2 2 L2 ð2LÞ2 2 L2
6EI 6EI
k51 ¼ k21 ¼ 2 , k61 ¼ k41 ¼ 2 , k71 ¼ 0
ð3Þ ð3Þ
L L

The elements of the 2nd column of K are computed by referring to
Fig. E11.6b. The displacement u2 ¼ 1 activates (deforms) the elements 3, 4,
and 6. Their matching with the beam element is:
Element 3: u1 ¼ 0, u2 ¼ 0, u3 ¼ u2 , u4 ¼ 0
Element 4: u1 ¼ u2 , u2 ¼ 0, u3 ¼ 0, u4 ¼ 0
Element 6: u1 ¼ u2 , u2 ¼ 0, u3 ¼ 0, u4 ¼ 0
12EI 12EI 12EI 36EI
k22 ¼ k33 + k11 + k11 ¼ 3 + 3 + 3 ¼ 3
ð3Þ ð4Þ ð6Þ
L L L L
6EI
k32 ¼ 0, k42 ¼ 0, k52 ¼ k23 ¼ 2
ð3Þ
L
6EI 6EI 6EI
k62 ¼ k43 + k21 ¼ 2 + 2 ¼ 0, k72 ¼ k21 ¼ 2
ð3Þ ð4Þ ð6Þ
L L L
@CivilMethod
The elements of the 3rd column of K are computed by referring to

Fig. E11.6c. The rotation u3 ¼ 1 activates (deforms) only element 1. Its match-
ing with the beam element is:
Element 1: u1 ¼ 0, u2 ¼ 0, u3 ¼ u2 , u4 ¼ 0
4EI EI 5 EI 2EI EI
k33 ¼ k44 + kR ¼
ð1Þ ð1Þ
+ ¼ , k 43 ¼ k24 ¼ ¼
2L 2L 2 L 2L L

k 53 ¼ 0,
k 63 ¼ 0,
k 73 ¼ 0
The elements of the 4th column of K are computed by referring to
Fig. E11.6d. The rotation u4 ¼ 1 activates (deforms) the elements 1 and 2. Their
matching with the beam element is:
Element 1: u1 ¼ 0, u2 ¼ u4 , u3 ¼ 0, u4 ¼ 0
Element 2: u1 ¼ 0, u2 ¼ u4 , u3 ¼ 0, u4 ¼ 0
4EI 4ð8EI Þ 34EI
k44 ¼ k22 + k22 ¼
ð1Þ ð2Þ
+ ¼
2L L L
2ð8EI Þ 16EI
k54 ¼ k42 ¼ k64 ¼ 0, k74 ¼ 0
ð2Þ
¼ ,
L L
Fig. E11.6e. The rotation u5 ¼ 1 activates (deforms) the elements 2 and 3. Their
Element 2: u1 ¼ 0, u2 ¼ 0, u3 ¼ 0, u4 ¼ u5
Element 3: u1 ¼ 0, u2 ¼ u5 , u3 ¼ 0, u4 ¼ 0
4ð8EI Þ 4EI 36EI
k55 ¼ k44 + k22 ¼
ð2Þ ð3Þ
+ ¼
L L L
2EI
k65 ¼ k42 ¼ k75 ¼ 0
ð3Þ
,
L
Fig. E11.6f. The rotation u5 ¼ 1 activates (deforms) the elements 3, 4, and 5.
Their matching with the beam element is:
Element 3: u1 ¼ 0, u2 ¼ 0, u3 ¼ 0, u4 ¼ u6
Element 4: u1 ¼ 0, u2 ¼ u6 , u3 ¼ 0, u4 ¼ 0
Element 5: u1 ¼ 0, u2 ¼ u6 , u3 ¼ 0, u4 ¼ 0
@CivilMethod

4EI 4EI 4ð8EI Þ 88 EI
k66 ¼ k44 + k22 + k22 ¼
ð3Þ ð4Þ ð5Þ
+ + ¼
L L 1:5L 3 L
2ð8EI Þ 32 EI
k76 ¼ k42 ¼
ð5Þ
¼
1:5L 3 L
Finally, the elements of the 7th column of K are computed by referring to
Fig. E11.6g. The rotation u7 ¼ 1 activates (deforms) the elements 5 and 6. Their
Element 5: u1 ¼ 0, u2 ¼ 0, u3 ¼ 0, u4 ¼ u7
Element 6: u1 ¼ 0, u2 ¼ u6 , u3 ¼ 0, u4 ¼ 0
4ð8EI Þ 4EI 76 EI
k77 ¼ k44 + k22 ¼
ð5Þ ð6Þ
+ ¼
1:5L L 3 L
Therefore, the stiffness matrix is
ð2Þ
Mass matrix. The lumped mass assumption for the elements, Fig E11.7a,
concentrates the mass at the nodes of the frame as shown in Fig. E11.7b.
Obviously, inertial forces arise only in the directions of the translational degrees
of freedom u1 and u2 . Thus, we have
ð1Þ ð2Þ ð2Þ ð3Þ

m11 ¼ m1 + m1 + m2 + m1 ¼ 11:5mL
ð3Þ ð4Þ ð5Þ ð5Þ ð6Þ

m22 ¼ m2 + m1 + m1 + m2 + m1 ¼ 16:5mL
Therefore the mass matrix of the structure reads
ð3Þ
@CivilMethod
(a) (b)
Fig. E11.7 Lumped masses of the elements (a) and nodal masses of the frame (b) in Example
11.5.2.
Vector of the external nodal loads. This includes the nodal loads that are
directly applied to the nodes of the frame as well as those resulting from
the contribution of the equivalent nodal loads due to the element loads. In
our problem, only element 6 is loaded. To be consistent with the mass
matrix, the equivalent nodal load will be obtained from Eq. (11.3.51).
Hence 8 9
>
2 >


> =
ð6Þ p1 ðt Þ 3
p ðt Þ ¼ ¼
p3 ðt Þ >
> >
: 1 P ðt Þ >
;
3
Therefore, the vector of the external loads of the structure reads
Static condensation. This includes the elimination of the five rotations

u3 , u4 ,…, u7 . The required partitioning of K, M , and pðt Þ is indicated in
Eqs. (2), (3), (4). Thus, we have
2 3 2 3
27 3 3
tt ¼ EI 4 2 12 5, K
K tq ¼ EI 4 2 L 2 L 6L 6L 0 5
L3 L3
12 36 0 0 6L 0 6L
2 3
2 3 5 2 2
3
L 0 62L L 0 0 0 7
62 7 6 7
6 7 6 2 7
63 7 6L 34L2 16L2 0 0 7
6 L 0 7 6 7
EI 62 7 EI 60 7

Kqt ¼ 3 6 6 7
, Kqq ¼ 3 6 6 16L 2
36L 2
2L 2
0 7
L 6 6L 6L 7 7 L 6 7
88 32 7
6 7 60 0 2L2 L2 L2 7
6 6L 0 7
4 5
6
6 3 3 7 7
4 32 2 76 2 5
0 6L 0 0 0 L L
3 3
@CivilMethod
8 9
8 9 >
> 0 >
>
P ðt Þ = >
> >
>

mL < < 0 =
33 0
Mtt ¼ , pt ðt Þ ¼ 2 , pq ðt Þ ¼ 0
2 0 23 : P ðt Þ ; >
> >
>
>
> 0 >
>
3 : ;
2P ðt ÞL
Applying Eqs. (11.4.4a), (11.4.4b) give
" #

EI 10:209 10:344 0:546
K tt ¼
∗
,
p ∗ ðt Þ ¼ P ðt Þ
t
L3 10:344 33:131 0:837
Therefore the equation of motion reads

33 0
mL EI 10:209 10:344 0:546
+ 3 ¼ P ðt Þ
2 0 23 L 10:344 33:131 0:837
11.6 Reduction of the degrees of freedom due to constraints

The number of degrees of freedom for the dynamic analysis of plane and space
frames may be considerably reduced by introducing axial constraints as a
consequence of ignoring the axial deformation of the elements. Before studying
this particular problem, we will discuss the following problem.
We assume that the equations of motion have been derived with respect to
N generalized coordinates u, that is,
M€
u + Cu_ + Ku ¼pðt Þ (11.6.1)
and that the coordinates u are not independent, but they are constrained
by K < N linear constraint equations
Du ¼0 (11.6.2)
where D is a K N rectangular matrix.
The problem is to determine L ¼ N K independent equations of motion
with respect to L independent generalized coordinates, which will be solved
to obtain the dynamic response of the system.
We cope with this problem by proceeding as follows.
We choose L independent displacements (coordinates) and we partition the
displacement vector as

u
u ¼ a (11.6.3)
ub
where ua are the independent and ub the dependent displacements. Then we
write Eq. (11.6.2) as
@CivilMethod

u
½ Da Db a ¼0 (11.6.4)
ub
or
Da ua + Db ub ¼0 (11.6.5)
which gives
ub ¼ D1 a
b Da u (11.6.6)
From Eq. (11.6.6), we conclude that the selection of the independent
coordinates cannot be arbitrary, but they must be chosen so that the inversion
of the matrix Db is ensured. This is accomplished, as will be shown in
Section 11.7, by the Gauss-Jordan elimination method. Generally, after select-
ing the independent coordinates, it is necessary to rearrange the elements of
the vector u to comply with the partition suggested by Eq. (11.6.3). Then
M u€ + K
u€ + C u ¼
pðt Þ (11.6.7)
where M, K,
C, u, and pðt Þ are the modified matrices and vectors after the
rearrangement.
Eq. (11.6.3) can be further written

I
u ¼ ua (11.6.8)
D1
b Da
or
u ¼T
ua (11.6.9)
where I is the unit matrix with dimensions L L and

I
T¼ (11.6.10)
D1b Da
Eq. (11.6.9) expresses a linear transformation, which transforms the

independent coordinates ua into u. This transformation can be employed to
reduce the number of Eq. (11.6.1).
Indeed, the kinetic energy of the system is
1 T
T ¼ u_ M u_ (11.6.11)
2
1 T u_ a
T ¼ u_ a TT MT
2
1 T^
¼ u_ a M u_ a (11.6.12)
2
@CivilMethod
where
^ ¼ TT MT
M (11.6.13)
is the L L reduced transformed mass matrix.

is partitioned as indicated by the vectors ua and ub , we have
If the matrix M
" #
ab
aa M
M
¼
M (11.6.14)
ba M
M bb

" #( )
n
^ 1 T o Maa Mab I
M ¼ I Db Da
M ba M
bb D1b Da (11.6.15)

¼M ab D1 Da D1 Da T M
aa M ba + D1 Da T M
bb D1 Da
b b b b
Similarly, the elastic energy of the system is
1
U ¼ uT K
u
2
1 ua
¼ uTa TT KT (11.6.16)
2
1 ^
¼ uTa K
ua
2
where
K
^ ¼ TT KT (11.6.17)
is the L L transformed reduced stiffness matrix.
as M
Partitioning K in Eq. (11.6.14) we obtain from Eq. (11.6.17)
" #( )
K aa K
ab I
^¼ I
K D1
b Da
ba K
K bb D1 b Da (11.6.18)

aa K
¼K ab D Da D Da K
1 1 T
ba + D1 Da T K
bb D1 Da
b b b b
@CivilMethod
The transformed damping matrix is established by considering the virtual

work of the nodal damping forces
dWD ¼ d
uT f D
uT Cu_
¼ d
(11.6.19)
u_ a
uTa TT CT
¼ d
^ u_ a
uTa C
¼ d
Hence, we have
C
^ ¼ TT CT (11.6.20)
Similarly, we establish the transformed vector of the external loads that is,
uT pðt Þ
dWp ¼d
uTa TT pðt Þ
¼d (11.6.21)
^ ðt Þ
uTa p
¼d
Hence
^ðt Þ ¼TT pðt Þ
p
T (11.6.22)
pa ðt Þ D1
¼ b Da pb ðt Þ
11.7 Axial constraints in the plane frame

In Section 11.5, we examined the flexural vibrations of plane frames. We saw
that the mass, stiffness, and damping matrices of the elements have dimensions
4 4 if the axial deformation is ignored. Using these matrices in Example
11.5.2, we formulated the equations of motion of a plane frame with elements
parallel to the global axes. Obviously, the procedure we used is tedious even for
simple frames having elements parallel to the global axes because its implemen-
tation requires the inspection of the user, which, however, does not facilitate the
automation of the method. The difficulty becomes practically insurmountable
when the frame has elements inclined with respect to the global axes. Neglect-
ing the axial deformation implies the equality of the displacements u1 and u4
in local axes (see Fig. 11.7.1). This annuls the nodal elastic and damping forces
in those directions, ðfS1 ¼ fS4 ¼ 0, fD1 ¼ fD4 ¼ 0Þ, but not the inertial forces
ðfI 1 , fI 4 6¼ 0Þ. This becomes evident by setting u1 ¼ u4 in Eqs. (11.3.14),
(11.3.22), (11.3.39), and taking into account Eqs. (11.3.13a), (11.3.21),
(11.3.37a), respectively. The inertial behavior of the element in the direction
of its local x axis can be simulated with that of a one-dimensional rigid body
within the frame. To tackle this problem, we must start by considering the frame
elements with the six degrees of freedom.
@CivilMethod
Fig. 11.7.1 Degrees of freedom of an element with flexural vibrations, u4 ¼ u1 .
We consider the beam element with its six degrees of freedom with respect
to global axes, as shown in Fig. 11.7.1. If we want to study only the flexural
vibrations of the frame, we must neglect the axial deformation of the elements.
This entails the imposition of constraints on the translational degrees of free-
dom. These constraints result from the fact that the projections of end displace-
ments on the element axis must be equal [7,8], that is,
u1 cos f + u2 sin f ¼ u4 cos f + u5 sin f (11.7.1)
The degrees of freedom of a frame with n nodes are N ¼ 3n, of which

Nt ¼ 2n are translational degrees of freedom. However, these degrees are not
independent because they are subject to constraints of the form (11.7.1). The
maximum number of constraints is at first sight Kmax ¼ Ne , where Ne is the
number of the elements of the frame. Therefore the independent translational
degrees of freedom are 2n Ne 0. Obviously, this is true if the layout of
the elements is such that each of them constrains only one degree of freedom.
Nevertheless, this is not always true as one can readily understand from the
frames of Fig. 11.7.2. More specifically, the frame of Fig. 11.7.2a has Nt ¼ 4
translational degrees of freedom. The three elements impose three axial con-
straints, that is, it is Nc ¼ Ne ¼ 3, where Nc denotes the number of constraints.
Thus the remaining translational degrees are: Nt Nc ¼ 4 3 ¼ 1. In the frame
of Fig. 11.7.2b, the four elements impose four axial constraints, that is
Nc ¼ Ne ¼ 4, hence Nt Nc ¼ 4 4 ¼ 0, which means that all translational
degrees of freedom are constrained. Further, in the frame of Fig. 11.7.2c, the
fifth element is redundant because the translational degrees of freedom are
again four, Nc ¼ 4, while it is Ne ¼ 5. In this case, we say that the frame is over-
constrained. The frame of Fig. 11.7.2d has four elements, that is, Ne ¼ 4. How-
ever, the number of constraints is three, Nc ¼ 3, because, in this layout, the
elements 3 and 4 lie on a straight line and consequently cannot constrain the
normal displacement un . Finally, the number of constraints in the frame of
Fig. 11.7.2e is Nc ¼ 4, even it is Ne ¼ 5.
From the previous examples, it becomes apparent that the number of ele-
ments Ne cannot determine the number of active axial constraints. In simple
@CivilMethod
frames, this can be done by inspection. However, a general procedure should be

developed that would allow determining automatically the number of
active constraints, rejecting the redundant ones and separating the independent
degrees of freedom from the dependent ones. This procedure is presented
right below.
The constraint equation (11.7.1) for the typical element e starting at the point
j and ending at the point k is written
u3j2 cos fe + u3j1 sin fe u3k2 cos fe u3k1 sin fe ¼ 0 (11.7.2)
which for e ¼ 1, 2, …, Ne yields
ut ¼ 0
D (11.7.3)
where D is an Ne Nt matrix and ut the vector of the translational degree of
freedom. Matrix D can be formulated by establishing appropriate assembly
matrices.
As we will show below, the rank of the matrix D will determine the active
constraints.

u
½ Da Db a ¼0 (11.7.4)
ub
or
Da ua + Db ub ¼0 (11.7.5)
where ua the independent and ub the dependent displacements.
Eq. (11.7.5) yields
ub ¼ D1 a
b Da u (11.7.6)
The foregoing equation points out that the selection of ub cannot be arbi-
trary, but such to ensure invertibility of the matrix Db , that is, det ðDb Þ 6¼ 0.
Hence the dimensions of Db , which is a submatrix of D, cannot be greater
than the rank of D. The determination of the rank can be achieved using the
Gauss-Jordan elimination [9]. This method is a variation of the Gauss elimina-
tion. The main difference is that by the elimination procedure, zeros are gener-
ated both below and above each pivot element. Besides, all rows are normalized
via division by the pivoting elements. Therefore, the resulting matrix that
determines the rank of the constraint matrix is a unit matrix and not an upper
diagonal matrix. This matrix is also known as the reduced row echelon form.
There are ready-to-use computer functions that transform a matrix to this form,
for example, the function R ¼ rref (A) of MATLAB transforms the matrix A
into a reduced row echelon matrix R.
@CivilMethod
(a) (b)
(c) (d)
(e)
Fig. 11.7.2 Frames with axial constraints.
This method leads to an automatic selection of the independent node transla-

tions. The procedure is better understood by applying it to the frames of Fig. 11.7.2.
Frame of Fig. 11.7.2a
The Cartesian coordinates of the nodes are 1 ð0, 0Þ, 2 ð0, 3Þ, 3 ð4, 4Þ, 4 ð6, 0Þ.
The computed geometrical data of the elements are shown in Table 11.7.1.
@CivilMethod
TABLE 11.7.1 Geometrical data of the elements of the frame in Fig. 11.7.2a.
Number of
element e Node j Node k D
x D
y Le cosfe sin fe
1 1 2 0 3 3 0 1
2 2 3 4 1 4.123 0.970 0.243
3 3 4 2 4 4.472 0.447 0.894
The constraint equations are three, namely

8 9
2 3> u1 > 8 9
> >
0 1 0 0 < >
> = <0=
6 7 u2
4 0:970 0:243 0:977 0:243 5 ¼ 0 (11.7.7)
> u4 >
> > : ;
0 0 0:447 0:894 > : > ; 0
u5
Hence
2 3
0 1 0 0
6 7
D ¼ 4 0:970 0:243 0:970 0:243 5 (11.7.8)
0 0 0:447 0:894
Gauss-Jordan elimination. Before starting the elimination, it is convenient

to interchange the first with second row. Thus, we have
2 3
0:970 0 0:970 0:243
40 1 0 0 5
0 0 0:447 0:894
Dividing the second row by 0:970 gives

2 3
1 0 1 0:250
40 1 0 0 5
0 0 0:447 0:894
Then, dividing the third row by 0:447 gives

2 3
1 0 1 0:250
40 1 0 0 5
0 0 1 2:000
@CivilMethod
Finally, adding the third row to the first row gives

2 3
1 0 0 2:225
40 1 0 0 5 (11.7.9)
0 0 1 2:000
Therefore, the rank of the matrix D is 3, hence Nt ¼ 4, Nc ¼ 3, N ¼ 4 3 ¼ 1.
We can take u1 as the independent. Thus, referring to Eq. (11.7.7) we have
2 3
1:000 0 0
Db ¼ 4 0:243 0:977 0:243 5 (11.7.10a)
0 0:447 0:894
2 3
0
Da ¼ 4 0:970 5 (11.7.10b)
0
Note that we can also use Eq. (11.7.9) instead of Eq. (11.7.7) to obtain
2 3
1 0 0
Db ¼ 4 0 1 0 5 (11.7.11a)
0 0 1
2 3
0
Da ¼ 4 0:856 5 (11.7.11b)
0:572
because Eq. (11.7.7) are homogeneous and the Gauss-Jordan elimination does
not change the order of the unknowns.
Apparently, performing Gauss-Jordan elimination by inspection is a tedious
task and it becomes too complicated for large matrices. However, computer
codes have been developed to apply this technique and ready-to-use programs
are available.
Frame of Fig. 11.7.2b
The Cartesian coordinates of the nodes are 1 ð0, 0Þ, 2 ð0, 3Þ, 3 ð4, 4Þ, 4 ð6, 0Þ,
5ð4, 6Þ. The computed geometrical data of the elements are shown in
Table 11.7.2.
TABLE 11.7.2 Geometrical data of the elements of the frame in Fig. 11.7.2b.
Number of
x D
y Le cosfe sinfe
1 1 2 0 3 3 0 1
2 2 3 4 1 4.123 0.970 0.243
3 3 4 2 4 4.472 0.447 0.894

4 3 5 0 2 2 0 1
@CivilMethod
The constraint equations are four, namely

2 38 9 8 9
0 1:000 0 0 > u1 >
> > > >0> >
6 0:970 0:243 0:970 0:243 7< u2 = < 0 =
6 7 ¼ (11.7.12)
40 0:447 0:894 5>
0 > u4 >
: > >
;
0>
: >
> ;
0 0 0 1 u5 0
Hence
2 3
0 1:000 0 0
6 0:970 0:243 0:970 0:243 7
D¼6
40
7 (11.7.13)
0 0:447 0:894 5
0 0 0 1
The Gauss-Jordan elimination gives
2 3
1 0 0 0
60 1 0 07
D¼6 40 0
7 (11.7.14)
1 05
0 0 0 1
the rank of which is 4, thus Nt ¼ 4, Nc ¼ 4, N ¼ 4 4 ¼ 0. Therefore, there is no
translational independent variable, which means that the number of active con-
strains is equal to the number of degrees of freedom.
Frame of Fig. 11.7.2c
5ð4, 6Þ, 6ð6, 4Þ. The computed geometrical data of the elements are shown in
Table 11.7.3.
TABLE 11.7.3 Geometrical data of the elements of the frame in Fig. 11.7.2c.
Number of
x D
y Le cos fe sinfe
1 1 2 0 3 3 0 1
2 2 3 4 1 4.123 0.970 0.243
3 3 4 2 4 4.472 0.447 0.894

4 3 5 0 2 2 0 1
5 3 6 2.5 0 2.5 1 0
The constraint equations are five, namely

2 3 8 9
0 1:000 0 0 8 9 >0>
> >
6 0:970 0:243 0:970 0:243 7> > u1 >
> >>0> >
6 
7> = >< > =
6 7 2
60 0 0:447 0:894 7 ¼ 0 (11.7.15)
6
1:000 5> > >
7> u4 > >0> >
40 0 0 : >
> ; >> > >
u5 : >
> ;
0 0 1:000 0 0
@CivilMethod
Hence
2 3
0 1:000 0 0
6 7
6 0:970 0:243 0:970 0:243 7
6 7
D¼6
60 0 0:447 0:894 7
7 (11.7.16)
6 7
40 0 0 1:000 5
0 0 1:000 0

2 3
1 0 0 0
60 1 0 07
6 7
D¼6
60 0 1 077 (11.7.17)
40 0 0 15
0 0 0 0
the rank of which is 4. Therefore, there is no translational independent variable.

Apparently, the fifth constraint is redundant.
Frame of Fig. 11.7.2d
5ð3, 6Þ. The computed geometrical data of the elements are shown in
Table 11.7.4.
TABLE 11.7.4 Geometrical data of the elements of the frame in Fig. 11.7.2d.
Number of
x D
y Le cosfe sin fe
1 1 2 0 3 3 0 1
2 2 3 4 1 4.123 0.970 0.243
3 3 4 2 4 4.472 0.447 0.894
4 3 5 1 2 2.236 0.447 0.894
The constraint equations are four, namely

2 38 9 8 9
u
> 1>
> > > > >
0 1:000 0 0 0
>
6 0:970 0:243 0:970 0:243 7> = > <0>=
6 7 2
6 7 ¼ (11.7.18)
40 0 0:447 0:894 5> > > >
u4 > >0>>
: >
> ; > : >;
0 0 0:447 0:894 u5 0
@CivilMethod
Hence
2 3
0 1:000 0 0
6 0:970 0:243 0:970 0:243 7
6 7
D¼6 7 (11.7.19)
40 0 0:447 0:894 5
0 0 0:447 0:894

2 3
1 0 0 2:250
60 1 0 0 7
6 7
D¼6 7 (11.7.20)
4 0 0 1 2:000 5
0 0 0 0
the rank of which is 3. Therefore, there is one independent displacement, which

means that the number of active constraints is three.
Frame of Fig. 11.7.2e
5ð3, 6Þ, 6(6,4). The computed geometrical data of the elements are shown
in Table 11.7.5.
TABLE 11.7.5 Geometrical data of the elements of the frame in Fig. 11.7.2e.
Number of
x D
y Le cos fe sinfe
1 1 2 0 3 3 0 1
2 2 3 4 1 4.123 0.970 0.243
3 3 4 2 4 4.472 0.447 0.894
4 3 5 1 2 2.236 0.447 0.894

5 3 6 2 0 2 1 0
The constraint equations are five, namely

8 9
2 38 9 > > 0>>
0 1:000 0 > u >0 >
> >
>
>
>
1 >
> >
> >
>
6 7>
> >
> >
> 0 >
>
6 0:970 0:243 0:970 0:243 7> > > >
6 7< u2 = < >
> > > =
6 7
60 0 0:447 0:894 7 ¼ 0 (11.7.21)
6 7>
> > >
> > > >
60
4 0 0:447 0:894 7 > u4 >
>
5> > >
> >
>
>
>
>
>
> >
> >
> 0 >
>
: ; > > >
>
0 0 1 0 u5 : >
> ;
0
@CivilMethod
Hence
2 3
0 1:000 0 0
6 0:970 0:243 0:970 0:243 7
6 7
D¼6
60 0 0:447 0:894 7
7 (11.7.22)
40 0 0:447 0:894 5
0 0 1 0
2 3
1 0 0 0
60 1 0 07
6 7
D¼6 60 0 1 077 (11.7.23)
40 0 0 15
0 0 0 0
the rank of which is 4. Therefore, there is no translational independent variable.
Apparently, the fifth constraint is redundant.
Example 11.7.1 Formulate the equations of motion of the frame in Example
11.3.2, when the axial constraints of the elements are taken into account.
Fig. E11.8 The translational degrees of freedom.
Solution
We note that in this example, two elements of the frame are not parallel to the
global axes. Therefore, the procedure for flexural vibrations described in
Example 11.5.2 is not convenient to formulate the equation of motion of the
structure. Consequently, the general procedure for the FEM analysis of a plane
frame should be employed.
First, the mass and stiffness matrices and the vector of the nodal loads
with respect to the seven free degrees of freedom are computed. Then, the rota-
tional degrees of freedom u3 , u6 , u7 are eliminated by static condensation. This
results in a system with four translational degrees of freedom (Fig. E11.8).
@CivilMethod
Subsequently, the three axial constraints are imposed as described in

Section 11.7, which leads to the elimination of three translational displace-
ments. Thus, the final equation will have only one degree of freedom. The rel-
evant matrices have been computed in Example 11.4.1, namely
2 3 2 3
7:5 0 0 0 3222:71 2389:02 3205:83 2387:19
60 0 7 EI 6 7
6 7:5 0 7 6 2389:02 4308:10 2396:65 1810:18 7
Mtt ¼ rA6 7, Ktt∗ ¼ 6 7
40 0 15 0 5 125 4 3205:83 2396:65 4101:75 1196:10 5
0 0 0 15 2387:19 1810:18 1196:10 3412:98
8 9
>
> 2:0P ðt Þ >>
>
< 0:08P ðt Þ >
=
∗
pt ðt Þ ¼
>
> 0 >
>
>
: >
;
1:08P ðt Þ
The translational displacements are shown in Fig. E11.8. Obviously, the

active constraints are three. Nevertheless, we apply the procedure described
in Section 1.7.
TABLE E11.3 Geometrical data of the elements.
Number of
Element e Node j Node k D
x D
y Le cos fe sinfe
1 1 2 0 5 5 0 1
2 2 3 4 5 5 0.8 0.6
3 3 4 6 8 10 0.6 0.8
On the base of the geometrical data shown in Table E11.3, the constraint
equations are
8 9 8 9
2 3> u1 > > 0 >
> > > >
0 1:0 0 0 < >
> = > <0>=
6 7 u2
4 0:8 0:6 0:8 0:6 5 ¼
> >
> u4 >
> >0>>
0 0 0:6 0:8 > : > ; > : >;
u5 0
hence
2 3
0 1:0 0 0
6 7
D ¼ 4 0:8 0:6 0:8 0:6 5
0 0 0:6 0:8
Using the Gauss-Jordan Elimination, we find out that the rank of the
matrix D is 3. Thus, the number of independent degrees of freedom is equal
@CivilMethod
to 4 3 ¼ 1. Taking u1 as the independent displacement, we can write the

constraint equations as
2 3 2 38 9 8 9
0 1:0 0 0 <ue2 = < 0 =
e1 + 4 0:6 0:8 0:6 5 u
4 0:8 5 u e4 ¼ 0
: ; : ;
0 0 0:6 0:8 e5
u 0
hence
2 3 2 3
0 1:0 0 0
Da ¼ 4 0:8 5, Db ¼ 4 0:6 0:8 0:6 5
0 0 0:6 0:8
Consequently, Eq. (11.6.10) gives
8 9
>
>
1 >
>

<0 >
> =
I
T¼ ¼
D1
b Da
>
>
> 0:64 >
>
>
: ;
0:48
The matrices Mtt , Ktt∗ , pt∗ ðt Þ are modified due to the axial constraints
by virtue of Eqs. (11.6.13), (11.6.17), (11.6.22). Thus, we have
M ¼ TT Mtt T ¼ 17:10rA
K ¼ TT Ktt∗ T ¼ 0:2306EI
pðt Þ ¼ TT pt∗ ðt Þ ¼ 1:4816P ðt Þ
Therefore the equation of motion of the frame is
17:10rAu€1 + 0:2306EI u1 ¼ 1:4816P ðt Þ
From the solution of the foregoing equation, we determine the displacement
e1 . Subsequently, the remaining translational displacements are computed from
u
the relation
e2 u
ub ¼ f u e5 gT ¼ T
e4 u u1
Then the rotations are computed from Eq. (11.4.2b)
uq ¼ K1
qq Kqt ut
11.8 The finite element method for the plane grid

11.8.1 Properties of the plane grid element
We consider the grid of Fig. 11.8.1, whose plane coincides with the global x y
plane. The axis of the e element defines its local x axis. The local y axis lies in
the x y plane while the local z axis coincides with the global axis z
@CivilMethod
(Fig. 11.8.2). The element cross-sections are symmetric with respect to the xz
plane. The nodal loads consist of forces in the z direction and of moments
about x and y axes (Fig. 11.8.1). The elements may be subjected to loads
normal to the xy plane and moments about x and y axes. Moreover, the axial
deformation of the elements is ignored. Under these assumptions, the nodes of
the grid undergo rotations about the global axes x, y and a translational
displacement normal to the x y plane. These global displacements produce
the rotations u1 , u4 about the local x axis, the rotations u2 , u5 about the local
y axis, and the translational displacements u3 , u6 in the direction of the z axis
at the ends of the element (see Fig. 11.8.2).
Thus the element displacement vector is
u ¼ f u1 u2 u3 u4 u5 u6 g (11.8.1)
The displacements u1 , u4 produce torsion of the element while the dis-
placements u2 , u3 , u5 , u6 produce bending in x z plane. The bending deforma-
tion of the element is given by Eq. (11.3.3) after changing appropriately the
role of the shape functions i ðx Þ to correspond to the nodal displacements.
Thus, we have
v ðx, t Þ ¼ u2 2 ðx Þ + u3 3 ðx Þ + u5 5 ðx Þ + u6 6 ðx Þ (11.8.2)
where now

2 ðx Þ ¼ L x 2x 2 + x 3 (11.8.3a)
3 ðx Þ ¼ 1 3x
2
+ 2x 3 (11.8.3b)

5 ðx Þ ¼ L x 2 + x3 (11.8.3c)
6 ðx Þ ¼ 3x 2x3
2
(11.8.3d)
Fig. 11.8.1 Plane grid.
@CivilMethod
Fig. 11.8.2 Nodal displacements of the grid element.
The sign of 2 ðx Þ and 5 ðx Þ has changed because the directions of the

corresponding displacements have changed with reference to Fig. 11.3.1.
In order to determine the torsional deformation of the element, we consider ele-
ments of constant cross-section. Regarding elements of variable cross-section,
we approximate them by a series of rigidly interconnected elements of constant
cross-section. This avoids the use of advanced methods employed for torsional
analysis to elements of variable cross-section.
The rotation qðx, t Þ about the x due to the rotations u1 and u4 is written in the
form
qðx, t Þ ¼ u1 1 ðx Þ + u 4 4 ðx Þ (11.8.4)
The shape functions 1 ðx Þ and 4 ðx Þ can be obtained from the solution of
the equilibrium equation of beam with constant cross-section subjected to the
end rotations u1 , u4 . Thus, referring to Fig. 11.8.3, we have
Fig. 11.8.3
dM t
¼0 (11.8.5)
dx
where Mt is the twisting moment given by
dq
Mt ¼ GI t (11.8.6)
dx
in which G ¼ E=2ð1 + n Þ is the shear modulus and It the torsional constant.
@CivilMethod

d 2q
¼0 (11.8.7)
dx 2
q ¼ c1 x + c2 (11.8.8)
The boundary conditions for Eq. (11.8.7) are qð0Þ ¼ u1 and qðLÞ ¼ u4 , which
are applied to Eq. (11.8.8) and give the arbitrary constants
u4 u1
c2 ¼ u4 , c1 ¼
L
Introducing these values in Eq. (11.8.8) yields
qðx, t Þ ¼ u1 ð1 xÞ + u4 x, x ¼ x=L (11.8.9)
Therefore, the shape functions defined by Eq. (11.8.4) are
1 ðx Þ ¼ 1 x, 4 ðx Þ ¼ x (11.8.10)
The value of It obtained on the base of the Saint-Venant theory depends on
the shape of the cross-section of the bar. For a circular cross-section, It is equal
to the polar moment of inertia with respect to its center. For a cross-section of
arbitrary shape, It is calculated from the expression [4]
Z
2
It ¼ y + z 2 zfy + yfz dydz (11.8.11)
A
where f ¼ fðy, z Þ is the warping function of the cross-section with respect to

the twist center and is obtained from the solution of the following boundary
value problem
∂2 f ∂2 f
+ ¼0 y, zA (11.8.12a)
∂y 2 ∂z 2
∂f
¼ zn y yn z y, zG (11.8.12b)
∂n
in which A is the cross-section of the bar and G its boundary; ny , nz are the
components of the unit vector normal to the boundary of the cross-section.
For the solution of the torsion problem, the reader is advised to look in the
literature. A detailed study of the Saint-Venant torsion problem is presented
in [4]. The Saint-Venant theory under certain support conditions, which
approach the Saint-Venant assumptions, gives acceptable results. This, how-
ever, does not always hold in actual structures. It is well known that the response
of a beam under general twisting loading and general boundary conditions leads
to the problem of nonuniform torsion. This problem has been extensively exam-
ined in the literature, for example, [10, 11].
@CivilMethod
Fig. 11.8.4 Positive directions of the nodal forces of the grid element.
The equivalent nodal forces of the grid element will be obtained using the
Lagrange equations. Fig. 11.8.4 shows the positive directions of element nodal
forces. The actions in the directions of u1 , u2 , u4 , and u5 are moments while in
the directions of u3 and u6 are forces.
(i) Nodal elastic forces and stiffness matrix of the grid element
The elastic energy is due to the torsional deformation of the element given
by the expressionf
1
Ut ¼ GI t ðq0 Þ L
2
(11.8.13)
2
as well as to the bending deformation of the element given by Eq. (10.5.2),
that is,
f. The elastic energy due to torsion is given by

Z
1
Ut ¼ G g 2yx + g 2zx dV , y,zA (1)
2 V

g yx ¼ q0 fy z , g zx ¼ q0 ðfz + y Þ (2)
and that q0 ¼ dq=dx is constant, Eq. (1) becomes

Z Z
1
Ut ¼ Gq0 L
2
f2y + f2z zfy + yfz dA + zfy + yfz + y 2 + z 2 dA (3)
2 A A
The second integral is equal to It while the first one is equal to zero. Indeed, the first integral is
transformed as
Z Z h i
R¼ f2y + f2z dA + ðzfÞy + ðyfÞz dA
A
Z A
Z Z
(4)
¼ f fyy + fzz dA + ffn dA f zn y yn z dA
A A A
¼0
The foregoing result is obtained by applying the Gauss-Green theorem to the first term and the
Gauss divergence theorem to the second term and taking into account that fyy + fzz ¼ 0 and
fn ¼ zn y yn z . Hence, Eq. (3) becomes
1 1
Ut ¼ GI t ðq0 Þ L ¼ Mt q0 L
2
(5)
2 2
@CivilMethod
Z L
1
EI ½v 00 ðx, t Þ dx
2
Ub ¼ (11.8.14)
2 0
Thus, the total elastic energy is

U ¼Ut + Ub
Z L (11.8.15)
1 1
¼ GI t ðq0 Þ L + EI ½v 00 ðx, t Þ dx
2 2
2 2 0

1
U ðu1 , u2 , …, u6 Þ ¼ GI t u1 01 ðx Þ + u4 04 ðx Þ L
2
Z
1 L 2
+ EI u2 002 ðx Þ + u3 003 ðx Þ + u5 00 00
5 ðx Þ + u6 6 ðx Þ dx
2 0
(11.8.16)
Differentiation of the foregoing equation with respect to ui , i ¼ 1, 2, …, 6
yields the nodal elastic forces. Thus, we obtain
8 9 2 38 9
> fS1 > k11 0 0 k14 0 0 > u1 >
> >
> > >
>
> fS2 >
> 6 0 k k 0 k k 7> >u > >
>
>
> >
> 6 22 23 25 26 7>>
> 2>>
>
< f = 6 0 k k 0 k k 7
> > 6 7 > =
¼6 7
S3 32 33 35 36 3
6 7 (11.8.17)
> f > 6 k41 0 0 k44 0 0 7> u >
> S4 >
>
> > 6
> 7 > 4>
>
> >
>
>
>
> fS5 >
>
> 4 0 k52 k53 0 k55 k56 5> >
> u5 >
>
>
>
: >
; : >
> ;
fS6 0 k62 k63 0 k65 k66 u6
in which
Z L
0 0
kij ¼ GI t i ðx Þ j ðx Þdx, i, j ¼ 1, 4 (11.8.18a)
0
Z L
00 00
kij ¼ EI i ðx Þ j ðx Þdx, i, j ¼ 2, 3, 5, 6 (11.8.18b)
0
or
f eS ¼ ke ue (11.8.19)
where u are the vectors of the nodal elastic forces and nodal displacements,
f eS , e
respectively, and ke the stiffness matrix of the e grid element. Further, substitut-
ing Eqs. (11.8.3a)–(11.8.3d) and (11.8.10) into Eqs. (11.8.18a), (11.8.18b) and
performing the integration gives
@CivilMethod
2 3
GI t GI t
6 L 0 0 0 7 0
6 L 7
6 4EI 6EI 2EI 6EI 7
6 0 2 0 7
6 L L L L2 7
6 7
6 6EI 12EI 6EI 12EI 7
6 0 2 0 2 3 7
6 L3 L 7
ke ¼ 6
6 GI t
L L 7
7 (11.8.20)
6 GI t 7
6 L 0 0 0 0 7
6 L 7
6 6EI 7
6 0 2EI

6EI 4EI 7
6 0 7
6 L L2 L L2 7
4 6EI 12EI 6EI 12EI 5
0 3 0
L2 L L2 L3
The superscript e has been dropped from Ite , I e , and Le for the sake of the
simplicity of the expressions.
(ii) Nodal inertial forces and mass matrix of the grid element
The equivalent inertial nodal forces result again on the base of two different
assumptions regarding the mass distribution along the length of the element,
namely the consistent mass assumption and the lumped mass assumption. The
inertial mass matrices resulting from both assumptions are derived right below.
During the motion, the infinitesimal mass m ðx Þdx of the element undergoes the
two displacements, that is, the rotation qðx, t Þ and the transverse displacement
v ðx, t Þ. Therefore, the kinetic energy of the grid element will be given by the
expression
Z
1 Ln 2 o
T¼ I0 ðx Þ q_ ðx, t Þ + m ðx Þ½v_ ðx, t Þ2 dx (11.8.21)
2 0
where I0 ðx Þ is the polar moment of inertia of the mass m ðx Þ ¼ rAðx Þ with
respect to the axis of the element, which is assumed to coincide with the
twist axis.
Using Eqs. (11.8.2), (11.8.4), Eq. (11.8.21) becomes
Z h i
1 L
T ðu_ 1 , u_ 2 , …, u_ 6 Þ ¼ I0 ðx Þ ½u_ 1 1 ðx Þ + u_ 4 4 ðx Þ2 dx
2 0
Z
1 L
+ m ðx Þ½u_ 2 2 ðx Þ + u_ 3 3 ðx Þ + u_ 5 5 ðx Þ + u_ 6 6 ðx Þ2 dx
2 0
(11.8.22)
@CivilMethod
The nodal inertial forces result from Eq. (11.2.9) for i ¼ 1, 2, …, 6. Thus,
after performing the differentiations, we obtain
8 9 2m 0 38 9
>
>
fI 1 >
>
11 0 m14 0 0 > u€1 >
>
>
> >
> 6 7>>
> >
>
>
> f >
>
I2 >
6 0 m m 0 m m 7 >
> u€2>
>
> > 6 22 23 25 26 7 > >
>
> >
> > 6 7>> >
> >
>
< fI 3 = 6 0 m32 m33 0 m35 m36 7 u€3 =
7 <
6
¼6 7 (11.8.23)
>
>
> fI 4 >
>
>
6 m41 0
6 0 m44 0 0 7 7
>
>
> u€4 >
>
>
>
> >
> 7>> >
>
> > 6 > >
> fI 5 >
>
> > 6
> 4 0 m52 m53 0 m55 m56 7 5 > u€5 >
>
> >
>
>
: ; > : >
> ;
fI 6 0 m62 m63 0 m65 m66 u€6
where
Z L
mij ¼ I0 ðx Þ i ðx Þ j ðx Þdx i, j ¼ 1, 4 (11.8.24a)
0
Z L
mij ¼ m ðx Þ i ðx Þ j ðx Þdx i, j ¼ 2, 3, 5, 6 (11.8.24b)
0

€e
f eI ¼ me u (11.8.25)
where € are the vectors of the nodal inertial forces and the nodal acceler-
u
f eI , e
ations, respectively, and me the mass matrix of the e grid element.

For an element with constant mass, m ðx Þ ¼ rA, rI ðx Þ ¼ rI 0 , Eqs. (11.8.24a),
(11.8.24b) are integrated analytically and the mass matrix becomes
2 3
140rg2 0 0 70rg2 0 0
6 7
60 4L2 3L2 13L 7
6 22L 0 7
6 7
6 7
m e 60 22L 156 0 13L 54 7
me ¼ 6 7 (11.8.26)
420 66 70rg2 0 0 140rg2 0 0 7
7
6 7
6 7
60 3L 13L 0
2
4L 22L 7
2
4 5
0 13L 156 54 0 22L
p ffiffiffiffiffiffiffiffiffiffi
where m e ¼ rALe is the mass of the element and rg ¼ I0 =A the radius of
gyration of the cross-section.
@CivilMethod
Z L
m1 ¼ m ðx Þð1 x Þdx (11.8.27a)
0
Z L
m2 ¼ m ðx Þxdx (11.8.27b)
0
Therefore, the kinetic energy of the element is

1 1
T ¼ m1 u_ 3 2 + m2 u_ 6 2 (11.8.28)
2 2
which by virtue of Eq. (11.2.9) for i ¼ 1, 2, …,6 yields
8 9 2 38 9
> fI 1 > 0 0 0 0 0 0 > u€1 >
> > > >
>f >
> > 60
> 0 7 > u€ >
> >
>
> I2 > 7>> 2>>
> 6
0 0 0 0
> 7> >
<f >
> = 660 0 m1 0 0
>
0 7 u€3 =
< >
¼6 7
I3
(11.8.29)
> > 6
> fI 4 > 60 0 0 0 0 0 77>> u€4 >
>
>
> >
> 6 7>> >
>
>
> >
>
fI 5 > 4 0 5 >
> € >
>
>
: >
> ;
0 0 0 0 0 > > >
:
u 5 >
;
fI 6 0 0 0 0 0 m2 u€6
For an element of constant cross-section, m ðx Þ ¼ rA, the mass matrix

becomes
2 3
0 0 0 0 0 0
60 0 0 0 0 07
6 7
6
e 60 0 1 0 0 07
7
m 6 7
me ¼ (11.8.30)
2 6 60 0 0 0 0 07
7
6 7
40 0 0 0 0 05
0 0 0 0 0 1
where m e ¼ rAe Le is the mass of e the element.

(iii) Nodal damping forces and damping matrix of the grid element
Following a procedure analogous to that for the plane frame, we can con-
struct a damping matrix for the grid element. However, as we will see in
Section 12.11, the practical significance of the damping matrices constructed
in this way is limited and therefore, for reasons of saving space, we avoid its
derivation here.
(iv) Equivalent nodal loads of the grid element
The equivalent nodal loads can be evaluated in two ways: (a) as generalized
forces in the direction of the nodal displacements ue and (b) as static equivalent
forces. In the first case, the shape functions of the stiffness matrix are used and
@CivilMethod
the resulting vector of the equivalent nodal forces is referred to as a consistent

nodal load vector.
We assume that the element is loaded by the transverse load pz ðx, t Þ and the
twisting moment mx ðx, t Þ. Referring to Fig. 11.8.5, the virtual work produced
during the deformation of a beam element dx is
Fig. 11.8.5 Loading and equivalent nodal loads of a plane grid element.
Z L Z L
p
dWnc ¼ mx ðx, t Þdqðx, t Þdx + pz ðx, t Þdv ðx, t Þdx (11.8.31)
0 0
which by virtue of Eqs. (11.8.2), (11.8.4) gives

p
dWnc ¼ p1 ðt Þdu1 + p2 ðt Þdu2 + p3 ðt Þdu3 + p4 ðt Þdu4 + p5 ðt Þdu5 + p6 ðt Þdu6
(11.8.32)
where
Z L
pi ðt Þ ¼ mx ðx, t Þ i ðx Þdx, i ¼ 1, 4 (11.8.33a)
0
Z L
pi ðt Þ ¼ pz ðx, t Þ i ðx Þdx, i ¼ 2, 3, 5, 6 (11.8.33b)
0
Consequently, the resulting vector of equivalent nodal loads of the e

element is
8 9
> p1 ðt Þ >
>
> >
>
>
>
> p2 ðt Þ >>
>
> >

> =
3
p ðt Þ ¼
e
(11.8.34)
>
> p ð t Þ >
>
>
>
4 >
>
>
>
>
> p5 ðt Þ >>
>
>
: ;
p6 ðt Þ
@CivilMethod

In this case, we assume that the grid element is a simply supported beam
under the load pz ðx, t Þ and the torsional moment mx ðx, t Þ. Actually, the com-
ponents p2 ðt Þ and p5 ðt Þ are zero. The remaining components are obtained as the
reactions of the simply supported beam
Z L
pi ðt Þ ¼ mx ðx, t Þ i ðx Þdx, i ¼ 1, 4 (11.8.35a)
0
Z L
p i ðt Þ ¼ pz ðx, t Þ i ðx Þdx, i ¼ 3, 6 (11.8.35b)
0
where now 3 ðx Þ ¼ 1 x, 6 ðx Þ ¼ x.Hence, we have

8 9
>
> p1 ðt Þ >
>
>
> >
>
>
> 0 >
>
< =
p3 ðt Þ
p ðt Þ ¼
e
(11.8.36)
> p4 ðt Þ >
>
> >
>
>
> >
>
>0
: >
;
p6 ðt Þ
The statically equivalent vector of the nodal loads is particularly useful
when the rotational inertial nodal forces of the element are ignored.
11.8.2 Transformation of the nodal coordinates of the plane

grid element
The displacements u1 , u2 are transformed into the global system of axes accord-
ing to Eq. (11.2.67), while the component u3 , which expresses the displacement
in the direction of the z axis, remains unaltered in the rotated system because the
local z axis and the global z axis are identical. Hence, we have

u1 cos f sin f u1
¼ (11.8.37a)
u2 sin f cos f u2
u3 ¼ u3 (11.8.37b)
The previous relations can be written as
8 9 2 38 9
< u1 = cos f sin f 0 < u1 =
u2 ¼ 4 sin f cos f 0 5 u2 (11.8.38a)
: ; : ;
u3 0 0 1 u3
8 9 2 38 9
< u4 = cosf sin f 0 < u4 =
u ¼ 4 sin f cos f 0 5 u5 (11.8.38b)
: 5; : ;
u6 0 0 1 u6
@CivilMethod
Eqs. (11.8.38a), (11.8.38b) are combined as

8 9 2 38 9
>
> u1 >
> cos f sin f 0 0 0 0 >> u1 >
>
>
> >
> 6 sin f cos f 0 >
> >
>
> u > 0 0 07 > u2 >
>
< >
> = 6 7< >
2
6 0 7 =
u3 0 1 0 0 0 7 u3
¼6 (11.8.39)
> u4 >
> > 6 6 0 0 0 cos f sin f 07 > u4 >
7> >
> > > >
> u5 >
> > 4 0
> 0 0 sin f cos f 0 5>
> u5 >
>
>
: ; : >
> ;
u6 0 0 0 0 0 1 u6
or
ue ¼ Re ue (11.8.40a)
ue ¼ ðRe ÞT ue (11.8.40b)

where
2 3
cos fe sin fe 0 0 0 0
6 7
6 sin f
e
cos fe 0 0 0 07
6 7
6 0 0 1 0 0 07
6 7
Re ¼ 6 7 (11.8.41)
6 0 0 0 cos fe sin f 0 7
e
6 7
6 7
4 0 0 0 sin fe cos fe 0 5
0 0 0 0 0 1
is the transformation matrix of the plane grid element e.
The global vectors of the nodal forces f S , f I , f D , and pe ðt Þ are defined in rela-
e e e
tion to u . Their transformation obeys the same law, that is, Eq. (11.8.40b).
e
Thus, we obtain
f e ¼ðRe ÞT f e (11.8.42)
S S
f e ¼ðRe ÞT f e (11.8.43)
I I
f e ¼ðRe ÞT f e (11.8.44)
D D
pe ðt Þ ¼ðRe ÞT pe ðt Þ (11.8.45)

Using Eqs. (11.8.19), (11.8.25), (11.3.41), (11.8.40a), the first three of the
foregoing equations are transformed into the global axes as
f e ¼ke ue (11.8.46)
S
f e e €e
u
¼m (11.8.47)
I
f e ¼e _e

D c u (11.8.48)
e e , and
where k , m ce represent the stiffness, mass, and damping matrices of the e
grid element in global axes, respectively, and are given by
k ¼ ðRe ÞT ke Re
e
(11.8.49)
@CivilMethod
m (11.8.50)

ce ¼ ð Re Þ T ce Re (11.8.51)
Note that the lumped mass matrix remains the same under this transformation.
Applying the procedure presented in Section 11.2.3 for the plane truss, we
obtain the equation of motion of the plane grid, which, if damping is taken into
account, is written as
ee
M u€ + Cu_ + Ke
ee e u ¼e
pð t Þ (11.8.52)
or after applying the support conditions
M u€f + C
e ff e u_ f + K
uf ¼ e (11.8.53)
M u€f + M
e sf e u€s + C
e ss e u_ f + C
e sf e u_ s + K
uf + K us ¼ e
ps ðt Þ (11.8.54)
where
pf∗ ðt Þ ¼ e
e fs e u_ s K
e fs e e fs e
us (11.8.55)
For convenience we will write Eq. (11.8.53) Check this as
M€
u + C€
u + Ku ¼ pðt Þ (11.8.56)
Example 11.8.1 Formulate the equation of motion of the plane grid of Fig. E11.9.
The grid is fixed at nodes 3 and 4. The element 1 is loaded by the transverse
t. A moment M ðt Þ ¼ 3p0 L2 sin w
load pðx, t Þ ¼ p0 sin w t acts at the node 2
t. The cross-section
while the support 1 is subjected to the rotation qg ðt Þ ¼ q0 sin w
of all elements is rectangular with aspect ratio h=b ¼ 2. Consider lumped mass
assumption for the elements. Data: L ¼ 3:0m, h ¼ 0:50m, a ¼ p=4,
E ¼ 2:1 107 kN=m2 , n ¼ 0:2, q0 ¼ 0:5p0 , mass density r.
Solution
The system has n ¼ 4 nodes. Hence, the free structure has N ¼ 3n ¼ 12 degrees
of freedom. The numbering of the nodes and the positive direction of the
elements are shown in Fig. E11.9
Fig. E11.9 Plane grid in Example 11.8.1.
@CivilMethod
1. Computation of ke , me , ce , pe ðt Þ, Re for e ¼ 1,2, 3.

Matrices ke . All elements have a rectangular cross-section with side
lengths h ¼ 0:50 m and b ¼ 0:25 m. Their cross-sectional moment of inertia
is I ¼ bh 3 =12. The torsional constant of the elements is given by the relation
It ¼ b3 h [12]. For h=b ¼ 2 it is ¼ 0:229. Hence, It ¼ 0:229hb3 ¼ 0:6866I .
Moreover, G ¼ E=2ð1 + n Þ ¼ 0:4166E. The stiffness matrices of the elements
are computed using Eq. (11.8.20)
2 3
2:572 0 0 2:572 0 0
6 0 36:0 18:0 0 18:0 18:0 7
6 7
EI 6 0 18:0 12:0 0 18:0 12:0 7
k1 ¼ k 2 ¼ k 3 ¼ 3 6 7
L 66 2:572 0 0 2:572 0 0 77
4 0 18:0 18:0 0 36:0 18:0 5
0 18:0 12:0 0 18:0 12:0
Matrices me . For lumped mass assumption, the mass matrices are computed
using Eq. (11.8.30)
2 3
0 0 0 0 0 0
60 0 0 0 0 0 7
6 7
6 0 0 1:5 0 0 0 7
m1 ¼ m2 ¼ m3 ¼ rA6 60 0 0 0 0 0 7
7
6 7
40 0 0 0 0 0 5
0 0 0 0 0 1:5
Vectors pe ðt Þ. To be consistent with the lumped mass assumption, the
vectors pe ðt Þ will be computed as the static equivalent nodal loads. Thus, using
Eq. (11.8.36), we have
8 9
> 0 >
> 0 >
>
> >
>
> >
< 1:5 >
> =
p ðt Þ ¼
1
t, p2 ðt Þ ¼ p3 ðt Þ ¼ 0
p0 sin w
>
> 0 >
>
>
> >
>
> >
: 0 >
> ;
1:5
Element
number e xj xk D
x yj yk D
y Le cosfe sinfe
1 0 0 0 0 3 3 3 0 1
2 0 2.12 2.12 3 5.12 2.12 3 0.707 0.707
3 0 2.12 2.12 3 5.12 2.12 3 0.707 0.707
@CivilMethod
On the base of the geometrical data given in Table E.11.4, we obtain

2 3 2 3
0 1 0 0 0 0 0:707 0:707 0 0 00
6 1 7 6 07
6 0 0 0 0 07 6 0:707 0:707 0 0 07
6 7 6 7
6 0 0 1 0 0 0 7 6 0 0 1 0 0 07
R1 ¼ 6
6 0
7 , R2 ¼ 6
7 6 0
7
6 0 0 0 1 07 6 0 0 0:707 0:707 0 77
6 7 6 7
4 0 0 0 1 0 0 5 4 0 0 0 0:707 0:707 0 5
0 0 0 0 0 1 0 0 0 0 0 1
2 3
0:707 0:707 0 0 0 0
6 0:707 0:707 0 0 07
6 0 7
6 7
6 0 0 1 0 0 07
R3 ¼ 6
6 0
7
6 0 0 0:707 0:707 077
6 7
4 0 0 0 0:707 0:707 05
0 0 0 0 0 1
e
e , pe ðt Þ for e ¼ 1, 2, 3
Matrices k . They are computed using Eq. (11.8.49)
e
2 3
36:0 0 18:0 18:0 0 18:0
6 0 0 2:57 0 7
6 2:57 0 7
6 7
EI 6 18:0 0 12:0 18:0 0 12:0 7

k ¼ 36
1 6 7
L 6 18:0 0 18:0 36:0 0 18:0 77
6 7
4 0 2:57 0 0 2:57 0 5
18:0 0 12:0 18:0 0 12:0
2 3
19:29 16:71 12:73 7:71 10:29 12:73
6 16:71 19:29 12:73 10:29 7:71 12:73 7
6 7
6 7
EI 6 12:73 12:73 12:0 12:73 12:73 12:0 7

k ¼ 36
2 6 7
L 6 7:71 10:29 12:73 19:29 16:71 12:73 7 7
6 7
4 10:29 7:71 12:73 16:71 19:29 12:73 5
12:73 12:73 12:0 12:73 12:73 12:0
2 3
19:29 16:71 12:73 7:71 10:29 12:73
6 16:71 19:29 12:73 10:29 7:71 12:73 7
6 7
6 7
EI 6 12:73 12:73 12:0 12:73 12:73 12:0 7
k3 ¼ 3 6 7
L 6 7:71 10:29 12:73 19:29 16:71 12:73 7
6
7
6 7
4 10:29 7:71 12:73 16:71 19:29 12:73 5
12:73 12:73 12:0 12:73 12:73 12:0
@CivilMethod
e . They are computed using

Matrices m Eq. (11.8.50)
2 3
0 0 0 0 0 0
60 0 7
6 0 0 0 0 7
6 7
60 0 1:5 0 0 0 7
m ¼ m ¼ m ¼ rA6
1 2 3
60 0
7
6 0 0 0 0 7
7
6 7
40 0 0 0 0 0 5
0 0 0 0 0 1:5
Vectors pe ðt Þ. They are computed using Eq. (11.8.45)
8 9
>
> 0 >>
> >
>
>
> 0 >>
>
< =
1:5

p ðt Þ ¼
1
t, p2 ðt Þ ¼ p3 ðt Þ ¼ 0
p0 sin w
>
> 0 >>
> >
>
>
> 0 >>
>
: ;
1:5
^ e, M
3. Computation of the enlarged matrices K ^ e, p
ê for e ¼ 1, 2, 3
Assembly matrices ae .
2 3
1 0 0 0 0 0 0 0 0 0 0 0
60 1 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 1 0 0 0 0 0 0 0 0 07
a ¼6
1
60
7,
6 0 0 1 0 0 0 0 0 0 0 07
7
6 7
40 0 0 0 1 0 0 0 0 0 0 05
0 0 0 0 0 1 0 0 0 0 0 0
2 3
0 0 0 1 0 0 0 0 0 0 0 0
60 0 0 0 1 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 1 0 0 0 0 0 07
a ¼6
2
60
7
6 0 0 0 0 0 1 0 0 0 0 07
7
6 7
40 0 0 0 0 0 0 1 0 0 0 05
0 0 0 0 0 0 0 0 1 0 0 0
2 3
0 0 0 1 0 0 0 0 0 0 0 0
60 0 0 0 1 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 1 0 0 0 0 0 07
a ¼6
3
60
7
6 0 0 0 0 0 0 0 0 1 0 077
6 7
40 0 0 0 0 0 0 0 0 0 1 05
0 0 0 0 0 0 0 0 0 0 0 1
@CivilMethod
Matrices K^ e . They are computed using the relation K^ e ¼ ðae ÞT ke ae (see
Eq. 11.2.92)
2 3
36:0 0 18:0 18:0 0 18:0 0 0 0 0 0 0
6 0 2:572 0 0 2:572 0 0 0 0 0 0 07
6 7
6 18:0 0 12:0 0 0 0 0 0 0 7
6 12:0 18:0 0 7
6 7
6 18:0 0 18:0 36:0 0 18:0 0 0 0 0 0 0 7
6 7
6 0 2:572 0 0 2:572 0 0 0 0 0 0 07
6 7
6 18:0 12:0 18:0 7
EI
^1 ¼ 6 6 0 0 12:0 0 0 0 0 0 0 7
K 7
L3 6 0 0 0 0 0 0 0 0 0 0 0 07
6 7
6 0 0 0 0 0 0 0 0 0 0 0 07
6 7
6 0 0 0 0 0 0 0 07
6 0 0 0 0 7
6 7
6 0 0 0 0 0 0 0 0 0 0 0 07
6 7
4 0 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 07
6 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 19:29 16:71 12:73 7:71 10:29 12:73 07
6 0 0 0 0 7
6 7
60 0 0 16:71 19:29 12:73 10:29 7:71 12:73 0 0 07
6 7
^ EI 6
60 0 0 12:73 12:73 12:0 12:73 12:73 12:0 0 0 077
K ¼ 36
2
7
L 60 0 0 7:714 10:29 12:73 19:29 16:71 12:73 0 0 07
6 7
60 10:29 7:71 12:73 16:71 07
6 0 0 19:29 12:73 0 0 7
6 7
60 0 0 12:73 12:73 12:0 12:73 12:73 12:0 0 0 07
6 7
60 07
6 0 0 0 0 0 0 0 0 0 0 7
6 7
40 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 7
6 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0
6 0 19:29 16:71 12:73 0 0 0 7:71 10:29 12:73 77
60 0
6 0 16:71 19:29 12:73 0 0 0 10:29 7:71 12:73 7
7
6 7
^ EI 60 0 0 12:73 12:73 12:0 0 0 0 12:73 12:73 12:0 7
7
K ¼ 36
3
7
L 660 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 7
6 0 7:714 10:29 12:73 0 0 0 19:29 16:71 12:73 7
6 7
40 0 0 10:29 7:71 12:73 0 0 0 16:71 19:29 12:73 5
0 0 0 12:73 12:73 12:0 0 0 0 12:73 12:73 12:0
@CivilMethod
Matrices M^ e . They are computed ^ e ¼ ð ae Þ T m

using the relation M e ae
(see Eq. 11.2.96)
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 0 0 1:5 0 0 07
6 0 0 0 0 0 0 7
60 0 0 0 0 07
6 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 1:5 0 0 0 0 0 07
^
M ¼ rA6
1 7,
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 07
6 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
40 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
6 07
^ 2 ¼ rA6 0
M
0 0 0 0 1:5 0 0 0 0 0 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 1:5 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
40 0 0 0 0 0 0 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
2 3
0 0 0 0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 1:5 0 0 0 0 0 0 7
^ 3 ¼ rA6
M 6
7
7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
40 0 0 0 0 0 0 0 0 0 0 0 5
0 0 0 0 0 0 0 0 0 0 0 1:5
@CivilMethod
ê ðt Þ. They are computed using the relation p

Vectors p ê ðt Þ ¼ ðae ÞT pe ðt Þ
(see Eq. 11.2.95)
8 9 8 9
> 0 > > 0>
>
> >
> >
> >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
>
> >
> >
> >
>
>
> 1:5 >
> >
> >
>
>
> >
> >
> 0 >
>
>
> >
> >
> >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
>
> >
> >
> >
>
>
> >
> >
> >
>
>
>
0 >
> >
>
0 >
>
> >
> > >
>
< 1:5 = <0>
> =
^ 1 ðt Þ ¼
p ^ 2 ðt Þ ¼ p
t, p
p0 sin w ^ 3 ðt Þ ¼
>
> 0 > >
>
>
> 0>>
>
>
> >
> > >
>
> >
>
> > > >
> 0 >
>
> >
> >0>
>
> >
>
>
> >
> >
> >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
>
> >
> >
> >
>
>
> 0 > >
> >
>0> >
>
> > >
> >
>
>
> >
> >
> >
>
>
> 0 >
> >
> 0 >
>
> >
> > >
>
: ; : >
> ;
0 0
P
3 P
3 P
3
¼
4. Computation of M M ¼
^ e, K ^ e , and p ¼ Pðt Þ +
K ^ e ðt Þ
p
e¼1 e¼1 e¼1
2 3
0 0 0 0 0 0 0 0 0 0 0 0
6 7
60 0 0 0 0 00 0 0 0 0 0 7
6 7
6 7
60 0 1:5 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 4:5 0 0 0 0 0 0 7
¼ rA ¼ 6
M 6
7
7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 0 0 0 7
4 5
0 0 0 0 0 0 0 0 0 0 1:5
@CivilMethod
2 3
36:0 0 18:0 18:0 0 18:0 0 0 0 0 0 0
6 0 2:572 0 0 2:57 0 0 0 0 0 0 0 7
6 7
6 18:0 0 12:0 0 7
6 12:0 18:0 0 0 0 0 0 0 7
6 7
6 18:0 0 18:0 74:57 0 18:0 7:71 10:29 12:73 7:71 10:29 12:73 7
6 7
6 0 2:572 0 0 41:14 25:46 10:29 7:71 12:73 10:29 7:71 12:73 7
6 7
6 18:0 0 12:0 18:0 25:46 12:73 12:0 12:73 12:73 12:0 7
EI
¼ 6 6 36:0 12:73 7
K 7
L3 6 0 0 0 7:71 10:29 12:73 19:29 16:71 12:73 0 0 0 7
6 7
6 0 0 0 10:29 7:714 12:73 16:71 19:29 12:73 0 0 0 7
6 7
6 0 0 12:73 12:73 12:0 12:73 12:73 12:0 0 7
6 0 0 0 7
6 7
6 0 0 0 7:71 10:29 12:73 0 0 0 19:29 16:71 12:73 7
6 7
4 0 0 0 10:29 7:714 12:73 0 0 0 16:71 19:29 12:73 5
0 0 0 12:73 12:73 12:0 0 0 0 12:73 12:73 12:0
8 9 8 9 8 9
>
> Mg >
> >
> 0 > > >
> Mg >
>
>
> >
> >
> >
> >
> >
>
>
> P >
> >
> 0 >
> >
> P >
>
>
>
2 >
> >
> >
> >
>
2 >
>
>
>
> P3 >
>
>
>
>
> 1:5 >
>
>
>
>
> t >
P3 1:5p0 sin w >
>
> > > > > >
>
>
> 3p L 2
sin
w t
>
>
> > 0 >
>
> >
>
>
>
> 27p sin
w t
>
>
>
>
> 0 >
> >
> >
> >
> 0 >
>
>
> >
> >
> >
> >
> >
>
>
> 0 >
> >
> 0 >
> >
> 0 >
>
X
3 >
<0 >
= < 1:5 >
> = >
< 1:5p sin w t
>
=
ðt Þ +
pðt Þ ¼ P ^ i ðt Þ ¼
p + t ¼
p0 sin w
0
>
> P7 >
> >
> 0 >
> >
> P7 >
>
i¼1 >
> >
> >
> >
> >
> >
>
>
> P >
> >
> 0 >
> >
> P8 >
>
>
> 8 >
> >
> >
> >
> >
>
>
> >
> >
> >
> >
> >
>
>
> P 9 >
> >
> 0 >
> >
> P9 >
>
>
> >
> >
> >
> >
> >
>
>
> P >
> >
> 0 >
> >
> P10 >
>
>
>
10 >
> >
> >
> >
> >
>
>
> >
> >
> >
> >
> >
>
>
> P 11 >
> >
> 0 >
> >
> P11 >
>
: ; : ; : ;
P12 0 P12
Note that Mg denotes the moment that produces the rotation qg ðt Þ. It is

unknown and will be established as a reaction force.
M,
6. Modification of the matrices K, pðt Þ due to support conditions of the
grid.
Referring to Fig. E11.9, the displacement vector should be modified as
e e1 u
uT ¼ f u e2 u
e3 u
e4 u
e5 u
e6 u
e7 u
e8 u
e9 u
e10 u
e11 u
e12 g
¼ f u4 u5 u6 u1 u2 u3 u7 u8 u9 u10 u11 u12 g
Hence, the matrix V that rearranges the matrices of the structure is
2 3
0 0 0 1 0 0 0 0 0 0 0 0
6 7
60 0 0 0 1 0 0 0 0 0 0 07
6 7
6 7
60 0 0 0 0 1 0 0 0 0 0 07
6 7
61 0 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 1 0 0 0 0 0 0 0 0 0 07
6 7
6 7
60 0 1 0 0 0 0 0 0 0 0 07
6 7
V¼6 7
60 0 0 0 0 0 1 0 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 1 0 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 1 0 0 07
6 7
6 7
60 0 0 0 0 0 0 0 0 1 0 07
6 7
60 0 0 0 0 0 0 0 0 0 1 07
4 5
0 0 0 0 0 0 0 0 0 0 0 1
e ¼ VT KV,
Thus, using the equations K e ¼ VT MV,
M pðt Þ ¼ VT pðt Þ,
e
we obtain
@CivilMethod
Following the indicated partitioning, we obtain

2 3
18:0 0 18:0
6 0 2:57 7
6 0 7
6 7
6 18:0 0 12:0 7
6 7
2 3 6 7
74:57 0 18:0 6 7:71 10:29 12:73 7
6 7
EI
e ff ¼ 4 06 7 e sf ¼ 6
EI 7
K 41:14 25:46 5, K 6 10:29 7:71 12:73 7
L 3 L 6
3 7
18:0 25:46 36:0 6 12:73 12:0 7
6 12:73 7
6 7
6 7:71 10:29 12:73 7
6 7
6 7
4 10:29 7:714 12:73 5
12:73 12:73 12:0
2 3
18:0 0 18:0 7:71 10:29 12:73 7:71 18:0 7:71
6 7
e fs ¼ EI 6
K 0 2:57 0 10:29 7:714 12:73 10:29 0 10:29 7
L 4
3 5
18:0 0 12:0 12:73 12:73 12:0 12:73 12:0 12:73
2 3
36:0 0 18:0 0 0 0 0 18:0 0
6 0 7
6 2:57 0 0 0 0 0 7 0 0
6 7
6 18:0 0 12:0 0 0 0 0 12:0 0 7
6 7
6 7
6 0 0 0 19:29 16:71 12:73 0 0 19:29 7
EI 6 7
e
Kss ¼ 3 6 0 0 0 16:71 19:29 12:73 0 0 16:71 7
L 66
7
7
6 0 0 0 12:73 12:73 12:0 0 0 12:73 7
6 7
6 0 0 0 0 0 0 19:29 0 0 7
6 7
6 7
4 0 0 0 0 0 0 16:71 0 0 5
0 0 0 0 0 0 12:73 0 0
@CivilMethod
2 3
0 0 0 0 0 0 0 0 0
60 0 0 0 0 0 0 0 0 7
6 7
60 0 1:5 0 0 0 0 0 0 7
2 3 6 7
0 0 0 60 0 0 0 0 0 0 0 0 7
6 7
e 4
Mff ¼ rA 0 0 0 5 e
, Mss ¼ rA6 0 7
60 0 0 0 0 0 0 0 7
0 0 4:5 60 0 0 0 0 1:5 0 0 0 7
6 7
60 0 0 0 0 0 0 0 0 7
6 7
40 0 0 0 0 0 0 0 0 5
0 0 0 0 0 0 0 0 1:5
M e sf ¼ 0
e fs ¼ M
8 9 8 9
>
> Mg >
> >
> q g ðt Þ >
>
>
> >
> >
> >
>
>
> P2 >
> >
> 0 >
>
>
> >
> >
> >
>
8 9 > 3
> P 1:5p0 sin w t >
> >
> 0 >
>
>
> >
> >
> >
>
< 27 = < 7
P = < 0 =
e
pf ðt Þ ¼ 0 t, e
p0 sin w p s ð t Þ ¼ P8 es ð t Þ ¼ 0
, u
: ; >
> >
> >
> >
>
1:5 >
> P9 >
> >
> 0 >
>
>
> >
> >
> >
>
>
> P >
> >
> 0 >
>
>
>
10
>
> >
> >
>
>
> P >
> >
> 0 >
>
: 11
; : ;
P12 0
e ff ¼ 0
The equation of motion results from Eq. (11.8.53) with C
M u€f + K
uf ¼ e
Taking into account that q0 ¼ 0:5p0 , Eq. (11.8.55) gives
8 9
< 18 =
p ∗ f ðt Þ ¼ e
e e fs u
pf ðt Þ M e sf us ¼ 0
€s K t
p sin w
: ; 0
7:5
Hence, the equation of motion is
5. Static condensation
The rotational degrees of freedom u4 , u5 are statically condensed. Obviously,
the displacements should be rearranged as u6 , u4 , u5 . This defines the
modification matrix
2 3
0 1 0
VC ¼ 4 0 0 1 5
1 0 0
@CivilMethod
which yields

EI EI EI 18:0
Ktt ¼ 36 , K ¼ ½ 18:0 25:45 , K qt ¼
L3 25:45
tq
L3 L3

EI 74:57 0
Kqq ¼ 3
L 0 41:14
Mtt ¼ 4:5rA, Mtq ¼ Mqt ¼ Mqq ¼ 0

27
t, pq ¼
pt ¼ 1:5p0 sin w t
p0 sin w
0
Hence
EI
Ktt∗ ¼ Ktt Ktq K1
qq Kqt ¼ 15:91
L3
pt∗ ¼ pt Ktq K1 t
qq pq ¼ 5:017p0 sin w
which are inserted into Eq. (11.4.3) to give the final form of the equation of
motion
EI
4:5rA u€6 + 15:91 t
u6 ¼ 5:017p0 sin w
L3
11.9 The finite element method for the space frame

11.9.1 Properties of the space frame element
We consider the space frame of Fig. 11.9.1, whose nodes are determined by
their Cartesian coordinates with respect to the global system of axes x y z.
The loads applied to the nodes may consist of forces or moments while the
elements may be loaded by distributed and/or concentrated loads (forces or
moments). The axis of the e element determines its local x axis while the y
and z axes lie on the principal planes of bending (see Fig. 11.9.2). The nodes
of the frame, hence the ends of the elements, undergo six displacements (three
translations and three rotations). Therefore, the nodal displacements of the
element are 12, six for each end, and are numbered as shown in Fig. 11.9.2.
They are represented by the vector
u ¼ f u1 u2 u3 u4 u5 u6 u7 u8 u9 u10 u11 u12 g (11.9.1)
@CivilMethod
Fig. 11.9.1 Space frame.
Fig. 11.9.2 Nodal displacements of a space frame element.
The displacements u1 , u7 produce axial deformation of the element, the

displacements u4 , u10 produce torsion, and the displacements u2 , u6 , u8 , u12
and u3 , u5 u9 , u11 produce bending in the planes x y and x z, respectively.
The elastic energy of the element is
U ¼ Ua + Ut + Uby + Ubz (11.9.2)
where Ua ,Ut ,Uby , Ubz represent the elastic energies due to axial and torsional
deformation as well as to bending deformation in the planes xy and xz, respec-
tively. Using appropriately the shape functions developed for the plane frame
and grid elements, we can express the displacements of a cross-section x of the
space frame element as
u ðx, t Þ ¼ u1 1 ðx Þ + u 7 7 ðx Þ (11.9.3a)
qðx, t Þ ¼ u4 4 ðx Þ + u10 10 ðx Þ (11.9.3b)
v ðx, t Þ ¼ u2 2 ðx Þ + u6 6 ðx Þ + u8 8 ðx Þ + u12 12 ðx Þ (11.9.3c)
w ðx, t Þ ¼ u3 3 ðx Þ + u5 5 ðx Þ + u9 9 ðx Þ + u11 11 ðx Þ (11.9.3d)
@CivilMethod
where u, v,w are the translations in the directions of the x, y, z axes, respec-
tively, and q the rotation about the x while i ðx Þ (i ¼ 1,2, …,12) are given by
1 ðx Þ ¼ 4 ðx Þ ¼ ð1 x Þ (11.9.4a)
7 ðx Þ ¼ 10 ðx Þ ¼ x (11.9.4b)
2 ðx Þ ¼ 3 ðx Þ ¼ 1 3x
2
+ 2x3 (11.9.4c)

6 ðx Þ ¼ 5 ðx Þ ¼ L x 2x 2 + x 3
(11.9.4d)
2
12 ðx Þ ¼ 11 ðx Þ ¼ L x + x
3
(11.9.4e)
9 ðx Þ ¼ 8 ðx Þ ¼ 3x 2x3
2
(11.9.4f)
On the base of Eqs. (11.3.8), (11.3.9), (11.8.13), Eq. (11.9.2) yields
Z Z
1 L 1 L
0
GI t ½q0 ðx Þ dx
2 2
U ðu1 , …, u12 Þ ¼ EA½u ðx, t Þ dx +
2 0 2 0
Z Z (11.9.5)
1 L 00 2 1 L 00 2
+ EI z ½v ðx, t Þ dx + EI y ½w ðx, t Þ dx
2 0 2 0
Further, for the consistent mass assumption, the kinetic energy is given by
Z n o
1 L
T ðu_ 1 , …, u_ 12 Þ ¼ m ðx Þ ½u_ ðx, t Þ2 + ½v_ ðx, t Þ2 + ½w_ ðx, t Þ2 dx
2 0
Z (11.9.6)
1 L 2
+ _
I0 ðx Þ q ðx, t Þ dx
2 0
while for the lumped mass assumption by
1 1
T ðu_ 1 , …, u_ 12 Þ ¼ m1 u_ 21 + u_ 22 + u_ 23 + m2 u_ 27 + u_ 28 + u_ 29 (11.9.7)
2 2
where
Z L Z L
m1 ¼ m ðx Þð1 xÞdx, m2 ¼ m ðx Þxdx (11.9.8)
0 0
The element mass and stiffness matrices as well as the equivalent nodal
forces are obtained using the method of the Lagrange equations as described
for the plane frame and the grid element. In the following, we consider an ele-
ment with constant cross-section.
(i) Nodal elastic forces and stiffness matrix of the space frame element
For an element with a constant cross-section, the stiffness matrix having dimen-
sions 12 12 results as
e e
k k
k ¼ jje jk
e
(11.9.9)
kkj kekk
@CivilMethod
where
2 3
L2 A 0 0 0 0 0
6 7
60 12Iz 0 0 0 6LI z 7
6 7
6 7
60 0 12Iy 0 6LI y 0 7
E 6 7
kejj ¼ 3 6 7 (11.9.10a)
L 660 0 0
G 2
L It 0 0
7
7
6 E 7
6 7
6 7
40 0 6LI y 0 4L2 Iy 0 5
0 6LI z 0 0 0 4L2 Iz
2 3
L2 A 0 0 0 0 0
6 7
60 12Iz 0 0 0 6LI z 7
6 7
6 7
60 0 12Iy 0 6LI y 0 7
E 6 7
kkk ¼ 3 6
e 7 (11.9.10b)
L 660 0 0
G 2
L It 0 0
7
7
6 E 7
6 7
60 7
4 0 6LI y 0 4L2 Iy 0 5
0 6LI z 0 0 0 4L2 Iz
2 3
L2 A 0 0 0 0 0
6 7
6 0 12Iz 0 0 0 6LI z 7
6 7
6 7
6 0 0 12Iy 0 6LI y 0 7
E 6 7
kekj ¼ 3 6 7 (11.9.10c)
L 66 0 0 0
G 2
L It 0 0
7
7
6 E 7
6 7
6 7
4 0 0 6LI y 0 2L2 Iy 0 5
0 6LI z 0 0 0 2L2 Iz
T
kejk ¼ kekj (11.9.10d)
In the previous matrices, A is the area of the cross-section of the element e,

L its length, Iy , Iz are the cross-sectional moments of inertia with respect to the
y and z axes, respectively, and It the torsional constant of the cross-section.
Note that the superscript e has been omitted from Ite , Iye , Ize , and Le for the
simplicity of the expressions.
(ii) Nodal inertial forces and mass matrix of the space frame element
The consistent matrix results as
" #
mejj mejk
m ¼
e
(11.9.11)
mekj mekk
@CivilMethod
where
2 3
140 0 0 0 0 0
60 156 0 0 0 22L 7
e 6
6 7
m 60 0 156 0 22L 0 7 7
mejj ¼ (11.9.12a)
420 660 0 0 140rg2 0 0 7 7
40 0 22L 0 4L2 0 5
0 22L 0 0 0 4L2
2 3
140 0 0 0 0 0
60 156 0 0 0 22L 7
6
e 6
7
m 6 0 0 156 0 22L 0 7 7
mkk ¼
e
(11.9.12b)
420 660 0 0 140rg2 0 0 7 7
40 0 22L 0 4L2 0 5
0 22L 0 0 0 4L2
2 3
70 0 0 0 0 0
60 54 0 0 0 13L 7
6
e 6
7
m 6 0 0 54 0 13L 0 7 7
mkj ¼
e
(11.9.12c)
420 660 0 0 70rg2 0 0 7 7
40 0 13L 0 3L2 0 5
0 13L 0 0 0 3L2
T
mejk ¼ mekj (11.9.12d)
pffiffiffiffiffiffiffiffiffiffi
m e ¼ rAL is the total mass of the e element and rg ¼ I0 =A is the radius of
gyration of the cross-section.
nodes, that is, they are obtained as the reactions of a simply supported beam
under the load m ðx Þ. Thus, for an element with constant mass, the mass lumped
assumption yields
2 3
m1 0 0 0 0 0
6 0 m1 0 0 0 0 7
6 7
6 0 0 m1 0 0 0 7
mjj ¼ 6
e 6 7 (11.9.13a)
7
60 0 0 0 0 07
40 0 0 0 0 05
0 0 0 0 0 0
2 3
m2 0 0 0 0 0
6 0 m2 0 0 0 0 7
6 7
6 0 0 m2 0 0 0 7
mkk ¼ 6
e 6 7 (11.9.13b)
7
60 0 0 0 0 07
40 0 0 0 0 05
0 0 0 0 0 0
mejk ¼ mekj ¼ 0 (11.9.13c)
@CivilMethod
(iii) Equivalent nodal loads of the space frame element

The element is subjected to the axial load px ðx, t Þ, the transverse loads
py ðx, t Þ, pz ðx, t Þ, and the torsional moment mx ðx, t Þ. The resulting equivalent
load vector is

e
pj
pe ð t Þ ¼ (11.9.14)
pek
where
pej ¼ f p1 p2 p3 p4 p5 p6 gT (11.9.15a)
pek ¼ f p7 p8 p9 p10 p11 p12 gT (11.9.15b)

The components pi (i ¼ 1, 2, …, 12) are given by
Z L
0
Z L
pi ðt Þ ¼ py ðx, t Þ i ðx Þdx, i ¼ 2, 6, 8, 12 (11.9.16b)
0
Z L
pi ðt Þ ¼ pz ðx, t Þ i ðx Þdx, i ¼ 3, 5, 9, 11 (11.9.16c)
0
Z L
pi ðt Þ ¼ mx ðx, t Þ i ðx Þdx, i ¼ 4, 10 (11.9.16d)
0

In this case, we assume that the element is a simply supported beam under
the loads px ðx, t Þ, py ðx, t Þ, pz ðx, t Þ and the torsional moment mx ðx, t Þ.
Apparently, the components p5 ðt Þ, p6 ðt Þ, p11 ðt Þ, and p12 ðt Þ are zero. The
remaining components result as
Z L
p i ðt Þ ¼ px ðx, t Þ i ðx Þdx, i ¼ 1, 7 (11.9.17a)
0
Z L
pi ðt Þ ¼ py ðx, t Þ i ðx Þdx, i ¼ 2, 8 (11.9.17b)
0
Z L
pi ðt Þ ¼ pz ðx, t Þ i ðx Þdx, i ¼ 3, 9 (11.9.17c)
0
Z L
pi ðt Þ ¼ mx ðx, t Þ i ðx Þdx, i ¼ 4, 10 (11.9.17d)
0
@CivilMethod
where i ðx Þ ¼ 1 x, i + 6 ðx Þ ¼ x, i ¼ 1, 2, 3, 4. Therefore
pej ðt Þ ¼ f p1 p2 p3 p4 0 0 gT (11.9.18a)
pek ¼ f p7 p8 p9 p10 0 0 gT (11.9.18b)
11.9.2 Transformation of the nodal coordinates of the space

frame element
We consider the two systems of axes: the global system x y z and the local system
xyz. The latter is defined by the axis x of the element and the principal axes y, z of
its cross-section, Fig. 11.9.3. The base unit vectors of the two systems are denoted
by ei and ei i ¼ 1, 2, 3, respectively. The projection of ei on the axes x, y z gives
e1 ¼ l11 e1 + l12 e2 + l13 e3 (11.9.19a)
e2 ¼ l21 e1 + l22 e2 + l23 e3 (11.9.19b)
e3 ¼ l31 e1 + l32 e2 + l33 e3 (11.9.19c)
The quantities lij are the direction cosines of the xyz axes with respect
to x y z. Because ei are unit vectors and normal to each other, we have
jei j2 ¼ l2i1 + l2i2 + l2i3 ¼ 1 and ei ej ¼ li1 lj1 + li2 lj2 + li3 lj3 ¼ 0, i ¼
6 j.
For the matrix of the direction cosines
2 3
l11 l12 l13
L ¼ 4 l21 l22 l23 5 (11.9.20)
l31 l32 l33
we can readily show that det ðLÞ ¼ 1. Hence, the matrix L is orthonormal and
consequently
L1 ¼ LT (11.9.21)
Fig. 11.9.3 Global and local systems of axes of the space frame element.
@CivilMethod
A vector a is written with respect to the two systems of axes

a ¼ a1 e1 + a2 e2 + a3 e3 (11.9.22)
a ¼ a1 e1 + a2 e2 + a3 e3 (11.9.23)
where a1 , a2 ,a3 and a1 , a2 , a3 denote the components of vector a with respect to
the local and global axes, respectively.
Substituting Eqs. (11.9.19a)–(11.9.19c) into Eq. (11.9.23) gives
a ¼ðl11 a1 + l21 a2 + l31 a3 Þ
e1 + ðl12 a1 + l22 a2 + l32 a3 Þ
e2
(11.9.24)
+ ðl13 a1 + l23 a2 + l33 a3 Þ
e3
which, when compared with Eq. (11.9.22), yield
a1 ¼ l11 a1 + l21 a2 + l31 a3
a2 ¼ l12 a1 + l22 a2 + l32 a3 (11.9.25)
a3 ¼ l13 a1 + l23 a2 + l33 a3
or
a ¼ LT a (11.9.26)
and by virtue of Eq. (11.9.21)
a ¼ L
a (11.9.27)
The matrix L is the transformation matrix of the components of a vector
from the axes x y z to xyz. In two dimensions, the matrix L is determined from
the Cartesian coordinates of the end points of the element. These are, however,
not adequate in three dimensions because an infinite number of systems may
have the common x axis. In order to define the system of local axes, we must
also know one of the principal axes of the cross-section. The simplest
way to
accomplish it is to specify the global coordinates of a point P xp , yp , zp on
one of the principal planes, say of xy.
The direction cosines of the x axis are computed from the relations
xk xj yk yj zk zj
l11 ¼ , l12 ¼ , l13 ¼ (11.9.28)
L L L

where xj , yj , zj , ðxk , yk , zk Þ are the global coordinates of the end points of the
h 2 2 2 i1=2
e element with length L ¼ xk xj + yk yj + zk zj . The points
j and P define the vector

r ¼ xp xj e1 + yp yj e2 + zp zj e3 (11.9.29)
The vector e3 is obtained from the vector product
e1 r
e3 ¼ (11.9.30)
je1 rj
@CivilMethod
This is a unit vector and as a vector product, it is normal to the plane defined
by the vectors e1 and r, that is, it is in the direction of the local z axis. Its com-
ponents give the direction cosines l31 , l32 , l33 .
Finally, the direction cosines l21 , l22 , l23 will result from the components
of the vector
e2 ¼ e3 e1 (11.9.31)
Apparently, the end displacements (translations and rotations) of the
element are transformed according to Eq. (11.9.27). Thus, we have
8 9 2 38 9
< u1 = l11 l12 l13 < u1 =
u ¼ 4 l21 l22 l23 5 u2 (11.9.32a)
: 2; : ;
u3 l31 l32 l33 u3
8 9 2 38 9
< u4 = l11 l12 l13 < u4 =
u5 ¼ 4 l21 l22 l23 5 u5 (11.9.32b)
: ; : ;
u6 l31 l32 l33 u6
8 9 2 38 9
< u7 = l11 l12 l13 < u7 =
u ¼ 4 l21 l22 l23 5 u8 (11.9.32c)
: 8; : ;
u9 l31 l32 l33 u9
8 9 2 38 9
< u10 = l11 l12 l13 < u10 =
u ¼ 4 l21 l22 l23 5 u11 (11.9.32d)
: 11 ; : ;
u12 l31 l32 l33 u12
which are combined to yield
ue ¼ Re ue (11.9.33)
where
2 3
L 0 0 0
60 L 0 07
R ¼6
e
40
7 (11.9.34)
0 L 05
0 0 0 L
Obviously, the matrix Re with dimensions 12 12 represents the transfor-
mation matrix of the e space frame element. This vector is orthonormal, hence
Eq. (11.9.33) is inverted as
ue ¼ ðRe ÞT ue (11.9.35)

The vectors of the equivalent nodal forces obey the same transformation
law. Namely
f e ¼ðRe ÞT f e (11.9.36)
S S
f e ¼ðRe ÞT f eI (11.9.37)
I
@CivilMethod
f e ¼ðRe ÞT f e (11.9.38)
D D
pe ðt Þ ¼ðRe ÞT pe ðt Þ (11.9.39)

e
It can be readily shown that the stiffness matrix k , damping matrix ce , and
mass matrix m e of the element e with respect to the global axes are computed
from the relations
k ¼ ðRe ÞT ke Re
e
(11.9.40)
e T e

ce ¼ ð R Þ c Re (11.9.41)
e T
e ¼ ð R Þ m e Re
m (11.9.42)
Applying the procedure presented in Section 11.2.3 for the plane truss,
we obtain the equation of motion of the space frame, which, if damping is taken
into account, is written as
M u_ + K
u€ + C u ¼
pðt Þ (11.9.43)
or after applying the support conditions and partitioning
M u€f + C
e ff e u_ f + K
uf ¼ e (11.9.44)
M u€f + M
e sf e u€s + C
e ss e u_ f + C
e sf e u_ s + K
uf + K us ¼ e
ps ðt Þ (11.9.45)
where
pf∗ ðt Þ ¼ e
e fs e u_ s K
e fs e e fs e
us (11.9.46)
For convenience, we will write Eq. (11.9.44) as
M€
u + Cu_ + Ku ¼ pðt Þ (11.9.47)
Example 11.9.1 Compute the transformation matrix L of the space frame

element, which is determined from the end coordinates j ð2, 1, 1Þ, k ð5, 2, 2Þ
and the point P ð7, 0, 0Þ of the principal xy plane.
Solution
First, we compute
Dx ¼ xk xj ¼ 3, Dy ¼ yk yj ¼ 1, Dz ¼ zk zj ¼ 1

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
L ¼ 32 + 11 + 12 ¼ 3:317
and using Eq. (11.9.28), we obtain
Dx Dy Dz
l11 ¼ ¼ 0:904, l12 ¼ ¼ 0:301, l13 ¼ ¼ 0:301 (1)
L L L
@CivilMethod
Then we compute

rp ¼ xp xj e1 + yp yj e2 + zp zj e3
e1 e2 e3
¼5
2 3
e1 e2 e3
6 7
rp ¼ 6
e1 4 0:904 0:301 0:301 7
5 ¼ 0
e1 + 2:412
e2 2:412
e3
5 1 1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
je1 rj ¼ 02 + 2:4122 + ð2:412Þ2 ¼ 3:141 (2)

e1 r
e3 ¼ ¼ 0
e1 + 0:707
e2 0:707
e3 (3)
je1 rj
Finally, using Eq. (11.9.31), we obtain
2 3
e1 e2 e3
6 7
e2 ¼ e3 e1 ¼ 6
40 0:707 0:707 7 5 ¼ 0:426
e1 + 0:639
e2 0:639
e3 (4)
0:904 0:301 0:301
Hence
2 3
0:904 0:301 0:301
6 7
L¼6 7
4 0:426 0:639 0:639 5 (5)
0 0:707 0:707
Example 11.9.2 Formulate the equation of motion of the space frame

of Fig. E11.10a. The nodal coordinates are 1 ð0:886L, 0, 0:5LÞ,
2 ð0, 0:5L, 0:5LÞ, 3 ð0, 0:5L, 0:5LÞ, 4 ð0:886L, 0, 0:5LÞ. The frame is
fixed at nodes 1 and 4. The local z axis of all elements is parallel to the
zx plane. The elements have a rectangular cross-section A with dimensions
h and b in the directions of y and z axes, respectively. The axial deformation
of the elements is ignored and the lumped mass assumption is adopted. The
element 2 is loaded by the distributed load py ðx, t Þ ¼ 5f ðt Þ while a
moment Mðt Þ ¼ f 0 0 200Lf ðt Þ gT is applied at node 3. Assume:
h=b ¼ 2, A ¼ 300Iy , elastic constants E, n ¼ 0:2, and mass density r.
@CivilMethod
(a)
Fig. E11.10a Space frame in Example 11.9.2.
(b)
Fig. E11.10b Translational and rotational degrees of freedom in Example 11.9.2.
Solution
The system has n ¼ 4 nodes. Thus, the free structure has N ¼ 6n ¼ 24
degrees of freedom. We set Iy ¼ I , hence A ¼ 300I , Iz ¼ 4I , and
It ¼ 0:229hb3 ¼ 2:748I . Moreover, it is G ¼ 0:4166E.
e xj xk D
x yj yk D
y zj zk D
z
1 0.886L 0 0.886L 0 0.5L 0.5L 0.5L 0.5L 0

2 0 0 0 0.5L 0.5L 0 0.5L 0.5L L
3 0 0.886L 0.886L 0.5L 0 0.5L 0.5L 0.5L 0
@CivilMethod
Computation of ke , me , pe , Re for e ¼ 1, 2, 3. Applying Eqs. (11.9.10a)–

(11.9.10d) we obtain
2 3
7500 0 0 0 0 0
60 48 0 0 0 120 7
6 7
EI 6 0 0 12 0 30 0 7
k1jj ¼ 3 6 7,
L 660 0 0 28:62 0 0 7 7
40 0 30 0 100 0 5
0 120 0 0 0 400
2 3
7500 0 0 0 0 0
60 48 0 0 0 120 7
6 7
EI 6 0 0 12 0 30 0 7
kkk ¼ 3 6
1 7
L 66 0 0 0 28:62 0 0 7 7
40 0 30 0 100 0 5
0 120 0 0 0 400
2 3
7500 0 0 0 0 0
6 0 48 0 0 0 120 7
6 7 T
EI 6 0 0 12 0 30 0 7
6
kjk ¼ 3 6
1 7, k 1 ¼ k1
L 6 0 0 0 28:62 0 0 7 7
kj jk
4 0 0 30 0 50 0 5
0 120 0 0 0 200
k2 ¼ k3 ¼ k1
The lumped mass assumption results in the matrices me
2 3 2 3
0:5 0 0 0 0 0 0:5 0 0 0 0 0
60 0:5 0 0 0 07 6 0 0:5 0 0 0 07
6 7 6 7
60 0 0:5 0 0 0 7 6 07
mjj ¼ rAL6
1 7, m1 ¼ rAL6 0 0 0:5 0 0 7
60 0 0 0 0 07 kk 60 0 0 0 0 07
6 7 6 7
40 0 0 0 0 05 40 0 0 0 0 05
0 0 0 0 0 0 0 0 0 0 0 0
T
m1jk ¼ m1kj ¼ 0 and m2 ¼ m3 ¼ m1
@CivilMethod
Because the loading is referred to the global axes, we obtain directly the
equivalent nodal loads pe ðt Þ
8 9 8 9 8 9
>
> 0 > > >
> 0 > > >
> 0>>
> >
> > >
> > >
>
> 2:5 >
>
>
> 2:5 >
>
>
> 0>>
>
< = >
< = < >
> =
0 0 0
pj ¼
2
Lf ðt Þ, pk ¼
2
Lf ðt Þ, pj ¼ pk ¼ pj ¼ pk ¼
1 1 3 3
> 0 >
> > >
> 0 > > >
> 0>>
> > > > > >
>
> 0 > >
>
>
> 0 > >
>
>
> 0>>
>
: ; >
: ; : >
> ;
0 0 0
Taking the positive direction of the local x to coincide with increasing node
number and the points P1 20 , P2 O, P3 30 to define the principal planes xy
of the elements 1, 2, 3, respectively, and Table E11.5, we obtain:
Element 1:
e1 ¼ ½ 0:866 0:500 0 , rP1 ¼ f 4:33 0 0 gT ,
e1 rP1
e3 ¼ ¼ ½ 0 0 1 , e2 ¼ e3 e1 ¼ ½ 0:5 0:866 0
je1 rP1 j
2 1 3
2 3 L 0 0 0
0:866 0:500 0 60
4 5 6 L1 0 0 77
L ¼ 0:5
1
0:866 0 , R ¼ 4
1
0 0 L1 0 5
0 0 1
0 0 0 L1
Element 2:
e1 rP2
e1 ¼ ½ 0 0 1 , rP2 ¼ ½ 0 2:5 2:5 , e3 ¼ ¼ ½ 1 0 0
je1 rP2 j
e2 ¼ ½ 0 1 0
2 3
2 3 L2 0 0 0
0 0 1 60 L2 0 0 7
L2 ¼ 4 0 1 0 5, R2 ¼ 6
40
7
0 L2 0 5
1 0 0
0 0 0 L2
@CivilMethod
Element 3:
e1 rP3
e1 ¼ ½ 0:866 0:500 0 , rP3 ¼ ½ 0 2:5 0 , e3 ¼ ¼ ½ 0 0 1
je1 rP3 j
e2 ¼ ½0:50 0:866 0
2 3 3
2 3 L 0 0 0
0:866 0:500 0 60 L3 0 0 7
L ¼ 4 0:500 0:866 0 5, R ¼ 6
3 3
40
7
0 L3 0 5
0 0 1
0 0 0 L3
Computation of total matrices M, K and total load vector p of the struc-

ture. First, we transform the element matrices and load vectors from the
local to global axes using Eqs. (11.9.39), (11.9.40), (11.9.42). Then we
determine the element assembly matrices ae ðe ¼ 1, 2, 3Þ, which are used
to evaluate the enlarged element matrices and load vectors. Finally, we eval-
uate the total matrices and the total load vector of the free structure from the
relations
X
3 X
3 X
3
¼ ðae ÞT k ae , M
¼
e
K e ae , pðt Þ ¼ Pext +
ða e ÞT m ðae ÞT pe
e¼1 e¼1 e¼1
where Pext ¼ f0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 200Lf ðt Þ 0 0 0 0 0 0gT is the vector

of the loads directly applied to the nodes. Subsequently, the transformation
matrix V is constructed by considering the support conditions. After
partitioning the transformed matrices, we compute the matrices M e ff and
e ff , K
e ∗
the load vector pf using Eqs. (11.9.44) and (11.9.45). This procedure yields
the equation of motion in terms of the 12 free degrees of freedom
ef ¼ f u7 u8 u9 u10 u11 u12 u13 u14 u15 u16 u17 u18 gT
u
M u€f + K
e ff e e ff e
uf ¼ pt∗
Then, reordering the displacements to separate the translations from the

rotations and eliminating the six rotational degrees of freedom yields the fol-
lowing equation with respect to the six translational degrees of freedom
f u7 u8 u9 u13 u14 u15 gT
@CivilMethod
2 38 € 9
1:0 0 0 0 0 >
> u7 >
0 >
>
> >
>
6 7>> >
60 7
1:0 0 0 0 0 7>> u8 >
€ >
>
6 >
> >
>
6 7>> >
>
60 7 <
0 1:0 0 0 0 7 u€9 =
6
rAL6 7
60 0 0 1:0 0 0 7 > u€13 >
6 7>> >
>
6 7>>
>
>
>
>
60 7
0 0 0 1:0 0 5>> >
4 > 14 >
> u€ >
>
>
> >
:€ > ;
0 0 0 0 0 1:0 u15
2 38 9
225L2 + 7:68 130L2 6:17 3:17 4:68 0:976 3:17 >
> u7 >>
6 130L2 6:17 75L2 + 13:6 7>> >
6 1:68 0:976 1:26 1:68 7>> u8 >>
>
6 7 >
< >
=
EI 6 6:17 1:68 2
300L + 7:66 3:17 1:68 300L 0:88 7 u9
2
+ 36 7
L 6 4:68 3:17 225L2 + 7:68 130L2 + 6:17 7> u13 >
7> >
0:976 3:17
6 >
> >
>
4 0:976 1:26 1:68 1:68 2
75L + 13:6 1:68 5 >
> u14 >
>
>
: >
;
3:17 1:68 300L2 0:88 3:17 1:68 300L2 u15
8 9
>
> 9:39 >
>
>
> >
>
> 16:26 2:5L >
>
>
>
< >
=
0
¼ f ðt Þ
>
> 140:61 >
>
>
> >
>
>
> 243:54 2:5L >
>
>
: >
;
0
To save space, the intermediate computed matrices are not given here
because they are very large. The reader can find them on this book’s companion
website.
Reduction of the degrees of freedom due to axial constraints. The equations
of the constraints are
l111 u7 + l112 u8 + l113 u9 ¼ 0

l211 u7 + l212 u8 + l213 u9 ¼ l211 u13 + l212 u14 + l213 u15
l311 u13 + l312 u14 + l313 u15 ¼ 0
Using the computed values for lij the constraint matrix is
2 3
0:866 0:5 0 0 0 0
6 7
D ¼ 40 0 1 0 0 15
0 0 0 0:866 0:5 0
The Gauss-Jordan elimination yields

2 3
1 0:577 0 0 0 0
6 7
4 0 0 1 0 0 1 5
0 0 0 1 0:577 0
We readily conclude that the rank of the constraint matrix is 3. Therefore,

the independent displacements are 3. In the following, we take u8 u9 u14 as
independent displacements. This requires the rearrangement of the equations
using the transformation matrix
2 3
0 0 0 1 0 0
61 0 0 0 0 07
6 7
6 7
60 1 0 0 0 07
^
V¼6 6 7
7
60 0 0 0 1 07
6 7
40 0 1 0 0 05
0 0 0 0 0 1
which transforms the constraint equations to
@CivilMethod
Thus the constraint matrix becomes
Hence
2 3 2 3
0:5 0 0 0:866 0 0
Da ¼ 4 0 1 0 5, Db ¼ 4 0 0 1 5,
0 0 0:5 0 0:866 0
2 3
1 0 0
6 7

6 0 1 0 7
I 6 0 0 1 7
T¼ ¼ 6 7
1
Db Da 6 7
6 0:577 0 0 7
4 0 0 0:577 5
0 1 0
2 3
1:33 0 0
^ ¼ TT V
M ^TM
e tt V
^ T ¼ rAL4 0 2 0 5
0 0 1:33
2 3
23:3 3:35 1:43
^ ¼ TT
K ^TK
V e ∗V ^ T ¼ EI 4 3:35 13:5 3:35 5
tt
L3
1:43 3:35 23:3
8 9
T ∗ < 21:68 2:5L =
p ^ p ðt Þ ¼
^ðt Þ ¼ TT V 0 f ðt Þ
t
: ;
324:73 2:5L
and the sought equation of motion reads
2 38 9 2 38 9
1:33 0 0 < u€8 = EI 23:3 3:35 1:43 < u8 =
rAL4 0 2 0 5 u€9 + 4 3:35 13:5 3:35 5 u9
: € ; L3 : ;
0 1:33 u14 1:43 3:35 23:3 u14
8 9
< 21:68 2:5L =
¼ 0 f ðt Þ
: ;
324:73 2:5L
@CivilMethod
11.10 The finite element method for the space truss

11.10.1 Properties of the space truss element
In a space truss (Fig. 11.10.1), the rods are interconnected at the nodes, which
are spherical hinges and as such allow freely the relative rotation of the end
cross-sections. In the deformed state at time t, the ends of the element j, k
are displaced to points j 0 , k 0 and the element occupies the position j 0 k 0 .
Therefore, the vector of nodal coordinates is (see Fig. 11.10.2)
u ¼ f u1 u2 u3 u4 u5 u6 gT (11.10.1)
According to what we mentioned for the plane truss, a point on the x
local axis of the element will undergo displacements along the x, y and z axes,
that is,
u ðx, t Þ ¼ u1 1 ðx Þ + u4 4 ðx Þ (11.10.2a)
v ðx, t Þ ¼ u2 2 ðx Þ + u 5 5 ðx Þ (11.10.2b)
w ðx, t Þ ¼ u3 3 ðx Þ + u6 6 ðx Þ (11.10.2c)
Fig. 11.10.1 Space truss.
Fig. 11.10.2 Nodal displacements of a space truss element.
@CivilMethod
where
1 ðx Þ ¼ 2 ðx Þ ¼ 3 ðx Þ ¼ 1 x, x ¼ x=L (11.10.3a)
4 ðx Þ ¼ 5 ðx Þ ¼ 6 ðx Þ ¼ x (11.10.3b)
Because the element does not undergo bending, the axes on the cross-section
of the rod may have any orientation. Nevertheless, it is recommended to take
them in the directions of the cross-sectional principal axes.
The elastic energy of the element is due only to the axial deformation.
Therefore, for an element with variable cross-section Aðx Þ and mass m ðx Þ, it is
Z
1 L
EAðx Þ½u 0 ðx, t Þ dx
2
U ðu1 , u2 , …, u6 Þ ¼ (11.10.4)
2 0
The kinetic energy for consistent mass assumption is
Z n o
1 L
T ðu_ 1 , u_ 2 , …, u_ 6 Þ ¼ m ðx Þ ½u_ ðx, t Þ2 + ½v_ ðx, t Þ2 + ½w_ ðx, t Þ2 dx
2 0
(11.10.5)
while for lumped mass assumption is
1 1
T ðu_ 1 , u_ 2 , …, u_ 6 Þ ¼ m1 u_ 21 + u_ 22 + u_ 23 + m2 u_ 24 + u_ 25 + u_ 26 (11.10.6)
2 2
where
Z L Z L
m1 ¼ m ðx Þð1 x Þdx and m2 ¼ m ðx Þxdx (11.10.7)
0 0
Applying the procedure developed for the plane truss, we obtain the
stiffness matrix, the mass matrix, and the vector of equivalent nodal forces
of the space truss.
(i) Nodal elastic forces and stiffness matrix of the space truss element
The relation between the elastic nodal forces and nodal displacements is
f eS ¼ ke ue (11.10.8)
in which
8 9
> fS1 >
> >
>f >
> >
>
> S2 >
>
> >
<f >
> =
S3
fS ¼
e
(11.10.9a)
> fS4 >
> >
>
> >
>
>
> >
>
> f >
: >
> S5
;
fS6
@CivilMethod
8 9
> u1 >
> >
> >
>
>
> u2 >
>
>
> >

> =
3
u ¼
e
(11.10.9b)
>
> u4 >
>
>
> >
>
>
> >
>
> u >
: >
> 5
;
u6
2 3
k11 0 0 k14 0 0
60 0 07
6 0 0 0 7
6 7
60 0 0 0 0 07
ke ¼ 6
6k
7 (11.10.9c)
6 41 0 0 k44 0 07
7
6 7
40 0 0 0 0 05
0 0 0 0 0 0
where
Z L
0 0
kij ¼ EAðx Þ i j dx, i, j ¼ 1, 4 (11.10.10)
0
For Aðx Þ ¼ A ¼ constant it is

2 3
1 0 0 1 0 0
6 0 0 0 0 0 07
6 7
e6
6
7
EA 0 0 0 0 0 07
ke ¼ e 6 7 (11.10.11)
L 66 1 0 0 1 0 0 7
7
6 7
4 0 0 0 0 0 05
0 0 0 0 0 0
The relation between the inertial nodal forces and nodal displacements is
€e
f eI ¼ me u (11.10.12)
in which
8 9
> fI 1 >
> >
>f >
> >
>
> I2 >>
> >
<f >
> =
I3
fI ¼
e
(11.10.13a)
>
> fI 4 >
>
>
> >
>
>
> >
>
> f >
: >
> I 5
;
fI 6
@CivilMethod
8 9
> u€1 >
> >
> >
>
>
> u€2 >
>
>
> >
< u€ >
> =
3
€ ¼
u e
(11.10.13b)
>
> u€4 >
>
>
> >
>
>
> € >
>
> u >
: >
> 5
;
u€6
and me is the mass matrix.
For the consistent mass assumption, we obtain
2 3
m11 0 0 m14 0 0
60 m22 0 0 m25 0 7
6 7
6 0 0 m 0 0 m36 7
m ¼6
e 6 33 7 (11.10.14)
6 m41 0 0 m44 0 0 7 7
40 m52 0 0 m54 0 5
0 0 m63 0 0 m66
where
Z L
mij ¼ mi + 1, j + 1 ¼ mi + 2, j + 2 ¼ m ðx Þ i ðx Þ j ðx Þdx, i, j ¼ 1, 4
0
(11.10.15)
¼ constant the matrix
For m ðx Þ ¼ m in Eq. (11.10.14) becomes
2 3
2 0 0 1 0 0
60 2 0 0 1 07
6 7
me 6
60 0 2 0 0 17 7
m ¼
e
(11.10.16)
6 6
61 0 0 2 0 07 7
40 1 0 0 2 05
0 0 1 0 0 2
¼ rAL is the total mass of the element.
where m e ¼ mL
For lumped mass assumption, the mass matrix results as
2 3
m11 0 0 0 0 0
60 m22 0 0 0 0 7
6 7
6 0 0 m33 0 0 0 7
me ¼ 6
60
7 (11.10.17)
6 0 0 m44 0 0 7 7
40 0 0 0 m55 0 5
0 0 0 0 0 m66
where m11 ¼ m22 ¼ m33 ¼ m1 , m44 ¼ m55 ¼ m66 ¼ m2 and
Z L Z L
m1 ¼ m ðx Þð1 xÞdx, m2 ¼ m ðx Þxdx (11.10.18)
0 0
@CivilMethod
¼ constant, Eq. (11.10.17) becomes

When m ðx Þ ¼ m
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
1 60 0 1 0 0 07
me ¼ m e 6 7 (11.10.19)
2 6 60 0 0 1 0 07
7
40 0 0 0 1 05
0 0 0 0 0 1
(iii) Equivalent nodal loads of the space frame element

If the element is subjected to the axial load pðx, t Þ, the equivalent nodal load
vector is
8 9
>
> p1 ðt Þ >
>
>
> >
>
>
> 0 >
>
< =
0
p ðt Þ ¼
e
(11.10.20)
>
> p4 ðt Þ >
>
>
> >
>
>
> 0 >
>
: ;
0
where
Z L
pi ðt Þ ¼ px ðx, t Þ i ðx Þdx, i ¼ 1, 4 (11.10.21)
0
11.10.2 Transformation of the nodal coordinates of the space

truss element
By selecting the local axes y, z to coincide with the principal axes of the cross-
section, we can compute the matrix of the direction cosines of the element
exactly as for the space frame element. Then, the nodal coordinates at the
end points are transformed according to Eqs. (11.9.32a), (11.9.32b). Hence,
we will have
8 9 2 38 9
< u1 = l11 l12 l13 < u1 =
u ¼ 4 l21 l22 l23 5 u2 (11.10.22a)
: 2; : ;
u3 l31 l32 l33 u3
8 9 2 38 9
< u4 = l11 l12 l13 < u4 =
u ¼ 4 l21 l22 l23 5 u5 (11.10.22b)
: 5; : ;
u6 l31 l32 l33 u6
which are combined to
ue ¼ ðRe ÞT ue (11.10.23)

@CivilMethod
where

L 0
Re ¼ (11.10.24)
0 L
is the transformation matrix of the e element of the space truss with dimensions
6 6. The remaining procedure for formulating the equation of motion is the
same as for the plane truss.
Example 11.10.1 Formulate the equation of motion of the space truss of
Fig. E11.11. All bars have the same cross-sectional area A. The distributed load
on element 1 is pðx, t Þ ¼ ð1 x=LÞP ðt Þ=L. Assume: modulus of elasticity E,
material density r, lumped mass assumption.
Solution
Fig. E11.11 Space truss in Example 11.10.1.
The free structure has n ¼ 4 nodes and N ¼ 3n ¼ 12 degrees of freedom. We

assume that the principal planes xy of all elements are normal to the plane
x y. Thus, we take the projection of the node 4 on the plane x y as point P, that
is, P ð1:5, 2, 0Þ, which is common for all elements. Then, taking into account the
geometrical data of Table E11.6, we obtain

e xj xk D
x yj yk D
y zj zk D
z L
1 3 1.5 1.5 0 2 2 0 4 4 4.717

2 0 1.5 1.5 0 2 2 0 4 4 4.717
3 3 1.5 1.5 4 2 2 0 4 4 4.717
@CivilMethod
Computation of the transformation matrices Re, e¼1,2,3

Element 1
e1 ¼ f 0:318 0:424 0:848 gT , rP1 ¼ f 1:5 2 0 gT

e1 rP1 T T
e3 ¼ ¼ f 0:8 0:6 0 g , e2 ¼ e3 e1 ¼ f 0:508 0:678 0:530 g
je1 rP1 j
2 3
0:318 0:424 0:848 " #
6 7 L 1
0
L1 ¼ 4 0:508 0:678 0:530 5, R1 ¼
0 L1
0:800 0:600 0
Element 2
e1 ¼ f 0:318 0:424 0:848 gT , rP2 ¼ f 1:5 2 0 gT

e1 rP2
e3 ¼ ¼ f 0:8 0:6 0 gT , e2 ¼ e3 e1 ¼ f 0:508 0:678 0:530 gT
je1 rP2 j
2 3
0:318 0:424 0:848 2
L 0
L ¼ 4 0:508 0:678 0:530 5, R2 ¼
2
0 L2
0:800 0:600 0
Element 3
e1 ¼ f 0:318 0:424 0:848 gT , rP3 ¼ f 1:5 2 0 gT

e1 rP3
e3 ¼ ¼ f 0:8 0:6 0 gT , e2 ¼ e3 e1 ¼ f 0:508 0:678 0:530 gT
je1 rj
2 3
0:318 0:424 0:848 3
L 0
L3 ¼ 4 0:508 5
0:678 0:530 , R ¼
3
0 L3
0:800 0:600 0
Computation of ke , me , pe ðt Þ, e ¼ 1, 2, 3
2 3
1 0 0 1 0 0
6 0 0:5 0 0 0 0 7
6 7
6 7
EA 6 0 0 0:5 0 0 0 7
k1 ¼ k2 ¼ k3 ¼ 6 7
L 6 1 0 0 1 0 0 7
6 7
4 0 0 0 0 0:5 0 5
0 0 0 0 0 0:5
2 3
0:5 0 0 0 0 0
60 0:5 0 0 0 0 7
6 7
6 0 0 0:5 0 0 0 7
m1 ¼ m2 ¼ m3 ¼ rAL6 60
7
6 0 0 0:5 0 0 7 7
40 0 0 0 0:5 0 5
0 0 0 0 0 0:5
@CivilMethod
Z L
p11 ðt Þ ¼ pðx, t Þ 1 ðx Þdx ¼ 0:333P ðt Þ,
0
Z L
p14 ðt Þ ¼ pðx, t Þ 4 ðx Þdx ¼ 0:167P ðt Þ
0
8 9
>
> 0:333 >
>
>
> >
>
>
> 0 >
>
< =
0
p ðt Þ ¼
1
P ðt Þ, p2 ðt Þ ¼ p3 ðt Þ ¼ 0
>
> 0:167 >
>
>
> >
>
>
>0 >
>
: ;
0
Computation of k , m
e
e , pe ðt Þ, e ¼ 1, 2, 3
T
k1 ¼ R1 k1 R1
2 3
0:101 0:135 0:27 0:101 0:135 0:27
6 0:135 0:18 0:135 0:18 0:36 7
6 0:36 7
6 7
EA 6
6
0:27 0:36 0:719 0:27 0:36 0:719 7
7
¼
L 6
6 0:101 0:135 0:27 0:101 0:135 0:27 7
7
6 7
4 0:135 0:18 0:36 0:135 0:18 0:36 5
0:27 0:36 0:719 0:27 0:36 0:719

k2 ¼ R2 k2 R2
T
2 3
0:101 0:135 0:27 0:101 0:135 0:27
6 0:135 0:36 0:36 7
0:135 0:18
6 0:18 7
6 7
EA 6
6
0:27 0:36 0:719 0:27 0:36 0:719 7
7
¼ 6
L 6 0:101 0:135 0:27 0:101 0:135 0:27 77
6 7
4 0:135 0:18 0:36 0:135 0:18 0:36 5
0:27 0:36 0:719 0:27 0:36 0:719
T
k3 ¼ R3 k3 R3
2 3
0:101 0:135 0:27 0:101 0:135 0:27
6 0:135 0:36 0:135 0:18 0:36 7
6 0:18 7
6 7
EA 6
6
0:27 0:36 0:719 0:27 0:36 0:719 7
7
¼ 6
L 6 0:101 0:135 0:27 0:101 0:135 0:27 77
6 7
4 0:135 0:18 0:36 0:135 0:18 0:36 5
0:27 0:36 0:719 0:27 0:36 0:719
@CivilMethod
2 3
0:5 0 0 0 0 0
6 0 0:5 0 0 0 0 7
6 7
6 7
6 0 0 0:5 0 0 0 7
1 ¼m
m 2 ¼m 3 ¼ rAL66 0 0 0 0:5 0 0 7,
7
6 7
6 7
4 0 0 0 0 0:5 0 5
0 0 0 0 0 0:5
8 9
> 0:106 >
>
> >
>
>
>
> 0:141 >
>
>
> >
< 0:283 >
> =
p1 ðt Þ ¼ P ðt Þ, p2 ðt Þ ¼ p3 ðt Þ ¼ 0
>
> 0:053 >
>
>
> >
>
>
>
>
> 0:070 >
>
>
>
: ;
0:141
Subsequently, the assembly matrices ae , ðe ¼ 1, 2, 3Þ are formulated
2 3 2 3
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
60
6 1 0 0 0 0 0 0 0 0 0 077
60
6 0 0 0 1 0 0 0 0 0 0 077
60 0 1 0 0 0 0 0 0 0 0 07 6
7, a2 ¼ 6 0 0 0 0 0 1 0 0 0 0 0 077
a1 ¼ 6
60
6 0 0 0 0 0 0 0 0 1 0 077
60
6 0 0 0 0 0 0 0 0 1 0 077
40 0 0 0 0 0 0 0 0 0 1 05 40 0 0 0 0 0 0 0 0 0 1 05
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1
2 3
0 0 0 0 0 0 1 0 0 0 0 0
60
6 0 0 0 0 0 0 1 0 0 0 077
60 0 0 0 0 0 0 0 1 0 0 077
a3 ¼ 6
60
6 0 0 0 0 0 0 0 0 1 0 077
40 0 0 0 0 0 0 0 0 0 1 05
0 0 0 0 0 0 0 0 0 0 0 1
and the total matrices of the structure are computed, namely
X
3 X
3 X
3
¼ ðae ÞT k ae , M
¼
e
K e ae , pðt Þ ¼ Pðt Þ +
aT m ðae ÞT pðt Þe
e¼1 e¼1 e¼1
e M
Modified matrices K, e and load vector p eðt Þ due to support conditions
The equation of motion will be derived with respect to the three free degrees
of freedom of node 4. Due to the support conditions, the displacement vector
should be modified as
e e1 u
uT ¼ f u e2 u e3 ue4 ue5 ue6 ue7 ue8 ue9 ue10 ue11 u
e12 g
¼ f u10 u11 u12 u1 u2 u3 u4 u5 u6 u7 u8 u9 g
@CivilMethod
Therefore, the matrix V defined by Eq. (11.2.111) is

2 3
0 0 0 1 0 1 0 0 0 0 0 0
60 0 0 0 1 0 1 0 0 0 0 07
6 7
61 0 0 0 0 1 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 0 0 0 0 07
6 7
61 0 0 0 0 0 0 0 0 0 0 07
6 7
60 1 0 0 0 0 0 0 0 0 0 07
V¼6 60
7
6 0 1 0 0 0 0 0 0 0 0 077
60 0 0 1 0 0 0 0 0 0 0 07
6 7
60 0 0 0 1 0 0 0 0 0 0 07
6 7
60 0 0 0 0 1 0 0 0 0 0 07
6 7
40 0 0 0 0 0 1 0 0 0 0 05
0 0 0 0 0 0 0 1 0 0 0 0
After partitioning the modified matrices, the equation of motion results from
Eq. (10.2.117) which in terms of the original numeration reads
2 38 9 2 38 9
1:5 0 0 < u€10 = EA 0:303 0:135 0:270 < u10 =
rAL4 0 1:5 0 5 u€11 + 4 0:135 0:539 0:359 5 u11
:€ ; L : ;
0 0 1:5 u12 0:270 0:359 2:157 u12
8 9
< 0:053 =
¼ 0:929 P ðt Þ
: ;
1:859
11.11 Rigid bodies within flexible skeletal structures

Skeletal structures may often embrace bodies of finite dimensions whose defor-
mation is very small compared to that of the elements of the structure. These
bodies may be considered as rigid bodies. Fig. 11.11.1 shows examples of such
structures. During motion, these bodies oblige the ends of the elements con-
nected with them to ride the displacements of the rigid body, which are three
for the plane body and six for the three-dimensional body.
11.11.1 Rigid bodies in spaces frames

First, we will derive the equations of motion of a space frame embracing rigid
bodies. We consider an element e, whose ends j and k are connected to the bod-
ies ðJ Þ and ðK Þ, respectively, Fig 11.11.2. Points J and K of the bodies, which
in general do not coincide with the center of mass of the bodies, are taken as
points of reference of their motion. The axes x y z are parallel to the global axes
of the structure. All points of reference of the motion are considered as virtual
(fictitious) nodes of the frame, the displacements of which will be the
@CivilMethod
Fig. 11.11.1 Rigid bodies B in skeletal structures.
Fig. 11.11.2 Element e of a space frame with rigid bodies at its ends.
coordinates of the motion. This imposes the transfer of the element end
forces from its ends j and k to J and K , respectively. This is achieved by work-
ing as follows.
The positive direction of the element is from point j to point k. We denote by
Dx j ¼ xj xJ , D
y j ¼ yj yJ , D
z j ¼ zj zJ the coordinates of point j with
respect to the system J x y z (see Fig. 11.11.3). The position vector defined
by the relation
rj ¼ D
x j i + D
y j j + D
zj k (11.11.1)
will be referred to as the translation (transfer) vector or the offset vector and is
directed from point J to point j.
(i) Transfer of the nodal forces from j to J
@CivilMethod
Fig. 11.11.3 Rigid body (J ).
The nodal element forces at the end j and those transferred to point J are
given by
ð11:11:2a; bÞ
in which f aj , f aJ denote forces while f bj , f bJ moments all referred to the global

system of axes. Nevertheless, the overbars are omitted here for the sake of sim-
plifying the symbols. Therefore, the vector f j has dimensions 6 1 and may
represent any of the vectors f I , f S , pe ðt Þ at j. The damping is not considered
e e
here but it will be introduced as proportional damping in the final equation

of motion (see Section 12.11).
Due to transfer, we have
f aJ ¼ f aj (11.11.3a)
f bJ ¼ f bj + rj f aj (11.11.3b)
After expanding the vector product using vector components, the foregoing
equations are written as
ð11:11:4Þ
or
@CivilMethod f J ¼ Tj f j (11.11.5)
where

I 0
Tj ¼ (11.11.6)
ej I
represents the transformation matrix of the end forces from point j to J .
I represents 3 3 unit matrix and
2 3
0 Dzj Dyj
ej ¼ 4 D zj 0 Dxj 5 (11.11.7)
Dyj Dxj 0
is the offset (or eccentricity) matrix, which is equal to zero in the absence of a
rigid body because rj ¼ 0. Obviously, the matrix ej is antisymmetric, namely
eTj ¼ ej (11.11.8)
Eq. (11.11.5) can be readily inverted by taking point j as a reference point
and repeating the previous procedure. This yields
f j ¼ TTj f J (11.11.9)
where

I ej I eTj
TTj ¼ ¼ ¼ (11.11.10)
0 I 0 I
(ii) Transfer of the displacements from J to j

The nodal displacements at J and those transferred to point j are denoted in
correspondence to Eq. (11.11.2a) , Eq. (11.11.2b) by
ð11:11:11a; bÞ
in which uaj ,uaJ denote translations while ubj , ubJ rotations.

The displacements at j in terms of the displacements J are given by the
expressions
uaj ¼ uaJ (11.11.12a)
ubj ¼ ubJ + uaJ rj (11.11.12b)

The term uaJ rj in Eq. (11.11.12b) results from Eq. (Α.2.10) of Appendix Α
by setting w ¼uaJ and R ¼ rj .
After expanding the vector product using vector components, the foregoing
equations yield
@CivilMethod
uj ¼ TTj uJ (11.11.13)
Similar relations are obtained for the nodal displacements and forces at end
k. Namely
uk ¼ TTk uK (11.11.14)
f k ¼ TTk f K (11.11.15)
where
I eTk
TTk ¼ (11.11.16)
0 I
2 3
0 Dzk D
yk
ek ¼ 4 Dzk 0 xk 5
D (11.11.17)
Dy k Dxk 0
x k , D
The quantities D y k , Dz k are the coordinates of the point k with respect
to the system of axes K x y z (see Fig. 11.11.2). Thus, the offset vector is
rk ¼ D
x k i + D
y k j + D
zk k (11.11.18)
Eqs. (11.11.5), (11.11.15) are combined as
f 0 e ¼ Te f e (11.11.19)
where

f e ¼ fj 0e f T 0
, f ¼ J , Te ¼ j (11.11.20)
fk fK 0 Tk
Similarly, Eqs. (11.11.13), (11.11.14) are combined as

u0 ¼ Te ue
e
(11.11.21)
where

uj uJ
, u0 ¼
e
ue ¼ (11.11.22)
uk uK
The quantities provided with a prime are referred to the virtual nodes
J and K .
(iii) Transformation of the stiffness matrix, mass matrix, and equivalent nodal
load vector due to the offset
f 0 e ¼ Te f e
S S
¼ Te k ue
e
(11.11.23)
e e e T 0e
¼ T k ðT Þ u
hence the transformed stiffness matrix is
0e
k ¼ Te k ðTe ÞT
e
(11.11.24)
@CivilMethod
Applying the same procedure yields the transformed mass matrix

0 ¼ Te m
e
m e ðTe ÞT (11.11.25)
The vector of the equivalent nodal forces is transformed according to
Eq. (11.11.19). Thus, we have
p0 ðt Þe ¼ Te pðt Þe (11.11.26)
The remaining procedure to formulate the equation is known. That is,
^0 , M
^ 0 , and p
e e
^0 ðt Þ using the assembly
e
we formulate the enlarged matrices K
matrices and we establish the equation
0 u€0 + K
M 0 u0 ¼ p0 ðt Þ (11.11.27)
where
X
Ne X
Ne X
Ne
^0 , K ^ 0 , p0 ðt Þ ¼ P0 ðt Þ +
e e
0¼
M M 0 ¼ K ^ 0 ðt Þe
p (11.11.28)
e¼1 e¼1 e¼1
The components of the vector P0 ðt Þ represent the loads that are applied
directly to the virtual nodes. But in real structures, the external loads are usually
applied to points not coinciding with the virtual nodes, therefore they must
previously be transferred to these points. The procedure is the same as that
employed for the element nodal forces. Thus, if P is the point of application
of the load with an offset vector.
rp ¼ D
x p i + D
y p j + D
zpk (11.11.29)
its transfer from P to J is performed using the relation
PJ ðt Þ ¼ Tp Pp (11.11.30)
Eq. (11.11.27) is not the equation of motion of the structure because so far
the inertial forces produced by the motion of the rigid bodies have not been
included in it. Apparently, these forces are the most important in the considered
type structures.
This is achieved if the motion of the rigid bodies is taken into account.
In Appendix, it is shown that the motion of a rigid body with respect to its center
of mass is described by Eqs. (A.2.2), (A.2.8), namely
€c
F ¼ mR (11.10.31a)
Mc ¼ H
_ c (11.10.31b)
where F is the resultant force, Mc the resultant moment with respect to the
center of mass, and Hc the angular momentum.
Introducing Eq. (A.2.14) in Eq. (11.11.31b), using (Α.2.11) and neglecting
the nonlinear terms (the products of rotations) because we examine small
displacements, we obtain the linearized Euler’s equations with respect to the
principal axes of the body
@CivilMethod
€c
F ¼ mR (11.11.32a)

Mc ¼ Ix w_ x + Ixy w_ y + Ixz w_ z e1

+ Iyx w_ x + Iy w_ y + Iyz w_ z e2 (11.11.32b)

+ Izx w_ x + Izy w_ y + Iz w_ z e3
The quantities Ix , Ixy , …, Iz are the moments and the products of inertia,
which are given by Eq. (Α.2.15). Using the notation of the current section,
we write the forgoing equations
€c
f c ¼ Mc u (11.11.33)
where
8 9 8 9 2 3
>
> fc1 >
> >
> uc1 >> m 0 0 0 0 0
> >
> > >
> >
> 60 0 7
> f > > u > m 0 0 0
< >
> < >
> 6 7
c2 c2
= = 60
f uc3 0 m 0 0 0 7
f c ¼ c3 , uc ¼ , Mc ¼ 6
60
7 (11.11.34)
> fc4 >
> > >
> uc4 >> 6 0 0 Ix Ixy Ixz 7
7
>
> >
> >
> >
> 40
> fc5 >
> > > u > 0 0 Iyx Iy Iyz 5
: ; : c5 >
> ;
fc6 uc6 0 0 0 Izx Izy Iz
f c , uc , and Mc the vector of the inertia forces applied at the center of mass C , the
displacement vector of the center of mass and the mass (inertia) matrix of
the body with respect to center of mass, respectively. If the point of reference
of the motion is not the center of mass, but an arbitrary point P, then Mc must
be transformed from C to P. The transformation of f c and uc is achieved by
Eqs. (11.11.5), (11.11.13). Thus, we have
f p ¼ Tc f c (11.11.35)
uc ¼ TTc up (11.11.36)
where
2 3
0 D
zc D
yc
I 0
Tc ¼ , ec ¼ 4 Dzc 0 xc 5
D (11.11.37)
ec I
Dy c D
xc 0
and D x c ¼ xc xp , D
y c ¼ yc yp , D
z c ¼ zc zp are the components of the
offset vector rc from the center of mass to the reference point P.
After that, we may write
f p ¼ Tc f c
€c
¼ Tc M c u
€p
¼ Tc Mc TTc u
Hence
Mp ¼ Tc Mc TTc (11.11.38)
of 0
Then the mass matrices of all rigid bodies are added to mass matrix M
Eq. (11.11.27) after they are enlarged by appropriate assembly matrices. There-
@CivilMethod
fore, we obtain the equation of motion
∗ u€0 + K
M 0 u0 ¼ p0 ðt Þ (11.11.39)
where now
X
Nb X
Ne
^0
e
∗¼
M î +
M M (11.11.40)
p
i¼1 e¼1
in which M^ i are the enlarged mass matrices of the Nb rigid bodies. The proce-
p
dure of determining M ∗ is illustrated with Example 11.11.1.
11.11.2 Rigid bodies in spaces trusses, plane grids, plane frames,

and plane trusses
Rigid bodies enclosed in the other types of skeletal structures can be studied
following the procedure developed for the space frame. They differ only in
the transformation matrix Tj and in the inertia matrix Mc . These matrices
for all types of structures are summarized in Table 11.11.1. The transformation
matrices can be readily found from that valid for the space frame (Type 5 in
Table 11.11.1) by deleting appropriate rows and columns. For example, point
TABLE 11.11.1 Transformation matrices Tj and rigid body mass-inertia

matrix Mc of skeletal structures.
Type of structure Tj Mc
2 3 2 3
1. Plane truss 1 0 m 0 0
(structure in 4 0 1 5 40 m 05
xy-plane) D yj D xj 0 0 Iz
2 3 2 3
2. Plane frame 1 0 0 m 0 0
(bending in 4 0 1 05 40 m 05
xy-plane) D yj D xj 1 0 0 Iz
2 3 2 3
3. Grid (structure 1 0 D yj Ix Ixy 0
in xy-plane) 4 0 1 D xj 5 4 Iyx Iy 0 5
0 0 1 0 0 m
2 3 2 3
4. Space truss 1 0 0 m 0 0 0 0 0
6 0
6 1 0 7 7
60
6 m 0 0 0 0 7 7
6 0
6 0 1 7 7
60
6 0 m 0 0 0 7 7
6 0
6 D z j Dy 7
j7
60
6 0 0 Ix Ixy Ixz 7
7
4 D zj 0 D xj 5 40 0 0 Iyx Iy Iyz 5
D yj D xj 0 0 0 0 Izx Izy Iz
2 3 2 3
5. Space frame 1 0 0 0 0 0 m 0 0 0 0 0
6 0
6 1 0 0 0 077
60
6 m 0 0 0 0 7 7
6 0
6 0 1 0 0 077
60
6 0 m 0 0 0 7 7
6 0
6 D zj Dyj 1 0 077
60
6 0 0 Ix Ixy Ixz 7
7
4 D zj 0 D xj 0 1 05 40 0 0 Iyx Iy Iyz 5
D yj D xj 0 0 0 1 0 0 0 Izx Izy Iz
@CivilMethod
J for a plane body in a plane truss has three displacements, two translations in
the x and y directions and a rotation about the z axis. However, a joint j of the
truss has only two displacements, the translations in the x and y directions.
Therefore, we keep only the first, second, and sixth rows and the first and sec-
ond columns in the 6 6 matrix. Thus, a 3 2 transformation matrix results for
the plane truss. It should be noted that the 3 3 matrix pertaining to the grid
requires not only the deletion of rows and columns but also their rearrangement.
The inertia matrix Mc results from the inertia properties of the rigid body mov-
ing within the structure. It can also be found from that valid for the space frame
(Type 5 in Table) by deleting appropriate rows and columns.
Example 11.11.1 Formulate the equation of motion of the truss of Fig. E11.12.
Coordinates of nodes: 1ð0, 0Þ, 2ð4, 0Þ, 3ð0, 3Þ, 4ð4, 3Þ, 5ð8, 3Þ; cross-section of
bars: A1 ¼ A2 ¼ 1:5A, A3 ¼ A5 ¼ A, A4 ¼ A6 ¼ A7 ¼ 2:1A, A ¼ 27cm2 ; damp-
ing coefficient: cs ¼ 0; modulus of elasticity: E ¼ 2:1 108 kN=m2 ; accelera-
tion of gravity g ¼ 9:81m=s2 material density r ¼ 7:55kNm1 s2 =m3 ;
consistent mass assumption. The plane rigid body B has dimensions 3:0 3:0
0:02 m3 and specific weight 24kN=m3 .
Fig. E11.12 Plane truss in Example 11.5.2.
Solution
The structure has n ¼ 5 nodes. We take node 1 as the point of reference of the
motion of the rigid body. The free structure has N ¼ 9 degrees of freedom, namely
the translational displacements u1 , u2 and the rotation u3 of the rigid body about
point 1 and the six displacements of nodes 2, 4, 5 (see Fig. E11.12). Thus, it is
u ¼ f u1 u2 u3 u4 u5 u6 u7 u8 u9 gT

Computation of ke , me , pe ðt Þ for e ¼ 1, …,7
2 3 2 3
0:375 0 0:375 0 0:250 0 0:250 0
6 0 0 0 07 6 07
k1 ¼ k2 ¼ EA6 7, k3 ¼ EA6 0 0 0 7
4 0:375 0 0:375 0 5 4 0:250 0 0:250 0 5
0 0 0 0 0 0 0 0
@CivilMethod
2 3 2 3
0:420 0 0:420 0 0:20 0 0:20 0
6 0 0 0 07 6 07
6
k ¼ k ¼ EA4
4 7 7, k5 ¼ EA6 0 0 0 7
0:420 0 0:420 05 4 0:20 0 0:20 05
0 0 0 0 0 0 0 0
2 3
0:70 0 0:70 0
6 0 0 0 07
k6 ¼ EA6
4 0:70
7
0 0:70 05
0 0 0 0
2 3 2 3
2 0 1 0 1:333 0 0:666 0
6 0 2 0 17 6 0:666 7
m1 ¼ m2 ¼ rA6 7, m3 ¼ rA6 0 1:333 0 7
41 0 2 05 4 0:666 0 1:333 0 5
0 1 0 2 0 0:666 0 1:333
2 3 2 3
3:50 0 1:75 0 1:667 0 0:833 0
6 0 3:50 0 1:75 7 6 0:833 7
m4 ¼ m7 ¼ rA6 7, m5 ¼ rA6 0 1:667 0 7
4 1:75 0 3:50 0 5 4 0:833 0 1:667 0 5
0 1:75 0 3:50 0 0:833 0 1:667
2 3
2:10 0 1:05 0
6 0 2:10 0 1:05 7
m6 ¼ rA6
4 1:05
7
0 2:10 0 5
0 1:05 0 2:10
p1 ðt Þ ¼ p2 ðt Þ ¼ p3 ðt Þ ¼ p4 ðt Þ ¼ p5 ðt Þ ¼ p6 ðt Þ ¼ p7 ðt Þ ¼ 0
Computation of Re
2 3 2 3
1 0 0 0 0:8 0:6 0 0
60 1 0 07 6 0:6 0:8 0 0 7
R ¼R ¼R ¼6
1 2 3
40
7, R ¼ R ¼ 6
4 7 7
0 1 05 4 0 0 0:8 0:6 5
0 0 0 1 0 0 0:6 0:8
2 3 2 3
0:8 0:6 0 0 0 1 0 0
6 0:6 0:8 0 0 7 6 1 0 0 07
R ¼6
5
4 0
7, R ¼ 6
6 7
0 0:8 0:6 5 4 0 0 0 15
0 0 0:6 0:8 0 0 1 0
Computation of k , m
e
e , pe ðt Þ
2 3 2 3
0:375 0 0:375 0 0:250 0 0:250 0
6 0 0 0 07 6 0 0 0 07
k ¼ k ¼ EA6
1 2
4 0:375
7, k ¼ EA6
3 7
0 0:375 05 4 0:250 0 0:250 05
0 0 0 0 0 0 0 0
@CivilMethod
2 3 2 3
0:268 0:202 0:268 0:202 0 0 0 0
6 0:202 0:151 0:202 0:151 7 6 0 0:7 0 0:7 7
6
k ¼ k ¼ EA4
4 7 7 6
, k ¼ EA4
6 7
0:268 0:202 0:268 0:202 5 0 0 0 0 5
0:202 0:151 0:202 0:151 0 0:7 0 0:7
2 3
0:128 0:096 0:128 0:096
6 0:096 0:072 0:096 0:072 7
k5 ¼ EA64 0:128
7
0:096 0:128 0:096 5
0:096 0:072 0:096 0:072
2 3 2 3
2 0 1 0 1:333 0 0:666 0
60 2 0 17 60 1:333 0 0:666 7
m ¼m
1
¼ rA6
2 7 ¼ rA6
4 1 0 2 0 5, m
3
4 0:666 0
7
5
1:333 0
0 1 0 2 0 0:666 0 1:333
2 3 2 3
3:50 0 1:75 0 1:667 0 0:833 0
6 0 3:50 0 1:75 7 6 0 1:667 0 0:833 7
7 ¼ rA6
4 ¼ m
m 4 1:75 0
7, m 5 ¼ rA6 7
3:50 0 5 4 0:833 0 1:667 0 5
0 1:75 0 3:50 0 0:833 0 1:667
2 3
2:10 0 1:05 0
60 2:10 0 1:05 7
¼ rA6
m6
4 1:05
7
0 2:10 0 5
0 1:05 0 2:10
Computation of Te
Because point 1 has been chosen as the point of reference, only the elements 2
and 5 are affected by the motion of the rigid body. Referring to Table 11.11.1
and Eqs. (11.11.24), (11.11.25), we obtain 2 3
2 3 1 0 0 0
1 0 6 0 1 0 07
6 7
Element 2. D x 3 ¼ 0, D y 3 ¼ 3, T2j ¼ 4 0 1 5 and T2 ¼ 66 3 0 0 0 7
7
3 0 4 0 0 1 05
0 0 0 1
2 3
0:375 0 1:125 0:375 0
6 0 0 0 0 07
02 6 7
k ¼ EA6 6 1:125 0 3:375 1:125 077
4 0:375 0 1:125 0:375 0 5
0 0 0 0 0
@CivilMethod
2 3
2 3 1 0 0 0
1 0 60 1 0 07
6 7
Element 5. D x 3 ¼ 0, T5k ¼ 4 0 1 5 and T5 ¼ 6
y 3 ¼ 3, D 60 0 1 077
3 0 40 0 0 15
0 0 3 0
2 3
0:128 0:096 0:128 0:096 0:384
6 0:096 0:072 0:096 0:072 0:288 7
05 6 7

k ¼ EA6 7
6 0:128 0:096 0:128 0:096 0:384 7
4 0:096 0:072 0:096 0:072 0:288 5
0:384 0:288 0:384 0:288 1:152
0e
For the remaining elements e ¼ 1, 3, 4, 6, 7 it is k ¼ k , m
e
0e ¼ m
e , p0 e ¼ pe .
Computation of the mass matrix M1
It is convenient to express the area AB and the density rB of the plane
AB ¼ 3333:33A, rB ¼ 0:324r. This yields
body in terms of A, r. Thuswe obtain
m ¼ 21:6rA, Ic ¼ 21:6rA 32 + 32 =12 ¼ 32:4rA. Then from Table 11.11.1,
we have
2 3 2 3 2 3
21:6 0 0 1 0 0 1 0 0
Mc ¼ rA4 0 21:6 0 5 Tc ¼ 4 0 1 05 ¼ 4 0 1 05
0 0 32:4 Dy c Dxc 1 1:5 1:5 1
and by virtue of Eq. (11.11.38), we obtain
2 3
21:6 0 32:4
M1 ¼ rA4 0 21:6 32:4 5
32:4 32:4 128:9
Finally, the load P ðt Þ is transferred from point A to 1
2 3 8 9
1 0
< 0=
0
P1 ðt Þ ¼ 4 0 1 5 ¼ 1 P ðt Þ
P ðt Þ : ;
3 3 3
@CivilMethod
Assembly matrices ae ðe ¼ 1, 2, …, 7Þ. Referring to Fig. E11.12, we have

2 3
2 3 1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 60 1 0 0 0
60 1 0 0 0 0 0 0 07 2 6 0 0 0 07 7
a1 ¼ 6 7 6
4 0 0 0 1 0 0 0 0 0 5, a ¼ 6 0 0 1 0 0 0 0 0 07 7
40 0 0 0 0 1 0 0 05
0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 1 0 0
2 3 2 3
0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 1 0 0 7 6 0 0 0 07
a3 ¼ 6 7 4 60 1 0 0 0 7
4 0 0 0 0 0 0 0 1 0 5, a ¼ 4 0 0 0 0 0 1 0 0 05
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0
2 3
0 0 0 1 0 0 0 0 0 2 3
60 0 0 0 1 0 0 0 07 0 0 0 1 0 0 0 0 0
6 7 6 60 0 0 0 1 0 0 0 07
a ¼6
5 7 6
6 1 0 0 0 0 0 0 0 0 7, a ¼ 4 0 0 0 0 0
7
40 1 0 0 0 0 0 0 05 1 0 0 05
0 0 0 0 0 0 1 0 0
0 0 1 0 0 0 0 0 0
2 3
0 0 0 1 0 0 0 0 0
60 0 0 0 1 0 0 0 07
a7 ¼ 6
40 0 0 0 0 0 0 1 05
7
0 0 0 0 0 0 0 0 1
The assembly matrix for the plane rigid body is
2 3
1 0 0 0 0 0 0 0 0
ap ¼ 4 0 1 0 0 0 0 0 0 05
0 0 1 0 0 0 0 0 0
Note that the dimensions of a2 and a5 are 5 9 while the dimensions of ap
are 39.
The enlarged element matrices and load vectors are computed using
Eqs. (11.2.92), (11.2.95), (11.2.96), namely
^ 0 ¼ ðae ÞT m ^ 0 ¼ ðae ÞT k0 e ae , p
e e
M 0 ae , K
e e e ^ 1 ¼ ðap ÞT M1 ap
^0 ðt Þ ¼ ðae ÞT p0 ðt Þ, M
The total stiffness matrix K and load vector pðt Þ are computed using
Eqs. (11.2.100), (11.2.107), respectively, namely
X
Ne X
Ne
^ 0 , pðt Þ ¼ P
e
¼
K K ðt Þ + ^ 0 ðt Þ
p
e
e¼1 e¼1
ðt Þ is the vector of the external loads, that is,

where P
ðt Þ ¼ P ðt Þf 0 1 3 0 0 0 1 0 5 gT
P
@CivilMethod
Regarding the total mass matrix, it is computed using Eq. (11.11.40)

X
7
^0
e
∗ ¼M
M ^1+ M
e¼1
u 2 ¼ u4 ¼ u5 ¼ 0), the following

Taking into account the support conditions (
transformation matrix V is constructed
2 3
1 0 0 0 0 0 0 0 0
60 0 0 0 0 0 1 0 07
6 7
60 1 0 0 0 0 0 0 07
6 7
60 0 0 0 0 0 0 1 07
6 7
V¼6 60 0 0 0 0 0 0 0 17
7
60 0 1 0 0 0 0 0 07
6 7
60 0 0 1 0 0 0 0 07
6 7
40 0 0 0 1 0 0 0 05
0 0 0 0 0 1 0 0 0
and the load vector pðt Þ yielding the
∗, K
which modifies the matrices M
equation of motion
M u€f + K
uf ¼ e
where
2 3
30:77 43:40 2:75 0 0 0
6 43:40 162:60 3:00 0 0 0 7
6 7
6 3:00 8:93 0 0 7
e ff ¼ rA6 2:75
M
0:67 7:
6 0 0 0 8:93 0 0:67 7
6 7
4 0 0 0:67 0 4:83 0 5
0 0 0 0:67 0 4:83
2 3
1:15 1:51 0:64 0:20 0 0
6 1:51 4:53 0:12 0 0 0 7
6 7
6 0:64 0:12 0:89 0:20 0:25 0 7
e ff ¼ EA6
K 7,
6 0:20 0 0:20 0:85 0 0 7
6 7
4 0 0 0:25 0 0:52 0:20 5
0 0 0 0 0:20 0:15
pf∗ ðt Þ ¼ P ðt Þf 0 3 0 1 0 5 gT , e
e uf ðt Þ ¼ f u3 u4 u6 u7 u8 u9 gT
Remark. The solution of the equation of motion gives the components of the
displacement euf ðt Þ due to the dynamic loads. However, the weight Wc ¼ 3 3
0:02 24 ¼ 4:32kN of the rigid body produces a static deformation, which
should be added to the dynamic one in order to obtain the total deformation. This
is achieved as follows.
@CivilMethod
First, the weight Wc is transferred from point C to point 1

2 3 8 9
1 0
< 0 =
0
W1 ¼ 4 0 1 5 ¼ 4:32
Wc : ;
1:5 1:5 6:48
T
This yields the load vector Pst ¼ 0 4:32 6:48 0 0 0 0 0 0 , which is
modified due to the supports to e st ¼ VPst ¼ 0 6:48 0 0 0 0 4:32 0 0 T .
P

Then we obtain P e stf ¼ 0 6:48 0 0 0 0 T and finally
1 1
ust ¼ K e ff Pstf ¼ f 2:34 3:07 3:45 1:37 3:45 4:60 gT
EA
11.12 Problems
Problem P11.1 A plane truss element of length L has variable cross-section
A ¼ A0 1 0:05x2 , x ¼ x=L. The element is loaded by the distributed axial load
t and the suddenly applied concentrated load P ¼ P0 at
pðx, t Þ ¼ p0 ð1 + xÞ sin w
the cross-section x ¼ 0:25L. Compute: (a) the matrices ke , me and the vector pe .
For the mass matrix consider (i) consistent mass assumption and (ii) lumped
mass assumption. Compare ke , me with those resulting when the exact shape
functions are employed. (b) Transform ke , me , pe to the
global
pffiffiffi system of axes,
if the coordinates of element end points j and k are 1 + 3 L=2, 3L=2 and
ðL=2, LÞ, respectively.
Problem P11.2 The height of the cross-section Aðx Þ ¼ bh ðx Þ of a plane frame
element of length L varies linearly so that h ðLÞ ¼ 0:5h ð0Þ. The element is
t, the distributed bend-
py ðx, t Þ ¼ p0 ð1 x Þsin w
loaded by the distributed load
ing moment mz ðx, t Þ ¼ m0 1 x + x 2 cos w t as well as by the concentrated
load Py ðL=3Þ and the concentrated moment Mz ð3L=4Þ, which are applied sud-
denly. Compute the matrices ke , me and the vector pe ðt Þ and compare them with
those resulting when the exact shape functions are employed. Consider the
lumped mass assumption for the mass matrix.
Problem P11.3 Considering only flexural vibrations, formulate the equation of
motion of the frame of Fig. P11.3. Use lumped mass assumption for the
element mass.
Fig. P11.3 Frame in problem P11.3.
@CivilMethod
Problem P11.4 Considering only flexural vibrations, formulate the equation of

motion of the frame of Fig. P11.4. Assume CT ¼ EI =L and lumped mass
assumption for the element mass.
Problem P11.5 Find the number of the active axial constraints of the frame of
Fig. P11.5 and determine its degrees of freedom after their reduction due to
static condensation and axial constraints. The coordinates of the nodes are:
Að0, 0Þ, B ð7, 0Þ, 1ð0, 3Þ, 2ð0, 8Þ, 3ð4, 8Þ, 4ð6, 8Þ.
Fig. P11.5 Frame in problem P11.5
Problem P11.6 Neglecting the axial deformation, formulate the equation of

motion of the frames of Fig. P11.6.
@CivilMethod
Fig. P11.6 Frames in problem P11.6.
Problem P11.7 Neglecting the axial deformation of the beam elements, formu-
late the equation of motion of the frame of Fig. P11.7. The elements have a
length L ¼ 4a, cross-sectional area A, moment of inertia I , modulus of elasticity
E, and line density m. The rigid bodies B1 and B2 have dimensions a a
and 1:5a 0:5a, respectively, and unit thickness. Their material density is r.
Consider lumped mass assumption for the elements.
Problem P11.8 Formulate the equation of motion of the grid of Fig. P11.8.
Assume the data: coordinates of nodes 1ð0, 0Þ, 2ð2, 0Þ, 3ð3, 2Þ, 4ð3, 2Þ;
rectangular cross-section of all elements h b ¼ 0:50 0:30m2 ; material con-
stants E ¼ 2:1 107 , G ¼ 0:4E, r ¼ 2:4kNm1 s2 /m3; loading pðx, t Þ ¼ p0 f ðt Þ,
M ðt Þ ¼ p0 L2 f ðt Þ. Consider lumped mass assumption for the elements.
@CivilMethod
Fig. P11.8 Grid in problem P11.8.
Problem P11.9 Formulate the equation of motion of the space frame of

Fig. P11.9. The principal plane xy of the elements is vertical. Assume data:
pðx, t Þ ¼ p0 sin w t, G ¼ 0:4E, 1 ð2a, 0, 0Þ, 2 ð2a, 1:5a, a Þ, 3 ða, 1:5a, a Þ,
4 ð0, 1:5a, a Þ, 5 ða, 5a, 0Þ. The elements have a square cross-section with side
length a=15.
Fig. P11.9 Space frame in problem P11.9.
Problem P11.10 A rigid circular plate of radius R and thickness 0:1a is supported
on the ground by truss bars as shown in Fig. P11.10. The plate is horizontal
and stands off at a distance h ¼ 0:5a from the ground. The support points on the
ground lie on the vertices of an equilateral triangle. A horizontal force
P ¼ 100kN produced by a rotating mass is applied at the center of the plate.
Formulate the equation of motion of the structure. Assume: a ¼ 20m, R ¼ 4m,
cross-sectional area of the bars A; density of the plate r; modulus of elasticity
E; load of the plate q ¼ 10kN=m2 ; lumped mass assumption for the elements.
Fig. P11.10 Structure in problem P11.10.
@CivilMethod
Problem P11.11 Formulate the equation of motion of the space frame of

Fig. P11.11. The axial deformation of the elements is neglected and the lumped
mass assumption is adopted. All elements have a rectangular cross-section with
dimensions h and b in the directions of the principal axes y and z, respectively.
The plane xy is vertical. The shape of the rigid body is a rectangular parallel-
epiped with dimensions a 0:5a 0:4a. The force P ðt Þ is applied at the middle
300 of the edge. The material constants are: modulus of elasticity E, Poisson’s
ratio n ¼ 0:2, and density r.
Fig. P11.11 Space frame in problem P11.11.

[1] J.T. Katsikadelis, A generalized Ritz method for partial differential equations in domains of
arbitrary geometry using global shape functions. Eng. Anal. Bound. Elem. 32 (5) (2008)
353–367, https://doi.org/10.1016/j.enganabound.2007.001.
[2] O. Zienkiewicz, R. Taylor, The Finite Element Method, seventh ed., Butterworth-Heinemann,
Oxford, UK, 2013.
[3] T.J.R. Hughes, The Finite Element Method: Linear Static and Dynamic Finite Element Anal-
ysis, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1987
[4] J.T. Katsikadelis, The Boundary Element Method for Engineers and Scientists, second ed.,
Academic Press, Elsevier, Oxford, UK, 2016.
[5] D.O. Brush, B.O. Almroth, Buckling of Bars, Plates and Shells, McGraw-Hill, New York,
1975.
[6] R.G. Guyan, Reduction of stiffness and mass matrices, AIAA J. 3 (1965) 380.
@CivilMethod
[7] W. Weaver Jr., P.R. Johnston, Structural Dynamics by Finite Elements, Prentice Hall,
Englewood Cliffs, NJ, 1987.
[8] W. Weaver Jr., M. Eisenberger, Dynamics of frames with axial constraints, ASCE J. Struct.
Eng. 109 (3) (1983) 773–784.
[9] S. Lipschutz, M. Lipson, Linear Algebra (Fourth Edition), Schaum’s Outline Series, McGraw-
Hill Companies, Inc., New York, 2009
[10] V.Z. Vlasov, Thin-walled Elastic Beams, second ed., Israel Program for Scientific Transla-
tions, Jerusalem, Israel, 1961.
[11] E.J. Sapountzakis, V.G. Mokos, Warping shear stresses in nonuniform torsion, Comput.
Methods Appl. Mech. Eng. 192 (2003) 4337–4353.
[12] S. Timoshenko, G.N. Goodier, Theory of Elasticity, second ed., McGraw-Hill, NY, 1951.
[13] R. Debasish, G.V. Rao, Elements of Structural Dynamics: A New Perspective (1st Edition),
John Wiley & Sons, 2012.
@CivilMethod
Chapter 12
systems: Free vibrations
Chapter outline
12.1 Introduction 523 12.10 Solution of the vibration
12.2 Free vibrations without problem with damping 570
damping 524 12.10.1 Direct solution of
12.3 Orthogonality of eigenmodes 532 the differential
12.4 Eigenmodes of systems with equation 571
multiple eigenfrequencies 534 12.10.2 Linearization of
12.5 The linear eigenvalue the quadratic
problem 541 eigenvalue problem 575
12.5.1 The standard 12.10.3 The use of a
eigenvalue problem proportional viscous
of linear algebra 541 damping matrix 578
12.5.2 Properties of the 12.11 Construction of a
eigenvalues and proportional damping
eigenvectors 544 matrix 584
12.5.3 The generalized 12.11.1 Rayleigh damping 584
eigenvalue problem 552 12.11.2 Additional
12.6 The Rayleigh quotient 558 orthogonality
12.7 Properties of conditions: Caughey
eigenfrequencies and damping matrix 587
modes of MDOF systems 12.11.3 Construction of
without damping: theo proportional
A summary 563 damping matrix
12.8 Solution of the vibration using the modal
problem without damping 563 matrix 590
12.9 The method of mode 12.12 Problems 597
superposition 565 References and further reading 600
12.1 Introduction
In this chapter, we study the dynamic response of multi-degree-of-freedom
(MDOF) systems in the absence of external forces. Like the SDOF systems,
the MDOF systems perform oscillations caused by initial displacements and/
or initial velocities. In accordance with the SDOF systems, we call these

@CivilMethod
oscillations free vibrations of MDOF systems. In this case, it is pðt Þ ¼ 0 and the
equation of motion (see Eq. 10.2.6) becomes
u + Cu_ + Ku ¼0
M€ (12.1.1)
where M is the mass matrix, C the damping matrix, K the stiffness matrix, and
u ¼ uðt Þ the displacement vector. When damping is neglected, C ¼ 0, we talk
about free vibrations without damping or free undamped vibrations. Their study
plays an important role in the analysis of the MDOF systems because, in addi-
tion to facilitating the understanding of their dynamic response by introducing
the physical concepts of the eigenfrequencies, periods, and mode shapes, it also
serves to develop methods for solving free vibration problems with damping as
well as forced vibration problems.
As mentioned, the equation describing the free undamped vibrations results
from Eq. (12.1.1) by setting C ¼ 0. Thus, we have
M€
u + Ku ¼0 (12.1.2)
The solutions of Eqs. (12.1.1) and (12.1.2) under the specified initial con-
ditions uð0Þ and u_ ð0Þ are the subject that will be discussed in this chapter.
All presented material is illustrated by appropriately chosen examples. The per-
tinent bibliography with recommended references for further study is also
included. The chapter is enriched with problems to be solved.
12.2 Free vibrations without damping

We look for a solution of Eq. (12.1.2) in the form
u1 ¼ b1 T ðt Þ, u2 ¼ b 2 T ðt Þ, …,uN ¼ bN T ðt Þ (12.2.1)
where b1 , b2 ,…,bN are constants and T ðt Þ a function of time, which is com-
mon for all displacements. If we define the vector b ¼ fb 1 , b2 , …, b N gT , then
we can write Eq. (12.2.1) in the form
u ¼bT ðt Þ (12.2.2)
MbT€ ðt Þ + KbT ðt Þ ¼0 (12.2.3)
Τ
Premultiplying the foregoing equation by b gives
bΤ MbT€ ðt Þ + bΤ KbT ðt Þ ¼0 (12.2.4)
Taking into account that both quantities bΤ Mb, bΤ Kb are constants and
excluding the value T ðt Þ ¼ 0 (see Section 8.3.3), we may write Eq. (12.2.4) as
T€ ðt Þ bΤ Kb
¼ (12.2.5)
T ðt Þ bΤ Mb
Because the right side of Eq. (12.2.5) is a constant, say l, it implies that also
the left side is a constant, namely
@CivilMethod
Multi-degree-of-freedom systems: Free vibrations Chapter 12 525
T€ ðt Þ
¼l (12.2.6)
T ðt Þ
or
T€ ðt Þ + lT ðt Þ ¼ 0 (12.2.7)
The constant l is unknown in the first instance and will be determined
subsequently.
Eq. (12.2.7) allows us to determine the time-dependent function T ðt Þ. To
this end, we look for solutions representing oscillatory motion. We distinguish
three cases:
(i) l ¼ 0. In this case, Eq. (12.2.7) becomes
T€ ðt Þ ¼ 0 (12.2.8)
which is integrated to yield
T ðt Þ ¼ C 1 t + C 2 (12.2.9)
where C1 , C2 are arbitrary integration constants. Obviously, the function
T ðt Þ obtained from Eq. (12.2.9) implies that Eq. (12.2.2) does not repre-
sent an oscillatory motion. Therefore, the value l ¼ 0 must be excluded.
(ii) l ¼ w2 < 0. In this case, Eq. (12.2.7) becomes
T€ ðt Þ w2 T ðt Þ ¼ 0 (12.2.10)
whose solution is (see Chapter 2)
T ðt Þ ¼ C1 ewt + C2 ewt (12.2.11)
From the foregoing expression of T ðt Þ, it follows that Eq. (12.2.2)
does not represent an oscillatory motion. Consequently, the negative
values of the constant l must also be excluded.
(iii) l ¼ w2 > 0. In this case, Eq. (12.2.7) becomes
T€ ðt Þ + w2 T ðt Þ ¼ 0 (12.2.12)
whose solution is (see Chapter 2)
T ðt Þ ¼ c cos wt + d sin wt
(12.2.13)
¼ a cos ðwt qÞ
where c,d or a,q are arbitrary integration constants.
From the foregoing expression of T ðt Þ, it follows that the displacements
obtained by Eq. (12.2.2) are bounded and the function T ðt Þ expresses a har-
monic oscillation with the natural frequency w.
Eq. (12.2.12) gives T€ ðt Þ ¼ w2 T ðt Þ, which is substituted into Eq. (12.2.3)
to yield

K w2 M bT ðt Þ ¼ 0 (12.2.14)
@CivilMethod
Because the foregoing equation must hold for all values of t > 0, it must be

K w2 M b ¼0 (12.2.15)
Eq. (12.2.15) represents a system of N linear algebraic equations, which can
be solved to determine the vector b ¼ fb1 , b 2 , …, bN gT , that is, the amplitude
of the displacements. The system of Eq. (12.2.15) is homogeneous; therefore it
has a nontrivial solution only if the determinant of the coefficient matrix van-
ishes, namely

det K w2 M ¼0 (12.2.16)
or

k11 w2 m11 k12 w2 m12 ⋯ k1N w2 m1N

k21 w2 m21 k22 w2 m22 ⋯ k2N w2 m2N

⋯ ⋯ ⋯ ⋯ ¼0 (12.2.17)

k N 1 w2 m N 1 k N 2 w2 m N 2 ⋯ kNN w2 mNN
The expansion of the determinant in Eq. (12.2.17) produces a polynomial of
the N degree with respect to w2 . Therefore Eq. (12.2.16) may hold true for its N
values w2 . These values are the eigenfrequencies or natural frequencies of the
system. Eq. (12.2.16), from which the eigenfrequencies are determined, is
referred to as the frequency equation of the system. As we will show below,
the roots of Eq. (12.2.16) are real and positive for the problems of dynamics
we examine. The eigenfrequencies are arranged in the order of magnitude,
w1 < w2 … < wN , the smallest of which is called the fundamental eigenfre-
quency. The eigenfrequencies may be all distinct or some of them may be mul-
tiple. First, we assume that the system has discrete eigenfrequencies. For each
value wi , we obtain a system of linear algebraic equations of the form (12.2.15),
which allows the determination of the corresponding vector bðiÞ . Due to the fact
that the determinant is equal to zero, the number of independent equations is
ði Þ ði Þ ði Þ
N 1, which implies that one of the components b 1 , b2 ,…,b N of the vector
ði Þ ði Þ
b can be determined arbitrarily. In this respect, we may take b 1 ¼ 1, and
write the system of Eq. (12.2.15) as
ð12:2:18Þ
ði Þ ði Þ
Partitioning the matrix A ¼ K w2i M
and the vector b as indicated in
Eq. (12.2.18) and defining the matrices
ði Þ
A11 ¼ k11 w2i m11 11 (12.2.19a)
@CivilMethod
ði Þ
A12 ¼ k12 w2i m12 ⋯ k1N w2i m1N 1ðN 1Þ (12.2.19b)
2 3
k21 w2 m21
A21 ¼ 4
ði Þ 5
⋮ (12.2.19c)
kn1 wi mN 1 ðN 1Þ1
2
2 3
k22 w2i m22 ⋯ k2N w2i m2N
A22 ¼ 4
ði Þ 5
⋮ ⋮ ⋮ (12.2.19d)
kN 2 wi mN 2 ⋯ kNN wi mNN ðN 1ÞðN 1Þ
2 2
8 ðiÞ 9
< b2 =
ði Þ
b2 ¼ ⋮ (12.2.19e)
: ðiÞ ;
bN ðN 1Þ1
we can write Eq. (12.2.18) as

" #
ði Þ ði Þ
A11 A12 1 0
ði Þ ¼ (12.2.20)
ði Þ
A21 A22
ði Þ b 2 0
which after performing the multiplication gives
A11 + A12 b2 ¼ 0 (12.2.21a)
A21 + A22 b2 ¼0 (12.2.21b)
The second of the foregoing equations, that is, Eq. (12.2.21b), can be used to
ði Þ
establish the vector b2 . Thus we have
h i1
b2 ¼ A22 A21 (12.2.22)
ði Þ
Eq. (12.2.21a) must hold true if the value of b2 obtained from Eq.
(12.2.21b) is inserted in it. Therefore, this equation can be employed to verify
ði Þ
the computed value of b2 .
Eq. (12.2.18) states that for each value wi , we obtain a vector bðiÞ , which
henceforth will be denoted by bi for convenience. These vectors are called
eigenvectors and are displayed in a square matrix B of dimensions N N , each
column of which is an eigenvector
2 3
1 1 ⋯ 1
6 b21 b 22 ⋯ b2N 7
B ¼ ½ b1 b2 ⋯ bN ¼ 6 4⋯ ⋯ ⋯ ⋯ 5
7 (12.2.23)
bN 1 b N 2 ⋯ bNN
For each value of wi , Eq. (12.2.2) gives a solution of the form
ui ¼ bi ðci cos wi t + d sin wi t Þ
(12.2.24)
¼ bi ai cos ðwt qi Þ
which is called an eigensolution.
@CivilMethod
The superposition of the solutions given by Eq. (12.2.24) for i ¼ 1, 2…, N ,

gives the general solution for the free vibration problem of the MDOF system
with N degrees of freedom. Thus we have
X
N
u¼ bi ðci cos wi t + di sin wi t Þ (12.2.25)
i¼1
or
X
N qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u¼ bi ai cos ðwi t qi Þ, ai ¼ ci2 + di2 , qi ¼ tan 1 ðdi =ci Þ (12.2.26)
i¼1
The 2N arbitrary constants ci ,di or ai ,qi are established from the initial con-
ditions using the method we will describe after discussing the properties of the
eigenvectors.
From Eq. (12.2.24), we conclude that the deformation pattern of the system
due to an eigensolution is given by a certain shape, which is expressed by the
components of the eigenvector bi multiplied each instant by the value of the
corresponding common time function T ðt Þ ¼ ai cos ðwi t qi Þ. The deformation
shape defined by vector bi is called the i-eigenmode of the free vibration.
As previously mentioned, the components of the eigenvector are determined
relative to one of its components, which can be arbitrarily chosen. Therefore,
the absolute magnitude of the eigenvector is not determined, that is, the eigen-
vectors define only directions in the N -dimensional space while their measure
remains undetermined. In view of this fact, it is appropriate to adopt a standard
scale-setting procedure for measuring it. This process is called normalization.
A common method of normalization is the use of the nondimensional
vectors
2 3 2 3
f1i b1i
6 f2i 7 1 6 7
fi ¼ 6 7 6 b2i 7
4 ⋯ 5 ¼ max b 4 ⋯ 5 (12.2.27)
ki
fNi bNi
The vectors fi (i ¼ 1,2…,N ) are called the normal modes or normal
eigenmodes of the vibration of the system. The shape of deformation defined
by the eigenmode fi is also referred to as the i-mode shape or simply the i-mode
of the free vibration.
The N normal modes can then be displayed in a single square matrix, each
column of which is a normal mode, namely
2 3
f11 f12 ⋯ f1N
6 f21 f22 ⋯ f2N 7
F ¼ ½ f1 f2 ⋯ fN ¼ 6 4⋯ ⋯ ⋯ ⋯ 5
7 (12.2.28)
fN 1 fN 2 ⋯ fNN
The matrix F is called the modal matrix.
@CivilMethod
Another method of normalization of the eigenvectors, which is usual in the-

oretical discussions and computer programs, is to normalize the modes with
respect to the mass. In this method, the scale-setting for measuring the compo-
nents of the eigenvector is chosen so that the following relation is satisfied
fTi Μfi ¼ 1 (12.2.29)
where fi represents the i normalized vector and Μ the mass matrix of the
system. This is achieved as follows:
Let mi be the number that must multiply the elements of bi in order to
normalize it. Thus, we will have
fi ¼ mi bi
Substituting it into Eq. (12.2.29) gives
1 ¼ fTi Μfi ¼ m2i bTi Μbi
1
mi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (12.2.30)
bTi Μbi
Hence
bi
fi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi (12.2.31)
bTi Μbi
qffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffi
If it is M ¼ I, then the scalar quantity bTi Ιbi ¼ bTi bi is the magnitude of
bi , that is, the normalized eigenvector fi is a unit vector. The described process
is also called orthonormalization with respect to mass.
The establishment of the natural frequencies wi and mode shapes fi can be
achieved when the flexibility matrix F ¼ K1 is known without inverting it to
obtain the stiffness matrix.
For this purpose, Eq. (12.2.15) is premultiplied by the matrix F to yield

FMw2 FK b ¼0 (12.2.32)
then taking into account that FK ¼ I, Eq. (12.2.32) becomes

1
FM 2 I b ¼0 (12.2.33)
w
which allows the establishment of the inverse of the eigenfrequencies, 1=w2i ,
and the eigenvectors bi . It should be noted that the matrix FM is not in general
symmetric.
Example 12.2.1 Compute the eigenfrequencies and mode shapes of the two-
story shear frame of Fig. E12.1a. Assume data: h1 ¼ 3:5m, h2 ¼ 4:0m,
q1 ¼ 40kN=m, q2 ¼ 50kN=m, and L ¼ 6:0m. The cross-sectional area of the
@CivilMethod
columns of the second floor is 25 25cm2 while that of the first floor is
30 30cm2 . The material of the structure is reinforced concrete with a specific
weight g b ¼ 24kN=m3 and modulus of elasticity E ¼ 2:1 107 kN=m2 . The
dead weight of the beams is included in the loads q1 ,q2 . Consider lumped mass
assumption of the columns.
Solution
(a) (b)
FIG. E12.1 Two-story shear frame.
The system has two degrees of freedom. Its dynamic model is shown in
Fig. E12.1b.
(i) Computation of the mass matrix M.

3:5
6:0 40 + 2 0:25 0:25 24
2
m11 ¼ ¼ 25kNm1 s2
9:81

3:5 4:0
6:0 50 + 2 0:25 0:25 + 0:30 0:30 24
2 2
m22 ¼
9:81
¼ 32 kNm1 s2
Hence

25 0
M¼ (1)
0 32
(ii) Computation of the stiffness matrix K.

Employing the method used in Example 11.5.2, we obtain
ð1Þ ð2Þ 12EI 1
k11 ¼ k11 + k11 ¼ 2
h13
12 2:1 107 0:254
¼2 ¼ 3826:5kN=m
3:53 12
@CivilMethod
ð1Þ ð2Þ 12EI 1

k21 ¼ k31 + k31 ¼ 2
h13
12 2:1 107 0:254
¼ 2 ¼ 3826:5kN=m
3:53 12
ð1Þ ð2Þ ð3Þ ð4Þ 12EI 1 12EI 3
k22 ¼ k33 + k33 + k11 + k11 ¼ 2 3 + 2 3
h1 h2
12 2:1 107 0:254 12 2:1 107 0:304
¼2 + 2 ¼ 9142:1kN=m
3:53 12 43 12
Hence

3826:5 3826:5
K¼ (2)
3826:5 9142:1
(iii) Computation of the eigenfrequencies.

The frequency equation results from Eq. (12.2.16) as

3826:5 25w2 3826:5
det K w2 M ¼
2¼0 (3)
3826:5 9142:1 32w
Expansion of the determinant yields the polynomial

w4 438:7506w2 + 25425:1792 ¼ 0 (4)
whose roots are
w21 ¼ 68:7089, w22 ¼ 370:0416
and the eigenfrequencies
w1 ¼ 8:289, w2 ¼ 19:236 (5)
(iv) Computation of the eigenmodes.

For N ¼ 2, Eqs. (12.2.21) become

3826:5 25w2i 3826:5b2i ¼ 0

3826:5 + 9142:1 32w2i b2i ¼ 0
The first of the foregoing equation
gives:
For i ¼ 1, b21 ¼ 3826:50=9142:1 32 8:2892 ¼ 0:5511
For i ¼ 2, b22 ¼ 3826:50= 9142:1 32 19:2362 ¼ 1:41751
Therefore, the matrix of the eigenvectors is

1 1
B¼ (6)
0:5511 1:41751
@CivilMethod
FIG. E12.2 Mode shapes of the shear frame in Example 12.2.1.
The normalization based on Eq. (12.2.27) gives the modal matrix

1 0:7054
F¼ (7)
0:5511 1
while the normalization based on Eq. (12.2.31) gives
b1
f1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
bT1 Μb1

1 1 0:1697
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi 0:5511 ¼ 0:0935
25 0 1
f 1 0:5511 g
0 32 0:5511
b2
f2 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffi
bT2 Μb

1 1 0:1061
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
1:41276 ¼
25 0 1 0:1499
f 1 1:41276 g
0 32 1:41276
Hence the modal matrix is

0:1697 0:1061
F¼ (8)
0:0935 0:1499
Fig. E12.2 presents the normal modes of the dynamic model of the frame
resulting from the first normalization, Eq. (7).
12.3 Orthogonality of eigenmodes

As we have already pointed out and will prove in Section 12.5, the eigenfre-
quencies w2i resulting from the frequency equation (12.2.16) are real and pos-
itive for the problems we encounter in the dynamics of structures. In the
@CivilMethod
following, we will derive the orthogonality conditions satisfied by the eigen-

modes. For the simplicity of the proof, we will assume that the eigenfrequencies
are distinct.
We write Eq. (12.2.15) for the two eigenfrequencies wn and wi , where
wn 6¼ wi , respectively
Kbn ¼w2n Mbn (12.3.1)
Kbi ¼w2i Mbi (12.3.2)

or after normalization
Kfn ¼w2n Mfn (12.3.3)
Kfi ¼w2i Mfi (12.3.4)

Premultiplying now both sides of Eq. (12.3.3) by fTi gives
fTi Kfn ¼w2n fTi Mfn (12.3.5)

Transposing both sides of the foregoing equation, we obtain
fTn KT fi ¼w2n fTn MT fi (12.3.6)

Further, using the symmetry of the matrices K and M, that is, KT ¼ K and
M ¼ M, we obtain
T
fTn Kfi ¼w2n fTn Mfi (12.3.7)

Next, premultiplying both sides of Eq. (12.3.4) by fTn gives
fTn Kfi ¼w2i fTn Mfi (12.3.8)

Finally, by subtracting Eq. (12.3.8) from (12.3.6), we obtain
2
wn w2i fTn Mfi ¼ 0 (12.3.9)
or
fTn Mfi ¼ 0, wi 6¼ wn (12.3.10)
Eq. (12.3.10) states that the eigenmodes are orthogonal with respect to the
mass matrix. This relation expresses the first orthogonality condition of the
eigenmodes. It is evident from Eq. (12.3.8) that
fTn Kfi ¼ 0, wi 6¼ wn (12.3.11)

Eq. (12.3.11) states that the eigenmodes are also orthogonal with respect to
the stiffness matrix. This relation expresses the second orthogonality condition
of the eigenmodes.
@CivilMethod
Example 12.3.1 Verify the orthogonality conditions for the eigenmodes in

Example 12.2.1:

0:1697 0:1061
f1 ¼ , f2 ¼
0:0935 0:1499
Solution
The computed mass and stiffness matrices are

25 0 3826:5 3826:5
M¼ , K¼
0 32 3826:5 9142:1
(a) The first orthogonality condition gives

T 25 0 0:1061
f1 Mf2 ¼ f 0:1697 0:0935 g
T
¼0 (1)
0 32 0:1499
(b) The second orthogonality condition gives

3826:5 3826:5 0:1061
fT1 Kf2 ¼ f 0:1697 0:0935 gT ¼0 (2)
3826:5 9142:1 0:1499
Because the eigenmodes have been orthonormalized with respect to the
mass, we may write
T
0:1697 0:1061 25 0 0:1697 0:1061 1 0
¼
0:0935 0:1499 0 32 0:0935 0:1499 0 1
or
FT MF ¼ I (3)
12.4 Eigenmodes of systems with multiple eigenfrequencies

In complex systems with many degrees of freedom, it is possible that a number
of eigenfrequencies are equal or differ very little from each other. In these cases,
we say that the system has multiple eigenfrequencies. Systems including a rigid
body motion are treated as systems with multiple eigenfrequencies (see
Example 12.4.2).
Let wn be the eigenfrequency, which has a multiplicity p, namely
wn ¼ wn + 1 ¼ wn + 2 ¼ ⋯ ¼ wn + p1 (12.4.1)
As we will show in the next section, in a system with p equal eigenfrequen-
cies, there are p linearly independent eigenmodes corresponding to the multiple
eigenfrequency, provided that its algebraic multiplicity is equal to its geometric
@CivilMethod
multiplicity, that is, if the eigenvalue problem has no degenerate eigenvalues.a

These eigenmodes can be determined as follows.
The rank of the matrix Aðwn Þ defined by Eq. (12.2.18) for w ¼ wn is N p.
We partition this matrix as
" #( )
ðn Þ ðn Þ ðn Þ
A11 A12 b1 0
¼ (12.4.2)
ðn Þ
A21 A22
ðn Þ ðn Þ
b2 0
where
ðn Þ
A11 : has dimensions : p p (12.4.3a)
ðn Þ
A12 : has dimensions : p ðN pÞ (12.4.3b)
ðn Þ
A21 : has dimensions : ðN pÞ p (12.4.3c)
ðn Þ
A22 : has dimensions : ðN pÞ ðN pÞ (12.4.3d)
ðn Þ
b1 : has dimensions : p 1 (12.4.3e)
ðn Þ
b2 : has dimensions : ðN pÞ 1 (12.4.3f)
Eq. (12.4.2) after performing the multiplication gives
ðn Þ ðn Þ ðn Þ ðn Þ
A11 b1 + A12 b2 ¼ 0 (12.4.4a)
A21 b1 + A22 b2 ¼0 (12.4.4b)
ðn Þ ðn Þ
Because A22 is not singular, Eq. (12.4.4b) is solved for b2
1
b2 ¼ A22 A21 b1 (12.4.5)
If we now choose arbitrarily p linearly independent vectors, each of them
with dimension p
ðn Þ ðn + 1Þ ðn + p1Þ
b1 ,b1 , ⋯,b1
then Eq. (12.4.5) can be used to determine the p vectors
b2 ,b2 , ⋯,b2
For example, a set of p arbitrary linearly independent vectors is
8 9 8 9 8 9
>
> 1>> >
> 0>> >
> 0>
>
>
> >
> >
> >
> >
> >
< =
0 < =
1 <0>=
b1 ¼ 0 ,b1 ¼ 0 ,⋯, b1 ¼ 0 (12.4.6)
>
> > > > > >
>
> ⋮>> >
> ⋮>> >
> ⋮>
>
: > ; : >
> ; : >
> ;
0 p1 0 p1 1 p1
The so-obtained arbitrary p vectors bðnÞ ,bðn + 1Þ , ⋯,bðn + p1Þ corresponding

to the multiple eigenfrequency wn are obviously linearly independent; however,
a. If the eigenvalue problem has degenerate

eigenvalues,
p the eigenvectors
are established from the
solution of the eigenvalue problem K w2n M b ¼0, instead of K w2n M b ¼0 [1].
@CivilMethod
they do not satisfy in general the orthogonality condition between each other.
As we will see in Section 12.8, The fulfillment of the orthogonality condition is
necessary for the solution of the free vibration problem of the MDOF systems.
This can be achieved using the Gram-Schmidt orthogonalization process, which
is described right below
FIG. 12.4.1 Geometrical interpretation of Gramm-Schmidt orthogonalization.
Let x1 ,x2 ,…, xN be a set of N linear independent but not orthogonal vectors.
These vectors constitute the basis of a space with N dimensions. We can readily
show that the vectors defined by the relations
e
x1 ¼ x1
e
xT1 x2
e
x2 ¼ x2 e
x1
x1 j2
je
e
xT1 x3 e
xT2 x3
e
x3 ¼ x3 2
e
x2e
x1 (12.4.7)
je
x1 j x2 j2
je
⋯ ¼ ⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
e
xT1 xN e
xT2 xN e
xTN 1 xN
e
xN ¼ xN e
x1 e
x2 ⋯ e
xN 1
x1 j2
je x2 j2
je xN 1 j2
je
are orthogonal to each other. From the geometrical point of view, the vector e xk
is equal to xk minus its projection on the space defined by the vectors
e
x1 ,e
x2 ,…,exk1 . For example, the vector ex2 is equal to x2 minus its projection
on the vector ex1 . This is shown in Fig. 12.4.1.
The orthogonality of the eigenvectors with respect to the mass, which is also
needed, is achieved using the following relations.
e
x1 ¼ x1
e
xT1 Mx2
e
x2 ¼ x2 e
x1
e
xT1 Me
x1
e
xT1 Mx3 e
xT2 Mx3
e
x3 ¼ x3 e
x 1 e
x2 (12.4.8)
e
xT1 Me
x1 e
xT2 Me
x2
⋯ ¼ ⋯⋯⋯⋯⋯⋯⋯⋯⋯
e
xT1 MxN e
xT MxN e
xT MxN
e
xN ¼ xN e
x1 2T x2 ⋯ TN 1
e e
xN 1
e
x1 Me
T
x1 e
x2 Me
x2 e
xN 1 Me
xN 1
@CivilMethod
Example 12.4.1 Compute the eigenfrequencies and eigenmodes of a structure

whose stiffness and mass matrices are:
2 3 2 3
3826 1913 7652 25 0 25
K ¼ 4 1913 2285:5 6484 5, M ¼ 4 0 8 16 5 (1)
7652 6484 21307:08 25 16 67
Solution
(i) Computation of the eigenfrequencies.
The frequency equation is given by the determinant

3826 25w2 1913 7652 + 25w2

det ½AðwÞ ¼ 1913 2285:5 8w 2
6484 16w 2 ¼0 (2)

7652 + 25w2 6484 16w2 21307:08 67w 2
which after expanding yields the characteristic polynomial

w6 507:4355w4 + 55567:85905w2 1746816:390 ¼ 0 (3)
from which we obtain the roots
w21 ¼ w22 ¼ 68:7088, w23 ¼ 370:0179 (4)
and the eigenfrequencies
w1 ¼ w2 ¼ 8:2891, w3 ¼ 19:2358 (5)
We observe that the structure has two equal eigenfrequencies, that is,
p ¼ 2.
(ii) Computation of the eigenmodes.
For w1 ¼ w2 ¼ 8:2891 we have
Because the rank of the matrix is N p ¼ 3 2 ¼ 1, the nonvanishing

determinants have dimensions 1 1 and Að1Þ is partitioned as shown above.
Thus partitioning Að1Þ as indicated, we have

ðn Þ 2108:2705 1913:0000
A11 ¼ (6)
1913:0000 1735:8266

ðn Þ 5934:2705
A12 ¼ (7)
5384:6531
ðn Þ
A21 ¼ ½ 5934:2705 5384:6531 (8)
ðn Þ
A22 ¼ ½16703:5650 (9)
@CivilMethod
For n ¼
1,we obtain
ð1Þ 1 ð1Þ
b1 ¼ and Eq. (12.4.5) gives b2 ¼ f0:3553g. Hence
0 8 9
<1 =
b1 ¼ 0 (10)
: ;
0:3553
For n ¼
2,we obtain
ð2Þ 0 ð2Þ
b1 ¼ and Eq. (12.4.5) gives b2 ¼ f0:322365g. Hence
1 8 9
< 0 =
b2 ¼ 1 (11)
: ;
0:322365
For n ¼ 3 the eigenfrequency is discrete, w3 ¼ 19:2358, the rank of the
matrix is 3 1 ¼ 2, and the matrix Að3Þ is partitioned as
ð12Þ
Thus we have
ð3Þ
A11 ¼ ½ 5424:4 (13)
ð3Þ
A12 ¼ ½ 1913 1598:4 (14)

ð3Þ 1913
A21 ¼ (15)
1598:4

ð3Þ 674:6280 563:7440
A22 ¼ (16)
563:7440 3483:9921
h i1
ð3Þ ð3Þ ð3Þ ð3Þ 2:8357
b1 ¼ f1g, b2 ¼ A22 A21 ¼ . Hence
8 90
< 1 =
b3 ¼ 2:8357 (17)
: ;
0
(iii) Orthogonalization of the eigenmodes.
The computed eigenvectors b1 , b2 , and b3 are linearly independent.
This is readily verified by computing the determinant of the eigenvectors,
which apparently does not vanish

1 0 1

det ðBÞ ¼ 0 1 2:8356 ¼ 1:26936 (18)
0:3553 0:322365 0
@CivilMethod
However, they do not satisfy the orthogonality condition. They are made
orthogonal using the Gram-Schmidt process described previously. Thus
applying Eq. (12.4.8), we obtain
e1 ¼ b1 ¼ f 1 0 0:3553 gT
b
eT Mb2
b
e
b2 ¼ b2 1T e1 ¼ f 0:3868 1 0:4598 gT
b (19)
e Mb
b e1
1
e3 ¼ b3 ¼ f 1 2:8357 0 gT
b
and after their orthonormalization with respect to the mass matrix, we obtain
2 3
0:2524 0:2551 0:1058
F ¼ 4 0:0000 0:6595 0:3000 5 (20)
0:0897 0:3032 0:0000
Example 12.4.2 Determine the eigenfrequencies and eigenmodes of the free

moving flexible bar of Fig. E12.3a by considering only flexural vibrations.
The bar is approximated by two equal elements. The element properties
r, A,E,I are assumed constant. Consider lumped mass assumption.
FIG. E12.3a Moving flexible bar in Example 12.4.2.
Solution
For lumped mass assumption, the dynamic model of the system is shown in
Fig. E12.3b. The system has six degrees of freedom, which are reduced to three,
u1 , u2 ,u3 , after the static condensation of the rotations.
FIG. E12.3b Dynamic model in Example 12.4.2.
Thus the mass matrix results as

2 3
1 0 0
rAL 4
M¼ 0 2 05 (1)
4
0 0 1
@CivilMethod
and the stiffness matrix as 2 3

1 2 1
12EI
K ¼ 3 4 2 4 2 5 (2)
L
1 2 1
The frequency equation is

1 l 2 1

det ½AðlÞ ¼ 2 4 2l 2 ¼ l 2 ð l 4Þ ¼ 0 (3)

1 2 1l
where l ¼ w2 rAL4 =48EI .
The roots of Eq. 3 are
l1 ¼ l2 ¼ 0, l3 ¼ 4 (4)
Obviously, there is a double eigenfrequency equal to zero, which is due to
the rigid body motion of the bar. The eigenvectors corresponding to the double
eigenfrequency are established using the procedure presented in Section 12.4.
Thus, for l1 ¼ l2 ¼ 0 it is
ð5Þ
Because the rank of the matrix is 3 2 ¼ 1, the nonvanishing determinants

have dimensions 1 1.
Assuming the partitioning of Aðl1 Þ and Aðl1 Þ shown in Eq. 5, we have

ð1Þ 1 2 ð1Þ 1 ð1Þ ð1Þ
A11 ¼ , A12 ¼ , A21 ¼ ½ 1 2 , A22 ¼ ½1 (6)
2 4 2
For n ¼ 1
ð1Þ 1 ð1Þ
l1 ¼ 0, b1 ¼ and Eq. (12.4.5) gives b2 ¼ f1g. Hence
0 8 9
< 1=
bð1Þ ¼ 0 (7)
: ;
1
For n ¼ 2
ð2Þ 1 ð2Þ
l2 ¼ 0, b1 ¼ and Eq. (12.4.5) gives b2 ¼ f1g. Hence
1 8 9
<1=
bð2Þ ¼ 1 (8)
: ;
1
For n ¼ 3 it is l3 ¼ 4, the rank of the matrix is 3 1 ¼ 2, and the matrix
Aðl3 Þ is partitioned as
@CivilMethod
Therefore

ð3Þ ð3Þ ð3Þ 2 ð3Þ 4 2
A11 ¼ ½3, A12 ¼ ½ 2 1 , A21 ¼ , A22 ¼
1 2 3
and Eq. (12.2.22) give
h i1
ð3Þ ð3Þ ð3Þ 1
b2 ¼ A22 A21 ¼ . Hence
1
8 9
< 1=
bð3Þ ¼ 1 (9)
: ;
1
The graphical representations of the eigenmodes are shown in Fig. E12.3c.
FIG. E12.3c Eigenmodes in Example 12.4.2.
We readily find out that the orthogonality condition with respect to the mass
matrix is satisfied.
12.5 The linear eigenvalue problem

12.5.1 The standard eigenvalue problem of linear algebra
As already mentioned, the problem of determining the eigenfrequencies and
eigenmodes is directly related to the linear eigenvalue problem of the linear
algebra [1, 2],
which
is stated as follows.
Let A ¼ aij be a square N N matrix and x an N 1 vector, whose com-
ponents with respect to a certain base are x1 ,x2 ,…, xN . The relation
y ¼Ax (12.5.1)
represents a linear transformation, which transforms the vector x to the vector y
with components
y1 ¼ a11 x1 + a12 x2 + ⋯ + a1N xN
y2 ¼ a21 x1 + a22 x2 + ⋯ + a2N xN
(12.5.2)
⋯ ¼ ⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
yN ¼ aN 1 xN 1 + aN 2 x2 + ⋯ + aNN xN
In general, the matrix A is complex. However, in the problems we are study-
ing, it is real. For example, we consider the matrix

2 1
A¼ (12.5.3)
5:5 3
@CivilMethod
Eq. (12.5.1) transforms the vector x ¼ f 2 3 gT to

y1 2 1 2 1
¼ ¼ (12.5.4)
y2 5:5 3 3 2
The vectors x and y are shown in Fig. 12.5.1.
FIG. 12.5.1 Transformation of a vector x to a vector y.
In general, the direction of y is different than that of x. The question now

arises whether for the given square matrix A a vector x exists, which when pre-
multiplied by the matrix produces a vector that is proportional to the original
vector. The answer is affirmative. The constant of proportionality is called
an eigenvalue of the matrix. Mathematically, this is expressed as
Ax ¼ lx (12.5.5)
or
ðA lIÞx ¼ 0 (12.5.6)
where I is the N N unit matrix and l is the corresponding eigenvalue.
Eq. (12.5.6) is written in expanded form as
2 38 9 8 9
A11 l A12 ⋯ A1N >
> x1 > > 0>
6 A21 7 < > = > < > =
6 A l ⋯ A 7 x 0
4⋯
22 2N 2
¼
5> ⋯ > > ⋯ > (12.5.7)
⋯ ⋯ ⋯ >
: > ; > : > ;
AN 1 AN 2 ⋯ ANN l xN 0
The foregoing relation represents a system of N linear algebraic equations,
which can be solved to determine the vector x. Because the system is homoge-
neous, a nontrivial solution can be obtained only if the determinant of coeffi-
cient matrix is equal to zero, that is,

A11 l A12 ⋯ A1N

A21 A22 l ⋯ A2N
det ðA lIÞ ¼ ¼0
(12.5.8)
⋯ ⋯ ⋯ ⋯
AN 1 AN 2 ⋯ ANN l
@CivilMethod
Expanding the determinant yields a polynomial of N degree with respect to

l, namely
PðlÞ ¼ a0 lN + a1 lN 1 + ⋯ + aN 1 l + aN (12.5.9)
which is referred to as the characteristic polynomial of the matrix A.
The equation
det ðA lIÞ ¼ PðlÞ ¼ 0 (12.5.10)
is called the characteristic equation of A. Because it was assumed that A is a
real matrix, the coefficients of PðlÞ are also real. It is known from the funda-
mental theorem of algebra that PðlÞ has N roots l1 ,l2 ,…,lN , which may
be real or complex conjugate while some of them may be multiple. The roots
lk ðk ¼ 1, 2, …, N Þ are called the eigenvalues of A. Understandably, for each
lk , the system of Eq. (12.5.7) yields a vector xk , which is real or complex
depending on whether lk is real or complex. The vectors xk (k ¼ 1, 2, …, N )
are called the eigenvectors of A. Obviously, if xk is an eigenvector then rxk ,
with r being a scalar, is also an eigenvector. This means that only the line of
action of the eigenvector is determined while its magnitude is undetermined.
The pairs ðlk , xk Þ, k ¼ 1, 2, …,N , are referred to as eigenpairs.
Example 12.5.1 Determine the eigenvalues and the eigenvectors of the matrix
given by Eq. (12.5.3).
Solution
The characteristic equation is

2 l 1

5:5 3 l ¼ l + 5l + 0:5 ¼ 0
2
(1)
The degree of the characteristic equation is 2, thus two eigenvalues are

obtained
l1 ¼ 0:1021, l2 ¼ 4:8980 (2)
The corresponding equation will be obtained from Eq. (12.5.7) for N ¼ 2.
Thus for l1 ¼ 0:1021, we have

2 ð0:1021Þ 1 x1 0
¼ (3)
5:5 3 ð0:1021Þ x2 0
which gives

1
x1 ¼ (4)
1:8779
and for l2 ¼ 4:8980, we have

2 ð4:8980Þ 1 x1 0
¼ (5)
5:5 3 ð4:8980Þ x2 0
@CivilMethod
which gives

1
x2 ¼ (6)
2:8978
12.5.2 Properties of the eigenvalues and eigenvectors

In this section, we will present some properties of the eigenvalues and eigen-
vectors that are useful for the development of numerical methods for calculating
the eigenfrequencies and eigenmodes of the free vibrations of the MDOF sys-
tems, but also in general for the study and understanding of their dynamic
response.
Let x ¼ fx1 , x2 , …, xN gT and y ¼ fy1 , y2 , …, yN gT be two, in general, com-
plex eigenvectors of the N dimensional space. Their inner or scalar product is
defined by the relation
X
N
xk yk ¼ xT y
(12.5.11)
k¼1
is the vector with components yk , the complex conjugate of yk .

where y
We readily verify that the inner product obeys the following rules:
(i) > 0 if x 6¼ 0 and xT x
xT x ¼ 0 if x ¼ 0
(ii) x y
T
¼ yT x

(iii) ðaxÞT y ¼ a x T y

(iv) xT ðay
Þ ¼ a xT y where a is a complex scalar and a its conjugate.
¼ x, y
If x and y are real then x ¼ y and therefore xT y ¼ yT x.
1. The matrix A and its transpose AT have the same eigenvalues.
This follows directly from the fact that the determinant of the matrix
does not change if its lines are interchanged with its columns. Therefore,
the matrices A and AT have the same characteristic polynomial, that is,

det ðA lIÞ ¼ det AT lI ¼ PðlÞ ¼ 0 (12.5.12)
hence the same eigenvalues.
2. If lr and ls are two distinct eigenvalues, lr 6¼ ls , of a real matrix A and ls
the complex conjugate of ls , then the eigenvectors xr of A corresponding to
lr and x0s of AT corresponding to ls are orthogonal, xTr x 0s ¼ 0.
To prove this property, we formulate the inner product of the vectors
yr ¼ Axr and x0s in two ways:
Besides, Taking into account that Axr ¼ lr xr and property (iii) of the
inner product, we have
0s ¼ðAxr ÞT x
yTr x 0s
0s
¼ ðlr xr ÞT x (12.5.13a)
0s
¼ lr xTr x
@CivilMethod
Taking into account that the matrix A is real, hence A T ¼ AT ,

AT x0s 0
¼ ls xs , we obtain on the basis of property (iv) of the inner product
0s ¼ ðAxr ÞT x
yTr x 0s

¼ xTr ATx 0s

0 0
¼ xTr AT x s s (12.5.13b)
0
s x
¼ xTr l s

¼ ls xTr x
Subtracting Eq. (12.5.13b) from Eq. (12.5.13a) yields
0s ¼ 0
ðlr ls ÞxTr x
Apparently, because it was assumed that lr 6¼ ls , we obtain
0s ¼ 0
xTr x (12.5.14)
When all eigenvalues are distinct, this property produces N 2 N
orthogonality relations between the eigenvectors of the matrices A and AT .
3. If the eigenvalues l1 ,l2 ,…,lN of the matrix A are distinct, then the eigen-
vectors x1 ,x2 , …,xN are linearly independent.
Let k < N eigenvectors x1 ,x2 , …,xk be linear independent. Then the
eigenvectors xk + 1 ,xk + 2 ,…,xN can be expressed as linear combinations
of the independent ones, namely
X
k
xj ¼ ci xi , j ¼ k + 1, k + 2, …,N (12.5.15)
i¼1
which when multiplied by A, gives

X
k
Axj ¼ ci Axi
i¼1
(12.5.16)
X
k
¼ ci li xi
i¼1
On the other hand, we have

Axj ¼ lj xj
X
k (12.5.17)
¼ lj ci xi
i¼1
Subtracting Eq. (12.5.17) from (12.5.16), we obtain by virtue of

Eq. (12.5.15)
X
k
lj li ci xi ¼ 0 (12.5.18)
i¼1
@CivilMethod
Because it was assumed that lj 6¼ li and xi are linearly independent, it

implies that all coefficients ci are zero, thus Eq. (12.5.15) gives xj ¼ 0. This,
however, is absurd because xj as an eigenvector cannot be equal to zero.
Therefore, x1 , x2 ,…,xN are linearly independent.
4. If the matrix A is symmetric, then its eigenvalues are real.
To prove this property, we formulate the inner product of the vectors
y ¼ Ax and x in two ways as previously.
First, we obtain on the basis of Eq. (12.5.13a)
¼lxT x
yT x (12.5.19)
Then taking into account that the matrix A is real and symmetric, it holds
T ¼ AT ¼ A, AT x ¼ l
A s x0 , Ax ¼ lx, and on the basis of property (iv) of
s
the inner product, we obtain
¼ðAxÞT x
yT x

¼ x T AT x
¼ xT A
x
(12.5.20)
¼ x Ax
T

¼ xT lx
Tx
¼ lx
Subtracting Eq. (12.5.20) from Eq. (12.5.19) yields

ll xT x
¼0 (12.5.21)
> 0 and
and on the basis of property (i) it is xT x

l¼l (12.5.22)
hence the eigenvalue l is real.
The orthogonality properties are simplified significantly for real and
symmetric matrices because, in addition to real eigenvalues, they have real
eigenvectors, too. Moreover, the matrices A and AT , in addition to having
the same eigenvalues, have the same eigenvectors, x0 ¼ x. Thus
Eq. (12.5.14) can be written as
xTr xs ¼ 0 (12.5.23)
5. If the real matrix A is symmetric and positive definite, then its eigenvalues
are positive.
Consider the matrix A and an arbitrary vector x ¼ fx1 , x2 , …, xN gT . The
homogeneous polynomial of the second degree defined by
N X
X N
Pðx1 , x2 , …, xN Þ ¼ aik xi xk (12.5.24)
i¼1 k¼1
@CivilMethod
or
PðxÞ ¼ xT Ax (12.5.25)
is referred to as the quadratic form of the matrix A. For a real and symmetric
matrix A, the coefficients of the polynomial PðxÞ are real and aik ¼ aki .
The quadratic form PðxÞ is called positive definite, if
T
x Ax > 0, if x 6¼ 0
PðxÞ ¼ (12.5.26)
xT Ax ¼ 0, if x ¼ 0
A matrix A is called positive definite if its quadratic form is positive.
A direct consequence of this is that the diagonal elements of A are positive.
Indeed, if we take xi ¼ 1 and xk ¼ 0, k 6¼ i, then Eq. (12.5.24) gives
Pðx1 , x2 , …, xN Þ ¼ aii > 0 (12.5.27)
Let x be an eigenvector of A. Because A is real and symmetric, the
eigenpair ðl, xÞ is real according to the previous property. To prove that
l > 0 we write Eq. (12.5.25) as
T
xT Ax ¼ AT x x
¼ ðAxÞT x (12.5.28)
¼ lxT x
Because xT x > 0, hence xT Ax > 0, the foregoing relation gives
xT Ax
l¼ >0 (12.5.29)
xT x
6. To each eigenvalue of a symmetric matrix there correspond as many line-
arly independent eigenvectors as the multiplicity of the eigenvalue.
Let A be an N N symmetric matrix. We assume that we have com-
puted the eigenvalue lk and the corresponding eigenvector xk . Note that
the eigenvalue problem has no degenerate eigenvalues because the matrix
is assumed symmetric. We construct now an N N orthonormal matrix Q,
whose first column is the vector xk , namely

Q ¼ xk Q ^ , QT Q ¼ I (12.5.30)
^
Apparently, the dimensions of Q are N ðN 1Þ. The construction of Q
is always possible because the vectors in Q provide a basis of the N dimen-
sional space defined by A. This is understood from Example 12.5.2, which
is presented below.
First we evaluate
" #
xTk
Q AQ ¼
T
A xk Q^
^
Q T
" # (12.5.31)
xTk Axk xTk AQ^
¼
^ T Axk Q
Q ^ T AQ^
@CivilMethod
^ T Axk ¼
Then taking into account that Axk ¼ lk xk , xTk xk ¼ 1, Q
^ T T ^ ^ T T
lk Q xk ¼ 0, xk AQ ¼ Q A xk ¼ 0 because it was assumed AT ¼ A,
we write the foregoing relation as

l 0
QT AQ ¼ k ^ (12.5.32)
0 A
where
^ ¼Q
A ^
^ T AQ (12.5.33)
^ ¼ c, that is, a con-
is a fully populated ðN 1Þ ðN 1Þ. If N ¼ 2, then A
stant, and the matrix Q AQ becomes diagonal
T

l 0
QT AQ ¼ k (12.5.34)
0 c
Premultiplying the previous equation by Q gives

lk 0
AQ ¼ Q
0 c
(12.5.35)
lk 0
¼ x Q ^
0 c
Hence Q ^ is another eigenvector of A and c is the other eigenvalue
regardless of whether lk is a multiple or a distinct eigenvalue. For arbitrary
N the proof is achieved by mathematical induction. Taking into account the
previous proof, the Gramm-Schmidt orthogonalization, and property (4), we
deduce that an N N symmetric matrix has N orthonormal eigenvectors.
Example 12.5.2 Construct the orthonormal matrix Q ¼ ½ x1 x2 x3 , from

xT1 ¼ f 1:5 3 2 g.
Solution
We normalize the vector x1 with respect to its magnitude jx1 j ¼ 3:9051. Thus
we obtain
xT1 ¼ f 0:3841 0:7682 0:5121 g (1)
The vector xT2 ¼ f x12 x22 x32 g can be computed from the orthogonality
relation
xT1 x2 ¼ 0:3841x12 0:7682x22 + 0:5121x32 ¼ 0 (2)
Thus, by determining the two components arbitrarily, then the third compo-
nent is computed from Eq. (2). In this respect, we take arbitrarily
x12 ¼ 5, x22 ¼ 1. Then Eq. (2) gives x32 ¼ 5:25 and
xT2 ¼ f 5 1 5:25 g (3)
@CivilMethod
which is normalized with respect to its magnitude jx2 j ¼ 7:3186 and becomes
xT2 ¼ f 0:6832 0:1366 0:7173 g (4)
The vector xT3 ¼ f x13 x23 x33 g can be computed from the orthogonality
relations
xT1 x3 ¼ 0:3841x13 0:7682x23 + 0:5121x33 ¼ 0 (5)
xT2 x3 ¼ 0:6832x13 0:1366x23 0:7173x33 ¼ 0 (6)
Thus, by determining x13 arbitrarily, say x13 ¼ 1, the foregoing relations
yield the linear system of equations
0:7682x23 + 0:5121x33 ¼ 0:3841 (7)
0:1366x23 0:7173x33 ¼ 0:6832 (8)
from which we obtain x23 ¼ 1:0070, x33 ¼ 0:7606, hence
x3 ¼ f 1 1:0070 0:7606 gT (9)

which is normalized with respect to its magnitude jx3 j ¼ 1:6102 and becomes
x3 ¼ f 0:6211 0:6254 0:4724 gT (10)

Therefore the requested orthonormal matrix is
2 3
0:3841 0:6832 0:6211
Q ¼ 4 0:7682 0:1366 0:6254 5 (11)
0:5121 0:7173 0:4724
The vectors x1 ,x2 , x3 provide a new basis in the three-dimensional space.
7. Reduction of a square matrix to diagonal form

From the previous properties, it can be concluded that the real and sym-
metric matrix A with dimensions N N has always N eigenvectors, which
are linearly independent and orthogonal to each other. Therefore, the eigen-
vectors can provide a new basis in the N dimensional space. A vector x
represented with respect to the new basis will have components
x^1 , x^2 ,…, x^N . On this basis, we can write
x ¼ x^1 x1 + x^2 x2 + … + x^N xN (12.5.36)
or
x ¼ X^
x (12.5.37)
where
2 3
x11 x12 ⋯ x1N
6 x21 x22 ⋯ x2N 7
X ¼ ½ x1 x2 ⋯ xN ¼ 6
4⋯
7 (12.5.38)
⋯ ⋯ ⋯ 5
xN 1 xN 2 ⋯ xNN
@CivilMethod
represents the transformation matrix. It should be noted that matrix X can

be inverted,because the vectors xi are linearly independent, that is,
det ðXÞ 6¼ 0.
Next, we transform the relation y ¼Ax with respect to the new base.
According to Eq. (12.5.37), the vector y is written
^ ¼ X1 y
y
¼ X1 Ax
(12.5.39)
¼ X1 AX^
x
^x
¼ A^
where
^ ¼ X1 AX
A (12.5.40)
But it holds
AX ¼ A½ x1 x2 ⋯ xN
¼ ½ Ax1 Ax2 ⋯ AxN
(12.5.41)
¼ ½ l1 x1 l2 x2 ⋯ lN xN
¼ XL
where
2 3
l1 0 ⋯ 0
60 l2 ⋯ 0 7
L ¼6
4⋯
7 (12.5.42)
⋯ ⋯ ⋯ 5
0 ⋯ 0 lN
is a diagonal matrix with elements the N eigenvalues of A.
Finally, Eq. (12.5.40) by virtue of Eq. (12.5.41) becomes
^ ¼ X1 XL ¼ L
A (12.5.43)
L ¼ X1 AX (12.5.44)
that is, the matrix A in the space defined by the eigenvectors takes a
diagonal form.
Further, solving Eq. (12.5.44) with respect to A gives
A ¼ XLX1 (12.5.45)
The forgoing representation of A is called the spectral decomposition of
the matrix.
Taking into account the orthogonality property of the eigenvectors,
we obtain
@CivilMethod
2 3
x1 T x1 0 ⋯ 0
6 0 x2 T x2 ⋯ 0 7
XT X ¼ XXT ¼ 6
4⋯
7
5 (12.5.46)
⋯ ⋯ 0
0 0 ⋯ xN xN
T
which,
p ffiffiffiffiffiffiffiffiffiffiffiif the eigenvectors are normalized with respect to their magnitude,
xi T xi ¼ 1, becomes
XT X ¼ XXT ¼ I (12.5.47)
Hence
X1 ¼ XT (12.5.48)
A ¼ XLXT (12.5.49)
8. Similar matrices have the same eigenvalues.

We consider that the change of the base is performed by the
transformation
x ¼T^
x (12.5.50)
where T is a nonsingular matrix.
The matrix A is transformed in the space defined by the new base
according to Eq. (12.5.40), when we set X ¼ T. Hence
^ ¼ T1 AT
A (12.5.51)
The above transformation is referred to as a similarity transformation
^ are similar matrices.
while A and A
The eigenvalue problem for the matrix A^ is

A^ lI x^¼0 (12.5.52)
and its characteristic polynomial

^ ðlÞ ¼ det A
P ^ lI

¼ det T1 ðA lIÞT
(12.5.53)
¼ det T1 det ðA lIÞ det ðTÞ

¼ det T1 det ðTÞPðlÞ

Because det T1 and det ðTÞ are nonzero constants, we obtain
^ ðlÞ ¼ PðlÞ ¼ 0. Hence the characteristic polynomials P
P ^ ðlÞ and PðlÞ
^
have the same roots, which implies that the matrices A and A have the same
eigenvalues. Note that the property of the product of determinants has been
@CivilMethod
employed to reach the foregoing conclusion, that is, det ðABCÞ ¼

det ðAÞdet ðBÞ det ðCÞ. Regarding the eigenvectors, they are transformed
according to Eq. (12.5.50).
12.5.3 The generalized eigenvalue problem

The problem of determining the eigenvalues and eigenvectors of a matrix as
stated above represents the typical or standard eigenvalue problem. However,
the eigenvalue problem for determining the eigenfrequencies and mode shapes
has a more general form
ðA lBÞx ¼ 0 (12.5.54)
This problem is known as the generalized eigenvalue problem of linear alge-
bra. In the literature, it is also referred to as the linearized eigenvalue problem.
The study of the properties of the eigenvalues and eigenvectors of the general-
ized eigenvalue problem, Eq. (12.5.54), is facilitated if it is transformed to the
standard eigenvalue problem, Eq. (12.5.6). Thus, the properties that apply to the
standard eigenvalue problem can be transferred to the generalized eigenvalue
problem. Without excluding the generality, the discussion will be restricted
to real, symmetric, and positive definite matrices A and B because in free vibra-
tions, they represent the stiffness and mass matrices, that is, A ¼ K and B ¼ M.
Their positive definiteness results from the fact that the elastic energy U ðuÞ and
the kinetic energy T ðu_ Þ are expressed by positive definite quadratic forms,
that is,
8
> 1
< uT Ku > 0, if u 6¼ 0
U ðuÞ ¼ 2 (12.5.55)
>
: 1 uT Ku ¼ 0, if u ¼ 0
2
8
> 1
< u_ T Mu_ > 0, if u_ 6¼ 0
T ðu_ Þ ¼ 2 (12.5.56)
>
: 1 u_ T Mu_ ¼ 0, if u_ ¼ 0
2
Applying Eq. (12.5.49) to matrix B we have
eL
B¼X eX
eT (12.5.57)
where L e is the diagonal matrix of the eigenvalues of B and X e the matrix
of its eigenvectors. Obviously, when the matrix B is diagonal, as in the case
of concentrated masses, then Xe ¼ I.
Further, we can set
B ¼ SST (12.5.58)
@CivilMethod
where
pffiffiffiffi
e
S¼X Le (12.5.59)
e are positive because B
The matrix S is real on account that the elements of L
was assumed positive definite.
Substituting Eq. (12.5.59) into Eq. (12.5.54), yields
Ax ¼ lSST x (12.5.60)
We define now the vector
^ ¼ ST x
x (12.5.61)
thenb
x ¼ ST x
^ (12.5.62)
1
Premultiplying Eq. (12.5.60) by S and using Eq. (12.5.62) yield

^ lI x
A ^¼0 (12.5.63)
where
^ ¼ S1 AST
A (12.5.64)
Obviously, the matrix A ^ is real and symmetric. Therefore, according to
property 3, its eigenvalues and eigenvectors x^ are real. Moreover, taking into
^ satisfy the orthogonality condition, x
account that the vectors x ^Ti x
^j ¼ 0, i 6¼ j,
we obtain.
T
^Ti x
x ^j ¼ ST xi ST xj
¼xTi SST xj , i 6¼ j
(12.5.65)
¼xTi Bxj
¼0
which implies that the eigenvectors of the generalized eigenvalue problem are
orthogonal with respect to the matrix B.
The eigenvalue problems (12.5.54) and (12.5.63) have the same eigen-
values. Indeed, we can set I ¼ S1 SST ST and write Eq. (12.5.63) by virtue
of Eqs. (12.5.64) and (12.5.58) as
1
S ðA lBÞST x ^¼0 (12.5.66)
If PðlÞ and P ^ ðlÞ are the characteristic polynomials of the eigenvalue
problems (12.5.54) and (12.5.63), respectively, we obtain
1
b. The notation ST ¼ ST is employed.
@CivilMethod

^ ðlÞ ¼ det A
P ^ lI

¼ det S1 ðA lBÞST
(12.5.67)

¼ det S1 det ðA lBÞdet ST

¼ det S1 PðlÞ det ST

Because det S1 6¼ 0 and det ST 6¼ 0, it implies that both eigenvalue
problems have the same characteristic equation, hence the same
eigenvalues.
The spectral decomposition of matrix B requires the complete solution of
the eigenvalue problem. Therefore, the transformation of the generalized eigen-
value problem on the basis of Eq. (12.5.59) is not the most convenient one.
A usual method to determine the matrix S is the Cholesky decomposition
method, or the square root method, in which the matrix B is written in the form
of a product, that is,
B ¼ UT U (12.5.68)
where U is an upper triangular matrix. Hence
S ¼ UT (12.5.69)
9. If the matrices A and B real, symmetric, and positive definite, then the
generalized eigenvalue problem has positive eigenvalues.
This is readily proved by premultiplying Eq. (12.5.54) by xT . This
gives
xT Ax ¼ lxT Bx (12.5.70)
Because A and B are positive definite, we obtain
xT Ax
l¼ >0 (12.5.71)
xT Bx
10. The eigenvectors xi of the generalized eigenvalue problem are orthogonal

with respect to the matrices A and B.
It was previously shown that
xTi Bxj ¼ 0, i 6¼ j (12.5.72)
Consequently, we obtain
xTi Axj ¼ lxTi Bxj
(12.5.73)
¼0
@CivilMethod
Example 12.5.3 Transform the generalized eigenvalue problem into a standard

eigenvalue problem using the spectral decomposition method, when

50 40 2:5 1
A¼ , B¼
40 90 1 3:2
Solution
The characteristic equation of B is

2:5 le 1
¼l
e2 5:7l
e + 7:0 ¼ 0 (1)
1 3:2 l
e
l e2 ¼ 3:909
e1 ¼ 1:790, l (2)
Hence

e 1:790 0
L¼ (3)
0 3:909
The eigenvectors are computed from the solution of the homogeneous linear
system

e 1
2:5 l x1 0
1 e x2 ¼ 0
3:2 l
(4)
e1 ¼ 1:790 and l
for l e2 ¼ 3:909.
Thus, we obtain the matrix of the eigenvectors normalized with respect to
their magnitude
" #
0:8157 0:5784
e¼
X (5)
0:5785 0:8158
Using Eq. (12.5.59) we obtain
pffiffiffiffi
e e 1:0913 1:1435 1 0:6097 0:4323
S¼X L¼ , S ¼ (6)
0:7740 1:6129 0:2926 0:4126
and on the basis of Eq. (12.5.64)

^ ¼ S1 AST ¼ 14:3213 2:1288
A (7)
2:1288 29:2564
11. If the real and symmetric matrix A is singular, then the generalized eigen-
value problem has at least one zero eigenvalue and the corresponding
eigenvector is different from zero.
@CivilMethod
First, we will show that this property holds for the standard eigenvalue
problem Ax ¼ lx. For this purpose, we write A in the form of its spectral
decomposition,
A ¼ XLXT (12.5.74)
Because it was assumed that A is
singular,
it implies that det ðAÞ ¼ 0.
Moreover, Eq. (12.5.47) gives det XT det ðXÞ ¼ 1, hence

det ðAÞ ¼ det XT det ðLÞdet ðXÞ
¼ det ðLÞ
(12.5.75)
¼l1 l2 ⋯lN
¼0
from which we conclude that at least one of the li is zero and the eigen-
value problem for this value becomes
Axi ¼ 0 (12.5.76)
which yields xi 6¼ 0 because det ðAÞ ¼ 0.
The generalized eigenvalue problem Ax ¼ lBx is transformed to the
standard eigenvalue problem A^^ x ¼ l^
x, hence it is

det A^ ¼ l1 l2 ⋯lN (12.5.77)

det A ^ ¼ det S1 det ðAÞdet ST
(12.5.78)
¼0
because it was assumed det ðAÞ ¼ 0. Eqs. (12.5.77) and (12.5.78) imply
that at least one of the eigenvalues li is zero. Moreover, it is x î 6¼ 0 and
by virtue of Eq. (12.5.62), we obtain xi ¼ ST x î 6¼ 0.
If the matrices A and B represent the stiffness and mass matrices of
the structure, that is, A ¼ K, B ¼ M, then the eigenvector xi correspond-
ing to the zero eigenvalue represents rigid body motion. This is shown
right below.
If we set xi ¼ ui , then Kui represents the vector of the elastic force f Si
corresponding to the displacement ui , that is
f Si ¼ Kui (12.5.79)
or because li ¼ 0
Kui ¼ li Mui
(12.5.80)
¼0
Hence f Si ¼ 0 while Eq. (12.5.76) yieldsui 6¼ 0, which is due to the
motion of the structure as a rigid body.
@CivilMethod
12. If the real and symmetric matrix B is singular, then the generalized eigen-
value problem has at least one infinite eigenvalue.
This is shown if the eigenvalue problem is written in the form
Bx ¼ mAx (12.5.81)
where m ¼ 1=l.
13. Any vector u with a dimension N can be represented as the superposition
of the eigenvectors of the eigenvalue problem.
We showed that the set of the eigenvectors x1 ,x2 ,…, xN of an N N
symmetric matrix is linearly independent and can be employed as a base of
the N dimensional space to represent an arbitrary vector u in that space.
Thus we may set
u ¼ a1 x1 + a2 x2 + ⋯ + aN xN (12.5.82)
or
u ¼ Xa (12.5.83)
where X is the matrix of the eigenvectors and a ¼ fa1 , a2 , …, aN gT the
vector of the coefficients. The matrix X is not singular because the eigen-
vectors xi are linearly independent. Hence
a ¼ X1 u (12.5.84)
a ¼ XT u (12.5.85)
or
ai ¼ xi T u, i ¼ 1, 2, …,N (12.5.86)
Similarly, we can use the eigenvectors of the generalized eigenvalue
problem to represent the vector u. The establishment of the coefficients
ai in that case is established by premultiplying both sides of
Eq. (12.5.82) by xTi B and noting that xTi Bxj ¼ 0 for i 6¼ j. This yields
xTi Bu
ai ¼ (12.5.87)
xTi Bxi
If the eigenvectors xi are normalized with respect to B so that
xTi Bxi ¼ 1, then Eq. (12.5.87) gives
ai ¼ xTi Bu, i ¼ 1, 2, …,N (12.5.88)
or
a ¼ XT Bu (12.5.89)
The representation of a vector as a superposition of the eigenvectors as
in Eq. (12.5.82) is known as the expansion theorem. As we will see in
@CivilMethod
Section 14.5, this theorem is a special case of a Ritz vector representation

when the eigenvectors are used as Ritz vectors.
14. If a real and symmetric matrix A is singular, then the quadratic form
PðuÞ ¼ uT Au is positive semidefinite, that is, PðuÞ ¼ uT Au 0 for u 6¼ 0.
We write the vector u in the form of Eq. (12.5.83) and the matrix A in
the form of Eq. (12.5.49). Thus we have
PðuÞ ¼uT Au
¼aT XT AXa
¼aT La (12.5.90)
X
N
¼ li ai2
i¼1
Because A is singular, at least on of its eigenvalues is zero, say lk ¼ 0.

If we take u ¼ xk , it will be ak 6¼ 0, ai6¼k ¼ 0 and Eq. (12.5.90) becomes
PðuÞ ¼lk ak2
(12.5.91)
¼0
Therefore
PðuÞ ¼ uT Au 0 for u 6¼ 0 (12.5.92)
12.6 The Rayleigh quotient

In Section 12.5, we discussed the standard and the generalized eigenvalue prob-
lems and we studied the properties of the eigenvalues and eigenvectors. In this
section, we will supplement these properties with Rayleigh’s quotient, which
yields further important properties that are very useful for the numerical com-
putation of eigenvalues and eigenvectors.
The Rayleigh quotient for an N N symmetric matrix A is the scalar
quantity defined by the relation
uT Au
rðuÞ ¼ (12.6.1)
uT u
where u is a vector with dimension N . We will show that the following
properties hold.
1. If l1 l2 l3 ⋯ lN are the eigenvalues of A, then
l1 rðuÞ lN (12.6.2)
and it is rðuÞ > 0, if A is positive definite and rðuÞ 0 if A is positive
semidefinite. Moreover, it is rðuÞ ¼ l1 if u ¼ x1 and rðuÞ ¼ lN if u ¼ xN .
@CivilMethod
On the basis of the expansion theorem, Eq. (12.5.82), we may set

X
N
u¼ ak xk (12.6.3)
k¼1
where the eigenvectors xk of A are assumed normalized with respect to their

magnitude, that is
xTi xj ¼ d ij (12.6.4)
where dij is the Kronecker delta defined as d ij ¼ 1 if i ¼ j and dij ¼ 0 if i 6¼ j.
Substituting Eq. (12.6.3) into Eq. (12.6.1) and taking into account
Eq. (12.6.4) and Axk ¼ lk xk yield
l1 a12 + l2 a22 + ⋯ + lN aN2
rðuÞ ¼ (12.6.5)
a12 + a22 + ⋯ + aN2
which for l1 6¼ 0 becomes
a12 + ðl2 =l1 Þa22 + ⋯ + ðlN =l1 ÞaN2
rðuÞ ¼ l1 (12.6.6)
a12 + a22 + ⋯ + aN2
Because lk =l1 1 (k ¼ 2, 3, …, N ), it follows that the nominator of
the foregoing fraction is greater or equal to the denominator, hence
rðuÞ l1 . If u ¼ x1 we have a1 ¼ 1, a2 ¼ a3 ¼ ⋯ ¼ aN ¼ 0 and Eq.
(12.6.5) gives rðuÞ ¼ l1 .
Similarly, for lN 6¼ 0 Eq. (12.6.5) is written as
a12 ðl1 =lN Þ + ðl2 =lN Þa22 + ⋯ + aN2
rðuÞ ¼ lN (12.6.7)
a12 + a22 + ⋯ + aN2
The nominator of the forgoing fraction is less or equal to the denominator
because lk =lN 1 (k ¼ 1, 2, …,N 1), hence rðuÞ lN . If u ¼ xN we
have a1 ¼ a2 ¼ ⋯ ¼ aN 1 ¼ 0, aN ¼ 1 and Eq. (12.6.7) gives rðuÞ ¼ lN .
2. If the vector u is chosen from a subset of vectors that is orthogonal to the
first k 1 eigenvectors of A, then
rðuÞ lk (12.6.8)
Applying the expansion theorem, Eq. (12.5.82), we have
u ¼ a1 x1 + a2 x2 + ⋯ + ak1 xk1 + ak xk + ⋯ + aN xN (12.6.9)
Due to the assumed orthogonality, it is uT xj ¼ 0, j ¼ 1, 2, …,k 1
and Eq. (12.5.86) gives aj ¼ 0, j ¼ 1, 2, …,k 1. Hence Eq. (12.6.9) is
written as
X
N
u¼ aj xj (12.6.10)
j¼k
@CivilMethod
Substituting the previous expression for u into Rayleigh’s quotient,

Eq. (12.6.1), yields
lk ak2 + lk + 1 ak2 + 1 + ⋯ + lN aN2
rðuÞ ¼
ak2 + ak2 + 1 + ⋯ + aN2
(12.6.11)
a 2 + ðlk + 1 =lk Þak2 + 1 + ⋯ + ðlN =lk ÞaN2
¼lk k
ak2 + ak2 + 1 + ⋯ + aN2
Inasmuch as lk + 1 =lk 1, ⋯,lN =lk 1, it follows that the nominator
of the foregoing fraction is greater or equal to the denominator, hence
rðuÞ lk . If u ¼ xk we have ak ¼ 1, ak + 1 ¼ ⋯ ¼ aN ¼ 0 and Eq.
(12.6.11) gives rðuÞ ¼ lk .
3. If the vector u deviates from the eigenvector xk by an error of order e, then
Rayleigh’s quotient approximates the eigenvalue lk with an error of order
e2 , namely

rðuÞ ¼ lk + O e2 (12.6.12)
To prove this statement we set
u
u ¼ xk + ee (12.6.13)
u is the deviation of u from xk and expresses the contribution of the
where ee
remaining eigenvectors.
Substituting Eq. (12.6.13) into Eq. (12.6.1) gives

xTk + eeuT Aðxk + ee uÞ
rðuÞ ¼
xTk + ee uT ðxk + eeuÞ
(12.6.14)
xTk Axk + ee T
u Axk + exTk Ae u + e2 e
uT Ae
u
¼
e + e2 u
uT xk + exTk u
xTk xk + ee eT u
e
Further, by virtue of the expansion theorem, Eq. (12.5.82), we can write
X
N
e
u¼ aj x j (12.6.15)
j ¼1
j 6¼ k
Substituting the foregoing expression for e
u into Eq. (12.6.14) and taking
into account the orthogonality of the eigenvectors, we obtain
X
N
lk + e2 lj aj2
j¼1, j6¼k
rðuÞ ¼ (12.6.16)
1 + e2 S
@CivilMethod
where it was set
X
N
S¼ aj2 (12.6.17)
j ¼1
j 6¼ k
Expanding the denominator in Eq. (12.6.16) in series using the

polynomial theorem for a negative exponent, we obtain
1
1 + e2 S ¼ 1 e2 S + e4 S 2 e6 S 3 + ⋯ (12.6.18)
which is introduced into Eq. (12.6.16) to give

0 1
B X C
B N C
rðuÞ ¼B l
B k + e 2
l a 2C
j j C 1e S +e S e S +⋯
2 4 2 6 3
@ j ¼1 A
j 6¼ k
0 1 (12.6.19)
B X C
B N2B
C
¼lk + e B lj aj lk S C
2
C + higher order terms
@j ¼1 A
j 6¼ k
Obviously, Eq. (12.6.19) proves Eq. (12.6.12).

The Rayleigh’s quotient for the standard eigenvalue problem and the
properties resulting thereof can be extended to the generalized eigenvalue
problem as well. This becomes obvious if the generalized eigenvalue prob-
lem Ax ¼ lBx is transformed into the standard one A^ ^ x ¼ l^
x. Thus writing
Rayleigh’s quotient in the form
û
^T A^
u
^Þ ¼
rðu (12.6.20)
^ u
u T ^
and taking into account that (see Eqs. 12.5.58, 12.5.61, and 12.5.64)
^ ¼ S1 AST
^ ¼ ST u, A
ST S ¼ B, u
uT Au
rðuÞ ¼ (12.6.21)
uT Bu
@CivilMethod
Example 12.6.1 Determine the eigenfrequencies and eigenmodes of the two-

story building in Example 12.2.1 using the properties of Rayleigh’s quotient.
Solution
The mass and stiffness matrices of the building were computed in Example
12.2.1, that is,

25 0 3826:5 3826:5
M¼ , K¼ (1)
0 32 3826:5 9142:1
We take

1
u¼ (2)
x
where x is an unknown. We will determine x so that Rayleigh’s quotient takes
an extreme value. For this purpose, we formulate Rayleigh’s quotient,
Eq. (12.6.21), for A ¼ K and B ¼ M. Thus we obtain
uT Ku
rðx Þ ¼
uT Mu
3826:5 3826:5 1
½1 x
3826:5 9142:1 x
¼ (3)
25 0 1
½1 x
0 32 x
9142:1x 2 7653:0x + 3826:5
¼
32x 2 + 25
The value of x is obtained by establishing the extreme values of the function
rðx Þ. Thus we obtain
drðx Þ 244896x 2 + 212209x 191325
¼
dx ð32x 2 + 25Þ2 (4)
¼0
Eq. (4) has two roots: x1 ¼ 0:5511 kai x2 ¼ 1:4176. For these values, the
second derivative
d 2 rðx Þ 107 ð1:567x 3 + 2:037x 2 3:673x 0:530Þ
¼ (5)
dx 2 ð32x 2 + 25Þ3
gives
d 2 rðx 1 Þ d 2 rðx 2 Þ
¼ 399:98 > 0, ¼ 60:4469 < 0 (6)
dx 2 dx 2
Therefore, x1 yields the minimum and x2 the maximum. On the base of
Eq. (3), we obtain

1
rðx1 Þ ¼ l1 ¼ w21 ¼ 68:7089, w1 ¼ 8:289, u1 ¼ (7)
0:5511
@CivilMethod

1
rðx2 Þ ¼ l2 ¼ w22 ¼ 370:0416, w2 ¼ 19:236, u2 ¼ (8)
1:4127
Normalizing the eigenmodes with respect to the mass, we obtain

0:1697 0:1061
F¼ (9)
0:0935 0:1499
Obviously, the obtained eigenfrequencies and mode shapes are identical to
those obtained by the method in Section 12.2.
12.7 Properties of eigenfrequencies and modes of MDOF

systems without damping: A summary
The properties of the generalized eigenvalue problem are applied to the problem
of the free vibrations of MDOF systems if the matrices A and B in Eq. (12.5.54)
represent the stiffness and mass matrices, respectively, that is, A ¼ K, B ¼ M.
Then the eigenvalues express the squares of the eigenfrequencies, li ¼ w2i , and
the eigenvectors the normalized eigenmodes of the vibrations. In Table 12.7.1,
these properties are summarized using the terminology employed in free vibra-
tion analysis.
TABLE 12.7.1 Properties of the eigenfrequencies and eigenmodes in free

vibrations of MDOF systems without damping.
1. When K and M are real, symmetric, and positive definite, then all eigenfrequencies are
positive.
2. When K is singular, det ðKÞ ¼ 0, at least one eigenfrequency is zero and the corresponding
eigenmodes express rigid body motion while the elastic energy is positive semidefinite,
U ðuÞ ¼ 12 uT Ku 0 for u 6¼ 0.
3. When M is singular, det ðMÞ ¼ 0, at least one eigenfrequency is infinite and the kinetic
energy is positive semidefinite, T ðu Þ ¼ 12 u_ T Mu_ 0 for u_ 6¼ 0.
4. To each eigenfrequency there correspond as many linearly independent eigenvectors
as the multiplicity of the eigenfrequency.
5. The eigenmodes, including those corresponding to a repeated eigenfrequency, are
linearly independent and orthogonal with respect to M and K, that is
fTm Mfn ¼ 0, m 6¼ n
fTm Kfn ¼ 0, m 6¼ n
6. Any arbitrary vector u with dimension N can be expressed as a superposition of the
N eigenmodes
u ¼ Fa, a ¼ FT Mu
12.8 Solution of the vibration problem without damping

In Section 12.2, it was shown that the solution of Eq. (12.1.2) is given by
Eq. (12.2.25) or Eq. (12.2.26), which we rewrite here for convenience
@CivilMethod
X
N
uðt Þ ¼ fi ðci cos wi t + di sin wi t Þ (12.8.1)
i¼1
or
X
N
uðt Þ ¼ fi ai cos ðwi t qi Þ (12.8.2)
i¼1
in which wi and fi are the eigenfrequencies and eigenmodes, respectively,

obtained from the solution of the eigenvalue problem
Kf ¼ lMf (12.8.3)
The use of the eigenmodes fi instead of the eigenvectors bi is permitted
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
because the normalization coefficient mi ¼ 1= fTi Mfi can be incorporated
in the arbitrary constants ci, di or ai ,qi . The solution of the vibration problem
is accomplished when the arbitrary constants are determined. This is achieved
by using the 2N initial conditions specified by the displacement vector uð0Þ and
velocity vector u_ ð0Þ, each of dimension N , at t ¼ 0.
Eq. (12.8.1) for t ¼ 0 gives
uð0Þ ¼ f1 c1 + f2 c2 + ⋯ + fN cN (12.8.4)
or
uð0Þ ¼ Fc (12.8.5)
T
where F is the modal matrix and c ¼ fc1 , c2 , ⋯, cN g the vector of the
unknown coefficients ci. Eq. (12.8.5) provides a system of N linear algebraic
equations with respect to N unknowns, whose solution results in the vector c,
namely
c ¼ F1 uð0Þ (12.8.6)
The constants di are determine by differentiating Eq. (12.8.1) and using an
analogous procedure.
The foregoing solution requires the inversion of the matrix F, which is not a
simple task. This can be avoided if the orthogonality of the eigenmodes with
respect to the mass is employed. For this purpose both sides of Eq. (12.8.4)
are premultiplied by fTi M. Thus we obtain
fTi Muð0Þ ¼ fTi Mfi ci (12.8.7)
because the terms fTi Mfj ¼ 0, i 6¼ j, vanish. Therefore
fTi Muð0Þ
ci ¼ (12.8.8)
fTi Mfi
which is further written as
ci ¼ fTi Muð0Þ (12.8.9)
@CivilMethod
if the eigenmodes have been orthonormalized with respect to the mass, that is,
fTi Mfi ¼ 1 (see Eq. 12.2.29).
The same procedure is used to evaluate the coefficients di . Differentiating
Eq. (12.8.1) with respect to time yields
X
N
u_ ðt Þ ¼ fi wi ðci sin wi t + di cos wi t Þ (12.8.10)
i¼1
and for t ¼ 0
u_ ð0Þ ¼ f1 w1 d1 + f2 w2 d2 + ⋯ + fN wN dN (12.8.11)
Premultiplying the foregoing equation by fTi M yields
1 fTi Mu_ ð0Þ
di ¼ (12.8.12)
wi fTi Mfi
and if fTi Mfi ¼ 1
1 T
di ¼ f Mu_ ð0Þ (12.8.13)
wi i
The constants ai ,qi are evaluated from the relations
ai ¼ ci2 + di2 , qi ¼ tan 1 ðdi =ci Þ (12.8.14)
12.9 The method of mode superposition

The solution to the problem of free vibrations can be obtained in another way,
which is based on the expansion theorem. Thus according to Eq. (12.5.82) the
displacement vector can be represented as
uðt Þ ¼f1 Y1 ðt Þ + f2 Y2 ðt Þ + ⋯ + fN YN ðt Þ
(12.9.1)
¼FYðt Þ
The foregoing equation transforms the components of the vector
u ¼ f u1 u2 ⋯ uN gT referred to a system with base vectors e1 ð1, 0, …, 0Þ,
e2 ð0, 1, …, 0Þ, …,eN ð0, 0, …, 1Þ into the components Y ¼ f Y1 Y2 ⋯ YN gT
referred to the system with base vectors the eigenmodes f1 , f2 ,⋯,fN . The compo-
nents Yn , n ¼ 1, 2, …,N are generalized coordinates and are referred to as modal or
normal coordinates.

M f1 Y€ 1 + f2 Y€ 2 + ⋯ + fN Y€ N + Kðf1 Y1 + f2 Y2 + ⋯ + fN YN Þ ¼ 0
(12.9.2)
Then, premultiplying the foregoing equation by fTn
and taking into account
the orthogonality conditions fTn Mfj ¼ 0 and fTn Kfj ¼ 0 for n 6¼ j, we obtain
@CivilMethod
fTn Mfn Y€ n + fTn Kfn Yn ¼0, n ¼ 1, 2, …, N (12.9.3)

Setting further
Mn ¼ fTn Mfn (12.9.4)
Kn ¼ fTn Kfn (12.9.5)
Mn Y€ n + Kn Yn ¼0, n ¼ 1, 2, …,N (12.9.6)
The quantities Mn and Kn express the generalized mass and the generalized
stiffness of the nth eigenmode, respectively, and are called the generalized or
modal mass and generalized or modal stiffness. Further, taking into account that
Kfn ¼ w2n Mfn (see Eq. 12.3.3), we obtain
Kn ¼ w2n Mn (12.9.7)
and observing that Mn 6¼ 0, Eq. (12.9.6) becomes
Y€ n + w2n Yn ¼ 0, n ¼ 1, 2, …,N (12.9.8)
The transformation of the natural coordinates un to modal coordinates
reduces the system of the N coupled equations to a set N uncoupled equations
of a SDOF system. Hence, the solution of the nth equation is given by
Y_ n ð0Þ
Yn ¼ sin wn t + Yn ð0Þcoswn t, n ¼ 1, 2, …,N (12.9.9)
wn
The quantities Yn ð0Þ, Y_ n ð0Þ represent the initial conditions of the trans-
formed equations and can be obtained from the specified initial conditions
for the physical displacements, that is, uð0Þ, u_ ð0Þ, as follows.
Premultiplied Eq. (12.9.1) by fTn M yields
fTn Muðt Þ
Y n ðt Þ ¼ , n ¼ 1, 2, …,N (12.9.10)
Mn
and differentiating it with respect to t gives
f Mu_ ðt Þ
T
Y_ n ðt Þ ¼ n , n ¼ 1, 2, …,N (12.9.11)
Mn
For t ¼ 0, Eqs. (12.9.10) and (12.9.11) give the sought initial conditions
fTn Muð0Þ
Y n ð 0Þ ¼ , n ¼ 1, 2, …,N (12.9.12)
Mn
f Mu_ ð0Þ
T
Y_ n ð0Þ ¼ n , n ¼ 1, 2, …,N (12.9.13)
Mn
@CivilMethod
The foregoing relations are further simplified if the eigenmodes are normal-
ized on the basis of Eq. (12.2.29). This gives
Mn ¼ fTn Μfn ¼ 1 (12.9.14)
Kn ¼ w2n (12.9.15)
that is, the modal masses become equal to the unity and the modal stiffnesses
equal to the square of the corresponding eigenfrequency. Thus, we have
Yn ð0Þ ¼ fTn Muð0Þ, n ¼ 1, 2, …,N (12.9.16)
Y_ n ð0Þ ¼ fTn Mu_ ð0Þ, n ¼ 1, 2, …,N (12.9.17)
The vector un ¼ fn Yn , n ¼ 1, 2, …,N expresses the contribution of the nth
eigenmode to the displacement vector u and is referred to as the nth modal
component of the displacement vector. Hence, Eq. (12.9.1) is written as
u ¼ u 1 + u2 + ⋯ + u N (12.9.18)
which implies that the displacement vector is the superposition of the modal
components or, in other words, the displacement vector is the superposition
of the eigenmodes multiplied by a time-dependent weight coefficient.
This method of solving the equation of motion for the free vibration problem
is known as the modal superposition method or the method of the superposition
of the eigenmodes. It is used not only for free undamped vibrations but also for
damped vibrations as well as for forced vibrations. The superposition is
illustrated in Fig. 12.9.1, where the mode shapes correspond to those of the
two-story shear frame in Example 12.2.1.
FIG. 12.9.1 Modal superposition.
Example 12.9.1 Determine the displacements of the frame in Example 12.2.1

using the modal superposition method to solve the equation of motion. Assumed
initial conditions uð0Þ ¼ f0:03 0:01gT , u_ ð0Þ ¼ 0.
Solution
In Example 12.2.1, it was found

25 0 3826:5 3826:5
M¼ , K¼ (1)
0 32 3826:5 9142:1
@CivilMethod
w1 ¼ 8:289, w2 ¼ 19:236

0:1697 0:1061
f1 ¼ , f2 ¼ (2)
0:0935 0:1499
Because the eigenmodes are orthonormalized with respect to the mass, the
modal matrices resulting from Eq. (12.9.14) are
M1 ¼ fT1 Μf1 ¼ 1, M2 ¼ fT2 Μf2 ¼ 1 (3)
Applying Eqs. (12.9.12) and (12.9.13) gives

25 0 0:03
Y1 ð0Þ ¼ ½ 0:1697 0:0935 ¼ 0:15720 (4)
0 32 0:01

25 0 0:03
Y2 ð0Þ ¼ ½ 0:1061 0:1499 ¼ 0:0316 (5)
0 32 0:01
Y_ 1 ð0Þ ¼ Y_ 2 ð0Þ ¼ 0 (6)
Thus Eq. (12.9.9) gives
Y1 ¼ 0:15720cos 8:289t (7a)
Y2 ¼ 0:0316cos 19:236t (7b)
Then Eq. (12.9.1) gives the displacement vector

u 1 ðt Þ 2:6677 0:3353
¼ 102 cos 8:289t + 102 cos 19:236t
u 2 ðt Þ 1:4698 0:4700
(8)
Example 12.9.2 The loads P1 and P2 applied to the structure in Fig.E12.4a are
removed suddenly at t ¼ 0. Determine the ratio P1 =P2 so that the structure
vibrates in the form of the second eigenmode. The axial deformation of the col-
umn is neglected. Consider lumped mass assumption. Assume a ¼ 1, m ¼ 1,
E ¼ 2:1 107 kN=m2 , and column cross-section 0:20 0:20 m2 .
(a) (b)
FIG. E12.4 Structure in Example 12.9.2 (a); Parameters of motion (b).
@CivilMethod
Solution
The system has two degrees of freedom, the horizontal displacement u and the
rotation f at the top of the column shown in Fig. E12.4b. The equation of
motion can be obtained either by applying Eqs. (A.4.8) or by considering the
structure as a plane frame consisting of one element and including a rigid body
at its free end (see Section 11.11). Here, Eqs. (A.4.8) are employed. Thus,
setting yc ¼ 0, XP ¼ u, YP ¼ 0, w_ ¼ f,€ Fx ¼ fS , Fy ¼ 0, MP ¼ MS , IP ¼ IO ,
we obtain
m u€ + fS ¼ 0 (1)
Fy ¼ mx c f€ (2)
IO f€ + MS ¼ 0 (3)
The elastic forces are applied at point O and are given by the relations
fS ¼ k11 u + k12 f
MS ¼ k21 u + k22 f
Eqs. (1) and (3) are combined in matrix form
M€
u + Ku ¼ 0 (4)
where

m 0 k11 k12
M¼ , K¼
0 Io k21 k22
Note that Eq. (2) will be used to evaluate the unknown force Fy once u,f are
established from the solution of Eq. (4).
Stiffness matrix: The stiffness matrix will result from Eq. (11.5.2) for
u3 ¼ u4 ¼ 0. Thus, we have
2 3
12EI 6EI 2 3 3a 3
6 ð4a Þ3 ð4a Þ2 7 EI
6 7 6 16 8 7
K¼6 7¼ 4 5 (5)
4 6EI 4EI 5 a 3 3a 2
a
ð4a Þ2 4a 8
Mass matrix: We have
1
m ¼ ð4a Þm + ð3a Þ10m
¼ 32a m

2
Z 2a
Io ¼ 2 dx ¼ 30ma
10mx 3
a
hence

m 0 32 0
M¼
¼ ma (6)
0 Io 0 30a 2
@CivilMethod
Eigenvalues and eigenmodes: They are obtained from the eigenvalue

problem
4 =EI
ðK lMÞb ¼ 0, l ¼ w2 ma (7)
which yields
l1 ¼ 3:606, l2 ¼ 106:133 (8)

0:1654 0:0625
F¼ (9)
0:0645 0:1708
Initial conditions: There are only initial displacements, which are equal to
the static displacements due to the loads P1 , P2 . They are evaluated from the
relation

P1 0:7619P1 + 0:5714P2
uð0Þ ¼ K1 ¼ 102 (10)
2P2 a 0:2857P1 0:2857P2
In order that the structure vibrates in the form of the second eigenmode, it
must be Y1 ðt Þ 0, which holds if Y1 ð0Þ ¼ Y_ 1 ð0Þ ¼ 0. Inasmuch as u_ ð0Þ ¼ 0,
Eq. (12.9.17) gives Y_ 1 ð0Þ ¼ 0. If we take uð0Þ ¼ kf2 , with k being an arbitrary
constant, then Eq. (12.9.16) gives
Y1 ð0Þ ¼fT1 Muð0Þ
¼kfT1 Mf2 (11)
¼0
Therefore, it must be

2 0:7619P1 + 0:5714P2 0:0625
10 ¼k (12)
0:2857P1 0:2857P2 0:1708
P1 ¼ 212:1k, P1 ¼ 271:9k
P1 =P2 ¼ 0:7801 (13)
12.10 Solution of the vibration problem with damping

In the case of free vibrations with damping, the damping matrix C does not
vanish. Thus the equation that must be solved is
u + Cu_ + Ku ¼0
M€ (12.10.1)
subject to the specified initial conditions uð0Þ, u_ ð0Þ.
In the following, three methods are presented for the solution of this
problem.
@CivilMethod
12.10.1 Direct solution of the differential equation

We look for a solution of the form
u ¼ belt (12.10.2)
which is substituted into Eq. (12.10.1) to yield
2
l M + lC + K b ¼0 (12.10.3)
The foregoing equation represents a homogeneous system of N linear
algebraic equations, which can be solved to give a nontrivial vector b, provided
that the determinant of the coefficient matrix
SðlÞ ¼ l2 M + lC + K (12.10.4)
is zero. Namely

det l2 M + lC + K ¼ 0 (12.10.5)
Expansion of the determinant yields a real polynomial of 2N degree in l,
whose vanishing gives the characteristic equation of the differential equation
of motion Eq. (12.10.1), namely
PðlÞ ¼ a0 l2N + a1 l2N 1 + ⋯ + a2N ¼ 0 (12.10.6)
According to the fundamental theorem of algebra, Eq. (12.10.6) has 2N
roots l1 ,l2 ,…,l2N . Because the coefficients of the polynomial are real, the
roots are either real or complex conjugate pairs. Eq. (12.10.3) represents an
eigenvalue problem referred to as a quadratic eigenvalue problem. For each
value of ln a vector bn is obtained, which is real or complex depending on
whether ln is real or complex. Assuming that the roots are discrete, the general
solution of Eq. (12.10.1) is given by the superposition
X
2N
u¼ an bn eln t (12.10.7)
n¼1
where an are 2N arbitrary constants, which are determined from the initial
conditions.
The response of the system depends on the type of roots ln . We distinguish
the following cases:
(i) Real roots
Let ln be a real root, the corresponding solution (12.10.2) is
un ¼ bn eln t (12.10.8)
Obviously, for ln > 0 the solution diverges exponentially while for
ln < 0 it converges exponentially. In both cases, the motion of the system
is not an oscillation. If the root has a multiplicity k, then the solution
(12.10.2) becomes [3]
@CivilMethod

un ¼ bn a0 + a1 t + ⋯ + t k1 eln t (12.10.9)
which again diverges or converges exponentially depending on whether
ln > 0 or ln < 0, respectively, and the motion of the system is not an
oscillation.
(ii) Complex roots.
Let lm be complex, then its complex conjugate l m is also a root.
We set
m ¼ m iwm
lm ¼ mm + iwm and l (12.10.10)
m
The respective vectors bm and b m are also complex conjugate and their
contribution to the general solution is

m e lm t
um ¼am bm elm t + am∗ b

¼emm t am bm eiwm t + am∗ b m eiwm t (12.10.11)

¼emm t am bm + am∗ b m cos wm t + i am bm a ∗ b

m m sin wm t
The constants am , am∗ are arbitrary constants, hence they can be chosen so that
they are complex conjugates, that is
am ¼ cm + id m and am∗ ¼ cm id m (12.10.12)
where cm and dm are also arbitrary constants.
Further, we set
m ¼ pm iqm
bm ¼ pm + iqm and b (12.10.13)
Substituting Eqs. (12.10.12) and (12.10.13) into Eq. (12.10.11) gives
um ¼ 2emm t ½ðcm pm dm qm Þ cos wm t ðcm qm + dm pm Þsin wm t (12.10.14)
The factor within the square bracket expresses a harmonic vibration. Its
amplitude diverges exponentially if mm > 0 and converges exponentially if
mm < 0. Finally, if mm ¼ 0, the roots are imaginary and the amplitude of vibra-
tion remains constant. The vibration with mm > 0 is known as flutter or negative
damping. The solution is obtained as a superposition of the terms given by Eq.
(12.10.14), namely
X
N
u¼ um (12.10.15)
m¼1
Eq. (12.10.14) implies that the natural components of the displacements are
real, and as would be anticipated, the general solution is real.
@CivilMethod
Eq. (12.10.14) is further written

20 8 9 8 91 0 8 9 8 91 3
>
> p1m > > >
> p1m >> >
> q1m >> >
> p1m >>
6B > > >
> >
> >
> C B >
> >
> >
> >
> C 7
6B > > >
> >
> >
> C B > > >
> >
> >
> C 7
6B > >
> p >
>
>
>
>
> p >
>
> C B >
>
> q >
>
>
>
>
> p >
>
> C 7
6B > > 2m >
> >
> 2m >C
> B > > 2m >
> >
> 2m >C
> 7
6B > > > > C B > > > > C 7
6B > > >
> >
> >
> C B >
> >
> >
> >
> C 7
6B > < ⋮ > = >
< >
=C B > < ⋮ > = >
< ⋮ > =C 7
m t 6B C B C 7
um ¼2e m 6Bcm dm C cos wm t Bcm + dm C sin wm t 7
6B > > > > C B > > > > C 7
6B > >
> p >
>
>
>
>
> p >
>
> C B >
> q >
> >
> p >
> C 7
6B > >
im >
> >
>
im >
> C B > >
>
im >
>
>
>
>
>
im >
>
> C 7
6B > > >
> >
> >
> C B > > >
> >
> >
> C 7
6B > > ⋮ > > >
> >
> C B > > ⋮ >
> >
> ⋮ >
> C 7
6B > > > > C B > > > > C 7
4@ > >
>
>
>
>
>
>
>
>
>
> A @ > >
>
>
>
>
>
>
>
>
>
> A 5
>
: >
; >
: >
; >
: >
; >
: >
;
pNm pNm qNm pNm
8 9
>
> r1m cos ðwm t q1m Þ > >
>
> >
>
>
> >
>
>
> ð Þ >
>
>
> r 2m cos wm t q 2m >
>
>
> >
>
>
> >
>
>
< ⋮ >
=
¼2emm t
> >
>
>
> r cos ðwm t qim Þ > >
>
> im
> >
>
>
> >
>
>
> ⋮ >
>
>
> >
>
>
> >
>
>
: >
;
rNm cos ðwm t qNm Þ
(12.10.16)
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rim ¼ ðcm pim dm qim Þ2 + ðcm qim + dm pim Þ2 (12.10.17a)

1 ðcm qim + dm pim Þ
qin ¼ tan (12.10.17b)
cm pim dm qim
It becomes obvious from Eq. (12.10.17b) that the displacement components
in a vibration mode have different phase angles, contrary to the undamped
vibrations where the phase angles are the same within a vibration mode (see
Eq. 12.8.2).
in the presence of damping by direct solution of the equation of motion.
Assume:

30 5
C¼ , uð0Þ ¼ f0:03 0:01gT , u_ ð0Þ ¼ 0
5 40
Solution
The matrices M kai K were computed in Example 12.2.1. Hence, we have

25 0 3826:5 3826:5
M¼ , K¼ (1)
0 32 3826:5 9142:1
@CivilMethod
The characteristic equation results from Eq. (12.10.5)

25 0 30 5 3826:5 3826:5
det l2 +l + ¼0 (2)
0 32 5 40 3826:5 9142:1
which after expanding gives
l4 + 2:45l3 + 440:219375l2 + 486:3225l + 25425:17925 ¼ 0 (3)
The roots of the polynomial (3) are
8 9 8 9
>
> l1 >
> >
> 0:52764 + 8:27232i >>
< = < =
l2 0:52764 8:27232i
¼ (4)
>
> l > > 0:69736 + 19:22372i >
: 3> ; > : >
;
l4 0:69736 19:22372i

+0:87585 0:69134 103 i
b1 ¼
+0:48259 + 0:12547 102 i

+0:87585 + 0:69134 103 i
b2 ¼
+0:48259 0:12547 102 i

+0:57638 + 0:30308 102 i
b3 ¼ (5)
0:81719 + 0:21377 102 i

+0:57638 0:30308 102 i
b4 ¼
0:81719 0:21377 102 i
The general solution is given by Eq. (12.10.7) for N ¼ 2
X
4
u¼ an bn eln t (6)
n¼1
The previous equation for t ¼ 0 gives

X
4
uð0Þ ¼ an bn (7)
n¼1
X
4
u_ ð0Þ ¼ an ln bn (8)
n¼1
which are further written in matrix form

B uð0Þ
a¼ (9)
BL u_ ð0Þ
where B is the matrix of the vectors bn with dimensions 2 4, L the diagonal
matrix of ln with dimension 4 4, and a the vector of the coefficients an .
@CivilMethod
Eq. (9) for the data of the problem gives

8 9 8 9
>
> a1 > > +0:15232 101 0:99900 103 i >
< > = > < >
=
a2 +0:15232 101 + 0:99900 103 i
¼ (10)
>
> a > > +0:28787 102 0:93751 104 i >
: 3> ; > : >
;
a4 +0:28787 102 + 0:93751 104 i
Finally, applying Eq. (12.10.15) for N ¼ 1, 2, and taking into account
Eq. (12.10.14) give

u1 0:52764t +0:26681 101
¼e cos 8:27232t
u2 +0:14704 101

+0:17710 102
+ sin 8:27232t
+0:92599 103

+0:33190 102
+e0:69736t cos 19:22372t
0:47044 102

+0:90624 104
+ sin 19:22372t
0:16553 103
12.10.2 Linearization of the quadratic eigenvalue problem

As mentioned in the previous section, Eq. (12.10.3) represents the second-order
eigenvalue problem, commonly known as the quadratic eigenvalue problem.
The 2N values ln , which make the determinant SðlÞ equal to zero, are its eigen-
values. Moreover, the 2N vectors bn obtained from Eq. (12.10.3) are its eigen-
vectors. In the previous section, we presented a direct method for determining
these quantities. Another method is to transform the quadratic eigenvalue prob-
lem to a standard eigenvalue problem. This method is of great interest because it
allows methods applied to undamped vibrations to be applied also to damped
vibrations. It is based on the transformation of the equation of motion to two
first-order equations. To this end, we set
u_ ¼ x1 and u ¼ x2 (12.10.18)
hence
x_ 2 ¼ x1 (12.10.19)
or premultiplying by M
Mx_ 2 Mx1 ¼ 0 (12.10.20)
Moreover, Eq. (12.10.1) is written as
Mx_ 1 + Cx_ 2 + Kx2 ¼0 (12.10.21)
@CivilMethod
Eqs. (12.10.20) and (12.10.21) are combined to

0 M x_ 1 M 0 x1 0
+ ¼ (12.10.22)
M C x_ 2 0 K x2 0
or
^ x_ + Kx
M ^ ¼0 (12.10.23)
where

^¼ 0 M ^ M 0 x1 u_
M , K¼ , x¼ ¼ (12.10.24)
M C 0 K x2 u
Eq. (12.10.23) accepts a solution of the form
x ¼ belt (12.10.25)
which is substituted into Eq. (12.10.23) to give

^ lM
K ^ b¼0 (12.10.26)
namely, the quadratic eigenvalue problem of order N is transformed into a gen-
eralized linear eigenvalue problem of order 2N . The matrices M ^ and K ^ are real
and symmetric. Thus, according to the first property in Table 12.7.1, the eigen-
value problem in Eq. (12.10.26) has real eigenvalues but not positive because
these matrices are not positive definite. The transformation to a standard eigen-
value problem using the method presented in Section 12.5.3 produces a matrix
^ in Eq. (12.5.64) that is not real. Therefore, if the matrix M
A ^ is not singular, it is
convenient to transform Eq. (12.10.26) as

A^ lI b ¼ 0 (12.10.27)
where
A ^ 1 K
^ ¼M ^ (12.10.28)
Eq. (12.10.27) gives 2N eigenvalues and the respective eigenvectors bn that
^ and K
satisfy the orthogonality condition with respect to M ^ (see Section 12.5.2,
property 2). The eigenvalues of the problems we encounter are complex with a
negative real part and express convergent vibrations.
^ we will have
If the eigenvectors are orthonormalized with respect to M,
^ n ¼1
fTn Mf (12.10.29)
^ n ¼ ln
fTn Kf (12.10.30)
The general solution of Eq. (12.10.23) is obtained as a superposition of the
solutions of Eq. (12.10.25), namely
X
2N
x¼ an fn eln t (12.10.31)
n¼1
@CivilMethod
where an are arbitrary constants determined from the following initial

conditions

u_ ð0Þ
xð0Þ ¼ (12.10.32)
uð0Þ
Eq. (12.10.31) for t ¼ 0 gives
X
2N
xð0Þ ¼ an fn (12.10.33)
n¼1
^ and by using Eq. (12.10.29) becomes

which if premultiplied by fTn M
^ ð 0Þ
an ¼ fTn Mx (12.10.34)
The coefficients an , the eigenvalues ln , and the eigenvectors fn are
complex conjugate. Hence we may set
an ¼ cn + id n , an ¼ cn id n (12.10.35)
ln ¼ mn + iwn , n ¼ m iwn
l (12.10.36)
n
fn ¼ pn + iqn , n ¼ pn iqn
f (12.10.37)
Apparently, for each pair of eigensolutions, we will have
n
xm ¼ xn + x
(12.10.38)
¼ 2emn t ½ðcn pn dn qn Þcos wn t ðcn qn + dn pn Þ sin wn t
and the general solution is obtained as
X
N
x¼ xm (12.10.39)
m¼1
The solution of Eq. (12.10.23) may result by considering the transformation

of xðt Þ in terms of the modal coordinates Yðt Þ
x ¼ FY (12.10.40)
If L represents the matrix of the eigenvalues and F the matrix of the
^ we have
eigenvectors orthonormalized with respect to M,
^ ¼I
FT MF (12.10.41)
^ ¼L
FT KF (12.10.42)
^ Y_ + KFY
MF ^ ¼0 (12.10.43)
which is further premultiplied by F and gives by virtue of Eqs. (12.10.41) and
T
(12.10.42)
Y_ LY ¼ 0 (12.10.44)
@CivilMethod
or
Y_ n ln Yn ¼ 0, n ¼ 1, 2, …,2N (12.10.45)
that is, the system of the 2N coupled equations is transformed to 2N
SDOF equations with respect to the modal coordinates. Integration of
Eq. (12.10.45) gives
Yn ¼ an elt , n ¼ 1, 2, …,2N (12.10.46)
Thus the general solution is given by Eq. (12.10.40). The vector
a ¼ fa1 , a2 , …, a2N gT of the arbitrary constants is obtained from the relation
^ ð 0Þ
a ¼ FT Mx (12.10.47)
In summary, the solution of the vibration problem with damping presented
in this section can be obtained by adhering to the following steps:
1. Formulation of the matrices M^ and K
^ defined by Eq. (12.10.24).
2. Computation of the eigenvalues and eigenvectors of the eigenvalue
problem, Eq. (12.10.26).
^
3. Orthonormalization of the eigenvectors with respect to M.
4. Computation of the arbitrary constants using Eq. (12.10.34).
5. Computation of the solution using Eq. (12.10.31).
6. Separation of the vectors u and u_
X
N X
N
u_ ¼ an fðn1Þ eln t , u¼ an fðn2Þ eln t (12.10.48)
n¼1 n¼1
where fðn1Þ and fðn2Þ are the upper and lower half of fn , respectively.
12.10.3 The use of a proportional viscous damping matrix

It becomes obvious from the previous example that the solution to the problem
of free vibrations with viscous damping through directly solving the quadratic
eigenvalue problem exhibits significant computational difficulties. On the other
hand, transforming it to a linear eigenvalue problem increases the number of
computations in addition to the complexity of handling complex numbers.
For these reasons, the previously discussed methods are rather theoretical
and, except in special cases, they do not represent the common praxis in the
dynamic analysis of structures. In practice, the modal damping assumption is
adopted, according to which the damping affects separately each eigenmode
by a specified damping coefficient. This is, of course, true when the modes
of the free vibration without damping are orthogonal with respect to the
damping matrix, that is,

Cn if n ¼ m
fn Cfm ¼
T
(12.10.49)
0 if n 6¼ m
@CivilMethod
Using Eq. (12.9.1), which represents the transformation from the physical
coordinates to modal coordinates of the undamped vibrations, we may write
Eq. (12.10.1) as
€ + CFY_ + KFY ¼0
MFY (12.10.50)
which is premultiplied by F to yield T
€ + FT CFY_ + FT KFY ¼0
FT MFY (12.10.51)
Taking into account the orthogonality of the eigenmodes with respect to M,
K, and assuming that Eq. (12.10.49) holds, we obtain
Mn Y€ n + Cn Y_ n + Kn Yn ¼ 0, n ¼ 1, 2, …,N (12.10.52)
namely, the uncoupling of equations of motion is achieved when damping is
present.
Further, setting
Cn ¼ 2x n Mn wn (12.10.53)
and taking into account that Kn ¼ Mn w2n Eq. (12.10.52) becomes
Y€ n + 2x n wn Y_ n + w2n Yn ¼ 0, n ¼ 1, 2, …, N (12.10.54)
whose solution is given by Eq. (2.3.18), namely
_
Y n ð0Þ + Yn ð0Þxn wn
Yn ðt Þ ¼ exn wn t sin wDn t + Yn ð0Þcos wDn t (12.10.55)
wDn
The quantity xn defined by Eq. (12.10.53) expresses the damping ratio of the
n-eigenmode. The initial conditions Yn ð0Þ, Y_ n ð0Þ are evaluated from the vec-
tors uð0Þ and u_ ð0Þ using Eqs. (12.9.12) and (12.9.13). After determining
Yn ðt Þ,n ¼ 1, 2, …,N , the solution of Eq. (12.10.1) is obtained from
X
N
uðt Þ ¼ f n Yn ð t Þ (12.10.56)
n¼1
This solution is referred to as the mode superposition method with modal

damping.
Eq. (12.10.55) by virtue of Eq. (2.3.19) is written as
Yn ðt Þ ¼ rn exn wn t cos ðwDn t qn Þ (12.10.57)
where
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
_ 2
Y n ð0Þ + Yn ð0Þx n wn
rn ¼ + ½Yn ð0Þ2 (12.10.58a)
wDn
_
Y n ð0Þ + Yn ð0Þxn wn
qn ¼ tan 1 (12.10.58b)
wD Y ð 0Þ
@CivilMethod
From the last equation, one concludes that the displacements in each mode
are in phase.
The above solution resulted by assuming that the transformation u ¼ FY
diagonalizes also the damping matrix, that is,
^
FT CF ¼ C (12.10.59)
where
2 3
2x1 w1 0 ⋯ 0 ⋯ 0
6⋯ 2x2 w2 ⋯ 0 ⋯ 0 7
6 7
6⋯ ⋯ ⋯ ⋯ ⋯ ⋯ 7
^ ¼6
C 6
7
7 (12.10.60)
6⋯ ⋯ ⋯ 2x n wn ⋯ ⋯ 7
6 7
4⋯ ⋯ ⋯ ⋯ ⋯ ⋯ 5
0 0 ⋯ 0 ⋯ 2xN wN
The damping matrix satisfying Eq. (12.10.49) is called proportional.
The damping that satisfies the orthogonality condition of the eigenmodes of
the undamped free vibrations is referred to in the literature as classical damping
versus nonclassical damping where this condition is not satisfied. Methods of
constructing proportional damping matrices will be presented in the next
section.
in the presence of damping by the method of linearization of the quadratic
eigenvalue problem. Assume:

30 5
C¼ , uð0Þ ¼ f0:03 0:01gT , u_ ð0Þ ¼ 0
5 40
Solution
The matrices M and K were computed in Example 12.2.1. Thus we have

25 0 3826:5 3826:5
M¼ , K¼ (1)
0 32 3826:5 9142:1
Eq. (12.10.24) gives
2 3
0 0 25 0
0 M 6 0 0 0 32 7
M^¼ ¼6 7 (2)
M C 4 25 0 30 5 5
0 32 5 40
2 3
25 0 0 0
6 7
^ ¼ M 0 ¼ 6 0 32 0
K
0 7 (3)
0 K 4 0 0 3826:5 3826:5 5
0 0 3826:5 9142:1
@CivilMethod
The characteristic equation is

02 3 2 31
25 0 0 0 0 0 25 0
B6
6 0 32 0
7 6 0 0 32 7 C
det K ^ ¼B
^ lM B6
0 7 6
7 l6
0 7C
7C ¼ 0
@4 0 0 3826:5 3826:5 5 4 25 0 30 5 5A
0 0 3826:5 9142:1 0 32 5 40
(4)
which after expanding gives
l4 + 2:45l3 + 440:219375l2 + 486:3225l + 25425:17925 ¼ 0 (5)

We observe that the characteristic polynomial (5) is identical to that in
Example 12.10.1. Hence, it gives the same eigenvalues c
8 9 8 9
>
> l1 >> >
> 0:52764 + 8:27232i >>
<l >
> = > < 0:52764 8:27232i > =
2
¼ (6)
>
> l3 >> >
> 0:69736 + 19:22372i >>
> ;
: > : > >
;
l4 0:69736 19:22372i
The eigenvectors are computed from Eq. (12.10.26) and are then normalized
^ Thus
with respect to mass M.
For l1 ¼ 0:52764 + 8:27232i we obtain
8 9
>
> +0:259428 0:228755i > >
< +0:143371 0:125558i >
> =
f1 ¼ (7)
>
> 0:029533 0:029477i > >
>
: >
;
0:016218 0:016297i
For l2 ¼ 0:52764 8:27232i we obtain
8 9
>
> +0:259428 + 0:228755i >
>
< +0:143371 + 0:125558i >
> =
f2 ¼ (8)
>
> 0:029533 + 0:029477i >
>
>
: >
;
0:016218 + 0:016297i
For l3 ¼ 0:69736 + 19:22372i we obtain
8 9
>
> +0:241638 0:222178i >
>
>
< >
0:340096 + 0:317685i =
f3 ¼ (9)
>
> 0:011998 0:012135i >
>
>
: >
;
+0:017145 + 0:017070
^ K
c. The matrices M, C,K are positive definite. However, the matrices M, ^ are not positive definite.
This justifies that the eigenvalue problem has complex eigenvalues and complex eigenvectors.
@CivilMethod
For l4 ¼ 0:69736 19:22372i we obtain

8 9
>
> +0:241638 + 0:222178i >
>
< 0:340096 0:317685i >
> =
f4 ¼ (10)
>
> 0:011998 + 0:012135i >
>
>
: >
;
+0:017145 0:017070
The vector xð0Þ representing the initial conditions is computed from
Eq. (12.10.32)
8 9
>>
<
0 > >
=
u_ ð0Þ 0
x ð 0Þ ¼ ¼ (11)
uð0Þ >
> 0:03 >
>
: ;
0:01
and Eq. (12.10.47) gives the coefficients an
8 9
>
> 0:211292 + 0:240875i >>
< 0:211292 0:240875i >
> =
T ^
a ¼ F Mxð0Þ ¼ (12)
>
> 0:066486 + 0:071021i >>
>
: >
;
0:066486 0:071021i
Finally, Eqs. (12.10.48) give
8 9
<u = X
2N
1 ð2Þ
¼ an fn eln t
:u ;
2 n¼1
08 9 8 9 1
< +0:26681 101 = < +0:17710 102 =
¼e0:52764t @ cos 8:27232t + sin 8:27232t A
: +0:14704 101 ; : +0:92599 103 ;
08 9 8 9 1
< +0:33190 102 = < +0:90624 104 =
+e 0:69736t @ cos 19:22372t + sin 19:22372t A
: 0:47044 102 ; : 0:16553 103 ;
8 9
< u_ = X
2N
1 ð1Þ
¼ an f n e ln t
: u_ ;
2 n¼1
08 9 8 9 1
< +0:57263 103 = < 0:22165 =
¼e0:52764t @ cos 8:27232t + sin 8:27232t A
: 0:98528 104 ; : 0:12213 ;
08 9 8 9 1
< 0:57241 103 = < 0:63866 101 =
+e 0:69736t @ cos 19:22372t + sin 19:22372t A
: +0:98450 104 ; : +0:90551 101 ;
Note that the obtained solution by this method is identical to that obtained in
Example 12.10.1.
in the presence of classical damping. Assume:
@CivilMethod

30 5
C¼ , uð0Þ ¼ f0:03 0:01gT , u_ ð0Þ ¼ 0
5 40
Solution
The eigenfrequencies and mode shapes were computed in Example 12.2.1,
namely

w1 8:289 0:1697 0:1061
¼ , F¼ (1)
w2 19:236 0:0935 0:1499
The transformation (12.10.49) gives
T
0:1697 0:1061 30 5 0:1697 0:1061
FT CF ¼
0:0935 0:1499 5 40 0:0935 0:1499

1:05496 0:05712
¼ (2)
0:05712 1:39556
The orthogonality assumption with respect to the damping matrix requires

that the off-diagonal terms are omitted. This is permitted here because these
terms are small compared to the diagonal ones. Thus, taking into account that
M1 ¼ M2 ¼ 1, we can set
^ 1 ¼ 2x w1 ¼ 1:05496
C (3a)
1
^ 2 ¼ 2x w2 ¼ 1:39556
C (3b)
2
which yield
x1 ¼ 0:0636, wD1 ¼ w1 1 x21 ¼ 8:272 (4a)
x2 ¼ 0:0363, wD2 ¼ w2 1 x22 ¼ 19:223 (4b)
Eqs. (12.9.16) and (12.9.17) give

Y1 ð0Þ ¼ fT1 Muð0Þ ¼ 0:157195 (5a)
Y2 ð0Þ ¼ fT2 Muð0Þ ¼ 0:031607 (5b)
Y_ 1 ð0Þ ¼ Y_ 2 ð0Þ ¼ 0 (6)
Y1 ðt Þ ¼ e0:52748t ð0:010018 sin 8:272t + 0:157195 cos 8:272t Þ (7a)
Y2 ðt Þ ¼ e0:69778t ð0:001148 sin 19:223t + 0:031607 cos 19:223t Þ (7b)
@CivilMethod
Finally, we obtain from Eq. (12.9.1)

( )
u1 X
2
¼ f n Yn
u2 n¼1
( ) ( ) !
0:52748t
0:02667 0:1697 102
¼e cos 8:272t + sin 8:272t
0:01469 0:9366 103
( ) ( ) !
0:69778t
0:3353 102 1:218 104
+e cos 19:223t + sin 19:223t
0:4737 102 0:172 103
(8)
Comparing the above solution with that obtained in Example 12.10.1, we
conclude that omitting the diagonal terms in Eq. (2) negligibly influences the
solution.
12.11 Construction of a proportional damping matrix

12.11.1 Rayleigh damping
When the solution is obtained by the method of superposition of eigenmodes
and the modal damping ratios are known (see Eq. 12.10.54), it is obvious that
the knowledge of a proportional damping matrix is not required. However,
when other methods for solving the equation of motion are employed, for exam-
ple, numerical integration methods, the knowledge of the damping matrix is
necessary. Thus, the problem of constructing the damping matrix arises when
all or some of the modal damping ratios are specified. This problem will be the
subject of this section.
It is obvious that the damping matrix satisfies Eq. (12.10.49) if it is propor-
tional to the mass or stiffness of the structure.
For a mass-proportional damping matrix it is
C ¼ a0 M (12.11.1)
where a0 is a constant.
Premultiplying the foregoing relation by fTn and postmultiplying it by fm
yield

a0 Mn if n ¼ m
fTn Cfm ¼ a0 fTn Mfm ¼ (12.11.2)
0 if n 6¼ m
Hence
Cn ¼ a0 Mn (12.11.3)
a0 ¼ 2x n wn (12.11.4)
@CivilMethod
The foregoing relation allows the determination of a0 if xn is specified. After

that, the damping ratios xk of the remaining eigenmodes (k 6¼ n) can be estab-
lished from the relation
a0
xk ¼
2wk , k 6¼ n
wn (12.11.5)
¼ xn
wk
For a stiffness-proportional damping matrix, we may set
C ¼ a1 K (12.11.6)
Following the previous procedure we obtain

a1 w2n Mn if n ¼ m
fTn Cfm ¼ a1 fTn Kfm ¼ (12.11.7)
0 if n 6¼ m
Hence
a1 w2n Mn ¼ Cn
(12.11.8)
¼ 2xn Mn wn
which yields
2xn
a1 ¼ (12.11.9)
wn
The foregoing relation allows the determination of a1 if xn is specified.
After that, the damping ratios x k of the remaining eigenmodes (k 6¼ n) can be
established from the relation
a 1 wk
xk ¼
2 , k 6¼ n
wk (12.11.10)
¼ xn
wn
Obviously, the linear combination
C ¼ a0 M + a1 K (12.11.11)
satisfies also the orthogonality condition (12.10.49). Inasmuch as this relation
involves two arbitrary constants, a0 and a1 , it is possible to establish matrix C
so that two of its eigenmodes have specified damping ratios. Indeed,
Eq. (12.11.11) gives
fTn Cfn ¼ 2x n Mn wn ¼ a0 Mn + a1 Mn w2n (12.11.12a)
fTm Cfm ¼ 2x m Mm wm ¼ a0 Mm + a1 Mm w2m (12.11.12b)
which provide the linear system of equations
@CivilMethod
2 1 3
w
1 6 wn n
7 a0 xn
4 1 5 ¼ (12.11.13)
2 a1 xm
wm
wm
from which the coefficients a0 and a1 are obtained
2wn wm ðwn xm wm xn Þ
a0 ¼ (12.11.14a)
w2n w2m
2 ð wn x n wm x m Þ
a1 ¼ (12.11.14b)
w2n w2m
If xn ¼ xm ¼ x, which is a reasonable assumption, the foregoing relations
yield
2xwn wm
a0 ¼ (12.11.15a)
wn + wm
2x
a1 ¼ (12.11.15b)
wn + wm
Once the coefficients a0 and a1 are established, the damping ratios of the
remaining eigenmodes (k 6¼ m, n) can be evaluated from the relation

1 1
xk ¼ a0 + a1 wk , k 6¼ m,n (12.11.16)
2 wk
xk = a0 /2wk
xk = xk (wk )
xk = (a0 /wk + a1wk)/2
xk = a1wk /2
wk
FIG. 12.11.1 Damping ratio versus frequency for Rayleigh damping.
@CivilMethod
The proportional damping determined by Eq. (12.11.11) is known as Ray-

leigh damping. Fig. 12.11.1 shows the variation of xk ðwk Þ, as obtained from
Eqs. (12.11.5), (12.11.10) and (12.11.16).
Example 12.11.1 Construct a proportional damping matrix of the frame in
Example 12.2.1, so that x1 ¼ 0:08 and x2 ¼ 0:06.
Solution
The mass and stiffness matrices as well as the natural frequencies were
computed in Example 12.2.1, namely

25 0 3826:5 3826:5 w1 8:289
M¼ , K¼ , ¼ (1)
0 32 3826:5 9142:1 w2 19:236
Using these data, Eq. (12.12.13) becomes
2 1 3
8:289
1 6 8:289 7 a0 0:08
4 5 ¼ (2)
2 1 a1 0:06
19:236
19:236
which yields a0 ¼ 1:1023 and a1 ¼ 0:003259. Then using Eq. (12.12.11) we
obtain

25 0 3826:5 3826:5
C ¼ 1:1023 + 0:003259
0 32 3826:5 9142:1
(3)
40:029 12:4717
¼
12:4717 65:0705
12.11.2 Additional orthogonality conditions: Caughey

damping matrix
Rayleigh damping enables us to construct a proportional damping matrix by
specifying the damping ratio of only two eigenmodes. However, if we are inter-
ested in constructing a proportional damping matrix by specifying the damping
ratio of more than two eigenmodes, we must use matrices beyond the mass and
stiffness matrices that satisfy the orthogonality condition. This is achieved by
formulating orthogonality conditions more than those defined by Eqs. (12.3.10)
and (12.3.11). These conditions are called additional orthogonality conditions
and can be derived as follows [4–6].
We assume that the eigenmodes fn have been orthonormalized with respect
to the mass, hence Mn ¼ 1. Starting from the relation
Kfn ¼ ln Mfn (12.11.17)
where ln ¼ w2n , and assuming that M is not singular, the multiplication of both
sides of the foregoing equation by fTm KM1 yields
@CivilMethod
fTm KM1 Kfn ¼ ln fTm Kfn

0 n 6¼ m (12.11.18)
¼
l2n n ¼ m
Further, multiplying Eq. (12.11.17) by fTm KM1 KM1 gives
fTm KM1 KM1 Kfn ¼ ln fTm KM1 Kfn
which by virtue of Eq. (12.11.18) yields

2 0 n 6¼ m
fTm KM1 Kfn ¼ (12.11.19)
l3n n ¼ m
Repeating the previous procedure results in the orthogonality relations

p 0 n 6¼ m p ¼ 1, 2, …1
fTm KM1 Kfn ¼ (12.11.20)
lpn + 1 n ¼ m p ¼ 1, 2, …1
Observing that MM1 ¼ I, we obtain
p p
KM1 K ¼ MM1 KM1 K

¼ MM1 KM1 KM1 ⋯ KM1 K
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
p

¼ M M K M1 K ⋯ M1 K
1
p+1
1
p + 1
¼M M K (12.11.21)

1
p + 1 0 n 6¼ m p ¼ 1, 2, …1
fTm M M K fn ¼ (12.11.22)
lpn + 1 n ¼ m p ¼ 1, 2, …1
Because the above relation is valid for p ¼ 1 we can write it as

1 q 0 n 6¼ m q ¼ 0, 1, 2, …1
fm M M K fn ¼
T
(12.11.23)
lqn n ¼ m q ¼ 0, 1, 2, …1
If matrix K is not singular, it holds ln 6¼ 0, n ¼ 1, 2, …,N and Eq.
(12.11.17) is written as
K1 Mfn ¼ l1
n fn (12.11.24)
or
1 1
M K fn ¼ l1
n fn (12.11.25)
Multiplication of both sides of the foregoing equation by fTm M yields
1
fTm M M1 K fn ¼ l1 n fm Mfn
T

0 n 6¼ m (12.11.26)
¼ 1
ln n ¼ m
@CivilMethod
1
Further, multiplying Eq. (12.11.25) by fTm M M1 K yields
2 1
fTm M M1 K fn ¼ l1 n fm M M K
T 1
fn (12.11.27)

2 0 n 6¼ m
fTm M M1 K fn ¼ (12.11.28)
l2
n n¼m
Repeating the previous procedure results in the orthogonality relations

1 q 0 n 6¼ m q ¼ 1, 2, …1
fm M M K fn ¼
T
(12.11.29)
lq
n n ¼ m q ¼ 1, 2, …1
Eqs. (12.11.22) and (12.11.29) are combined to

q 0 n 6¼ m q ¼ 0, 1, 2, … 1
fTm M M1 K fn ¼ (12.11.30)
lqn n ¼ m q ¼ 0, 1, 2, … 1
Therefore, the matrices defined by the relation
q
Cq ¼ M M1 K , q ¼ 0, 1, 2, … 1 (12.11.31)
satisfy the orthogonality condition. Evidently, any linear combination of k N
terms
X
k X q
C¼ aq Cq ¼ aq M M1 K (12.11.32)
q q
can be employed as a proportional damping matrix with k specified modal

damping ratios. Eq. (12.11.32) is known as the Caughey series and the damping
matrix C obtained from this equation is known as the Caughey damping
matrix [7]. Obviously, for k ¼ 2 and q ¼ 0,1 the Rayleigh damping matrix is
obtained. The summation index k may be arbitrary. Let us assume that
q ¼ 0, 1, 2, …, k 1. Then
X
k1 q
C¼ aq M M1 K (12.11.33)
q¼0
The coefficients aq are evaluated from the relation

X
k 1 q
fTn Cfn ¼ aq fTn M M1 K fn (12.11.34)
q¼0
which by virtue of Eqs. (12.10.53) and (12.11.30) becomes

X
k 1
2xn wn ¼ aq lqn (12.11.35a)
q¼0
@CivilMethod
or because ln ¼ w2n

1 1
xn ¼ a0 + a1 wn + a2 wn + ⋯ + ak1 wn
3 2k3
(12.11.35b)
2 wn
Applying the foregoing equation for n ¼ 1, 2,…,k provides the system of
linear equations
2 3
1
6 w1 w1 w31 ⋯ w2k3
1 78 9 8 9
6 7 > a0 > > x 1 >
61 7> > > >
16 w2 w32 ⋯ w2k3 7 < a1 = < x 2 =
6 7
26 w 2
7> ⋮ > ¼ > ⋮ > (12.11.36)
6 ⋯2 ⋯ ⋯ ⋯ ⋯ 7 > > > >
6 7: ak1 ; : xk ;
41 5
wk w3k ⋯ w2k3
k
wk
which yields the coefficients ak when xk is specified.
12.11.3 Construction of the proportional damping matrix

using the modal matrix
An alternative way of constructing a proportional damping matrix follows
directly from the orthogonality relation (12.10.59) if it is premultiplied by
FT and postmultiplied by F1 . Thus, we obtain
^ 1
C ¼ FT CF (12.11.37)
The foregoing formulation requires the inversion of the modal matrix. This
can be achieved using the relation
FT MF ¼I (12.11.38)
which gives
FT ¼MF (12.11.39a)
and
F1 ¼FT M (12.11.39b)
The previous equations are substituted into Eq. (12.11.37) to yield
^ TM
C ¼ MFCF (12.11.40)
We can further write
2 3
2x 1 w1 ⋯ 0 ⋯ 0
6 7
6⋯ ⋯ ⋯ ⋯ ⋯ 7
6 7h i
6 7 T
^ T ¼ f1 ⋯ fn ⋯ fN 6 ⋯
FCF ⋯ 2x n wn ⋯ ⋯ 7 f ⋯ fT ⋯ fT
6 7 1 n N
6 7
6⋯ ⋯ ⋯ ⋯ ⋯ 7
4 5
0 ⋯ 0 ⋯ 2x N wN
X
N
¼ 2x n wn fn fT
n
n¼1
@CivilMethod

!
X
N
C¼M 2xn wn fn fTn M (12.11.41)
n¼1
Eq. (12.11.41) shows clearly the contribution of each eigenmode to the pro-
portional damping matrix.
When using a proportional damping constructed by the previous methods,
we should take into account the following:
1. The damping matrix C obtained from Eq. (12.11.41) is in general complete.
The same holds for the damping matrix obtained using the additional
orthogonality conditions for k > 2. Apparently, the use of a complete damp-
ing matrix increases the computational cost.
2. The linear system of Eq. (12.11.36) exhibit ill conditioning because the
coefficients 1=wn ,wn ,w3n ,⋯,w2k3
n may differ significantly. As a remedy,
it is recommended that the values of series exponents in Eq. (12.11.33)
are taken near zero. For example, it is advisable to take q ¼ 2, 1,
0, 1, 2, when k ¼ 5.
3. In Rayleigh or Caughey damping, it is possible to obtain negative values for
x n . These values should be excluded; otherwise, the response of the struc-
ture will be divergent.
4. It is inappropriate when the system consists of parts with significantly dif-
ferent damping, for example, in the case of soil-structure interaction where
soil damping is much greater than that of the structure. The same is true
when energy-absorbing dampers are incorporated into the structure or
base isolation systems are used. The resulting damping matrix does not
satisfy the orthogonality condition; therefore, the method presented in
Section 12.10.2 should be employed. Another method to cope with this
problem is to use the method of substructures with a different proportional
damping matrix for each of them.
5. The use of Rayleigh damping, which offers a reasonable solution to the
damping problem, yields increased values of the damping ratios for
higher-order eigenmodes.
Example 12.11.2 The computed mass and stiffness matrices of a single-story

building are
2 3 2 3
52 0 0 1472 407 5553
M ¼ 4 0 52 0 5, K ¼ 4 407 1001 4154 5
0 0 441 5553 4154 43516
Determine a proportional damping matrix for the structure using (i) addi-
tional orthogonality conditions, (ii) the alternative method based on the modal
matrix, and (iii) Rayleigh damping. Assume: x 1 ¼ 0:06, x2 ¼ 0:08, x 3 ¼ 0:10
@CivilMethod
Solution
To obtain reliable results, the computations were performed using arithmetic
with 10 significant figures.
The solution of the eigenvalue problem ðK w2 MÞf ¼ 0 gives
8 9 8 9 2 3
< w1 = < 2:9290 = 0:07127 0:10731 0:05133
w ¼ 3:9429 , F ¼ 4 0:10257 +0:08574 0:03682 5 (1)
: 2; : ;
w3 11:0502 +0:02068 +0:00653 0:04239
The eigenmodes are orthonormalized with respect to the mass.
(i) Additional orthogonality conditions
The damping matrix will be computed using Eq. (12.11.32) for
q ¼ 0, 1, 2
2
C ¼ a0 M + a1 K + a2 M M1 K
(2)
¼ a0 M + a1 K + a2 KM1 K
The coefficients a0 ,a1 ,a2 result from Eq. (12.11.36) for N ¼ 3
2 38 9 8 9
1=w1 w1 w31 < a0 = < 0:06 =
14
1=w2 w2 w32 5 a1 ¼ 0:08 (3)
2 : ; : ;
1=w3 w3 w33 a2 0:10
Substituting the values of the eigenfrequencies from Eq. (1) and solving
the system give
a0 ¼ 0:22227 101 , a1 ¼ 0:45470 101 , a2 ¼ 0:22666 103 (4)
and Eq. (2) gives
2 3
40:219 2:549 88:247
C ¼ 4 2:549 30:647 70:130 5 (5)
88:247 70:130 806:837
(ii) The alternative method based on the modal matrix

We compute the matrix C ^ using Eq. (12.10.60) for N ¼3
2 3 2 3
2x1 w1 0 0 0:351 0 0
C^ ¼ 40 2x 2 w2 0 5 ¼ 40 0:631 0 5 (6)
0 0 2x 3 w3 0 0 2:210
2 3
40:219 2:549 88:247
C ¼ MFCF ^ T M ¼ 4 2:549 30:647 70:130 5 (7)
88:247 70:130 806:837
@CivilMethod
(iii) Rayleigh damping

The damping matrix will be computed from Eq. (12.11.11) while the
coefficients a0 ,a1 are obtained from the solution of the system
2 3
1
w n 7
166 wn
a
7 0 ¼
xn
(8)
24 1 5 a
1 x m
wm
wm
and the third eigenfrequency from Eq. (12.11.16), that is,

1 1
xk ¼ a0 + a1 wk , k 6¼ m,n (9)
2 wk
There are three possibilities for computing the damping matrix.
(a) n ¼ 1, m ¼ 2. The system (8) is written

1 1=w1 w1 a0 0:06
¼ (10)
2 1=w2 w2 a1 0:08
which gives
a0 ¼ 0:74714 102 , a1 ¼ 0:40098 101 (11)

Then substituting these values in Eq. (12.11.11) yields the damping
matrix
2 3
59:41 16:32 222:66
6 7
C ¼ a0 M + a1 K ¼ 4 16:32 40:52 166:56 5
222:66 166:56 1748:22
and Eq. (9) gives x3 ¼ 0:221.
(b) n ¼ 1, m ¼ 3. The system (8) is written

1 1=w1 w1 a0 0:06
¼ (12)
2 1=w3 w3 a1 0:10
which gives
a0 ¼ 0:21103, a1 ¼ 0:16371 101 (13)

Then Eq. (12.11.11) gives the damping matrix
2 3
35:07 6:66 90:90
6 7
C ¼ a0 M + a1 K ¼ 4 6:66 27:36 68:00 5 (14)
90:90 68:00 805:46
and Eq. (9) gives x 2 ¼ 0:059
@CivilMethod
(c) n ¼ 2, m ¼ 3. The system (8) is written

1 1=w2 w2 a0 0:08
¼ (15)
2 1=w3 w3 a1 0:10
which gives
a0 ¼ 0:40047, a1 ¼ 0:14819 101 (16)
Then Eq. (12.11.11) gives the damping matrix
2 3
42:63 6:03 82:29
C ¼ a0 M + a1 K ¼ 4 6:03 35:65 61:56 5 (17)
82:29 61:56 821:49
and Eq. (9) gives x 1 ¼ 0:090.
We observe that the damping ratio computed from Eq. (9) deviates signif-
icantly from the corresponding predetermined. One technique to restrict the
deviation is by best fitting a straight line x ¼ a
w + b through the pairs ðwi , xi Þ,
i ¼ 1, 2, …, k using the least square method and then taking the values xi .
Example 12.11.3 Determine the eigenfrequencies and eigenmodes of the truss
in Fig. E12.5 and construct a proportional damping matrix based on the modal
matrix. The truss is loaded by the static (gravity) load W ¼ 400 kN. Assumed
data: Nodal coordinates: 1ð0, 0Þ, 2ð3, 3:5Þ, 3ð6, 0Þ; Cross-sectional areas of
the bars: A1 ¼ 1:5A, A2 ¼ A3 ¼ A, and A ¼ 27 cm2 ; modulus of elasticity
E ¼ 2:1 108 kN=m2 ; material density: r ¼ 7:55 kNm1 s2 =m3 ; lumped
mass assumption; and modal damping ratios x1 ¼ 0:1, x2 ¼ 0:08, x3 ¼ x4 ¼
… ¼ xN ¼ 0:06.
FIG. E12.5 Plane truss in Example 12.12.3.
Solution
1. Global mass and stiffness matrices
The geometrical data of the truss and the element properties were com-
puted in Example 11.2.2 except for the mass of the gravity load W , which
constitutes an additional nodal mass in the directions 3, 4 of the global mass
matrix. Therefore, we have m33 ¼ m44 ¼ W =g ¼ 40:775. Hence
@CivilMethod
2 3
0 0 0 0 0 0
6 7
60 0 0 0 07
0
6 7
6 7
60 0 2000:0 0 0 07
W 6 7
M ¼ rA6 7
60 0 0 2000:0 0 0 7
6 7
6 7
60 0 0 0 0 07
4 5
0 0 0 0 0 0
and
2 3
6:805 0 0 0 0 0
60 7
6 6:805 0 0 0 0 7
X 6 7
3 60 0 2004:610 0 0 0 7
¼ MW
M + ^ e ¼ rA6
M 7
60 7
6 0 0 2004:610 0 0 7
e¼1
6 7
40 0 0 0 6:805 0 5
0 0 0 0 0 6:805
2 3
0:342 0:107 0:092 0:107 0:25 0
6 0:107 0:125 0:107 0:125 0 7
6 0 7
X 6 7
3 6 0:092 0:107 0:184 0 0:092 0:107 7

K¼ ^ e 6
K ¼ EA6 7
0:107 0:125 0:125 7
6 0 0:25 0:107 7
e¼1
6 7
4 0:25 0 0:092 0:107 0:342 0:107 5
0 0 0:107 0:125 0:107 0:125
M
2. Modification of the matrices K, due to support conditions
The truss is supported as in Example 11.2.2. Thus the matrix V that
M
reorders the matrices K, is the same. This yields
@CivilMethod
3. Eigenfrequencies and eigenmodes.

The eigenfrequencies and eigenmodes result from the solution of the
eigenvalue problem

e ff lM
K e ff f ¼0 (1)
where l ¼ w2 r=E. Taking into account the above-indicated partitioning

we have
2 3
10 4:610 0 0
Me ff ¼ rA6
40 104:610 0
7
5,
0 0 6:805
2 3
0:184 0 0:092
e ff ¼ EA6
K 4 0 0:25
7
0:107 5
0:092 0:107 0:342
The solution of the eigenvalue problem (1) gives

8 9 8 9 2 3
= >
< w1 >
> < 45:20 > = 0:0206 0:0086 0:0004
6 7
w2 ¼ 56:30 , F ¼ 4 0:0086 0:0206 0:0004 5
>
: ; :> > >
;
w3 1182:66 0:0082 0:0041 0:3832
Fig. E12.6 shows the eigenmodes
FIG. E12.6 Eigenmodes of the truss in Example 12.12.3.
@CivilMethod
4. Construction of the proportional damping matrix.

The matrix C ^ is
2 3
0:1 45:20 0 0
^ ¼ 24 0
C 0:8 56:30 0 5
0 0 0:06 1182:66
2 3
9:0392 0 0
¼ 40 9:0084 0 5
0 0 141:9195
2 3
181:77 0:98 2:43
^
C ¼ MFCF M ¼ 10 4 0:98
T 2
81:56 2:83 5
2:43 2:83 9:65
Note that, as was anticipated, it is
2 3
9:0392 0 0
FT CFT ¼ 4 0 9:0084 0 ^
5¼C
0 0 141:9195
12.12 Problems
Problem P12.1 Compute the eigenfrequencies and eigenmodes of the truss
shown in Fig. P12.1. W1 ¼ 400 kN and W2 ¼ 250 kN are gravity dead loads.
Assumed data: Coordinates of the nodes:1ð0, 0Þ, 2ð4, 0Þ, 3ð0, 4Þ,4ð4, 4Þ,
5ð8, 4Þ; cross-sectional area of bars:A1 ¼ A2 ¼ A3 ¼ 1:5A, A4 ¼ A5 ¼ A7 ¼ A,
A6 ¼ 2A, and A ¼ 27 cm2 ; modulus of elasticity E ¼ 2:1 108 kN=m2 ; and
material density r ¼ 7:55 kNs2 =m4 . Consistent mass assumption. The rigid plane
body B has dimensions 4 4 m2 , thickness h ¼ 0:10 m, and specific weight
24 kN=m3 .
FIG. P12.1 Plane truss in problem P12.1.
@CivilMethod
Problem P12.2 A single-story building whose plan form is an equilateral trian-

gle of side a is supported by three columns of a rectangular cross as shown in
Fig. P12.2. The floor plate is assumed rigid and the columns fixed at both ends.
The mass of the columns is neglected. Construct a proportional damping matrix
using (i) Rayleigh damping, (ii) three additional orthogonality conditions, and
(iii) the modal matrix method. Assume: a ¼ 10 m, height of columns h ¼ 5 m;
cross-sectional area of columns 30 60 cm2 ; modulus of elasticity
E ¼ 2:1 107 kN=m2 and Poisson’s ratio n ¼ 0:2; and total load of the plate
(including dead weight) p ¼ 12 kN=m2 . Damping ratios: x1 ¼ 0:10,
x2 ¼ x3 ¼ 0:08.
FIG. P12.2 Plan form of the single-story building in problem P12.2.
Problem P12.3 Formulate the equation of motion of the two-story shear build-
ing of Fig P12.3. Then (i) compute the eigenfrequencies and eigenmodes, and
(ii) construct a proportional damping matrix of the form C ¼ a0 M + a1 K with
x1 ¼ 1:2, x2 ¼ 0:1 and formulate the equation of motion with damping. Assumed
data: E ¼ 2:1 107 kN=m2 and g ¼ 25 kN=m3 . All columns have a square
cross-section 0:30 0:30m2 . The load of the slabs also includes their dead load.
FIG. P12.3 Two-story building in problem P12.3.
@CivilMethod
Problem P12.4 The industrial chimney of length L ¼ 75 m shown in Fig. P12.4

consists of the outer reinforced concrete shell, which supports the linings.
The thickness of the thermal insulation layer is ti ¼ 0:10 m and that of the
refractory layer tr ¼ 0:10 m. The chimney is fixed on the ground. The structure
is modeled by three constant elements. Adopting lumped mass assumption:
(i) formulate the equation of motion and compute the eigenfrequencies and
eigenmodes, and (ii) construct a proportional damping if x 1 ¼ 0:15, x 2 ¼ 0:1
and xk ¼ 0:8 for k > 2 using all discussed methods. Assume the following data:
Specific weight of reinforced concrete g b ¼ 24 kN=m3
Specific weight of thermal insulation g i ¼ 0:7 kN=m3
Specific weight of refractory bricks g r ¼ 20 kN=m3
Modulus of elasticity of reinforced concrete E ¼ 2:1 107 kN=m2
FIG. P12.4 Industrial chimney in problem P12.4.
Problem P12.5 Formulate the Rayleigh proportional damping matrix C for the
three-degree-of-freedom system with modal damping ratios x1 ¼ 1:5, x2 ¼ 0:10,
and x3 ¼ 0:08. Consider all different cases and compare them with that obtained
using the modal matrix. Data
2 3 2 3
1000 0 5000 120000 40000 0
6 7 6 7
M ¼ 40 500 4000 5, K ¼ 4 40000 40000 0 5,
5000 4000 77833:33 0 0 100000
8 9
< 1:120 =
w ¼ 11:553
: ;
13:541
@CivilMethod

[1] G. Strang, Linear Algebra and Its Applications, fourth ed., Cengage Learning, United States,
2005.
[2] S. Lipschutz, M.L. Lipson, Linear Algebra, fourth ed., Schaum’s Outline SeriesMcGraw-Hill
Companies, Inc., New York, 2009
[3] E.L. Ince, Ordinary Differential Equations, Dover Publications, New York, 1956.
[5] R.W. Clough, J. Penzien, Dynamics of Structures, second ed., McGraw-Hill, New York, 1993.
[6] K.-J. Bathe, E.L. Wilson, Numerical Methods in Finite Element Analysis, Prentice-Hall Inc.,
Englewood Cliffs, NJ, 1976
[7] T.K. Caughey, Classical normal modes in damped linear dynamic systems, J. Appl. Mech.
27 (1960) 269–271. ASME.
@CivilMethod
Chapter 13
Numerical evaluation of the

eigenfrequencies and
eigenmodes
Chapter outline
13.1 Introduction 601 13.3.1 The vector purification
13.2 The vector iteration method 603 method 609
13.2.1 The inverse vector 13.3.2 The inverse vector
iteration method 604 iteration method with
13.2.2 Convergence of the shifts 610
inverse vector iteration 13.4 Free or partially supported
method 606 structure 612
13.3 Computation of higher-order 13.5 Problems 617
eigenpairs 609 References and further reading 618
13.1 Introduction
From the study of the problem of free vibrations of MDOF systems presented in
Chapter 12, it becomes obvious that the computation of the eigenfrequencies
and eigenmodes, that is, the solution of the eigenvalue problem, plays a crucial
role in the dynamic analysis of structures. In the examples presented there, the
computation of the eigenvalues li was made by determining the roots of the
characteristic polynomial
PðlÞ ¼ a0 lN + a1 lN 1 + ⋯ + aN 1 l + aN (13.1.1)
and subsequently the eigenvectors fi were obtained from the solution of the
linear algebraic system
ðK li MÞfi ¼ 0 (13.1.2)
This method requires the computation of the coefficients of PðlÞ. As soon as
this is achieved, its roots are computed by a numerical method because an ana-
lytical determination of the roots is possible only for polynomials up to fourth
degree (N ¼ 4). The determination of the polynomial coefficients requires a large
number of computations because the roots of PðlÞ are very sensitive to small
changes of the coefficients, leading to a considerable inaccuracy of the roots.

@CivilMethod
Thus, this method is rather theoretical in nature and not suitable for solving the
eigenvalue problem, particularly for large degree-of-freedom systems.
The numerical solution of the eigenvalue problem is a difficult problem of
computational linear algebra. Great efforts have been made to develop reliable
methods for solving this problem during the last two centuries. The emergence
of computers in the 1950s gave a colossal impetus for research in this area.
Thus, since that time, many new methods have been developed for the solution
of the eigenvalue problem. Generally, these methods can be grouped into the
following three categories:
1. Transformation methods.
2. Iteration methods.
3. Determinant search method.
The transformation methods are applied when the size of the eigenvalue prob-
lem is comparatively small and the matrices are fully populated or banded with
large bandwidth. The underlying idea of these methods is the use of consecutive
transformations, which render the matrix A of the standard eigenvalue problem
ðA lIÞx ¼ 0 diagonal. These methods compute all the eigenvalues simulta-
neously while the eigenmodes are computed either by adhering to the inverse
transformations or by iteration procedures. The known transformation methods
are (i) Jacobi diagonalization, named after Carl Gustav Jacob Jacobi, who first
proposed the method in 1846, although it became widely used only in the 1950s
with the advent of computers, (ii) Givens triangularization, (iii) Householder
transformation, and (iv) QR and QL transformations.
The iteration methods are suitable for solving large eigenvalue problems
with banded or sparse matrices as those in the finite elements. Particularly, they
are applied in cases where it is necessary to compute only a few eigenvalues, the
smallest or the largest ones. The most known iteration methods that are effec-
tively applied to the solution of the vibration problem in engineering are: (i)
vector iteration, (ii) vector iteration with shift, (iii) subspace iteration, and
(iv) the Lanczos method.
Finally, the determinant search method is based on the search for the roots of
the function det ðA lBÞ. It is employed when the size of the eigenvalue
problem is comparatively large while the matrices are banded with a small
bandwidth and the computation of only a few first eigenpairs is required.
The previous methods can be combined to give the best solution for a partic-
ular problem. It should be emphasized that all methods are by nature iterative
because the solution of the eigenvalue problem is equivalent to that of determin-
ing the roots of a polynomial, which, as already mentioned, are established only
by numerical iterative methods for orders of the characteristic polynomial N > 4.
Although an iterative process is necessary for the computation of the eigen-
pair ðli , fi Þ, it should be noted that as soon as one of its members has been
computed, the other member can be computed without further iterative process.
Suppose that li is computed by an iteration method, then fi can be computed by
solving the linear system (13.1.2). On the other hand, if fi has been computed
@CivilMethod
Numerical evaluation of the eigenfrequencies and eigenmodes Chapter 13 603
by an iteration method, then li can be computed from the Rayleigh quotient

using the relations
li ¼ fTi Kfi , fTi Mfi ¼ 1 (13.1.3)
Therefore, when selecting the appropriate method, the question arises as to
whether it is computationally less expensive to compute li first and then fi , or
vice versa, or both at the same time. The answer to this question depends on the
requirements of the particular problem and the properties of the matrices K and
M, for example, the number of the desired eigenpairs, the dimensions and the
density of K and M, etc.
From the above discussion, we realize that there is not always a single algo-
rithm leading to an effective solution to the eigenvalue problem. Indeed, an
algorithm that can be effective for a particular problem may be totally inade-
quate for another one. The criteria for choosing the solution algorithm are
two: (a) the reliability of the method, and (b) its computational cost. Reliability
implies the computation of the solution with the desired precision without the
method collapsing. The computational cost is determined by the number of
numerical operations and the required computer memory.
All methods mentioned above are described in detail in many books of struc-
tural dynamics published earlier, for example, Refs. [1–4]. The reason is clear.
At that time, the ready-to-use computer codes and functions for the numerical solu-
tion of the eigenvalue problem were limited, if not unavailable, and the engineers
had to write their own computer codes to cope with the largely increasing needs for
determining the eigenfrequencies and mode shapes. The full and detailed develop-
ment of all the above methods requires extensive work, the presentation of which
goes beyond the objectives of this book. Besides, this issue is the subject of many
books on numerical analysis to which the interested reader may refer [4]. On the
other hand, mathematical libraries, for example, IMSL, MAPLE, MATLAB,
Numerical Recipes, etc., offer many ready-to-use subroutines and functions for
the solution of the eigenvalue problem. In the following, for educational purposes,
we limit our presentation only to the vector iteration method, which can be readily
used to solve small-sized eigenvalue problems by hand or by writing simple
computer code. The chapter closes with the discussion of the response of the
free or partially supported structures. The examples facilitate the understanding
of the presented material. References and related bibliography for further study
are also included. The chapter is enriched with problems to be solve.
13.2 The vector iteration method

In the vector iteration method, the starting point is the equation
Kf ¼ lMf (13.2.1)
aiming at approximating the vector f so that this equation is satisfied. Thus, we
assume arbitrarily a vector f, which we designate as x1 , and a value of l, say
l ¼ 1. Then we evaluate the right side of Eq. (13.2.1). This yields
R1 ¼ 1 Mx1 (13.2.2)
@CivilMethod
Inasmuch as x1 is arbitrary, it will not, in general, satisfy Kx1 ¼ R1 .

Nevertheless, if this happens, then x1 is an eigenvector. Therefore, a vector
x2 6¼ x1 should be determined to satisfy the equation
Kx2 ¼ R1 (13.2.3)
Eq. (13.2.3) can be considered as the static equilibrium of a system subjected
to load R1 . From intuition, we understand that the vector x2 resulting from the
solution of Eq. (13.2.3) is closer to the eigenvector f than x1 . Repeating
the above procedure with an initial vector x2 gives a better approximation of
the eigenvector. Thus, we can develop a simple iterative procedure that yields
f, when it converges. The assumption of an arbitrary value for l does not affect
the method because, as is known, the eigenvectors are determined to an arbitrary
multiplier. The iteration method resulting in this way is known as the direct
vector iteration method. The procedure we have described is the underlying
idea of the inverse vector iteration method, in which Eq. (13.2.1) is employed
inversely. Apparently, in the direct vector iteration method, the vector
R1 ¼ Kx1 is first computed and then a better approximation x2 is obtained from
the solution of the linear system Mx2 ¼ R1 .
13.2.1 The inverse vector iteration method

This method, also known as the Stodola-Vianello method, is very efficient for
the computation of the eigenvectors and the respective eigenvalues. Stodola
used it for the solution of the vibration problem of a rotating shaft in 1904. Ear-
lier in 1898, Vianello used it for determining the critical load in buckling for a
rotating shaft. It applies when K is not singular while M may be diagonal
including zeros elements, banded or fully populated. If K is singular, the inverse
vector iteration method with shifts is employed (Section 13.3.2). This happens
when the structure is free or partially supported (Section 13.4). In this method,
as mentioned, we start with an initial vector x1 and then adhere to the iteration
steps k ¼ 1, 2,… until convergence is achieved:
k + 1 by solving the linear system
1. We determine the vector x
K
xk + 1 ¼ Mxk (13.2.4)
2. We compute the approximate value of the eigenvalue lðk + 1Þ corresponding
k + 1 using the Rayleigh quotient, Eq. (12.6.21),
to x
Tk+ 1 K
x xk + 1 ðk + 1Þ
k + 1 Þ ¼
rðx , l k + 1 Þ
¼ rðx (13.2.5)
k + 1 M
x T xk + 1
3. We check for convergence by comparing two consecutive values of l

ðk + 1Þ
l lðk Þ
<e (13.2.6)
lðk + 1Þ
where e is a specified tolerance
@CivilMethod
k + 1 with respect
4. If the convergence criterion is not satisfied, we normalize x
to mass
k + 1
x
xk + 1 ¼ 1=2 (13.2.7)
Tk+ 1 M
x xk + 1
and go back to step 1.
5. If the criterion is satisfied, the procedure ends and we set
f1 xk + 1 , l1 lðk + 1Þ (13.2.8)
The normalization does not affect the convergence. However, if the vector xk + 1
is not normalized, its elements increase in each step, a fact that can cause numer-
ical problems. If the desired precision of l1 is 2d digits, then the tolerance should
be taken as e ¼ 102d . The resulting precision of the eigenvector will be d digits.
Computationally, it is more effective to formulate the previous algorithm as
follows.
We take an arbitrary vector x1 and compute the vector z1 ¼ Mx1 .
Subsequently, we compute for k ¼ 1, 2, … until convergence is achieved
K
xk + 1 ¼ zk (13.2.9)
zk + 1 ¼ M
xk + 1 (13.2.10)
Tk+ 1 zk
x
k + 1 Þ¼
rðx , lðk + 1Þ ¼ rðx
k + 1 Þ (13.2.11)
k + 1zk + 1
xT
zk + 1
zk + 1 ¼ 1=2 (13.2.12)
Tk+ 1zk + 1
x
If convergence is achieved after k iterations, we take
k + 1
x
f1 ’ 1=2 , l1 lðk + 1Þ (13.2.13)
Tk+ 1zk + 1
x
The method works provided that zT1 f1 6¼ 0.
Example 13.2.1 Compute the eigenpair ðl1 , f1 Þ of the eigenvalue problem
Kf ¼ lMf using the inverse vector iteration method, when
2 3 2 3
32 24 18 0:5 0 0
K ¼ 4 68 71 38 5, M ¼ 4 0 0:5 0 5
36 27 20 0 0 1
Solution
A convenient trial vector to start the procedure is the vector x1 ¼ f1 1 1gT . The
computed eigenpair for different values of k is shown in Table E13.1.
@CivilMethod
Apparently, after eight iterations, it coincides with that obtained by the

MATLAB function [V,D] ¼ eig(A,B).
TABLE E13.1 Computation of the first eigenpair using

the inverse vector iteration method.
k l1 fT1
1 1 0.1522 0.6875 0.8672
2 13.2829 0.0327 0.6113 0.9015
3 6.4215 0.0460 0.6103 0.9015
4 6.6418 0.0438 0.6112 0.9013
5 6.6439 0.0442 0.6110 0.9013

6 6.6416 0.0441 0.6110 0.9013
7 6.6421 0.0442 0.6110 0.9013
8 6.6420 0.0442 0.6110 0.9013
MATLAB 6.6420 0.0442 0.6110 0.9013
13.2.2 Convergence of the inverse vector iteration method

In the preceding section, we presented the algorithm of inverse vector iteration
assuming that the method converges to the eigenvector corresponding to the
smallest eigenvalue. We will now present a formal proof of the convergence
because it allows us to understand the extension of the method in determining
the higher-order eigenpairs.
To prove the convergence, we consider Eq. (13.2.4), which we write as
Kxk + 1 ¼ Mxk (13.2.14)
We assume that the eigenmodes have been computed and let
F ¼ ½f1 f2 ⋯fN be the modal matrix. Applying the expansion theorem
(12.5.82) for u ¼ xk ,xk + 1 , X ¼ F and a ¼ yk , yk + 1 yields
xk ¼ Fyk (13.2.15a)
xk + 1 ¼ Fyk + 1 (13.2.15b)
Substituting Eqs. (13.2.15a), (13.2.15b) into Eq. (13.2.14) and premultiply-
ing by FT gives
FT KFyk + 1 ¼ FT MFyk (13.2.16)
@CivilMethod
Using the orthogonality relations FT MF ¼ I and FT KF ¼ L, the forgoing

equation becomes
Lyk + 1 ¼ yk (13.2.17)
where L is the diagonal matrix including the eigenvalues.
Eqs. (13.2.14), (13.2.17) are equivalent, therefore the convergence proper-
ties of Eq. (13.2.14) are those of Eq. (13.2.17). The latter, however, simplifies
the proof of convergence because the elements of L are the eigenvalues li and
the eigenvectors are unit vectors ei , that is,
ð13:2:18Þ
In the inverse vector iteration method, the starting vector x1 should not be
M-orthogonal to f1 . Now y1 should not be M-orthogonal to e1 . Apparently, the
vector
y 1 ¼ f 1 1 ⋯ 1 gT (13.2.19)
satisfies this requirement.
Applying Eq. (13.2.17) for k ¼ 1, 2, …l gives

1 1 1 T
yl + 1 ¼ ⋯ (13.2.20)
ll1 ll2 llN
Because the eigenvector does not change if it is multiplied by a number,
we can write
l + 1 ¼ ll1 yl + 1
y
( l l )T
l1 l1
¼ 1 ⋯ (13.2.21)
l2 lN
Assuming that l1 < l2 < ⋯lN , we obtain

l + 1 ¼ e1
lim y (13.2.22)
l!1
Therefore, for l ! 1, yl + 1 converges to a multiple of e1 .

Using the Euclidian norm, the convergence of a sequence of vectors
y1 , y2 ,…,yl to a vector e1 is defined by the relation

yl + 1 e1
lim ¼r (13.2.23)
l!1 jy l e1 jp
where p is the number characterizing the rate of convergence, for example, for
p ¼ 1 the rate is linear, for p ¼ 2 quadratic and so forth. In the present case it is
@CivilMethod

y l + 1 e1 l1
lim ¼ (13.2.24)
l!1 l e1 j
jy l2
Hence, convergence is linear and the rate of convergence is r ¼ l1 =l2 . The
smaller this ratio, the faster the rate of convergence.
In the preceding proof, we have accepted that the eigenvalues are distinct.
Let us now consider the case of multiple eigenvalues, for example,
l1 ¼ l2 ⋯ ¼ lm . Then Eq. (13.2.21) becomes
( l )T
l1 l l1
l + 1 ¼ 1 1 ⋯
y ⋯ (13.2.25)
lm + 1 lN
from which we conclude that the iteration procedure converges with rate of con-
vergence l1 =lm + 1 . Hence, we can generally state that the rate of convergence
is given by the ratio of l1 to the nearest distinct eigenvalue.
The Rayleigh quotient for the vector yl + 1 is obtained by Eq. (12.6.1) for
A ¼ L and u ¼ yl + 1 . Thus we have
yT Ly
r yl + 1 ¼ l +T 1 l + 1
yl + 1 yl + 1
yT y
r yl + 1 ¼ T l + 1 l (13.2.26)
yl + 1 yl + 1
Using Eq. (13.2.21) to express yl and yl + 1 , and substituting into the
foregoing equation gives
X
N
l1 ðl1 =li Þ2l1
i¼1
r yl + 1 ¼ (13.2.27)
X
N
2l
ðl1 =li Þ
i¼1
which implies

lim r yl + 1 ¼ l1 (13.2.28)
l!1
The preceding proof was made by taking x1 ¼ f 1 1 ⋯ 1 gT as a starting

vector. However, we can show that the convergence properties apply for any
starting vector provided that it is not M-orthogonal to f1 . Nevertheless, in some
cases, the convergence may require a large number of iterations.
The convergence rate depends, as we have shown, on the ratio l1 =lm + 1 .
The convergence is very slow if l1 is almost equal to lm + 1 , but it can be
accelerated using the inverse vector iteration method with shifts presented in
Section 13.3.2.
@CivilMethod
13.3 Computation of higher-order eigenpairs

13.3.1 The vector purification method
When the starting vector x1 is selected to be M-orthogonal to f1 , that is
xT1 Mf1 ¼ 0, the inverse vector iteration method converges, at list theoretically,
to an eigenvector other than f1 , and indeed to that corresponding to the next
eigenvalue l2 . The selection of the starting vector may be again arbitrary
and become orthogonal to f1 by the Gram-Schmidt orthogonalization process
presented in Section 12.4. Thus, on the basis of the second of Eqs. (12.4.8),
we obtain
fT1 Mx1
e
x1 ¼ x1 f1 (13.3.1)
fT1 Mf1
which is orthogonal to f1 . Obviously, the contribution of the eigenmode f1
has been removed from the trial vector ex1 . The vector that remains is said to
be purified. Now, Eq. (13.3.1) can be further written as
!
f1 fT1 M
e
x1 ¼ I T x1
f1 Mf1 (13.3.2)
¼ S 1 x1
where
f1 fT1 M
S1 ¼ I (13.3.3)
fT1 Mf1
The matrix S1 is referred to as the sweeping matrix. It is used to purify the
vector x1 from f1 . Theoretically, the iterations start with the purified vector e
x1
and we expect convergence to the second eigenvector f2 . Unfortunately, during
the iterations, small arithmetic rounding errors are introduced, which result in
pollution of the vector x1 by f1 . This demands the purification procedure in
each iteration, which, in fact, increases the computational cost. For the compu-
tation of the nth order eigenpair ðln , fn Þ, the starting vector x1 must be free
from the contribution of the previous n 1 eigenmodes and M-orthogonal to
them. This is achieved by the Gram-Schmidt orthogonalization procedure.
Thus, by virtue of Eq. (12.4.8), we obtain
fT1 Mx1 fT2 Mx1 fTn Mx1
e
x1 ¼ x1 f 1 f 2 ⋯ fn
fT1 Mf1 fT2 Mf2 fTn Mfn
!
f1 fT1 M f2 fT2 M fn fTn M (13.3.4)
¼ I T ⋯ T x1
f1 Mf1 fT2 Mf2 fn Mfn
¼ Sn x1
@CivilMethod
where Sn represents the sweeping matrix

f1 fT1 M f2 fT2 M fn fTn M
Sn ¼ I ⋯ (13.3.5)
fT1 Mf1 fT2 Mf2 fTn Mfn
or
fn fTn M
Sn ¼ Sn1 , n ¼ 1, 2, …, S0 ¼ I (13.3.6)
fTn Mfn
Although Gram-Schmidt orthogonalization combined with the inverse vec-
tor iteration may yield the higher-order eigenpairs, it is not suitable for writing a
general computer program for the solution of the eigenvalue problem because
convergence becomes slow as the order of the eigenvector increases. Instead,
we prefer the vector iteration method with shifts, which is described in the next
section.
13.3.2 The inverse vector iteration method with shifts

The study of the convergence of the inverse iteration method presented in
Section 13.2.2 showed that for l1 < l2 , the iteration process converges to the
eigenvector f1 with a convergence rate l1 =l2 . Thus, the convergence is fast
when l1 is very small compared to l2 , for example, l1 =l2 ¼ 0:01, or extremely
slow when l1 is near l2 , for example, l1 =l2 ¼ 0:9999. Therefore, a problem
of improving or even controlling the rate of convergence arises. This is
accomplished by the vector iteration method with shifts, which can also be used
to determine any other eigenpair ðli , fi Þ. This method also addresses the case
where the matrix K is positive semidefinite, hence the solution of Eq. (13.2.4) is
not possible.
A shift m in K in the eigenvalue problem Kf ¼ lMf defines the matrix
^ ¼ K mM. The matrices K
K ^ and K have the same eigenvectors while the
eigenvalues l^ and l satisfy the relation
^ ¼lm
l (13.3.7)
^ is shifted by m with respect to the origin
That is, the eigenvalue l
(Fig. 13.3.1). Indeed
^ ¼ ðK mMÞf
Kf
¼ Kf mMf
(13.3.8)
¼ ðl mÞMf
^
¼ lMf
@CivilMethod
Π( ) = det (K − M)
| 2 − m|
| 1 − m| | 3 − m|
1 2 3 4
m
FIG. 13.3.1 Graph of the characteristic polynomial and shifts
We assume that all eigenvalues l ^ i of the eigenvalue problem (13.3.8) are

distinct. In this case, Eq. (13.2.20) becomes
T
1 1 1
yl + 1 ¼ ⋯ (13.3.9)
ðl1 mÞl ðl2 mÞl ðlN mÞl
where it was assumed that li m is nonzero, that is, it is a positive or a negative
number. We assume that l ^ i ¼ li m has the smallest absolute value and we
formulate the product
ð13:3:10Þ

In the foregoing relation, it is jli mj=lp m < 1 for all p 6¼ i. Hence,
l + 1 ! ei as l ! 1 and the iteration vector converges to fi . Moreover, we
y
obtain li ¼ l^ i + m. The rate of convergence is determined by the element
having the largest absolute value, that is,

li m
r ¼ max (13.3.11)
p6¼i lp m
Because li is closer to m, the rate of convergence of the iteration vector to fi

will be

li m li m
r ¼ max , (13.3.12)
li1 m li + 1 m
For the data shown in Fig 13.3.1, we have jl2 mj < jl1 mj < jl3 mj.
Therefore, the iterative procedure will converge to f2 with rate of convergence
r ¼ jl2 mj=jl1 mj.
Evidently, the selection of the shift very close to the eigenvalue we are
interested in permits, at least theoretically, achieving the desired rate of
@CivilMethod
convergence. In practice, however, this is difficult because the eigenvalue is not

known in advance. Often, in certain problems, we need to compute a few of the
first eigenpairs. In such cases, the vector iteration method with shifts is com-
bined with that without shifts. Namely, a few, say five, eigenpairs are computed
by applying directly the inverse vector iteration method in conjunction with the
Gram-Schmidt orthogonalization. The same procedure is employed for the
computation of the next eigenpair, that is, the sixth one. However, instead of
completing the iteration process until convergence is achieved, we stop it after
some iterations and obtain a first estimation of the sixth eigenvalue from the
Rayleigh quotient, Eq. (13.2.5). This value is used as a new shift and the next
five eigenpairs are computed using the inverse vector iteration method. The rate
of convergence can be improved if in each iteration, the shift m takes the value of
the Rayleigh quotient, namely if we set mk ¼ rðx k Þ in the k + 1 iteration. Thus
the algorithm described by Eqs. (13.2.9)–(13.2.13) is modified as
k ÞM
½K rðx x k + 1 ¼ zk (13.3.13)
zk + 1 ¼ M
xk + 1 (13.3.14)
rðxk + 1 Þ ¼ x
Tk+ 1 zk
(13.3.15)
k + 1zk + 1 + rðx
x T k Þ
zk + 1
zk + 1 ¼ 1=2 (13.3.16)
Tk+ 1zk + 1
x
where now
k + 1 Þ ! li when k ! 1
zk + 1 ! Mfi and rðx (13.3.17)
The inverse vector iteration method is very useful when the system is sub-
jected to a harmonic excitation with a specified frequency and we want to see if
there are frequencies close to the excitation frequency. This allows us to prevent
resonance by changing the dynamic characteristics of the structure.
13.4 Free or partially supported structure

When the rigid body motion of the structure is not restrained, that is, the struc-
ture is free to move or the restrained degrees of freedom are less than those
required to prevent rigid body motion, then the stiffness matrix K contains zero
eigenvalues, hence it is singular and cannot be inverted. Therefore, the methods
of computation of the eigenpairs that involve the inversion of the stiffness
matrix, for example, the inverse vector iteration, do not apply. This trouble-
shooting can be overcome in the following two ways:
(a) Shifting of the eigenvalue
This method is mathematical in nature and it is accomplished by virtue
of Eq. (13.3.8)
Kf ^
^ ¼ lMf (13.4.1)
@CivilMethod
where
K ^ ¼lm
^ ¼ K mM and l (13.4.2)
Obviously, K ^ is generally not singular and can be inverted. If the
selected value of m is negative and the matrix M is diagonal, this shift is
equivalent to adding a positive number to the diagonal elements of K,
which from a physical point of view expresses restraining each degree of
freedom by a spring. The advantage of this method is that the
eigenmodes remain unchanged.
(b) Restraining the free degrees using springs
The simplest method to treat the unrestrained structure that is also com-
putationally easy to implement is by restraining the degrees of freedom cor-
responding to rigid body motion by springs with a very small stiffness.
These fictitious springs are placed between the structure and the supports
and inhibit the rigid body motion. Computationally, these springs represent
additive terms to the diagonal elements of K corresponding to the unre-
strained degrees of freedom. The addition of a small stiffness, while mak-
ing the stiffness matrix nonsingular, modifies the structure. However, the
small stiffness of the springs negligibly affects the eigenmodes and the
eigenfrequencies while at the same time the eigenmodes corresponding
to the rigid body of motion are also computed. As is anticipated, the
computed eigenfrequencies corresponding to the rigid body motion are
very small (approximately zero). This is illustrated with the example that
follows.
Example 13.4.1 Compute the mode shapes and eigenfrequencies of the plane
structure in Fig. E13.1a. The density of the rigid bodies is r while the mass of
the column is negligible.
f
′
(a) (b) (c)

FIG. E13.1 Plane structure in Example 13.4.1 (a), degrees of freedom (b), and free body diagram (c).
@CivilMethod
Solution
The system has three degrees of freedom, u1 ,f, u3 (Fig. E13.1b). The equation
of motion can be formulated by different methods. Here, the method of equi-
librium of forces with respect to the center of mass C is employed. Referring
to Fig. E13.1c, we obtain
m1 u€1 + fS1 ¼ 0 (1a)
a
Ic f€ + fS2 + fS1 ¼ 0 (1b)
2
m3 u€3 + fS3 ¼ 0 (1c)
where
fS1 ¼ k11 ðu2 u3 Þ + k12 f (2a)
fS2 ¼ k21 ðu2 u3 Þ + k22 f (2b)
fS3 ¼ k31 ðu2 u3 Þ + k32 f (2c)
The coefficients kij in the previous equations are computed using
Eq. (11.5.2). Moreover, we have
1
m1 ¼ ra 2 , m3 ¼ 3ra 2 , Ic ¼ ra 4
6
Substituting Eqs. (2a), (2b), (2c) into Eqs. (1a), (1b), (1c) gives
M€u + Ku ¼ 0 (3)
where
2 3 2 3 8 9
1 0 0 1:5 2:25a 1:5 < u1 =
EI
M ¼ ra 2 4 0 a 2 =6 0 5, K ¼ 3 4 2:25a 3:875a2 2:25a 5, u ¼ f
a : ;
0 0 3 1:5 2:25a 1:5 u3
(4a--c)
It is convenient to eliminate the length a and make the matrices dimension-
less. To this end, the rotational coordinate f is replaced by the translational
coordinate u2 ¼ u1 + a2 f (see Fig. E13.1b). This is accomplished by the
transformation (see Section 10.7).
8 9 2 38 9
< u1 = 1 0 0 < u1 =
f ¼ 4 2=a 2=a 0 5 u2
: ; : ;
u3 0 0 1 u3
which yields
2 3
1:667 0:667 0
^ ¼ RT MR ¼ ra 2 6
M
7
4 0:667 0:667 0 5,
0 0 3
2 3
8:0 11:0 3:0
^ ¼ RT KR ¼ EI 6
K 4 11:0 15:5 4:5 5
7
a3
3:0 4:5 1:5
2 3
1 0 0
where R ¼ 4 2=a 2=a 0 5
0 0 1
@CivilMethod
Therefore, the eigenvalues and eigenmodes will be obtained from the solu-
tion of the eigenvalue problem

^ lM
K ^ f ¼ 0, l ¼ w2 ra5 =EI

^ ¼ 0. Hence the solution is obtained either by the inverse vector
It is det K
iteration method with shift or by restraining u3 using a spring of small stiffness.
1. The inverse vector iteration with shift
a. Computation of the first eigenpair.
We employ the algorithm presented in Section 13.3.2. We take
arbitrarily x1 ¼ f 1 2 1 gT as the starting vector, which gives
m1 ¼ rðx1 Þ ¼ 3:321. After five iterations we obtain l1 ¼ 0:868e 16 0
and f1 ¼ f 1 1 1 gT . The obtained values of the eigenpair were anticipated
because the structure has a rigid body motion.
b. Computation of the second eigenpair
We apply the vector purification method described in Section 13.3.1.
We take arbitrarily x1 ¼ f 1 1 2 gT as the trial vector. This gives
2 3
0:75 0:472 1015 0:75
f f M 6
T
7
S1 ¼ I 1T 1 ¼ 4 0:25 1:000 0:75 5,
f1 Mf1
0:25 0:482 1015 0:25
8 9
< 0:749 >
> =
e
x1 ¼ S1 x1 ¼ 0:749 , m1 ¼ rðx1 Þ ¼ 2:0:
>
: >
;
0:249
After four iterations we obtain
l2 ¼ 0:239 and f2 ¼ f 0:834 0:508 0:278 gT .
c. Computation of the third eigenpair.
We take arbitrarily x1 ¼ f 2 2 1 gT as the trial vector. This gives
2 3
0:128 0:182 0:053
f f M 6
T
7
S2 ¼ S1 2T 2 ¼ 4 0:784 1:111 0:325 5,
f2 Mf2
0:043 0:061 0:018
8 9
< 0:054 >
> =
e
x 1 ¼ S2 x 1 ¼ 0:325 , m1 ¼ rðe
x1 Þ ¼ 25:0096
>
: >
;
0:018
After five iterations we obtain
l3 ¼ 24:998 and f3 ¼ f 0:231 1:411 0:077 gT .
2. Restraining u3 using a spring of small stiffness
0
Adding the small stiffness k33 ¼ 0:0001EI =a 3 to the element k33 yields
2 3
8:0 11:0 3:0
^ ¼ EI 4 11:0 15:5 4:5 5
K
a3
3:0 4:5 1:5001
@CivilMethod
a. Computation of the first eigenpair.

It is now det ðKÞ ¼ 0:0003 6¼ 0, therefore we can directly apply the
inverse vector iteration method using the algorithm described in
Section 13.2.1. We take arbitrarily x1 ¼ f 1 2 1 gT as a trial vector.
After five iterations we obtain l1 ¼ 0:2499 104 0 and
f1 ¼ f 1 0:999 0:998 gT f 1 1 1 gT
b. Computation of the second eigenpair.
We take x1 ¼ f 1 1 2 gT as the trial vector, which gives
2 3 8 9
0:75 0:00 0:75 < 0:75 =
f1 fT1 M 4
S1 ¼ I T ¼ 0:25 1:00 0:75 5, e x1 ¼ S1 x1 ¼ 0:75
f1 Mf1 : ;
0:25 0:00 0:25 0:25
After four iterations we obtain
l2 ¼ 0:2399 and f2 ¼ f 0:834 0:508 0:278gT .
c. Computation of the third eigenpair.
We take x1 ¼ f 2 2 1 gT as the trial vector, which gives
2 3 8 9
0:128 0:181 0:053 < 0:054 =
f2 fT2 M 4
S2 ¼ S1 T ¼ 0:785 1:111 0:323 5, e
x1 ¼ S 2 x1 ¼ 0:325
f2 Mf2 : ;
0:043 0:061 0:018 0:018
After five iterations we obtain

l3 ¼ 25:009 and f3 ¼ f 0:231 1:411 0:077gT .
Table E13.2 summarizes the eigenpairs computed by the previous two methods
and the MATLAB function [V,D] ¼ eig(A,B). The obtained results illustrate
that the method of restraining the degrees corresponding to rigid body motion
by springs of small stiffness, besides being simple to implement, gives good
results.
TABLE E13.2 Eigenvalues and eigenmodes in Example 13.3.1.

Eigenpair Iteration with shift Restraining with a spring MATLAB
n ln fn ln fn ln fn
16 4
1 0:868 10 0:999 0:249 10 1:000 0.0000 1:000
0:999 0:999 1:000
1:000 0:998 1:000
2 0:239 0:834 0:2399 0:834 0.2399 0:834

0:508 0:508 0:508
0:278 0:278 0:278
3 24:998 0:231 25:009 0:231 24.998 0:231

1:411 1:411 1:411
0:077 0:077 0:077
@CivilMethod
13.5 Problems
Problem P13.1 The one-story building of Fig. P13.1 is subjected to the harmonic
excitation p ¼ p0 sin20t. Examine whether the structure is at resonance risk.
Assume: E ¼ 2:1 107 kN=m2 , n ¼ 0:2: Columns 1, 3, 6, 8 0:35 0:35 m2 ,
Columns 2, 4, 5, 7 0:20 0:40 m2 . The dead weight of the plate is included in q.
Problem P13.2 We consider the plane structure of Fig. P13.2. At time t the sup-
port A becomes a hinge. Compute the eigenfrequencies and eigenmodes of the
structure at t + .
FIG. P13.2 Plane structure in problem P13.2
Problem P13.3 We consider the plane structure of Fig. P13.3 whose base is iso-
lated. Compute the eigenfrequencies (i) if the stiffness of the isolators is small
compared with that of the columns, that is, kb =k ¼ 0:1, (ii) if the isolators break
at time t, that is, kb ¼ 0. Assume; m1 ¼ m2 , mb ¼ 0:8m1 .
@CivilMethod
FIG. P13.3 Plane structure in problem P13.3

[1] K.J. Bathe, E.L. Wilson, Numerical Methods in Finite Element Analysis, Prentice-Hall,
[3] J.T. Katsikadelis, Dynamic of Structures, Vol. II, Symmetria Publications, Athens, 2004 (in
Greek).
[4] W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterlin, Numerical Recipes in Fortran,
second ed., Cambridge University Press, New York, 1992.
[5] T.J.R. Hughes, The Finite Element Method: Linear Static and Dynamic Finite Element
Analysis, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1987.
@CivilMethod
Chapter 14
systems: Forced vibrations
Chapter outline
14.1 Introduction 620 14.9 Comparison of mode
14.2 The mode superposition superposition method
method 620 and Rayleigh-Ritz method 672
14.3 Modal contribution in the 14.10 Numerical integration
mode superposition method 629 of the equations of
14.3.1 Modal participation 629 motions—Linear MDOF
14.3.2 Static correction systems 675
method 641 14.10.1 The central
14.3.3 Error in mode difference method
superposition method (CDM)—Linear
due to truncation of equations 675
higher modes 647 14.10.2 The average
14.4 Reduction of the dynamic acceleration method
degrees of freedom 649 (AAM)—Linear
14.4.1 Static condensation 649 equations 677
14.4.2 Kinematic constraints 650 14.10.3 The analog equation
14.5 Rayleigh-Ritz method 651 method (AEM)—
14.5.1 Ritz transformation 651 Linear equations 679
14.5.2 Approximation using 14.11 Numerical integration of the
Ritz vectors 654 equations of motions—
14.6 Selection of Ritz vectors 655 Nonlinear MDOF systems 681
14.6.1 Method of natural 14.11.1 The average
mode shapes 656 acceleration method
14.6.2 The method of (AAM)—Nonlinear
derived Ritz vectors 659 equations 681
14.7 Support excitation 663 14.11.2 The analog equation
14.7.1 Multiple support method (AEM)—
excitation 663 Nonlinear equations 684
14.7.2 Uniform support 14.12 Problems 688
excitation 665 References and further reading 692
14.8 The response spectrum
method 668

@CivilMethod
14.1 Introduction
In this chapter, we study the dynamic response of MDOF systems under the
action of external forces. As was shown in Chapter 10, the equation of motion
of a linear system with viscous damping is
M€ (14.1.1)
where M is the mass matrix, C the damping matrix, K the stiffness matrix, u the
displacement vector, and pðt Þ the vector of the external forces in the directions
of the displacements. Besides, the system is subjected to the initial conditions
uð0Þ ¼u0 , u_ ð0Þ ¼u_ 0 (14.1.2)

where u0 and u_ 0 are known vectors.
The time-dependent forces pðt Þ may be due to moving loads, machine
excitations, explosions, wind pressure, aerodynamic actions, ground motions,
etc. Their determination, which depends on the type of structure and its use, con-
stitutes a special problem that does not concern the discussion that follows.
The methods that can be used to solve the initial value problem (14.1.1),
(14.1.2) are:
1. The mode superposition method.
2. The response spectrum method.
3. The direct time integration method of the equations of motion.
4. The analysis in the frequency domain.
None of the above methods is purely analytical because their implementation
includes numerical techniques, at least at a certain stage of their application.
Indeed, the mode superposition method and the response spectrum method
require the determination of the eigenfrequencies and eigenmodes, which is
accomplished, as we saw in Chapter 13, numerically. The analysis in the fre-
quency domain requires the use of numerical methods for computing the Fou-
rier transform and its inverse. Finally, the direct time integration method of the
equations of motion is predominantly a numerical method. Therefore, the
description of a method as analytical is only theoretical because the real prob-
lems of structural dynamics encountered by the engineer in practice require
numerical processes and their programming on a computer.
14.2 The mode superposition method

As was shown in Section 12.9, the displacement vector is written by virtue of the
expansion theorem as
uðt Þ ¼ f1 Y1 ðt Þ + f2 Y2 ðt Þ + ⋯ + fN YN ðt Þ (14.2.1)
where f1 ,f2 ,…, fN are the eigenmodes of the free undamped vibrations.
@CivilMethod
Multi-degree-of-freedom systems: Forced vibrations Chapter 14 621

M f1 Y€ 1 + f2 Y€ 2 + ⋯ + fN Y€ N + C f1 Y 1 + f2 Y 2 + ⋯ + fN Y N

+ Kðf1 Y1 + f2 Y2 + ⋯ + fN YN Þ ¼ pðt Þ
(14.2.2)
Premultiplying the previous equation by fTn
and assuming that the damping
matrix is proportional, we obtain by virtue of the orthogonality properties
Mn Y€ n + Cn Y n + Kn Yn ¼Pn ðt Þ, n ¼ 1, 2,…,N

(14.2.3)
where
Mn ¼ fTn Mfn (14.2.4a)
Cn ¼ fTn Cfn (14.2.4b)
Kn ¼ fTn Kfn (14.2.4c)
Pn ¼ fTn pðt Þ (14.2.4d)
The quantities defined previously are referred to as generalized or modal
mass, generalized or modal damping, generalized or modal stiffness, and
generalized or modal force of the nth mode shape, respectively. Taking into
account that
Kn ¼ w2n Mn (14.2.5)
Cn ¼ 2x n Mn wn (14.2.6)
1
Y€ n + 2xn wn Y n + w2n Yn ¼

Pn ðt Þ, n ¼ 1,2, …,N (14.2.7)

Mn
whose solution is given by Eq. (3.3.23)

" #
x n wn t Y n ð0Þ + Yn ð0Þx n wn
Y n ðt Þ ¼ e sin wDn t + Yn ð0Þcos wDn t
wDn
Z t
1
+ Pn ðτÞexn wn ðtτÞ sin ½wDn ðt τÞdτ (14.2.8)
Mn wDn 0

The initial conditions Yn ð0Þ, Y n ð0Þ are computed using Eqs. (12.9.12),
(12.9.13), namely
fTn Muð0Þ
Y n ð 0Þ ¼ , n ¼ 1, 2, …, N (14.2.9a)
Mn

fTn Mu_ ð0Þ
Y n ð 0Þ ¼ , n ¼ 1, 2, …, N (14.2.9b)
Mn
@CivilMethod
Once the modal coordinates Yn ðt Þ have been computed, the solution of

Eq. (14.1.1) is obtained from the superposition (14.2.1). The Duhamel integral
in Eq. (14.2.8) can be evaluated analytically using a symbolic language. Nev-
ertheless, this is not always possible or the obtained expressions are too com-
plicated for numerical computation, therefore it is preferable to solve
Eq. (14.2.7) directly numerically using any of the numerical methods presented
in Chapter 4.
The study of the dynamic response of a MDOF system under an external
force pðt Þ via the mode superposition method is accomplished by adhering
to the following steps:
1. Determine the dynamic model of the structure, hence the degrees of free-
dom, and define the nodal displacement vector uðt Þ and the nodal load vec-
tors pðt Þ.
2. Compute the mass matrix M and stiffness matrix K of the structure (see
Chapters 10 and 11).
3. Solve the eigenvalue problem ðK w2 MÞf ¼0 to compute the eigenfre-
quencies wn and mode shapes fn (Chapter 13).
4. Compute the modal masses Mn and modal loads Pn ðt Þ using Eqs. (14.2.4a),
(14.2.4d) and formulate Eq. (14.2.7) for the modal coordinate Yn ðt Þ.

5. Compute the initial conditions Yn ð0Þ, Y n ð0Þ from the natural initial condi-
tions uð0Þ, u_ ð0Þ using Eqs. (14.2.9a), (14.2.9b).
6. Compute the modal coordinates Yn ðt Þ either using Eq. (14.2.8) with analyt-
ical evaluation of the Duhamel integral or by direct numerical solution of
Eq. (14.2.7).
7. Compute the displacement vector uðt Þ from the modal contributions
fn Yn ðt Þ using the superposition relation (14.2.1).
8. Compute the nodal elastic forces f S ¼ Ku.
Example 14.2.1 Determine the dynamic response of the two-story shear frame
in Example 12.2.1 for the following two load cases:
(a) The horizontal loads p1 ¼ 190 kN and p2 ¼ 300 kN are suddenly applied
at t ¼ 0 while the structure is at rest.
(b) The ground undergoes the displacement ug ðt Þ ¼ 0:02 sin 10t.
More specifically, in case (a), compute the extreme values of the displacements,
shear forces, and bending moments at the top of the columns while in case (b),
compute the maximum shear force at the base of the frame and the maximum
overturning moment. Consider proportional damping with modal damping
ratios x1 ¼ 0:06, x 2 ¼ 0:04.
@CivilMethod
Solution
The structure properties, eigenfrequencies, and mode shapes are taken from
Example 12.2.1, namely

25 0 3826:5 3826:5
M¼ , K¼ ,
0 32 3826:5 9142:1

8:289 0:1697 0:1061
w¼ , F¼
19:236 0:0935 0:1499
Case (a)
The modal matrices are computed using Eq. (14.2.4a)

25 0 0:1697
M1 ¼ ½0:1697 0:0935 ¼1
0 32 0:0935

25 0 0:1061
M2 ¼ ½ 0:1061 0:1499 ¼1
0 32 0:1499
The load vector is

190
pðt Þ ¼
300
and Eq. (14.2.4d) gives the modal loads

190
P1 ¼ ½ 0:1697 0:0935 ¼ 4:1930
300

190
P2 ¼ ½ 0:1061 0:1499 ¼ 65:1290
300
The equations of motion for the modal coordinates Y1 ðt Þ, Y2 ðt Þ result from
Eq. (14.2.7), that is
Y€ 1 + 0:9947Y 1 + 68:7089Y1 ¼4:1930

Y€ 2 + 1:5389Y 2 + 370:0416Y2 ¼65:1290

Because the structure is at rest at t ¼ 0, we have uð0Þ ¼ 0, u_ ð0Þ ¼ 0 and

Eqs. (14.2.9a), (14.2.9b) give
fTn Muð0Þ
fTn Mu_ ð0Þ
Yn ð0Þ ¼ ¼0 Y n ð0Þ ¼ ¼0 n ¼ 1, 2
Mn Mn
The damped eigenfrequencies are
qffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffi
wD1 ¼ w1 1 x21 ¼ 8:2741, wD2 ¼ w2 1 x22 ¼ 19:2206
@CivilMethod
Thus, Eq. (14.2.8) gives

Z t
Pn
Yn ð t Þ ¼ exn wn ðtτÞ sin ½wDn ðt τÞdτ
Mn wDn 0
2 0 1 3
Pn 6 B xn C 7
¼ 2 41 @ cos wDn t + qffiffiffiffiffiffiffiffiffiffiffiffi sin wDn t Aexn wn t 5
M n wn 1x 2
n
The Duhamel integral was evaluated using MAPLE (see Eq. 3.4.7). Hence,
using the data of the problem, we obtain

Y1 ðt Þ ¼ 0:06125 0:06125 cos 8:2741t + 0:36817 102 sin 8:2741t e0:4973t

Y2 ðt Þ ¼ 0:17629 0:17629cos 19:2206t + 0:70572 102 sin 19:2206t e0:7694t
The displacements result from Eq. (14.2.1) for N ¼ 2

( ) ( ) ( )
u 1 ðt Þ 0:1697 0:1061
¼ Y1 ðt Þ + Y2 ðt Þ
u 2 ðt Þ 0:0935 0:1499
Fig. E14.1 shows the graphical representations of u1 ðt Þ and u2 ðt Þ in the

interval 0 t 5, from which we obtain
max u1 ¼ 0:0473 m, min u2 ¼ 0:04555 m
The shear forces at the top of columns 1 and 3 are
12EI 1 12EI 2
Q1 ¼ ðu1 u2 Þ, Q2 ¼ u2
h13 h23
u1(t)
u2(t)
t t
FIG. E14.1 Time history of u1(t) and u2(t) in Example 14.2.1.
@CivilMethod
Q1(t)
Q2(t)
t t
FIG. E14.2 Time history of Q1(t) and Q2(t) in Example 14.2.1
Fig. E14.2 shows the graphical representations of Q1 ðt Þ and Q2 ðt Þ in the

interval 0 t 5, from which we obtain
max |Q1 | ¼ 168:66 kN, max |Q2 | ¼ 11:3082 kN
The respective bending moments are
h1
max M1 ¼ max |Q1 | ¼ 295:16 kN m
2
h2
max M2 ¼ max |Q2 | ¼ 22:61 kN m
2
Case (b)
The equation of motion in terms of the relative displacements u 1 ¼ u1 ug ,
u 2 ¼ u2 ug (see Chapter 6) is
€ + Cu
Mu _ + K
u ¼
pð t Þ
where

1 50
ðt Þ ¼ M
p u€g ¼ sin 10t
1 64
The modal loads are

50
P1 ¼ ½ 0:1697 0:0935 sin 10t ¼ 14:4690 sin 10t
64

50
P2 ¼ ½ 0:1061 0:1499 sin 10t ¼ 4:2886 sin 10t
64
@CivilMethod
The initial conditions with respect to the relative displacements are

1 0
ð0Þ ¼ utot ð0Þ
u ug ð0Þ ¼
1 0

1 0:2
_ ð0Þ ¼ u_ tot ð0Þ
u u_ g ð0Þ ¼
1 0:2
The equations of motion for the modal coordinates result from
Eq. (14.2.7) as
Y€ 1 + 0:9947Y 1 + 68:7089Y1 ¼14:4690 sin 10t

Y€ 2 + 1:5389Y 2 + 370:0416Y2 ¼ 4:2886 sin 10t

The solution of the foregoing equations is obtained from Eq. (14.2.8). The
Duhamel integral is evaluated using MAPLE. Thus, setting Pn ðτÞ ¼ Pn sin w τ
we can obtain
Z t
Pn
Yn ð t Þ ¼ exn wn ðtτÞ sin w
τ sin ½wDn ðt τÞdτ
Mn wDn 0

Pn 20x n wn cos 10t + w2Dn + x2n w2n 100 sin 10t
¼
Mn x2 w2 + ðw 10Þ2 x2 w2 + ðw + 10Þ2
n n Dn n n Dn

x 2 w2 1000
20x n wn cos wDn t + 10wDn + 10 n n + sin wDn t exn wn t
Pn w wDn
+ Dn
Mn x n wn + ðwDn 10Þ
2 2 2
x n wn + ðwDn + 10Þ2
2 2
The initial conditions for the modal coordinates are

fT1 Muð0Þ fT2 Muð0Þ
Y1 ð 0Þ ¼ ¼ 0, Y 2 ð 0Þ ¼
¼0
M1 M2

25 0 0:2
Y 1 ð0Þ ¼ ½ 0:1697 0:0935 ¼ 1:4469
0 32 0:2

25 0 0:2
Y 2 ð0Þ ¼ ½ 0:1061 0:1499 ¼ 0:42886
0 32 0:2
Therefore, applying Eq. (14.2.8) for the data of the problem yields
Y1 ðt Þ ¼ 0:13348cos 10t 0:41996 sin 10t
+ ð0:1334 cos 8:2741t + 0:34063 sin 8:2741t Þe0:4973t
Y2 ðt Þ ¼ 0:9022 103 cos 10t 0:01583 sin 10t

+ 0:9022 103 cos 19:2206t + 0:03051 sin 19:2206t e0:7694t
@CivilMethod
The displacements result from Eq. (14.2.1) for N ¼ 2

u 1 ðt Þ 0:1697 0:1061
¼ Y 1 ðt Þ + Y2 ð t Þ
u 2 ðt Þ 0:0935 01499
The elastic forces applied at the roof levels are computed from
f S ¼ Ku
¼ KFYðt Þ

3826:5 3826:5 0:1697 0:1061 Y1 ðt Þ
¼
3826:5 9142:1 0:0935 0:1499 Y2 ðt Þ

291:5793Y1 ðt Þ + 979:584Y2 ðt Þ
¼
205:4293Y1 ðt Þ 1776:39244Y2 ðt Þ
Hence
fS1 ¼ 291:5793Y1 ðt Þ + 979:584Y2 ðt Þ
fS2 ¼ 205:4293Y1 ðt Þ 1776:3924Y2 ðt Þ
A particularly useful response parameter in earthquake analysis of buildings
is the base shear. This is the total horizontal force Q0 ðt Þ that must be resisted at
the foundation, and is found by summing the inertia (D’Alambert) forces over
the structure. These forces are equal to the elastic forces if damping is neglected.
Thus, we have
Q0 ðt Þ ¼ fS1 + fS2 ¼ 497:0Y1 ðt Þ 796:8084Y2 ðt Þ
and the overturning moment
M0 ðt Þ ¼ fS1 ðh1 + h2 Þ + fS2 h2 ¼ 3008:5619Y1 ðt Þ + 241:3104Y2 ðt Þ
The graphical representations of Q0 ðt Þ and M0 ðt Þ are shown in Fig. E14.3,
from which we obtain max jQ0 j ¼ 283:28 kN and max jM0 j ¼ 1783:83 kN m.
M0(t)
Q0(t)
t t
FIG. E14.3 Time history of Q0(t) and M0(t) in Example 14.2.1.
@CivilMethod
Example 14.2.2 Determine the steady-state response of the two-story shear

t at
frame in Example 12.2.1, when it is subjected to the load p1 ðt Þ ¼ po sin w
the roof level of the second floor. Assume: po ¼ 5 kN, w ¼ 10 s1 and (i)
x1 ¼ 0:06, x2 ¼ 0:04, and (ii) x 1 ¼ x 2 ¼ 0.
Solution
The load vector is
t0gT
pðt Þ ¼ f p0 sin w
Eq. (14.2.4d) gives the modal loads
t
P1 ¼ fT1 pðt Þ ¼ f11 p0 sin w
t
P2 ¼ fT2 pðt Þ ¼ f12 p0 sin w
f1n p0
Y€ n + 2xn wn Y n + w2n Yn ¼pn0 sin w

t, pn0 ¼ , n ¼ 1, 2 (1)
Mn
The steady-state response is obtained from Eq. (3.2.26) by setting u ¼ Yn
t qn Þ
Yn ðt Þ ¼ rn sin ðw (2)
where now
!
Pn0 h i 1
2 2 2x n b n
w
qn ¼ tan 1
2 2
rn ¼ 1 b n + ð 2x n b n Þ , , bn ¼
Mn w2n 1 b2n wn
(3)
The displacements of the steady state response result from Eq. (14.2.1).
t q1 Þ + f2 r2 sin ðw
u ¼ f1 r1 sin ðw t q2 Þ (4)
The eigenfrequencies and mode shapes are taken from Example 12.2.1,
namely
f1 ¼ f 0:1697 0:0935 gT , f2 ¼ f 0:1061 0:1499 gT

w1 ¼ 8:289, w2 ¼ 19:236, M1 ¼ M2 ¼ 1
(i) x 1 ¼ 0:06, x2 ¼ 0:04. We have

w
w
b1 ¼ ¼ 1:2064, b2 ¼ ¼ 0:5198
w1 w2
and Eqs. (3) give
r1 ¼ 0:02584, r2 ¼ 0:001961, q1 ¼ 0:3076, q2 ¼ 0:0569
@CivilMethod
Hence

u1 ðt Þ 0:438
¼ 102 sin ð10t + 0:3076Þ + 103
u2 ðt Þ 2:416

0:208
sin ð10t 0:0569Þ
0:294
(ii) x1 ¼ x2 ¼ 0. Eqs. (3) give
r1 ¼ 0:2561 102 , r2 ¼ 0:7636 103 , q1 ¼ q2 ¼ 0

Hence

u 1 ðt Þ 0:5156
¼ 103 sin 10t
u 2 ðt Þ 0:1250
14.3 Modal contribution in the mode superposition method

In the mode superposition method, the displacement vector uðt Þ results as a sum
of the individual modal displacements un , namely
X
N
uðt Þ ¼ un (14.3.1a)
n¼1
un ¼ fn Yn ðt Þ (14.3.1b)
The exact solution is obtained when the sum includes all the modal displace-
ments. However, the economy of the analysis requires the use of a few modal
displacements corresponding to the lower eigenmodes, at least those whose
contribution is not negligible, while cutting off higher order modal displace-
ments. Besides, the truncation is necessary for another important reason. The
modeling of the structure with finite elements reveals deviations between the
discretized and the actual structure, which become pronounced in higher-order
eigenmodes. Therefore, the additional error due to truncated higher eigenmodes
does not concern us very much. In order to estimate this error, we first need to
understand the contribution of each eigenmode to the overall response.
14.3.1 Modal participation

The forces produced by the modal component un ¼ fn Yn ðt Þ of the displace-
ment u are:
un ¼ Mfn Y€ n
The inertial force: ðf I Þn ¼ M€
The damping force: ðf D Þn ¼ Cu_ n ¼ Cfn Y n

The elastic force: ðf S Þn ¼ Kun ¼ Kfn Yn
@CivilMethod
Obviously, the resultant of the above forces equilibrates only a part pn ðt Þ of the
external force pðt Þ. Therefore, we have
pn ðt Þ ¼ ðf I Þn + ðf D Þn + ðf S Þn

¼ MY€ n + CY n + KYn fn

(14.3.2)
¼ Aðt Þfn
and consequently
X
N
pðt Þ ¼ pn ðt Þ (14.3.3)
n¼1
If the damping matrix C is proportional, for example, C ¼ a0 M + a1 K, then

the matrix Aðt Þ satisfies the relation

Aðt Þfn ¼ MY€ n + CY n + KYn fn

h i
¼ Y€ n + a0 + a1 w2n Y n + w2n Yn Mfn

(14.3.4)
¼ ðt ÞMfn
and Eq. (14.3.2) is written as
pn ðt Þ ¼ ðt ÞMfn
(14.3.5)
¼ ðt Þen
From the forgoing equation, it is evident that the vectors en ¼ Mfn
(n ¼ 1, 2, …, N ) do not change with time, therefore they provide a constant vec-
tor base to express the vector pðt Þ. Therefore, we can write
X
N
pðt Þ ¼ an ðt Þen (14.3.6)
n¼1
The coefficients an ðt Þ are evaluated using the orthogonality conditions

(12.3.10)

0 i 6¼ n
fTi en ¼ fTi Mfn ¼ (14.3.7)
Mn i ¼ n
Premultiplying Eq. (14.3.6) by fTn gives
fTn pðt Þ
an ðt Þ ¼ (14.3.8)
Mn
hence
fTn pðt Þ
pn ðt Þ ¼ en
Mn
(14.3.9)
fT pðt Þ
¼ n Mfn
Mn
@CivilMethod
The component pn ðt Þ of the load vector defined by the foregoing relation

produces the generalized forces according to Eq. (14.2.4d)
ðpn Þk ¼ fTk pn ðt Þ
fTn pðt Þ T (14.3.10)
¼ fk Mfn , k ¼ 1, 2, …, N
Mn
which by virtue of the orthogonality condition yields

Pn ð t Þ k ¼ n
ðpn Þk ¼
0 k 6¼ n
That is, only the component Pn ðt Þ does not vanish. Hence, we deduce that
pn ðt Þ excites only the nth eigenmode.
We shall examine now the case where the external load is of the form
pðt Þ ¼ Rf ðt Þ (14.3.11)
where R is a constant vector expressing the spatial distribution of the loading
and f ðt Þ a scalar function specifying the amplitude of the components pk ðt Þ
ðk ¼ 1, 2, 3, …, N Þ of pðt Þ with respect to the global system of coordinates
at time t. This type of loading is usual in structures, for example, in the case
of support excitation produced by seismic ground motion (see Section 14.7).
Substituting Eq. (14.3.11) into Eq. (14.3.3), gives
X
N
pðt Þ ¼ f ðt Þ Rn (14.3.12)
n¼1
and by virtue of Eq. (14.3.9)

fTn R
Rn ¼ Mfn
Mn (14.3.13)
¼ Gn Mfn
The quantity
fTn R
Gn ¼ (14.3.14)
Mn
provides a measure for determining the degree to which the nth eigenmode par-
ticipates in the dynamic response and it is called the modal participation factor
or simply the participation factor. However, this definition of Gn is not useful in
the study of the dynamic response of structures as it depends on how the modes
are normalized. Besides, it cannot be used as a measure of the modal contribu-
tion to a response quantity. Both drawbacks are overcome by introducing the
modal contribution described next [1].
When pðt Þ ¼ Rf ðt Þ, the equation of motion (14.2.7) for the modal coordi-
nate Yn is written as
@CivilMethod
Y€ n + 2xn wn Y n + w2n Yn ¼ Gn f ðt Þ

(14.3.15)
or
y€n + 2xn wn y_ n + w2n yn ¼ f ðt Þ (14.3.16)
where
Yn ¼ G n y n (14.3.17)
Eq. (14.3.16) is identical to Eq. (3.3.16) and expresses the motion of the
SDOF system when m ¼ 1, w ¼ wn , and pðt Þ ¼ f ðt Þ. The solution of
Eq. (14.3.16) gives yn ðt Þ and Eq. (14.3.1b) gives the modal displacement
un ¼ Gn fn yn ðt Þ (14.3.18)
The corresponding elastic force is
ðf S Þn ¼ Kun
(14.3.19)
¼ KGn fn yn ðt Þ
which, taking into account that Kfn ¼ w2n Mfn , becomes

ðf S Þn ¼ Gn Mfn w2n yn ðt Þ
(14.3.20)
¼ Rn w2n yn ðt Þ
The dynamic response at time t is obtained as the static responsea when the
structure is loaded by the elastic forces. We denote by qn ðt Þ the contribution of
the nth mode to a certain quantity q ðt Þ (representing deformation or stress).
Hence qn ðt Þ will be obtained as the static response of the structure produced
by the load ðf S Þn . If qnst denotes the static response due to static load Rn , then
we may write

qn ðt Þ ¼ qnst w2n yn ðt Þ (14.3.21)
The foregoing equation leads to the conclusion that the nth mode contribu-
tion qn ðt Þ is obtained as a result of two analyses: (1) the static analysis of the
structure subjected to the external load Rn , and (2) the dynamic analysis of the
nth mode SDOF system under the excitation force f ðt Þ. Therefore, the dynamic
analysis requires the static analysis of the structure by N loadings Rn ,
n ¼ 1, 2, …,N , and the dynamic analysis of N different SDOF systems under
the same excitation force f ðt Þ.
a. The term static response is used in the sense that time is involved as a parameter and no inertial
forces or damping forces are produced. This response is called also quasistatic.
@CivilMethod
The superposition of all response contributions gives the total response

X
N
q ðt Þ ¼ qn ðt Þ
n¼1
(14.3.22)
X
N 2
¼ qnst wn y n ð t Þ
n¼1
The previous modal analysis procedure provides a means to understand the

contribution of modes to the dynamic response of a quantity and it can be mea-
sured by defining the quantity
qnst X
N
g qn ¼ , q st ¼ qnst (14.3.23)
q st n¼1
Note that q st results as the static analysis of the structure subjected to the
load R. Evidently, this definition of g qn does not require the computation of
all qnst when the response is approximated by the first few mode shapes.
The quantity g qn defined by Eq. (14.3.23) is called the nth modal contribution
factor of the quantity q. This modal contribution factor has three important
properties, which show its advantage over the modal participation factor Gn
defined by Eq. (14.3.14), namely
(1) It is by definition dimensionless.
(2) It does not depend on how the modes are normalized because qnst is due to
the load Rn , which does not depend on the mode normalization method,
and the modal properties do not enter into q st .
(3) The sum of all participation factors is equal to unity
X
N
g qn ¼ 1 (14.3.24)
n¼1
On the basis of Eqs. (14.3.21), (14.3.23), the extreme value of qn ðt Þ is

qn0 ¼ g qn q st w2n yn0 (14.3.25)
where yn0 ¼ max t jyn ðt Þj.
The static response ynst ðt Þ is obtained from Eq. (14.3.16), if the inertial and
damping forces are neglected, that is,
f ðt Þ
ynst ðt Þ ¼ (14.3.26)
w2n
Hence its extreme value is
f0
ynst 0
¼ (14.3.27)
w2n
where f0 ¼ max t jf ðt Þj.
@CivilMethod
The quantity
yn0
Dn ¼ (14.3.28)
ynst 0
defines the dynamic magnification factor of the nth mode. Thus, Eq. (14.3.25)
by virtue of Eqs. (14.3.27), (14.3.28) becomes
qn0 ¼ g qn q st f0 Dn (14.3.29)
Eq. (14.3.29) shows that the extreme value of the nth mode contribution to
the quantity q ðt Þ is the product of four quantities:
(1) The dimensionless modal contribution factor g qn
(2) the static value of q due to the loading R
(3) the maximum value of the function f ðt Þ
(4) the nth mode dynamic magnification factor Dn .
It is emphasized that the quantities q st and g n depend only on the spatial
distribution R of the external forces while Dn and f0 only on f ðt Þ. When the
excitation is due to seismic ground motion, the quantity w2n yn0 , according to
Eq. (6.2.21), represents the pseudoacceleration, Spa ðTn , xn Þ ¼ w2n yn0 , hence
its value can be taken from the respective response spectrum (see Fig. 6.2.6).
Example 14.3.1 The chimney of the variable cross-section shown in
Fig. E14.4a is modeled with six beam elements of constant cross-section.
Compute the modal contribution factor of (i) the base shear force Qb , (ii)
the overturning moment Mb , and (iii) the top displacement u6 when the excita-
t. Assume: m1 ¼ 6m, m2 ¼ 5m,
tion is due to ground motion ug ðt Þ ¼ uo sin w
m3 ¼ 4m, m4 ¼ 3m, m5 ¼ 2m, m6 ¼ m, I1 ¼ 16I , I2 ¼ 11I , I3 ¼ 7I , I4 ¼ 4I ,
I5 ¼ 2I , and I6 ¼ I :
(a) (b) (c)

FIG. E14.4 Chimney in Example 14.2.2 (a); FEM model (b); Degrees of freedom (c).
@CivilMethod
Solution
The equation of motion is (see Section 14.7).
u + Cu_ + Ku ¼ M1u€g
M€ (1)
T
where 1 ¼ f 1 1 1 1 1 1 g , u€g ðt Þ ¼ t
2
w u0 sin w
1. Formulation of the matrices M, K and vector R
The structure has 12 degrees of freedom, 6 translational, and 6 rotational
(Fig. E14.4c). After the static condensation of the rotational degrees of freedom,
we obtain:
2 3
6 0 0 0 0 0
60 5 0 0 0 07
6 7
60 0 4 0 0 07
M ¼ m6 60 0 0 3 0 07
7 (2)
6 7
40 0 0 0 2 05
0 0 0 0 0 1
2 3
232:80 121:78 37:89 7:11 1:21 0:15
6 121:78 122:97 69:99 21:01 3:57 0:44 7
6 7
6 37:89 69:99 72:03 39:04 10:73 1:34 7
K ¼ k66 7 (3)
6 7:11 21:01 39:04 38:38 19:04 3:88 7 7
4 1:21 3:57 10:73 19:04 16:00 5:00 5
0:15 0:44 1:34 3:88 5:00 2:12
where
EI
k¼
ðL=6Þ3
Moreover, it is
8 9
> 6>
> >
> >
>
> 5>>
< >
> =
4
R ¼ M1 ¼ m , f ðt Þ ¼ u€g ðt Þ (4)
>
> 3>>
>
> >
>
>2>
>
: >
;
1
2. Computation of eigenfrequencies and eigenmodes

Using the MATLAB function [V, D]¼eig (K,M), we obtain
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
w1 ¼ 0:2030 k=m , w2 ¼ 0:7570 k=m , w3 ¼ 1:7928 k=m
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
w4 ¼ 3:3067 k=m , w5 ¼ 5:1964 k=m , w6 ¼ 7:9089 k=m
@CivilMethod
2 3
0:0161 0:0547 0:1123 0:1665 0:2106 0:2806
6 0:0645 0:1747 0:2478 0:1858 0:0205 0:2626 7
6 7
1 6 6 0:1469 0:2736 0:1305 0:1863 0:2569 0:1893 77
F ¼ pffiffiffiffiffi 6 7
m 6 0:2640 0:2310 0:2348 0:1831 0:3320 0:1067 7
6 7
4 0:4133 0:0662 0:2981 0:4140 0:2491 0:0498 5
0:5835 0:6010 0:4539 0:2807 0:1148 0:0178
(5)
3. Computation of the vectors Rn

The eigenmodes have been orthonormalized with respect to the mass. Hence,
Mn ¼ 1, n ¼ 1, 2, …, 6. Then Eq. (14.3.14) gives
pffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffi
G1 ¼ 3:2094 m , G2 ¼ 2:2555 m , G3 ¼ 1:5881 m
G4 ¼ 1:1813 m , G5 ¼ 0:9509 m , G6 ¼ 0:8893 m
The vectors Rn are computed from Eq. (14.3.13)
R1 =m R2 =m R3 =m R4 =m R5 =m R6=m
0:311 0:739 1:070 1:180 1:201 1:497
1:036 1:970 1:967 1:098 0:097 1:168
1:886 2:468 0:829 0:880 0:977 0:673
2:542 1:563 1:119 0:649 0:947 0:284
2:653 0:298 0:947 0:978 0:474 0:088
1:873 1:355 0:721 0:333 0:109 0:016
st st st
4. Computation of Qbn , Mbn , and u6n
The base shear is obtained as the sum of all the elastic forces while the over-
turning moment as the sum of the moments of all the elastic forces with respect
to the base of the chimney. Thus, we have
X
6
st
Qbn ¼ Rkn ¼ 1T Rn
k¼1
X
6
L
st
Mbn ¼ hk Rkn ¼ hT Rn , hT ¼ f 1 2 3 4 5 6 g
k¼1
6
The vector ust

n is computed by considering the static analysis of the structure
1 Gn
subjected to the external force Rn , that is, ust
n ¼ K Rn ¼ w2 fn . Hence
n
Gn
st
u6n ¼ f :
w2n 6n
@CivilMethod
Finally, the modal contribution factors are computed from Eq. (14.3.23):
st st st
Qbn Mbn u6n
n ¼
gQ , n ¼
gM , g un6 ¼
b b
X
6 X
6 X
6
st st st
Qbk Mbk u6k
k¼1 k¼1 k¼1
Their values are given in Table E14.1. From this table, we deduce that each
quantity requires a different number of contributing modes to achieve a spec-
ified degree of approximation. For example, four modes for the base shear, two
modes for the overturning moment, and one mode for the top displacement
should contribute in order to achieve an approximation over 90%.
TABLE E14.1 Modal contribution factors in Example 14.3.1.
Number Base shear force Overturning moment Top displacement

of mode P
K P
K P
K
gQ b
gM b
g un6
shape n gQ
n
b
n¼1
n
gM
n
b
n¼1
n
g un6 n¼1
1 0.4905 0.4905 0.7627 0.7627 1.0500 1.0500

2 0.2423 0.7328 0.1556 0.9183 0.0549 0.9954
3 0.1201 0.8529 0.0466 0.9648 0.0052 1.0006
4 0.0664 0.9193 0.0186 0.9834 0.0007 0.9999
5 0.0430 0.9623 0.0096 0.9930 0.0000 1.0000
6 0.0377 1.0000 0.0070 1.0000 0.0000 1.0000
Example 14.3.2 For the structure in Example 14.3.1, compute the extreme
values Qbn0 , Mbn0 , u6n0 when w ¼ 1:5w1 , m ¼ 31:056 kN m1 s2 ,
k ¼ 5486:208 kN=m, L ¼ 75 m, x1 ¼ 0:06, and x2 ¼ x3 ¼ ⋯ ¼ x6 ¼ 0:04,
u0 ¼ 1.
Solution
The extreme value qn0 of a quantity q is obtained from Eq. (14.3.29), namely
qn0 ¼ g qn q st f0 Dn
The modal contribution factors g qn (q Qb , Mb , u6 ) were computed in
Example 14.3.1 and are given in Table E14.1. It is f ðt Þ ¼ u0 w t, hence
2 sin w
. The quantities Qb , Mb are obtained from the relations:
f0 ¼ max jf ðt Þj ¼ u0 w 2 st st
X
6
Qbst ¼ Rk ¼ 1T R ¼ 652:1760 kN (1)
k¼1
@CivilMethod
X
6
Mbst ¼ hk Rk ¼ hT R ¼ 21739:2 kN m (2)
k¼1
The vector ust is obtained from the relation
ust ¼ K1 R ¼ f 0:0088 0:0331 0:0711 0:1209 0:1802 0:2451 gT
Hence
u6st ¼ 0:2451 (3)
The dynamic magnification factors Dn are computed from Eq. (14.3.28).

This requires the solution of Eq. (14.3.16), which is given by Eq. (14.2.8).
The initial conditions for the relative displacements are:
uð0Þ ¼ utot ð0Þ ug ð0Þ ¼ 0 u_ ð0Þ ¼ u_ tot ð0Þ u_ g ð0Þ ¼
wu0 1:
Therefore, the initial conditions yn ð0Þ, y_ n ð0Þ result from Eqs. (14.2.9a),
(14.2.9b) with Mn ¼ 1, n ¼ 1, 2, …,6:
Yn ð0Þ fTn Muð0Þ

yn ð0Þ ¼ ¼ ¼ 0, n ¼ 1, 2, …, 6 (4a)
Gn Gn

T
_ g ð0Þ
f n M u fTn M1
y_ n ð0Þ ¼ Y n ð0Þ ¼ Gn ¼
wu0 , n ¼ 1, 2, …,6 (4b)
Gn
or taking into account that Gn ¼ fTn R ¼ fTn M1, we obtain
y_ n ð0Þ ¼
wu0 , n ¼ 1, 2, …,6 (5)
Then Eq. (14.2.8) gives for Mn ¼ 1, Pn ðt Þ ¼ u0 w t

2 sin w

yn ðt Þ xn wn t
w
¼e sin wDn t
u0 wDn
Z (6)
2 t
w
τexn wn ðtτÞ sin ½wDn ðt τÞdτ
sin w
wDn 0
The eigenfrequencies wn are taken from Example 14.3.1, which for data of
this problem become
w1 ¼ 2:6977, w2 ¼ 10:0608, w3 ¼ 23:8283

w4 ¼ 43:9498, w5 ¼ 69:0658, w6 ¼ 105:1181
@CivilMethod
R1(t)
R2(t)
t t
R3(t)
R4(t)
t t
R5(t)
R6(t)
t t
FIG. E14.5 Time history of the response ratios and modal magnification factors in Example 14.4.2.
Eq. (6) demands the analytical evaluation of the Duhamel integral, which
leads to a complicated expression. To avoid this, the solution is obtained numer-
ically using the program called aem_lin.m described in Section 4.4 while the
modal dynamic magnification factors are obtained from the relation
Dn ¼ max ðjRn jÞt , where Rn ðt Þ ¼ yn =ðyn Þst ðn ¼ 1, 2, …, 6Þ represents the
respective response ratio within the integration interval. Their graphical
@CivilMethod
representation is shown in Fig. E14.5, in which the dynamic magnification fac-

tors are annotated. Table E14.2 shows the extreme values Qbn0 , Mbn0 , u6n0
TABLE E14.2 Extreme values of Qbn0 , Mbn0 , and u6n0 in Example 14.3.2.
Qbn0 ¼ Mbn0 ¼ u6n0 ¼

n tn Dn gQb st
n Qb f0 Dn gM st
n Mb f0 Dn
b
g un6 u6st f0 Dn
1 1.819 2.084 10916.50 565819.0 8.78
2 0.171 2.575 6663.12 142631.0 0.56
3 0.067 5.628 7218.45 93361.1 0.11
4 0.036 10.263 7277.62 67953.7 0.03
5 0.023 16.082 7385.09 54958.8 0.00
6 0.016 24.435 9837.86 60888.6 0.00
Example 14.3.3 The chimney of Example 14.3.1 is subjected to horizontal

loading pðt Þ ¼ Rf ðt Þ, where R ¼ lfk , that is, the spatial distribution is propor-
tional to the kth eigenmode when l is a constant. Compute the modal contribu-
tion factors of the base shear when k ¼ 1, 2.
Solution
The vector Rn is obtained from Eq. (14.3.13), namely
Rn ¼ Gn Mfn
where
Gn ¼ fTn R ¼ lfTn fk n ¼ 1, 2, …,6
In general it is 6¼ 0. Hence, Rn 6¼ 0 and the loading pðt Þ excites all the
fTn fk
mode shapes, in contrast to the free vibrations where only the corresponding
mode shape is excited when the initial conditions are proportional to a certain
of its mode shapes (see Example 12.9.2). The mass matrix, the stiffness matrix,
and the mode shapes are taken from Example 14.3.1.
(i) R ¼ lf1
Computation of Rn and g Q n
The mode shapes have been orthonormalized with respect to the mass, hence
Mn ¼ 1, n ¼ 1, 2, …, 6. Eq. (14.3.14) gives:
G1 ¼ 0:6071l m , G2 ¼ 0:2647l m , G3 ¼ 0:1167l m
G4 ¼ 0:0537l m , G5 ¼ 0:0186l m , G6 ¼ 0:0025l m
@CivilMethod
The vectors Rn are obtained from Eq. (14.3.13):

lR1 =m lR1 =m lR1 =m lR1 =m lR1 =m lR1 =m
0:0588 0:0868 0:0786 0:0536 0:0235 0:0043
0:1959 0:2312 0:1445 0:0499 0:0019 0:0034
0:3568 0:2897 0:0609 0:0400 0:0191 0:0019
0:4808 0:1835 0:0822 0:0295 0:0185 0:0008
0:5018 0:0351 0:0695 0:0445 0:0093 0:0002
0:3542 0:1591 0:0529 0:0151 0:0021 0:0000
The contribution factors are obtained from Eq. (14.3.23)
n ¼ ½1:30894, 0:40112, 0:12445, 0:04263, 0:01188, 0:00153

gQ
(ii) R ¼ lf2
Computation of Rn and g Qn
Eq. (14.3.14) gives:
G1 ¼ 0:2647l m , G2 ¼ 0:5273l m , G3 ¼ 0:22218l m
G4 ¼ 0:0895l m , G5 ¼ 0:0309l m , G6 ¼ 0:0040l m
Eq. (14.3.13) gives:
lR1 =m lR2 =m lR2 =m lR2 =m lR2 =m lR2 =m
0:0256 0:1729 0:1497 0:0894 0:0391 0:0067
0:0854 0:4605 0:2752 0:0832 0:0031 0:0052
0:1555 0:5770 0:1159 0:0667 0:0317 0:0030
0:2096 0:3654 0:1565 0:0491 0:0308 0:0012
0:2188 0:0698 0:1324 0:0741 0:0154 0:0003
0:1544 0:3168 0:1008 0:0251 0:0035 0:0000
and using Eq. (14.3.23), we obtain
n ¼ ½12:7194, 17:8053, 5:2824, 1:5837, 0:4405, 0:0532

gQ
14.3.2 Static correction method

When the time function f ðt Þ in Eq. (14.3.11) is harmonic, that is,
t, the dynamic load magnification factor Dn approaches the unity
f ðt Þ ¼ f0 sin w
for small values of the ratio b n ¼ w , where Tn ¼ 2p=wn is the nat-
=wn ¼ Tn =T

ural period of the nth eigenmode and T ¼ 2p= w is the period of the loading. This
becomes evident from Fig. 3.2.3. The same applies when the ratio Tn =ttot is
very small or the ratio ttot =Tn is too large, where ttot is the duration of loading.
This is shown in Fig. 14.3.1, which represents the variation of the dynamic mag-
nification factor Dn versus the ratio ttot =Tn when the loading is the acceleration
of the Athens earthquake, September 7, 1999, f ðt Þ ¼ u€g ðt Þ, of total duration
@CivilMethod
ttot ¼ 39 s. We observe that for ttot =Tn > 1000, hence for periods Tn < 0:039 s
or eigenfrequencies wn > 160s1 , we may ignore the dynamic effect and
assume Dn 1 for eigenmodes of order n > Nd , where Nd is the number of
eigenmodes with Tn > 0:039 s and Dn noticeably greater than the one. This
observation is the quintessence of the static correction method, presented next.
Suppose that we use N eigenmodes to approximate the dynamic response of
a quantity q ðt Þ. We split the contribution of eigenmodes into two parts. The first
Nd eigenmodes with Dn greater than one and the remaining N Nd with Dn
close to one. Then we can write
X
Nd
X
N
q ðt Þ ¼ qnst w2n yn ðt Þ + qnst w2n ynst ðt Þ (14.3.30)
n¼1 n¼Nd + 1
where yn ðt Þ is taken from the solution of Eq. (14.3.16) and ynst ðt Þ from
Eq. (14.3.26). Substituting Eq. (14.3.26) into Eq. (14.3.30) and using
Eq. (14.3.23) we obtain
X
Nd X
N
q ðt Þ ¼ qnst wn2 yn ðt Þ + f ðt Þ qnst
n¼1 n¼Nd + 1
( ) (14.3.31)
X
Nd 2 X
N
¼q st
gqn wn yn ðt Þ + f ðt Þ gqn
n¼1 n¼Nd + 1
3.5
3 x =0.05
x =0.1
2.5 x =0.15
2
Dn
1.5
0.5
0
0 200 400 600 800 1000 1200 1400 1600
t tot /Tn
FIG. 14.3.1 Variation of the dynamic magnification factor Dn versus the ratio ttot/Tn
Eq. (14.3.24) gives

X
N X
Nd
g qn ¼ 1 g qn (14.3.32)
n¼Nd + 1 n¼1
@CivilMethod

( !)
X
Nd
X
Nd
q ðt Þ ¼ q st g qn w2n yn ðt Þ + f ðt Þ 1 g qn (14.3.33)
n¼1 n¼1
or
( )
X
Nd

q ðt Þ ¼ q st f ðt Þ + g qn w2n yn ðt Þ f ðt Þ (14.3.34)
n¼1
It becomes obvious from the foregoing equation that the computation of q ðt Þ

requires the determination of only Nd eigenfrequencies and eigenmodes.
In Eq. (14.3.33), the term
!
X
Nd
f ðt Þ 1 q
gn (14.3.35)
n¼1
represents the static correction that should be added to the dynamic solution in
order to obtain the total dynamic response. This method is known as the static
correction method.
This static correction method is very useful when the contribution of
higher eigenmodes to a quantity q ðt Þ cannot be neglected in dynamic analysis,
but the corresponding dynamic magnification factors produced by the excita-
tion force f ðt Þ are close to one. In such cases, the combination of the dynamic
response of a few lower eigenmodes together with the static correction will
give results, which adequately approximate the response obtained by the mode
superposition method. The advantage of the static correction method offers a
significant reduction in the computational task, especially when the equations
are solved numerically. The example that follows illustrates the merits of this
method.
Example 14.3.4 The chimney in Example 14.3.1 is subjected to the wind blast
load pðt Þ ¼ Rf ðt Þ R ¼ f 1:5 6:0 15:0 25:0 40:0 60:0 gT . The time function
f ðt Þ is given in Fig. E14.6 (p0 ¼ 1 t1 ¼ 0:1). Compute the response of the base
shear force, the overturning moment, and the top displacement using the exact
mode superposition method as well as the static correction method by including
the dynamic response of the eigenmodes with Dn > 1:15. Compare the extreme
values obtained from the two methods. The equation of motion will be solved
numerically to obtain yn ðt Þ. L ¼ 75 m, x1 ¼ 0:06, x2 ¼ x3 ¼ ⋯ ¼ x6 ¼ 0:04,
m ¼ 31:056 kN m1 s2 , and k ¼ 5486:208 kN=m.
@CivilMethod
f(t)
t
FIG. E14.6 Time variation of the blast load in Example 14.3.4.
Solution
The mass matrix, stiffness matrix, eigenfrequencies, and mode shapes are taken
from Example 14.3.1 for the given values of k and m. The vectors Rn are
obtained from Eq. (14.3.13) with Mn ¼ 1. Thus, we obtain
R1 R2 R3 R4 R5 R6
5:888 9:085 8:7984 6:284 2:279 0:096
19:604 24:191 16:170 5:843 0:184 0:075
35:708 30:311 6:8147 4:685 1:853 0:043
48:124 19:197 9:196 3:454 1:796 0:018
50:225 3:6693 7:782 5:206 0:899 0:005
35:456 16:644 5:925 1:765 0:207 0:001
st st st
The quantities Qbn , Mbn , and u6n are computed from the relations
X
6
st
Qbn ¼ Rkn ¼ 1T Rn
k¼1
X
6
L
st
Mbn ¼ hk Rkn ¼ hT Rn , hT ¼ f 1 2 3 4 5 6 g
k¼1
6
Gn
st
u6n ¼ f
w2n 6n
The modal contribution factors of Qb , Mb are obtained from Eq. (14.3.23). The
computed values are given in Table E14.3. The dynamic magnification factors
are computed from the relation Dn ¼ max jRðt Þj, where Rðt Þ ¼ yn ðt Þ= ynst 0 is
t
the response ratio of the nth mode and ynst 0 ¼ f0 =w2n . For the given loading it
is f0 ¼ 1. Hence
@CivilMethod
st
y1 0 ¼ 0:1374, y2st 0
¼ 0:9870 102 , y3st 0
¼ 0:1761 102
st st
y4 0 ¼ 0:5170 103 , y5 0 ¼ 0:2096 103 , y6st 0
¼ 0:9045 104
TABLE E14.3 Modal contribution factors in Example 14.3.4.

Overturning moment Top
Base shear force Qb Mb displacement u6
Mode P
K P
K P
K
gQ
n
b
gM
n
b
g un6
shape n gQ
n
b
n¼1 gM
n
b
n¼1 g un6 n¼1
1 1.3221 1.3221 1.1254 1.1254 0.9652 0.9652

2 0.4235 0.8985 0.1489 0.9765 0.0325 0.9977
3 0.1405 1.0391 0.0298 1.0063 0.0021 0.9998
4 0.0503 0.9887 0.0077 0.9986 0.0002 0.9999
5 0.0116 1.0003 0.0014 1.00004 0.0000 0.9999
6 0.0003 1.0000 0.00004 1.0000 0.0000 1.000
Qbst ¼ 147:5 kN Mbst ¼ 8981:25 kN m u6st ¼ 0:156 m
The values max jyn ðt Þj are obtained from the solution of Eq. (14.3.16),
t
where f ðt Þ is obtained from Fig. E14.6. The initial conditions are
uð0Þ ¼ u_ ð0Þ ¼ 0, which yields yn ð0Þ ¼ y_ n ð0Þ ¼ 0. The solution of the equation
of motion is obtained numerically using the program aemlin.m developed in
Section 4.4. Thus, we obtain
D1 ¼ 0:2461, D2 ¼ 0:8803, D3 ¼ 1:4990
D4 ¼ 1:4787, D5 ¼ 1:1482, D6 ¼ 1:1169
We observe that the dynamic magnification factors of the last two modes
are close to one, so we can include the two higher modes in the static correc-
tion and obtain the total response from Eq. (14.3.34) with Nd ¼ 4. That is
( )
X
4 2
q ðt Þ ¼ q f ðt Þ +
st
g n wn yn ðt Þ f ðt Þ
q
n¼1
@CivilMethod
Static correction
Exact
Qb (t)
Static correction
Exact
Mb (t)
Static correction
Exact
u6(t)
FIG. E14.7 Time history of Qb , Mb , u6 .
where q Qb ,Mb , u6 . Fig. E14.7 gives the graphical representation of

Qb ,Mb ,u6 . They were obtained from the foregoing equation as well as from
the exact method, that is, by including all eigenmodes. The extreme values
max jQb j, max jMb j, and max ju6 j together with the instants tmax they occur
@CivilMethod
are given in Table E14.4. It becomes obvious that the static correction method
provides very good results.
TABLE E14.4 Extreme values max jQb j, max jMb j, max ju6 j in Example 14.3.4.
Base shear force Overturning moment Top displacement
max jQb j max jMb j max ju6 j
Static Static Static
Exact correction Exact correction Exact correction
Extreme 92.4876 91.500 3341.2 3341.4 0.0384 0.0384
value
tmax 0.634 0.634 0.620 0.620 0.751 0.751
14.3.3 Error in mode superposition method due to truncation

of higher modes
From the results in Examples 14.3.1 and 14.3.3, we conclude that for a given
spatial distribution R of the external load, the contribution of eigenmodes is
not the same for all quantities q ðt Þ. For instance, in Example 14.3.1, the con-
tribution of one eigenmode of the top displacement is sufficient to reach
100% of the total response while the contribution of five eigenmodes of the base
shear force and three eigenmodes of the overturning moment are required to
cover 95% of their total response. Similar findings result from Example
14.3.3, that is, more eigenmodes of the base shear force than of the top displace-
ment or the overturning moment should contribute in order to obtain the same
approximation of the response. However, we cannot reach decisive conclusions
because the above findings apply to structures of a specific geometry and sup-
port conditions (here, cantilever). But even in such cases, as can be seen from
Eq. (14.3.22), the modal contribution depends also on the factor w2n yn ðt Þ, that is,
the pseudoacceleration, which is influenced by the dynamic characteristics of
the eigenmode and the time function f ðt Þ with an extreme value f0 Dn . The
results in Example 14.3.2 illustrate this influence. Therefore, the answer to
the question as to how many eigenmodes must be included in the dynamic anal-
ysis by the mode superposition method to obtain an accurate solution is: Include
all of them at the first instance.
However, the need to reduce the computational task requires the truncation
of eigenmodes. We can adequately approximate the value of a quantity q ðt Þ by
including modes with a large modal contribution factor g qn and a dynamic mag-
nification factor Dn appreciably greater than one. For the eigenmodes with large
values of g qn but with Dn 1, the method of static correction limits the dynamic
analysis to a few lower eigenmodes. It should be noted that the extreme value
@CivilMethod
max jq ðt Þj cannot be obtained as a superposition of max jqn ðt Þj ¼ q st g qn f0 Dn

t t
because the values Dn do not occur at the same time. A measure to estimate
the truncation error produced by omitting higher-order eigenmodes may result
from the following consideration.
Let K N be the number of eigenmodes that we include in the superposi-
tion. This means that the displacement vector is approximated by the sum
X
K
u¼ un
n¼1
(14.3.36)
X
K
¼ fn Yn
n¼1
Substituting u into Eq. (14.1.1) gives

K
X
MY€ n + CY n + KYn fn ¼ e

pðt Þ (14.3.37)
n¼1
Obviously, it is e
pðt Þ 6¼ pðt Þ, when K < N . Taking into account Eq. (14.3.2),
X
K
pn ðt Þ ¼ e
pðt Þ (14.3.38)
n¼1
Namely, the K < N modal forces equilibrate a part of the external force
pðt Þ. This results in an error
e¼pe
p (14.3.39)
Then the error norm eK for e can be defined as
pT e
eK ¼ (14.3.40)
pT p
This gives eK ¼ 0, if K ¼ N , and e ¼ 1, K ¼ 0.
Obviously, when the loading is given by Eq. (14.3.11), the error takes the form
RT e
eK ¼ (14.3.41)
RT R
where now
X
K
e¼R Rn
n¼1
(14.3.42)
X
K
¼R Gn Mfn
n¼1
Table E14.5 gives the mode truncation error in Examples 14.3.1 and 14.3.3.
@CivilMethod
TABLE E14.5 Truncation error e ¼ RT e=RT R.
Number Example 14.3.3

of modes K Example 14.3.1 k ¼1 k ¼2
1 0.6770 3.275e5 1.000
2 0.3814 2.325e5 5.409e5
3 0.2160 2.000e5 8.891e5

4 0.1201 2.008e5 9.377e5
5 0.0565 2.057e5 9.505e5
6 3.59e11 2.048e5 9.485e5
14.4 Reduction of the dynamic degrees of freedom

The modeling of structures with finite elements leads to MDOF systems with a
large number of degrees of freedom. Although in the static analysis, a fine dis-
cretization is necessary to determine accurately the deformation and stress of
the structure, in dynamic analysis reliable results can be obtained without a fine
discretization. On the contrary, the large number of degrees of freedom makes
the dynamic analysis laborious and complicated because it requires the deter-
mination of a large number of eigenfrequencies and eigenmodes, whose contri-
bution to the response may be negligible.
We can use the discretization employed for the static analysis also for the
dynamic analysis, but after reducing considerably the degrees of freedom.
The usual methods of achieving that are: (a) by the static condensation,
(b) by imposing kinematic constraints, and (c) by the Rayleigh-Ritz method.
14.4.1 Static condensation

In the directions of some displacements of a structure, the inertial forces are zero
or very small, and therefore they can be neglected. The equations of motion in
these directions degenerate into static equations and are used to eliminate the
corresponding displacements. This is common when simulating structures with
finite elements. Because of the lumped assumption, the rotational inertia of the
concentrated masses is zero, hence the corresponding inertial quantities are also
zero. Therefore, the rotational degrees of freedom, though being necessary to
approximate accurately the stiffness of the structure, have a negligible contri-
bution to the dynamic response. Apparently, static condensation significantly
reduces the degrees of freedom. The static condensation and its implementation
procedure have already been described in detail in Section 11.4.
@CivilMethod
14.4.2 Kinematic constraints

The geometry of a structure and the properties of parts or members of it allow
for the introduction of the kinematic constraints, which express the displace-
ments of certain degrees of freedom in terms of a smaller set of them. We
encountered this problem in Section 11.7, where we discussed the axial con-
straints in frames, or in Section 11.11, where the general problem of rigid bodies
in flexible structures was studied. The importance of this reduction is illustrated
with the following two examples.
We consider the frame plane of Fig. 14.4.1, in which the infill wall B con-
tributes to the stiffness of the frame. The frame without the wall has 9 3 ¼ 27
degrees of freedom. The response of the wall is that of a plane elastic body.
Because its deformation is very small compared to that of the frame, it can
be considered as a plane rigid body. Therefore, the nodes a,b,c, d are con-
strained by the rigid body B and the 4 3 degrees of freedom are reduced to
3, thus the remaining degrees of freedom are 18ð¼ 5 3 + 3Þ. The static con-
densation of the 5 unconstrained rotational degrees of freedom reduces the
degrees of freedom to 18 5 ¼ 13. Finally, the axial constraints of the 11 beam
elements reduce the degrees of freedom to 13 11 ¼ 2. Hence, the remaining
degrees of freedom are the two horizontal displacements u1 ð¼ u2 Þ and u3 .
FIG. 14.4.1 Plane frame with an infill wall.
As a second example, we consider the 3-story building of Fig. 14.4.2 con-

sisting of 6 plane frames, 3 in the x-direction and 3 in the y-direction, and 3 hor-
izontal slabs (plates). The 9 vertical columns together with the beams
supporting the plates are simulated with a space frame of total nodes n ¼ 27,
each having 6 degrees of freedom, 3 translational and 3 rotational. Thus the
number of degrees of freedom are N 0 ¼ 27 6 ¼ 162. The floor slabs, also
called diaphragms, although flexible in the vertical direction, are usually very
stiff in their own plane and can be assumed as rigid plane bodies. This
assumption restrains the displacements in the xy-plane and rotations about
the vertical axis at each floor level and reduces their number to 3, that is,
equal to the degrees of freedom of a plane rigid body moving in its plane
@CivilMethod
(see Example 1.5.6). Therefore, 3 9 ¼ 27 degrees of freedom within the xy

plane are reduced to 3 and the total degrees of freedom are reduced to
N 00 ¼ 162 3 3 9 + 3 3 ¼ 90. Further, neglecting the axial deformation
of the columns, an assumption valid for a building of small height compared
to the other two dimensions, introduces additional 3 9 ¼ 27 constraints, which
reduce the degrees of freedom to N 000 ¼ 90 3 9 ¼ 63. Finally, static conden-
sation allows the elimination of the rotational degrees of freedom about x and y,
which amass 2 9 ¼ 18 degrees of freedom at each floor. Hence, the active
degrees of freedom are limited to N ¼ 63 3 2 9 ¼ 9, that is, equal to the
number of degrees of freedom of the three slabs. The procedures of reducing
the degrees of freedom were presented in Section 11.6.
FIG. 14.4.2 Three-story building with diaphragms.
14.5 Rayleigh-Ritz method

14.5.1 Ritz transformation
In Section 10.5.1, we described the Ritz method as the method of global shape
functions to approximate the dynamic response of continuous systems. In that
method, the function u ðx, t Þ representing the displacement of the points of the
structure was approximated by the superposition
X
K
u ðx, t Þ i ðx Þui ðt Þ (14.5.1)
i¼1
where i ðx Þ are functions representing deformation patterns of the structure and

ui ðt Þ time functions that play the role of generalized coordinates and can rep-
resent displacements of certain points or rotations of cross-sections. In order
that function u ðx, t Þ approximates the actual deformation of the structure,
the functions i ðx Þ must be (i) geometrically admissible and (ii) linearly inde-
pendent. Besides, they must be continuously differentiable up to the degree
required by the equation in which they are involved.
The Ritz method can also be applied to MDOF systems. Indeed, its applica-
tion to these systems does not encounter the difficulties of continuous systems,
where the establishment of shape functions is a very difficult mathematical
@CivilMethod
problem, especially for bodies of arbitrary shape [2]. In MDOF systems, the
shape functions are replaced by a set of discrete displacements, which define
admissible deformation patterns of the structure. Ritz applied this method in
1909 as an extension of Rayleigh’s method to establish the lower eigenfrequen-
cies, therefore it is also known as the Rayleigh-Ritz method.
For that reason, in the Ritz method for the MDOF systems, the displacement
vector is expressed as the superposition of a number of linearly independent
vectors representing possible deformation patterns of the structure. These vec-
tors are referred to as Ritz vectors. Thus, we can write
u ¼ y1 z1 ðt Þ + y2 z2 ðt Þ + ⋯ + yK zK ðt Þ
(14.5.2)
¼ xz
where z represent the vector of K generalized coordinates and C the matrix of
the Ritz vectors with dimensions N K ðK N Þ. The transformation defined
by Eq. (14.5.2) is known as the Ritz transformation. Apparently, the transfor-
mation u ¼ FY we met in Section 12.9 represents a special case of the Ritz
transformation when the vibration modes are used as the Ritz vectors.
Substituting the vector u from Eq. (14.5.2) into Eq. (14.1.1) gives
Mx€z + Cx z_ + Kxz ¼pðt Þ (14.5.3)
which, when premultiplied by x , gives
T
M€ e z + Kz
e z + C_ e ¼e
pðt Þ (14.5.4)
where
M e ¼ x T Cx, K
e ¼ x T Mx, C e ¼ x T Kx, e
pðt Þ ¼ x T pðt Þ (14.5.5)
Eq. (14.5.4) represents a system of K differential equations for the gener-
alized coordinates zk ðt Þ. The Ritz vectors generally do not satisfy the orthogo-
nality conditions, so consequently the matrices defined by Eq. (14.5.5) are not
diagonal; hence Eq. (14.5.4) is coupled. Nevertheless, their solution can be
obtained using the method of superposition presented in Section 14.2. Thus,
we have first to establish the eigenfrequencies w e n and eigenmodes zn from
the solution of the eigenvalue problem

Ke w e z¼0
e2 M (14.5.6)
Then the transformation

e ðt Þ
z ¼ ΖY (14.5.7)
reduces Eq. (14.5.4) to K SDOF equations, whose solution gives the modal
coordinates Ye ð t Þ ¼ fY e 2 ðt Þ ⋯ Y
e 1 ðt Þ Y e K ðt ÞgT .
The natural displacements result from Eq. (14.5.2), which by virtue of
Eq. (14.5.7) gives
@CivilMethod
e ðt Þ
u ¼ xΖY
(14.5.8)
¼FeY
e ðt Þ
where it was set

e ¼ xΖ
F (14.5.9)
or
e n ¼ xz
f (14.5.10)
n
The latter equation is analogous to Eq. (14.5.2) and expresses the Ritz trans-
formation of the mode shapes of the eigenvalue problem (14.5.6). We can read-
ily show that the transformed eigenmodes f e n satisfy the orthogonality
conditions with respect to the original mass and stiffness matrices M and K,
namely
f e i ¼ 0,
e T Mf e T Kf
f e i ¼ 0, n 6¼ i (14.5.11)
n n
Indeed, we have

e i ¼ xzn T M xzi
e T Mf
f n

¼ zTn x T Mx zi
e i
¼ zTn Mz
¼0 (14.5.12)
because zn and zi are mode shapes of the eigenvalue problem (14.5.6). Simi-
larly, we prove the second orthogonality condition. The orthogonality with
respect to the damping matrix holds only if it is a proportional damping matrix
as discussed in Section 12.11. If the eigenmodes are orthonormalized with
respect to mass, we have
e n ¼ 1,
zTn Mz e n ¼w
zTn Kz e 2n (14.5.13)
A consequence of this is
f e n ¼ 1,
e T Mf e T Kf
f en ¼ w
e 2n (14.5.14)
n n
It is reasonable to raise the question of why we use the Rayleigh-Ritz method

because it also demands the solution of an eigenvalue problem. The reason is
because the Rayleigh-Ritz method has important advantages over the mode
superposition method, the most noteworthy of which is the significant reduction
in the number of equations of motion if the number K of the Ritz vectors is
taken sufficiently smaller than N . The obtained solution by this method is
approximate in nature. Therefore, to be able to use the Rayleigh-Ritz method,
we have to answer first the following questions:
@CivilMethod
1. How good is the dynamic solution approximated when it is obtained using

the Ritz method?
2. How can we derive Ritz vectors for a given structure?
The answers are given immediately below.
14.5.2 Approximation using Ritz vectors

We assume that the displacement vector u defined by Eq. (14.5.2) represents a
deformation configuration of the structure. The Rayleigh quotient correspond-
ing to this vector is given by Eq. (12.6.21). Hence
uT Ku
rðuÞ ¼
uT Mu
zT x T Kxz
¼ T T
z x Mxz
e
zT Kz
¼
e
zT Mz
or
e
zT Kz
r ð zÞ ¼ (14.5.15)
e
z Mz
T
The value of the Rayleigh quotient is altered by changing the selected defor-
mation pattern, which occurs when one or more of the components of the vector
z change. The matrices Ke and Me are positive definite, hence rðzÞ has finite
values and according to Eq. (12.6.2), we have
0 < l1 rðuÞ lN < 1 (14.5.16)
We know that in the neighborhood of an eigenvalue, rðzÞ takes a stationary
value (see Section 12.6), which is a minimum near all eigenvalues except near
lN , where it is a maximum. The condition ensuring this is
∂r
¼ 0, k ¼ 1, 2, …, K (14.5.17)
∂zk
or

Te e
1 ∂ z Kz e
zT Kz ∂ zT Mz
2 ¼0 (14.5.18)
e
zT Mz ∂zk e ∂zk
zT Mz

e
∂ zT Kz e
∂ zT Mz
r ¼0 (14.5.19)
∂zk ∂zk
@CivilMethod
e and M
Taking into account that the matrices K e are symmetric and that
∂z=∂zk ¼ ∂z =∂zk ¼ I, we obtain
T

e
∂ zT Kz ∂zT e e ∂z
¼ Kz + zT K
∂zk ∂zk ∂zk (14.5.20a)
e
¼ 2Kz
and similarly

e
∂ zT Mz ∂zT e e ∂z
¼ Mz + zT M
∂zk ∂zk ∂zk (14.5.20b)
e
¼ 2Mz
Substituting Eqs. (14.5.20a), (14.5.20b) into Eq. (14.5.19) gives

e rM
K e z¼0 (14.5.21)
which proves that the stationary values of r are the eigenvalues of the trans-
formed eigenvalue problem (14.5.21). The eigenvalue problem is solved using
any of the methods presented in Chapter 13. The eigenvalues give the
eigenfrequencies
pffiffiffiffiffi
en ¼ rn , n ¼ 1, 2, …, K
w (14.5.22)
and the eigenvector zn gives the natural mode

e n ¼ xz
f (14.5.23)
n
The quantities w e n are the best approximation to the corresponding

e n and f
actual ones. It is observed (see Example 14.6.1) that the accuracy of the approx-
imation is better for the lower eigenfrequencies and mode shapes. According to
what we stated in Section 12.6, it holds
e1 ,
w1 w e2 , …
w2 w eK 1
wK 1 w and e K wK
w (14.5.24)
14.6 Selection of Ritz vectors

The efficiency of the Rayleigh-Ritz method depends on the appropriate selec-
tion of the Ritz vectors so that they can approximate the natural vibration
modes. Various methods have been developed for their selection. Two of them
are presented below. The first method is based on the knowledge of the shapes
of natural modes while in the second method the Ritz vectors are derived by a
computational process.
@CivilMethod
14.6.1 Method of natural mode shapes

In this method, we select the shapes of natural modes as Ritz vectors. Thus n is
selected so that it has the shape of fn . The eigenmode fn is not computed by
solving the eigenvalue problem, but its shape is known from the analysis of
similar structures. For example, we know that the first two natural modes of
a multistory shear frame have the shapes shown in Fig. 14.6.1. The components
i1 and i2 are arbitrary but with the limitation that their graphs give the
shapes of 1 and 2 shown in Fig. 14.6.1. For example, we can take
T T
1 ¼ f 0:25 0:5 0:75 1:00 g and 2 ¼ f 0:3 0:5 0:20 1:00 g .
FIG. 14.6.1 Shapes of natural modes of a multistory shear frame.
Apparently, the method of natural mode shapes requires the experience of its
user. It is clear that this method becomes difficult for complicated structures and
practically impossible for three-dimensional structures. The following example
illustrates the implementation of the method of natural mode shapes.
Example 14.6.1 Determine the eigenfrequencies and mode shapes of the chim-
ney in Example 14.3.1 using the Rayleigh-Ritz with (i) two, (ii) three, and (iii)
four Ritz vectors. Use the mode shapes of the cantilever with a constant cross-
section as Ritz vectors.
(a) (b) (c)

FIG. E14.8 Ritz vectors in Example 14.6.1.
@CivilMethod
Solution
The mode shapes of the cantilever with a constant cross-section are obtained
from Eq. (8.3.41). Because C1 is arbitrary, we may write
1
n ðx Þ ¼ ½ cosh an x cos an x Cn ð sinh an x sin an xÞ
2
where
cosh an + cos an
an ¼ bn L, x ¼ x=L, Cn ¼
sinh an + sin an
The values of an for n ¼ 1, 2, 3, 4 are taken from Table 8.3.1, That is
a1 ¼ 1:875, a2 ¼ 4:694, a3 ¼ 7:855, a4 ¼ 10:995
TABLE E14.6 Ritz vectors in Example 14.6.1.

xi 1 2 3 4
0.167 0.05 0.23 0.49 0.70

0.333 0.16 0.59 0.72 0.22
0.500 0.34 0.71 0.02 0.72
0.667 0.55 0.42 0.64 0.17
0.833 0.77 0.22 0.22 0.55
1.000 1.00 1.00 1.00 1.00
The values of the Ritz vectors are given in Table E14.6 and their graphs in
Fig. E14.8.
1. Mass and stiffness matrices
They are obtained from Example 14.3.1.
2 3
6 0 0 0 0 0
60 5 0 0 0 07
6 7
60 0 4 0 0 07
M ¼ m6 7
6 0 0 0 3 0 0 7,
6 7
40 0 0 0 2 05
0 0 0 0 0 1
2 3
232:80 121:78 37:89 7:11 1:21 0:15
6 121:78 122:97 69:99 21:01 3:57 0:44 7
6 7
6 37:89 69:99 72:03 39:04 10:73 1:34 7
K ¼ k66 7:11
7
6 21:01 39:04 38:38 19:04 3:88 7 7
4 1:21 3:57 10:73 19:04 16:00 5:00 5
0:15 0:44 1:34 3:88 5:00 2:12
@CivilMethod
e and K
(a) Computation of the matrices M e for two Ritz vectors
In this case, it is
2 3
0:05 0:23
6 0:16 0:59 7
6 7
6 0:34 0:71 7
x¼6
6 0:55 0:42 7
7
6 7
4 0:77 0:22 5
1:00 1:00
Hence

e ¼ x T Mx ¼ m 3:6987 0:8608 ,
M e ¼ x T Kx ¼ k 0:1911 0:3397
K
0:8608 5:7003 0:3397 4:5018

Thus, the reduced eigenvalue problem K e rMe z ¼ 0 becomes

0:1948 0:3305 3:6987 0:8608 z1 0 m
l ¼ , l¼r
0:3305 4:4985 0:8608 5:7003 z2 0 k
Using any of the methods presented in Chapter 12, we obtain l1 ¼ 0:0459,
l2 ¼ 0:7974. Hence

k k 1:0000 0:1257
r1 ¼ 0:0459 , r2 ¼ 0:7974 , z ¼
m m 0:0708 1:0000
(b) Computation of the approximate eigenfrequencies and mode shapes. They

are obtained from the relations
pffiffiffiffiffi
e n ¼ rn , fn ¼ xzn
w
The obtained results are given in Table E14.7 in juxtaposition to the exact ones.
Moreover, Tables E14.8 and E14.9 present the results obtained using three and
four Ritz vectors, respectively. We note that the use of four Ritz vectors pro-
duces very good results for the first two eigenfrequencies and mode shapes.
A practical rule is to use 2n Ritz vectors to achieve a good approximation of
n eigenfrequencies and mode shapes.
TABLE E14.7 Results obtained using two Ritz vectors in Example 14.6.1.
Rayleigh-Ritz Exact
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
e1 ¼ 0:2143 k=m
w w1 ¼ 0:2030 k=m
e2 ¼ 0:8930 k=m
w w2 ¼ 0:7570 k=m
2 3 2 3
0:0178 0:0950 0:0162 0:0547
6 0:0623 0:2421 7 6 0:0645 0:1747 7
6 7 6 7
6 7 6 0:1469 0:2736 7
e ¼ p1ffiffiffi 6 0:1526 0:2834 7
F F ¼ p1ffiffiffi 6 7
m 6 0:2740 0:1490 7 m 6 0:2640 0:2310 7
6 7 6 7
4 0:4137 0:1346 5 4 0:4133 0:0662 5
0:5639 0:4782 0:5835 0:6010
@CivilMethod
TABLE E14.8 Results obtained using three Ritz vectors in Example 14.6.1.
Rayleigh-Ritz Exact
e 1 ¼ 0:2130 k=m
w w1 ¼ 0:2030 k=m
e 2 ¼ 0:7787 k=m
w w2 ¼ 0:7570 k=m
e 3 ¼ 2:1869 k=m
w w3 ¼ 1:7928 k=m
2 3 2 3
0:0186 0:0548 0:1951 0:0161 0:0547 0:1123
6 0:0618 0:1859 0:2872 7 6 0:0645 0:1747 0:2478 7
6 7 6 7
6 0:1478 0:2910 0:0060 7 6 0:1469 0:2736 0:1305 7
e
F ¼ pffiffiffi
1 6 7 F ¼ p1ffiffiffi 6 7
m 6 0:2680 0:2149 0:2665 7 m 6 0:2640 0:2310 0:2348 7
6 7 6 7
4 0:4145 0:1058 0:1181 5 4 0:4133 0:0662 0:2981 5
0:5765 0:5563 0:3437 0:5835 0:6010 0:4539
TABLE E14.9 Results obtained using four Ritz vectors in Example 14.6.1.
Rayleigh-Ritz Exact
e 1 ¼ 0:2091 k=m
w w1 ¼ 0:2030 k=m
e 2 ¼ 0:7648 k=m
w w2 ¼ 0:7570 k=m
e 3 ¼ 1:8840 k=m
w w3 ¼ 1:7928 k=m
e 4 ¼ 4:0680 k=m
w w4 ¼ 3:3067 k=m
2 3 2 3
0:0171 0:0574 0:1166 0:2800 0:0161 0:0547 0:1123 0:1665
6 0:0623 0:1759 0:2742 0:1070 7 6 0:0645 0:1747 0:2478 0:1858 7
6 7 6 7
6 7 6 7
6 0:2745 7 6 0:1469 0:1863 7
e ¼ p1ffiffiffiffi 6 0:1482
F
0:2796 0:1056 7 ffi6
F ¼ p1ffiffiffi
0:2736 0:1305 7
m6 6 0:2616 0:2340 0:2867 0:0412 77 m6 6 0:2640 0:2310 0:2348 0:1831 77
6 7 6 7
4 0:4110 0:0838 0:2105 0:2419 5 4 0:4133 0:0662 0:2981 0:4140 5
0:5898 0:5784 0:4035 0:2215 0:5835 0:6010 0:4539 0:2807
14.6.2 The method of derived Ritz vectors

We consider the case where the external load is of the form given by
Eq. (14.3.11), namely
pðt Þ ¼ Rf ðt Þ (14.6.1)
The constant vector R may be used to derive a sequence of Ritz vectors,
which are orthonormal with respect to mass using the following procedure.
The first Ritz vector 1 is derived as the vector of the static displacements
produced by the load R, that is
y1 ¼ K1 R (14.6.2)
which is normalized with respect to mass to give
@CivilMethod
y1
1 ¼ 1=2 (14.6.3)
y1 My1
T
The second vector is derived from the vector y2 , which is obtained as the
static displacement vector under the load R ¼ M 1 , that is, the inertial forces
associated with the first Ritz vector. Thus, we have
y2 ¼ K1 M 1 (14.6.4)
The vector y2 , which is not orthogonal to 1 , is mass-orthogonalized to it
using the Gram-Schmidt method presented in Section 12.4. Thus, taking into
account that T1 M 1 ¼ 1, the second of Eq. (12.4.8) gives

e
y2 ¼ y2 yT2 M 1 1 (14.6.5)
which is normalized with respect to mass to give
e
y2
2 ¼ 1=2 (14.6.6)
ey2 Me
T
y2
Similarly, the vector k is obtained as
yk ¼ K1 M k1 (14.6.7)
X
k 1
e
yk ¼ yk ai i (14.6.8)
i¼1
where
ai ¼ yTk M i (14.6.9)
Hence
e
yk
k ¼ 1=2 (14.6.10)
eyk Me
T
yk
The procedure is repeated until we obtain the desired number of Ritz vectors.
The set of vectors 1 , 2 , …, k is orthogonal with respect to mass and linearly
independent, thus satisfying the requirement of the Ritz method. The Ritz
vectors derived by this procedure are also called load-dependent Ritz vectors.
Example 14.6.2 Determine the eigenfrequencies and mode shapes of the chim-
ney in Example 14.3.1 using the Rayleigh-Ritz with (i) two, (ii) three, and (iii)
four derived Ritz vectors.
Solution
The mass and stiffness matrices were computed in Example 14.3.1. We adopt a
load vector proportional to the mass of the structure, that is,
R ¼ m f 6 5 4 3 2 1 gT
@CivilMethod
a. Computation of the first Ritz vector 1

1
Taking m ¼ 31:056 kN m s and k ¼ 5486:208 kN=m (see Example
2
14.3.2), Eq. (14.6.2) gives
y1 ¼ K1 R ¼ f 0:0088 0:0331 0:0711 0:1209 0:1802 0:2451 gT

and substituting into Eq. (14.6.3)
y1
1 ¼ 1=2 ¼ f 0:0036 0:0135 0:0289 0:0491 0:0732 0:0996 gT
y1 My1
T
b. Computation of the second Ritz vector

Eq. (14.6.4) gives
y2 ¼ K1 M 1 ¼ f 0:0402 0:1605 0:3643 0:6521 1:0171 1:4316 gT 102
which is orthogonalized with respect to 1
a1 ¼ yT2 M 1 ¼ 0:1371
e
y2 ¼ y2 a1 1
¼ f 0:0885 0:2408 0:3182 0:2113 0:1351 0:6653 gT 103
which after normalization gives
e
y2
2 ¼ 1=2
e
y2 Me
T
y2
¼ f 0:0136 0:0371 0:0491 0:0326 0:0208 0:1026 gT
c. Computation of the third Ritz vector

Adhering to the previous steps we obtain
y3 ¼ K1 M 2
¼ f 0:0808 0:2363 0:3004 0:0638 0:6461 1:7681 g 103
T
a 1 ¼ yT
3M 1 ¼ 0:00649, a 2 ¼ yT
3M 2 ¼ 0:010059
e
y3 ¼ y 3 a 1 1 a2 2
¼ f 0:3343 0:5016 0:05419 0:5528 0:3853 0:9141 g 104
T
e
y3
3¼ T 1=2
ey3 Mey3
T
¼ f 0:03003 0:04510 0:00493 0:04965 0:03476 0:08186 g
@CivilMethod
d. Computation of the fourth Ritz vector

Similarly we obtain
y4 ¼ K1 M 3
¼ f 0:0255 0:0376 0:0298 0:1270 0:0612 0:2817 gT 103
a1 ¼ yT4 M 1 ¼ 5:8861 1016 , a2 ¼ yT4 M 2 ¼ 0:0011036

a3 ¼ yT4 M 3 ¼ 0:0018245
e
y4 ¼ y4 a1 1 a2 2 a3 3
¼ f 0:14213 0:037082 0:15304 0:0050278 0:20841 0:19202 g 104
T
ey4
4¼ 1=2
eT
y4 Mey4
T
¼ f 0:04338 0:01131 0:04671 0:00153 0:06361 0:05861 g
Hence
2 3
0:0200 0:0760 0:1674 0:2418
6 0:0751 0:2069 0:2514 0:0631 7
6 7
1 6 0:1610 0:2733 0:0275 0:2603 7
x ¼ pffiffiffiffiffi 6 7
m6 6 0:2737 0:1816 0:2767 0:0086 7
7
4 0:4080 0:1161 0:1937 0:3545 5
0:5550 0:5715 0:4562 0:3266
Applying the Rayleigh-Ritz method as described in Example 14.6.1, we
obtain the approximate eigenfrequencies and mode shapes. The obtained results
are given in Tables E14.10–E14.12 in juxtaposition to the exact ones.
TABLE E14.10 Results obtained using two derived

Ritz vectors in Example 14.6.2.
Rayleigh-Ritz Exact
e 1 ¼ 0:2030 k=m
w w1 ¼ 0:2030 k=m
e 2 ¼ 0:7946 k=m
w w2 ¼ 0:7570 k=m
2 3 2 3
0:0161 0:0769 0:0162 0:0547
6 0:0645 0:2104 7 6 0:0645 0:1747 7
6 7 6 7
6 7 6 0:1469 0:2736 7
e ¼ p1ffiffiffi 6 0:1470 0:2811 7
F F ¼ p1ffiffiffi 6 7
m 6 0:2642 0:1952 7 m 6 0:2640 0:2310 7
6 7 6 7
4 0:4134 0:0929 5 4 0:4133 0:0662 5
0:5832 0:5427 0:5835 0:6010
@CivilMethod
TABLE E14.11 Results obtained using three derived Ritz vectors in

Example 14.6.2.
Rayleigh-Ritz Exact
e 1 ¼ 0:2030 k=m
w w1 ¼ 0:2030 k=m
e 2 ¼ 0:7571 k=m
w w2 ¼ 0:7570 k=m
e 3 ¼ 1:9777 k=m
w w3 ¼ 1:7928 k=m
2 3 2 3
0:0162 0:0542 0:1761 0:0161 0:0547 0:1123
6 0:0645 0:1754 0:2769 7 6 0:0645 0:1747 0:2478 7
6 7 6 7
6 0:1469 0:2751 0:0643 7 6 0:1469 0:2736 0:1305 7
e
F ¼ pffiffiffi
1 6 7 F ¼ p1ffiffiffi 6 7
m 6 0:2640
6 0:2301 0:2486 7
7
m 6 0:2640 0:2310
6 0:2348 7
7
4 0:4133 0:0687 0:2048 5 4 0:4133 0:0662 0:2981 5
0:5835 0:5979 0:3804 0:5835 0:6010 0:4539
TABLE E14.12 Results obtained using four derived Ritz vectors in

Example 14.6.2.
Rayleigh-Ritz Exact
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
e1 ¼ 0:2030 k=m , w
w e 2 ¼ 0:7570 k=m w1 ¼ 0:2030 k=m , w2 ¼ 0:7570 k=m
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
e3 ¼ 1:7958 k=m , w
w e 4 ¼ 3:8410 k=m w3 ¼ 1:7928 k=m , w4 ¼ 3:3067 k=m
2 3 2 3
0:0161 0:0547 0:1120
0:2772 0:0161 0:0546 0:1123 0:1665
6 0:0645 0:1747 0:1281 7 6 0:0645 0:1747 0:2478 0:1858 7
6
6
0:2540 7
7 6 7
6 0:1469 0:1272 0:2378 7
0:2736
6 0:1469 0:2736 0:1305 0:1863 7
e ¼ p1ffiffiffiffi 6
F 7 ffi6
F ¼ p1ffiffiffi 6
7
7
m6 6 0:26400:2311 0:2421 0:0532 77 m 6 0:2640 0:2310 0:2348 0:1831 7
6 7 6 7
4 0:4133 0:0663 0:2854 0:2941 5 4 0:4133 0:0662 0:2981 0:4140 5
0:5835 0:6009 0:4457 0:2218 0:5835 0:6010 0:4539 0:2807
14.7 Support excitation

14.7.1 Multiple support excitation
When the motion of an MDOF system is due only to support excitation, the
external forces applied to the free nodes are zero, that is, is pf ðt Þ ¼ 0. As
was shown in Section 11.2.5, the equations of motion of the supported
structure are
€f + Cff u_ f + Kff uf ¼ Mfs u
Mff u €s Cfs u_ s Kfs us (14.7.1)
€f + Mss u
ps ðt Þ ¼ Msf u €s + Csf u_ f + Css u_ s + Ksf uf + Kss us (14.7.2)
@CivilMethod
where uf ¼ uf ðt Þ is the vector of the f unknown free displacements, us ¼ us ðt Þ

the vector of the s known support displacements, and ps ðt Þ the vector of the s
support reactions. We first examine the case where the components of us are s
time functions, different from each other. We refer to this type of support exci-
tation as multiple support excitation to distinguish it from the special case where
the vector us has the form us ¼ eus f ðt Þ, with e
us being a constant vector and f ðt Þ a
common time function, to which we refer as uniform support excitation.
Eqs. (14.7.1), (14.7.2) are transformed to another form, which is more prac-
tical in dynamic analysis. To this end we set
uf ¼ ust f
f +u (14.7.3)
The displacements ust f are obtained as the static solution of Eq. (14.7.1), that
is, when the accelerations and the velocities are zero. Thus, setting
u €s ¼ u_ f ¼ u_ s ¼ 0 and uf ¼ ust
€f ¼ u f into Eqs. (14.7.1), (14.7.2), we obtain
Kff ust
f + Kfs us ¼ 0 (14.7.4)
Ksf ust
f + Kss us ¼ ps
st
(14.7.5)
Assuming
that the supports do not permit rigid body motion, it is
det Kff 6¼ 0 and Eq. (14.7.4), gives
1
f ¼ Sus , S ¼ Kff Kfs
ust (14.7.6)
which is substituted into Eq. (14.7.5) to yield the support reaction due to the
static application of the displacements us

Ksf S + Kss us ¼ pst
s (14.7.7)
Moreover, if the matrix C is proportional to K, we can readily show that it
also holds
Cff u_ st _s ¼0
f + Cfs u (14.7.8)
Indeed, if C ¼ cK, where c is a constant, Eq. (14.7.8) holds true due to
Eq. (14.7.4). If C is not proportional to K, the contribution of the damping force
to the excitation force does not vanish. However, it is small compared with the
inertial force and it can be neglected.

Mff u€f + Cff u_ f + Kff u
f ¼ Mff u
€st
f Mfs u€s Cff u_ st _s
f + Cfs u

Kff ustf + K u
fs s (14.7.9)
which by virtue of Eqs. (14.7.4), (14.7.6), and (14.7.8) becomes

Mff u€f + Cff u_ f + Kff u
f ¼ Mff S + Mfs u€s (14.7.10)
@CivilMethod
The foregoing equations permit the establishment of the displacements u f .

On the basis of Eq. (14.7.1), Eq. (14.7.3) and Eq. (14.7.4), we can write the
elastic forces at the free nodes as
f Sf ¼ Kff uf + Kfs us

f + Kff ust
¼ Kff u f + Kfs us (14.7.11)
f
¼ Kff u
which states that the elastic forces are produced only by the dynamic dis-
placements u f , as they are defined by Eq. (14.7.3) and computed from
Eq. (14.7.10). Further, using Eq. (14.7.2) and Eq. (14.7.7) we can write
the elastic forces at the supports
f Ss ¼ Ksf uf + Kss us

f + Ksf ust
¼ Ksf u f + Kss us
(14.7.12)
f + Ksf S + Kss us
¼ Ksf u
f + pst
¼ Ksf u s
To simplify the expressions, we use the notation Mff ¼ M, Cff ¼ C,

f ¼ u
Kff ¼ K, u and Eq. (14.7.10) is written as

Mu€ + Cu_ + K u ¼ MS + Mfs u€s (14.7.13)
If the mass matrix is diagonal, then Mfs ¼ 0 and Eq. (14.7.13) reduces to
Mu€ + Cu_ + K
u ¼ MS€
us (14.7.14)
14.7.2 Uniform support excitation

We now examine the case where the displacement vector of the supports has
the form
us ¼ e
us f ðt Þ (14.7.15)
where, as stated in the previous section, u es is a constant vector specifying the
spatial distribution of the support displacements, and f ðt Þ a common time func-
tion specifying their amplitude at time t.
u ¼ Mrf€ðt Þ
Mu€ + Cu_ + K (14.7.16)
where it was set
r ¼ Se
us (14.7.17)
@CivilMethod
That is, the vector r represents the static displacements of the free nodes
produced by the vector e us .
When the supports are subjected to a uniform motion ug ðt Þ, as in the
seismic ground motion, the vector of the support displacements can be
written as
us ¼ 1ug ðt Þ (14.7.18)
and inserting it into Eq. (14.7.17) gives
r ¼ S1 (14.7.19)
T
in which 1 ¼ f 1 1 1 ⋯ 1 g is the vector with dimension s. In this case, the
vector r represents rigid body motion, therefore it is
f ¼ 1ug ðt Þ
ust (14.7.20)
Eqs. (14.7.4), (14.7.5) by virtue of Eqs. (14.7.18), (14.7.20) are written as
( )
0
^ g ðt Þ ¼
K1u (14.7.21)
pst
s
where
" #
Kff Kfs
^¼
K (14.7.22)
Ksf Kss

The matrix K^ is the stiffness matrix of the free structure, hence det K
^ ¼0
and ps ¼ 0, because Eq. (14.7.21) has a nontrivial solution. Hence, it is not
st
necessary to use Eq. (14.7.7) to compute pst s .
Example 14.7.1 Formulate the equation of motion of the bridge in Fig. E14.9
when the supports 1 and 2 are subjected to the displacements ug1 ¼ ug ðt Þ and
ug2 ¼ 2ug ðt Þ. The cross-sectional area of the cable is A and its modulus of
elasticity Es while the modulus of elasticity of the remaining structure
is E ¼ 2:1 107 kN=m2 . Assume: I1 ¼ I2 ¼ I3 ¼ 2:055 m4 , I4 ¼ 52 m4 ,
¼ 11:2 kN=m, and L ¼ 50 m. Adopt the lumped
I5 ¼ 31 m4 , Es A ¼ 0:4EI 1 , m
mass assumption.
Solution
By neglecting the axial deformations of members undergoing bending,
the structure has 9 degrees of freedom including those corresponding to
ground motion. The degrees of freedom are shown in Fig. E14.9. Their num-
bering was based on the reasoning to avoid renumbering after static conden-
sation. The lumped mass assumption is adopted. The cable is treated as a
truss member while its sag due to self-weight is neglected. Possible compres-
sion forces are encountered by appropriate prestress. The stiffness matrix
is formulated using the procedure described in Example 11.5.2. Thus,
we obtain
@CivilMethod
FIG. E14.9 Bridge with support excitations in Example 14.7.1
The static condensation (see Section 11.4) yields

2 3 2 3
105 0 0 0 201 134 226 24:7
60 560 0 0 7 6 160 7
Mtt ¼ 6 7, Ktt∗ ¼ 103 6 134 104 132 7
40 0 m3 0 5 4 226 132 308 82:3 5
0 0 0 m4 24:7 1:59 82:3 57:6
The displacements u3 , u4 are known. Thus, we can write

u1 u3 1
uf ¼ , us ¼ ¼ u ðt Þ
u2 u4 2 g

105 0 201 134 3 226 24:7
Mff ¼ , Kff ¼ 10 3
, Kfs ¼ 10
0 560 134 104 132 160
@CivilMethod
The displacements of free nodes are written in the form of Eq. (14.7.3), that
is, uf ¼ ust
f +uf , and Eq. (14.7.6) gives the static solution

1 1
ust
f ¼ K ff K fs u g ðt Þ
2

2:01 0:961 1
¼ u g ðt Þ
1:25 1:250 2

0:088
¼ ug ðt Þ
1:250
f is obtained from Eq. (14.7.14), that is
while u

105 0 u€1 201 134 u 1 10
+ 10 3
¼ u€ ðt Þ
0 560 u€2 134 104 u 2 674 g
14.8 The response spectrum method

The mode superposition method gives the dynamic response of a certain
quantity q ðt Þ as a superposition of its modal responses qn ðt Þ, namely
X
N
q ðt Þ ¼ qn ðt Þ (14.8.1)
n¼1
where

qn ðt Þ ¼ qnst w2n yn ðt Þ (14.8.2)
Eq. (14.8.1) expresses the time history of q ðt Þ. From Eq. (14.8.2), we deduce
that the peak value qn0 of the modal response qn ðt Þ occurs simultaneously with
the extreme value of the pseudoacceleration w2n yn ðt Þ, That is

qn0 ¼ qnst max w2n yn ðt Þ (14.8.3)
t
or as stated in Section 6.2

qn0 ¼ qnst Spa ðxn , Tn Þ (14.8.4)
It should be noted that the quantity Spa ðxn , Tn Þ is by definition positive
while the quantity qnst is algebraic. Because the value of the pseudoacceleration
depends on the period of the system (see Fig. 6.2.6), we can deduce from
Eq. (14.8.4) that the peak values of the modal quantities do not occur at the same
time, nor do they have the same phase. Therefore, the peak value of q ðt Þ cannot
be obtained by superposition. Nevertheless, because the quantities qn0 for a
given excitation are readily computed from the response spectrum without
dynamic analysis, researchers earlier sought methods of exploiting these quan-
tities to compute the peak value of q ðt Þ.
@CivilMethod
A first estimate is obtained from the relation

X
N
jq jmax jqn0 j (14.8.5)
n¼1
which provides an upper bound to jq jmax . Eq. (14.8.5) expresses the absolute
sum (ABSSUM) modal combination rule. It assumes that all modal peak values
occur at the same instant. It gives a too conservative estimate of the peak value.
Therefore, it is not popular in structural design.
Another estimate is obtained from the relation
!1=2
X N X N
jq jmax rnm qn0 qm0 (14.8.6)
n¼1 m¼1
where rmn is the correlation coefficient of the peak modal values qn0 and qm0 . Its
value varies between 0 and 1, with rmn ¼ 1 if m ¼ n. The foregoing equation is
written in matrix form
qffiffiffiffiffiffiffiffiffiffiffiffiffi
jq jmax qT0 rq0 (14.8.7)
where q0 ¼ f q10 q20 ⋯ qN 0 gT and r the matrix of the correlation

coefficients.
Various mathematical expressions for computing the correlation coefficients
are reported in the literature when the excitation is due to seismic ground motion.
Here, we quote only Der Kiureghian’s relation [3], which is widely used today
pffiffiffiffiffiffiffiffiffiffi
8 xn xm ðxn + b nm xm Þb3=2
rmn ¼ 2
nm
(14.8.8)
1 b nm + 4xn x m b nm 1 + b nm + 4 x 2n + x 2m b 2nm
2 2
where b nm ¼ wm =wn , and rmn ¼ rnm . The above relationship results from a
stochastic process starting from the mean square response, which we do not pre-
sent here because it goes beyond the scope of this book. The interested reader
should consult the related literature, for example, [4]. The method of computing
jq jmax using Eq. (14.8.7) is known as the complete quadratic combination
(CQC) rule. Eq. (14.8.7) is greatly simplified when the eigenfrequencies are
well separated, that is, there are no multiple eigenfrequencies nor are they close
to each other. In this case, the off-diagonal elements of the correlation matrix,
rmn ðm 6¼ n Þ, are negligible and we can set r I. Then Eq. (14.8.7) becomes
qffiffiffiffiffiffiffiffiffiffiffi
jq jmax qT0 q0 (14.8.9)
or
!1=2
X
N
jq jmax 2
qn0 (14.8.10)
m¼1
@CivilMethod
The method of computing jq jmax using Eq. (14.8.10) developed by Rosen-

blueth [4] is known as the square-root-of-sum-of-squares (SRSS) rule. It should
be emphasized that Eq. (14.8.10) can lead to unacceptable results when the
eigenfrequencies are not well separated.
Example 14.8.1 Using the response spectrum method, compute the peak values
of the base shear force Qb , the overturning moment Mb , and the top displace-
ment u6 of the chimney in Example 14.3.1 when subjected to the Athens earth-
quake, Sept. 7, 1999. Assume: m ¼ 31:056 kN m1 s2 , k ¼ 5486:208 kN=m,
L ¼ 75 m, x 1 ¼ 0:06, x2 ¼ x3 ¼ ⋯ ¼ x6 ¼ 0:04.
Solution
The eigenfrequencies and mode shapes were computed in Example 14.3.1.
The peak modal values will be computed using Eq. (14.8.4).
st st st
Computation Qbn , Mbn , and u6n
From Example 14.3.1, we obtain for m ¼ 31:056 kN m1 s2
R1 R2 R3 R4 R5 R6
9:66 22:98 33:23 36:65 37:32 46:49
32:17 61:18 61:09 34:10 3:02 36:27
58:57 76:65 25:75 27:33 30:34 20:92
78:94 48:54 34:75 20:15 29:41 8:86
82:39 9:28 29:40 30:37 14:71 2:75
58:17 42:08 22:39 10:30 3:39 0:49
st st st
The quantities Qbn , Mbn , u6n are computed from the relations (see Example
14.3.1)
Gn
st
Qbn ¼ 1T Rn , Mbn
st
¼ hT Rn , u6n
st
¼ f
w2n 6n
The computed values are given in Table E14.13.
st st st
TABLE E14.13 Modal quantities Qbn , Mbn , u6n in
Example 14.8.1.
Mode n st
Qbn ¼ 1T Rn st
Mbn ¼ hT Rn st
u6n ¼ Gwn2 f6n
n
1 319.89 16580.0 0.257

2 157.99 3381.8 0.013
3 78.32 1012.4 1.269e3
4 43.33 403.8 0.172e3
5 28.08 209.1 0.023e3
6 24.56 151.9 0.001e4
@CivilMethod
Computation of the modal pseudoacceleration Spa ðxn , Tn Þ

According to Eq. (14.8.4), we have
Spa ðxn , Tn Þ ¼ w2n max jyn ðt Þj
t
The peak value max t jyn ðt Þj is obtained from the solution of Eq. (14.3.16)
by setting f ðt Þ ¼ u€g ðt Þ, where u€g ðt Þ is given by the accelerogram of the spec-
ified earthquake. The computation of Spa ðx n , Tn Þ is performed numerically
using any of the methods described in Section 6.2. The computed values are
given in Table E14.14 together with the modal values of Qb , Mb , u6 .
Computation of the correlation matrix
Eq. (14.8.8) is used to compute the correlation coefficients. For the data of
the problem, it gives
3
r ¼ 102 3
1000:0 3:53 0:716 0:262 0:128 0:067
6 3:53 1000:0 7:34 1:91 0:85 0:423 7
6 7
6 0:716 7:34 1000:0 15:6 4:47 1:88 7
66 7
0:262 1:91 15:6 1000:0 29:1 7:15 7
6 7
4 0:128 0:85 4:47 29:1 1000:0 33:8 5
0:067 0:423 1:88 7:15 33:8 1000:0
Table E14.15 presents the peak values jQb jmax , jMb jmax , and ju6 jmax as com-
puted using the three modal combination rules in juxtaposition with the exact
values obtained by the response history analysis (RHA) using the relation

X 6 2

jq jmax ¼ max qn wn yn ðt Þ
st
t
n¼1
with numerical integration of Eq. (14.3.16).
TABLE E14.14 Modal quantities Qbn0 , Mbn0 , and u6n0 in Example 14.8.1.
Mode Spa ðx n , Tn Þ Qbn0 Mbn0 u6n0

n Tn xn ðcm=s2 Þ ðkNÞ ðkN mÞ ðcmÞ
1 2.2861 0:06 31.9908 102.33 5304.0 8.2324
2 0.6242 0:04 437.4272 691.10 14793.0 5.858

3 0.2634 0:04 412.2019 322.84 4173.0 0.523
4 0.1429 0:04 506.3626 219.42 2044.6 0.087
5 0.0909 0:04 540.8229 151.85 1131.0 0.012
6 0.0597 0:04 507.2341 124.58 770.52 0.0007
@CivilMethod
TABLE E14.15 Peak values in Example 14.8.1.

Method jQb jmax jMb jmax ju6 jmax
ABSSUM 1612.12 28216.12 0.147
CQC 830.73 16511.35 0.1013
SRSS 824.04 16444.95 0.1012
RHA 922.50 18545.48 0.1157
(exact)
The results in Table E14.15 lead us to certain conclusions. First, the

ABSSUM rule gives extremely conservative estimations (þ75%, þ52%, and
þ27% larger than the respective quantities jQb jmax , jMb jmax , and ju6 jmax u6
obtained by the RHA). Second, the SRSS and CQC rules give essentially the
same estimations for the extreme values. This is due to the fact that the eigen-
frequencies of the structure are well separated (w1 ¼ 2:69, w2 ¼ 10:06,
w3 ¼ 23:82, w4 ¼ 43:94, w5 ¼ 69:06, w6 ¼ 105:11), which produces negligible
off-diagonal terms of the correlation matrix. Third, the SRSS and CQC rules
give estimations of the extreme values smaller than the exact resulting from
RHA. Consequently, the estimation of the extreme values using the SRSS
and CQC is not in favor of safety for the analyzed structure. However, this is
not a general rule because larger values may result by SRSS and CQC for other
types of structures, for example, buildings. Nevertheless, the response spectrum
method is the most convenient method for dynamic analysis of structures under
seismic excitation and has been introduced in all earthquake codes. It should
not, however, be applied to any structure without a previous consideration
because, as we have shown, the estimated peak values may not ensure safety.
Because of the great interest in the response spectrum method, the search for
a rule combining extreme values to obtain a better approximation of the exact
values is still of interest for researchers. The reader may find improved rules in
the literature, mainly based on the theory of probability, which ensure a better
correlation of mode shapes. Nevertheless, increased and cheap computer power
facilitates the exact evaluation of the peak values, a fact that may reduce notice-
ably the importance of the combination rules.
14.9 Comparison of mode superposition method

and Rayleigh-Ritz method
The transformation of the equations of motion using K < N Ritz vectors
reduces the number of equations from N to K (Eq. 14.5.4), which, however,
are coupled because the resulting matrices M, e K
e C, e (Eq. 14.5.5) are not in
@CivilMethod
e ¼I
general diagonal, except when the Ritz vectors are load-dependent. Then M
because the Ritz vectors are orthonormalized with respect to mass.
According to Eqs. (14.3.12)–(14.3.14), we can write
X
K
e
pðt Þ ¼ f ðt Þ en M
Rn , Rn ¼ G en ¼
where G nR
T
n (14.9.1)
n¼1
The number K of the Ritz vectors that should be used in dynamic analysis
must yield a load e pðt Þ sufficiently close to pðt Þ. Hence, according to
P
K
Eq. (14.3.42), the deviation of Rn from R is
n¼1
X
K
e ¼R Rn
n¼1
(14.9.2)
X
K
¼R en M
G n
n¼1
and the truncation error

RT e
eK ¼ (16.9.3)
RT R
Before stating some general conclusions about the preference between
the mode superposition method and the Rayleigh-Ritz method, it is advisable
to examine the truncation error eK as a function of the number of eigenmodes
fn and Ritz vectors n ðn ¼ 1, 2, …, K Þ, respectively, for the chimney in of
Example 14.3.1, when it is subjected to loads pi ðt Þ ¼ Ri f ðt Þ (i ¼ 1, 2, 3, 4) for dif-
ferent spatial distributions of R: f 1 1 1 1 1 1 gT , f 0 0 0 0 1:5 0:5 gT ,
f 0 0 1:5 1 1 0:5 gT , and f 0 0 0 0 0 1 gT . The results are presented
in Fig. 14.9.1. We observe that the truncation error is noticeably smaller when
the analysis is done using Ritz vectors than mode shapes in all four cases. This
is because the Ritz vectors result from the spatial distribution R of the external
load. Although the Rayleigh-Ritz method is preferable on the base of the trunca-
tion error, however, the use of eigenmodes still has advantages because it leads to
uncoupled equations of motion and allows the use of the response spectrum
method for determining the peak values. Another advantage of the Rayleigh-Ritz
method is that static correction is unnecessary because this correction is contained
in the first Ritz vector resulting from the static analysis under the load R. As a
conclusion to the preceding analysis, we can state the following in terms of advan-
tages and disadvantages of the two approaches:
(a) Rayleigh-Ritz method
Advantages: It leads to a considerable reduction of the equations of
motion, thus it reduces the computational task, especially when the
@CivilMethod
FIG. 14.9.1 Error eK as a function of the number K of mode shapes fn and Ritz vectors c n .
equations of motion are solved numerically. The truncation error is notice-

ably smaller. The static correction is unnecessary.
Disadvantages: The transformed equations are coupled. Thus, the solu-
tion of the eigenvalue problem is necessary to determine the eigenfrequen-
cies and eigenmodes of the reduced equations. Only the lower
eigenfrequencies and mode shapes are accurately determined, so to obtain
K eigenfrequencies and eigenmodes with acceptable accuracy, we need
about a double number of Ritz vectors. It is not suitable for the response
spectrum method.
(b) The mode superposition method
Advantages: Leads to uncoupled equations of motion. Therefore, the
solution of a MDOF system with N degrees of freedom results from the
solution of N SDOF systems. It is suitable for the response spectrum
analysis.
Disadvantages: It requires the solution of large eigenvalue problems.
The truncation error is larger. Static correction is necessary. It is less suit-
able for numerical analysis.
@CivilMethod
14.10 Numerical integration of the equations

of motions—Linear MDOF systems
The method of mode superposition presented in Section 14.2 requires a propor-
tional damping matrix to uncouple the equations of motion. In this case, it is
possible to obtain an analytical solution when the load functions are simple.
However, when those functions are complicated or described by a set of discrete
values, such as for example by the seismic accelerogram, it is necessary to use
numerical methods (see Chapter 4) for the computation of the modal coordi-
nates Yn ðt Þ. In the general case where the damping matrix is not proportional
or the equations are nonlinear, recourse to direct time integration numerical
methods is inevitable.
The nonlinear equation of motion of a SDOF system is given by Eq. (5.1.5).
Apparently, for MDOF systems, this equation takes the form
u + f D ðu_ Þ + f S ðuÞ ¼ pðt Þ
M€ (14.10.1)
For the linear response of the system, the foregoing equation becomes (see
Section 10.2)
u + Cu_ + Ku ¼ pðt Þ
M€ (14.10.2)
When the number of equations is large, the numerical solution is simplified
if the degrees of freedom are previously reduced. In linear systems, this is
accomplished by the Rayleigh-Ritz method as discussed in Section 14.5. This
reduction procedure applies also to nonlinear system [5].
As in SDOF systems, the direct numerical integration method yields the
solution uðt Þ at time instants Dt apart starting from uð0Þ. Next, for conve-
nience, we denote the displacement vector at time tn ¼ nDt, n ¼ 0, 1, 2, … by
un . The literature on numerical integration of equations of motion is extensive.
Here, the discussion will be limited to the methods presented in Chapter 4,
which are modified to apply to MDOF systems.
14.10.1 The central difference method (CDM)—Linear equations

At time tn , Eq. (14.10.2) is written
un + Cu_ n + Kun ¼ pn
M€ (14.10.3)
Applying the procedure described in Section 4.2, the derivatives of the vec-
tor uðt Þ are approximated by the central differences
un + 1 un1
u_ n (14.10.4)
2Dt
un + 1 2un + un1
€n
u (14.10.5)
Dt 2
@CivilMethod
which are substituted into Eq. (14.10.3) to yield

1 1 2 1 1
M+ C un + 1 ¼ pn K 2 M un M C un1
Dt 2 2Dt Dt Dt 2 2Dt
(14.10.6)
or
^1 un k
^ n + 1 ¼ pn k
Ku ^2 un1 , n ¼ 0, 1, 2, … (14.10.7)
where
K^ ¼ 1 M+ 1 C (14.10.8a)
Dt 2 2Dt
^1 ¼ K 2 M
k (14.10.8b)
Dt 2
^2 ¼ 1 M 1 C
k (14.10.8c)
Dt 2 2Dt
Eq. (14.10.7) provides the recursive formula to compute un + 1 in terms of

un and un1 . For n ¼ 0 the vector u1 appears in the right side of the formula,
which can be computed by the truncated Taylor series
1
€0
u1 u0 Dt u_ 0 + Dt 2 u (14.10.9)
2
in which u0 , u_ 0 are known from the initial conditions while u
€0 is obtained from
the equation of motion for t ¼ 0

€0 ¼ M1 p0 Cu_ 0 Ku0
u (14.10.10)
The CDM is conditionally stable. The stability condition requires

Dt < TN =p, where TN is the smallest natural period. Obviously, for systems
with a large number of degrees of freedom, TN is very small so that the stability
criterion of CDM demands a very small time step Dt, which renders the method
inappropriate. This disadvantage is circumvented if the use of the method is pre-
ceded by a significant reduction of the degrees of freedom as described in
Section 14.5. The steps of the numerical scheme are given in Table 14.10.1
in the form of a pseudocode so that the reader can write a computer code in
the language of his/her preference. Adhering to the steps of Table 14.10.1, a
computer program called centr_diff_lin_MDOF.m has been written in
MATLAB for the numerical integration of the equation of motion using the cen-
tral difference method. The program is available on this book’s companion
website.
@CivilMethod
TABLE 14.10.1 Central difference method. Linear equations.

A. Data
1. Read: M, C, K, u0 , u_ 0 , pðt Þ, ttot
B. Initial computation
1. Select: Dt ¼ 0:1Tmin , Tmin ¼ 2p=wN
2. Compute: u €0 ¼ M1 ðp0 Cu_ 0 Ku0 Þ
3. Compute: u1 ¼ u0 Dt u_ 0 + Dt2 u
2
€0
^ ¼ 12 M + 1 C
4. Compute: K Dt 2Dt
^1 ¼ K 2 2 M, k
5. Compute: k ^2 ¼ 1 2 M 1 C
Dt Dt 2Dt
C. For each time step

1. Compute: p ^1 un k
^n ¼ pn k ^2 un1
^ ^n
2. Solve: Kun + 1 ¼ p
3. Compute: u_ n ¼ 2Dt
1 €n ¼ Dt1 2 ðun + 1 2un + un1 Þ
ðun + 1 un1 Þ, u
4. Set n ¼ n + 1 and check:
If tn ttot end. Else set un1 ¼ un , un ¼ un + 1 and go to step C.1.
14.10.2 The average acceleration method (AAM)—Linear

equations
The basic concept of the method is to approximate the acceleration vector in the
interval Dt with its mean value, that is,
1
€ ðt + τ Þ ¼ ½ u
u € ðt Þ + u
€ðt + Dt Þ 0 τ Dt (14.10.11)
2
Following the procedure in Section 4.3, we obtain identical relations

where, however, the displacement, velocity, and acceleration are replaced
with the respective vectors of the MDOF system. Thus, the vectors un + 1 ,
€n + 1 at instant tn + 1 ¼ ðn + 1ÞDt, n ¼ 0, 1, 2, … are given by the
u_ n + 1 , and u
relations
un + 1 ¼ Du + un (14.10.12a)
u_ n + 1 ¼ Du_ + u_ n (14.10.12b)
€n + 1 ¼ D€
u €n
u+u (14.10.12c)
@CivilMethod
TABLE 14.10.2 Average acceleration method. Linear equations.

A. Data
1. Select: Dt
2. Compute: u €0 ¼ M1 ðp0 Cu_ 0 Ku0 Þ
3. Compute: K ^ ¼ K + 2 C + 42 M
Dt Dt
4. Compute: ^c ¼ Dt
4 ^ ¼ 2M
M + 2C, m
C. For each step
pn ¼ Dpn + ^cu_ n + m€
1. Compute: Dpn ¼ pn + 1 pn , D^ ^ un
^
2. Solve: KDu ¼ D^
pn
3. Compute: Du_ ¼ Dt2
Du 2u_ n , D€u ¼ Dt4 2 Du Dt4 _
u n 2€
un
4. Compute: un + 1 ¼ Du + un , u_ n + 1 ¼ Du_ + u_ n , u€n + 1 ¼ D€ €n
u+u
If tn ttot end. Otherwise set un ¼ un + 1 , u_ n ¼ u_ n + 1 , u
€n ¼ u
€n + 1 and go to step C.1.
_ D€
The difference vectors Du, u are computed from the relations
2
Du_ ¼
Du 2u_ n (14.10.13)
Dt
4 4
u ¼ 2 Du u_ n 2€
D€ un (14.10.14)
Dt Dt
while Du from
^
KDu ¼ D^
pn (14.10.15)
where
^ ¼K+ 2 C+ 4 M
K (14.10.16)
Dt Dt 2
and
pn ¼ Dpn + ^cu_ n + m€
D^ ^ un (14.10.17)
in which
4
^c ¼ ^ ¼ 2M, Dpn ¼ pn + 1 pn
M + 2C, m (14.10.18)
Dt
The AAM, contrary to CDM, is unconditionally stable. Therefore, the time
step Dt can be chosen arbitrarily. Its size, however, is influenced by the accu-
racy of the method and its capability to describe an oscillatory motion. There-
fore, it must be small enough. The selection of Dt equal to 1/10 of the smallest
natural period of the system produces accurate results. Table 14.10.2 presents
@CivilMethod
the algorithm for the numerical implementation of the AAM in pseudocode-

type notation so that the reader can write a computer code in the language of
his/her preference.
Following the steps in Table 14.10.2, a computer program called av_ac-
c_lin_MDOF.m has been written in MATLAB for the numerical integration
of the equations of motion using the average acceleration method. The program
is available on this book’s companion website.
14.10.3 The analog equation method (AEM)—Linear equations

The underlying concepts of this method are presented in Section 4.4, where the
AEM was developed for a SDOF system. Apparently, the solution procedure
applies also to MDOF systems provided that the coefficients m, c, k in
Eq. (4.4.1) and the quantities u0 , u_ 0 , un , u_ n , qn , pn are replaced with the coefficient
matrices M,C,K and the vectors u0 , u_ 0 , un , u_ n , qn , pn , respectively, and the scalar
operations with matrix operations. Thus, Eqs. (4.4.17), (4.4.20a)–(4.4.20c) read
2 3 2 3
M C K 8 9 0 0 0 8 9 8 9
6 c1 7< qn = 6 c1 7< qn1 = < I =
6 7 I 0 I7
6 2 I hI I 7 u_ n ¼ 6 7 u_ n1 + 0 pn
6 7: ; 6 4
2
5: ; : ;
4 c2 5 un c2 un1 0
I I 0 I I 0
2 2
(14.10.19a)

q0 ¼ M1 p0 Cu_ 0 Ku0 , det ðMÞ 6¼ 0 (14.10.19b)
where qn ¼ u €n . Eq. (14.10.19a), when solved successively for n ¼ 1, 2, …, gives
the solution of the equation of motion (14.10.2). The AEM is self-starting and,
as shown in Section 4.4, unconditionally stable and accurate while conserving
energy. Therefore, it can be used as a practical method for integration of the
equations of motion in cases where widely used time integration procedures
become unstable [6]. Table 14.10.3 presents the algorithm for the numerical
implementation of the AEM in pseudocode-type notation so that the reader
can write a computer code in the language of his/her preference. Following
the steps in that table, a computer program called aem_lin_MDOF.m has been
written in MATLAB for the numerical integration of the equations of motion
using the AEM. The program is available on this book’s companion website.
Example 14.10.1 Solve the discretized equations of motion of the chimney in
Example 14.3.4 using the three numerical methods, CDM, AAM, and AEM.
Give the graphical representation of the base shear force Qb ðt Þ, overturning
moment Mb ðt Þ, and top displacement u6 ðt Þ. Assume: k ¼ 5486:208 kN=m,
m ¼ 31:056 kN m1 s2 , x1 ¼ 0:06, x2 ¼ x3 ¼ ⋯ ¼ x6 ¼ 0:04, pðt Þ ¼ Rf ðt Þ.
The time function f ðt Þ is given in Fig. E14.6 (t1 ¼ 0:1, p0 ¼ 1),
R ¼ f 1:5 6:0 15:0 25:0 40:0 60:0 gT .
@CivilMethod
TABLE 14.10.3 Analog equation method (AEM). Linear equations.

A. Data
1. Select: h ¼ Dt and compute ntot ¼ bttot =h c
2. Compute: q0 ¼ M1 ðp0 Cu_ 0 Ku0 Þ, c1 ¼ h 2 =2, c2 ¼ h
3. Formulate: U0 ≔f q0 u_ 0 u0 gT
2 31 2 3
M C K 0 0 0
6 c1 7 6 c1 7
6 I hI I 7 6 I 0 I 7
4. Compute: A ¼ 6 2 7 6 2 7
4 5 4 5
c2 c2
I I 0 I I 0
2 2
2 3
M C K 1 8 9
I
6 c1 7 >< >=
6 I hI I 7
b¼6 2 7 0
4 c 5 >: >;
2
I I 0 0
2
C. Compute solution
for n≔1 to ntot
Un ¼ AUn1 + bpn
end
Solution
The mass and stiffness matrices, eigenfrequencies, and mode shapes are taken
from Example 14.3.1, which for the given values of k and m give.
2 3
186:336 0 0 0 0 0
60 155:280 0 0 0 0 7
6 7
60 0 124:224 0 0 0 7
M¼6 6 7
0 0 0 93:168 0 0 7
6 7
40 0 0 0 62:112 0 5
0 0 0 0 0 31:056
2 3
1277189:2 668110:4 207872:4 39006:9 6638:3 822:9
6 668110:4 674639:0 383979:7 115265:2 19585:7 2413:9 7
6 7
6 207872:4 383979:7 395171:5 214181:5 58867:0 7351:5 7
6 7
K¼6 7
6 39006:9 115265:2 214181:5 210560:6 104457:4 21286:4 7
6 7
4 6638:3 19585:7 58867:0 104457:4 87779:3 27431:0 5
822:9 2413:9 7351:5 21286:4 27431:0 11630:7
The obtained eigenfrequencies and eigenmodes are

w1 ¼ 2:697, w2 ¼ 10:060, w3 ¼ 23:828
w4 ¼ 43:949, w5 ¼ 69:065, w6 ¼ 105:118
@CivilMethod
2 3
0:0028 0:0098 0:0201 0:0298 0:0377 0:0503
6 0:0115 0:0313 0:0444 0:0333 0:0036 0:0471 7
6 7
6 0:0263 0:0491 0:0234 0:0334 0:0460 0:0339 7
F¼6
6 0:0473
7
6 0:0414 0:0421 0:0328 0:0595 0:0191 7 7
4 0:0741 0:0118 0:0534 0:0742 0:0447 0:0089 5
0:1047 0:1078 0:0814 0:0503 0:0205 0:0031
The numerical methods require the damping matrix. This is constructed as a
proportional damping matrix from the eigenfrequencies, eigenmodes, and
modal damping ratios. Thus applying Eq. (12.11.40), we obtain
C ¼ MFCF^ TM
2 3
1152:72 396:84 59:28 6:35 1:97 0:19
6 396:84 657:43 289:32 2:82 1:69 7
6 37:54 7
6 7
6 59:28 289:32 441:10 192:30 23:00 0:12 7
¼66
7
6 6:35 37:54 192:30 277:56 113:97 13:23 77
6 7
4 1:97 2:82 23:00 113:97 148:39 44:83 5
0:19 1:69 0:12 13:23 44:83 35:59
The graphical representation of Qb ðt Þ, Mb ðt Þ, and u6 ðt Þ, obtained using the
three numerical methods with Dt ¼ 0:01, are shown in Fig. E14.10. Apparently,
the obtained results are graphically identical.
14.11 Numerical integration of the equations

of motions—Nonlinear MDOF systems
The solution to Eq. (14.10.1) is obtained by the step-by-step integration tech-
nique as already developed for the SDOF systems in Chapter 5. This technique
demands the fulfillment of the equation of motion at discrete time instants
Dt apart. Numerous methods using this concept are available in the literature
[7,8]. Here, we will limit our discussion to the average acceleration method
(AAM) in conjunction with the modified Newton-Raphson method for the con-
trol of the error in the tangential stiffness, and the analog equation method
(AEM) [6]. The AAM is very popular in numerical dynamic analysis while
the AEM, besides its efficiency and accuracy, performs well where other
well-known methods fail.
14.11.1 The average acceleration method (AAM)—Nonlinear

equations
The average acceleration method (AAM) developed for linear systems in
Section 14.10.2 can also be applied to nonlinear systems. We limit our presen-
tation of the AAM as applied to equation
u + Cu_ + f S ðuÞ ¼ pðt Þ
M€ (14.11.1)
@CivilMethod
FIG. E14.10 Time histories of Qb ðt Þ, Mb ðt Þ, and u6 ðt Þ in Example 14.10.1.
@CivilMethod
TABLE 14.11.1 Average acceleration method (AAM). Nonlinear equations.

A. Data
1. Read: M, C, f S ðuÞ, u0 , u_ 0 , ttot , a
B. Initial Computations
1. Select: Dt
2. Compute: u €0 ¼ M1 ½p0 Cu_ 0 f S ðu0 Þ
^
3. Compute: c ¼ Dt4
M + 2C m ^ ¼ 2M
C. In each step

1. Compute: ðkT Þn ¼ d ðfSi Þ=du j n ,
2. Compute: k^n ¼ ðkT Þ + 2 C + 4 2 M, Dp ¼ pn + 1 pn
n Dt Dt
^n ¼ Dp + ^cu_ n + m€
3. Compute: p ^ un
^n and p
4. Compute: Du from k ^n using the modified Newton-Raphson method
(see Table 14.11.2)
5. Compute: Du_ ¼ Dt 2
Du 2un
6. Compute: un + 1 ¼ un + Du, u_ n + 1 ¼ u_ n + Du_
u€n + 1 ¼ M1 ½pn + 1 Cu_ n + 1 f S ðun + 1 Þ
If tn ttot end. Else set un ¼ un + 1 , u_ n ¼ u_ n + 1 , u_ n ¼ u_ n + 1 and go to step C.1.
This equation represents a special case of Eq. (14.10.1), that is, only when the
stiffness vector is a nonlinear function of the displacements while the damping
vector is linear. The steps of the method are given in Table 14.11.1. Moreover,
Table 14.11.2 shows the steps for the modified Newton-Raphson method as
applied to MDOF systems. Following these steps, a computer program called
av_acc_nlin_MDOF.m has been written in MATLAB for the numerical integra-
tion of the nonlinear equations of motion using the average acceleration method.
The program list is available on this book’s companion website.
TABLE 14.11.2 Modified Newton-Raphson method.
A. Data
ð0Þ ð0Þ
1. Read: un + 1 ¼ un , f S ¼ f S ðun Þ, DRð1Þ ¼ D^
pn
B. For each iteration Compute i ¼ 1,2,3…
1. duðiÞ ¼ k ^1 DRðiÞ
n
ði Þ ði1Þ
2. un + 1 ¼ un + 1 + duðiÞ h i
ði Þ ði Þ
3. Dp ¼ f S f S + k
ði1Þ ^n ðkT Þ duðiÞ
n
4. DRði + 1Þ ¼ DRðiÞ DpðiÞ
5. Compute ^1 DRði + 1Þ
duði + 1Þ ¼ k
n
6. If duði + 1Þ =DuðnI Þ > a Set i ¼ i + 1 and go to step Β.2
@CivilMethod
14.11.2 The analog equation method (AEM)—Nonlinear equations

The solution procedure developed in Section 5.3 for nonlinear SDOF systems
can be straightforwardly extended to MDOF systems. The nonlinear initial
value problem for MDOF systems reads
M€ _ uÞ ¼ pðt Þ
u + Fðu, (14.11.2a)
uð0Þ ¼ u0 , u_ ð0Þ ¼ u_ 0 (14.11.2b)
where M is N N known coefficient matrix with det ðMÞ 6¼ 0; Fðu, _ uÞ is an
N 1 vector, whose elements are nonlinear functions of the components of
_ and pðt Þ is the vector of the N given load functions and u0 , u_ 0 given con-
u, u;
stant vectors.
TABLE 14.11.3 Analog equation method (AEM). Nonlinear equations.

A. Data
_ uÞ, u0 , u_ 0 , pðt Þ, ttot
1. Read: M, Fðu,
1. Choose: h≔Dt and compute ntot
2. Compute: c1 ≔h 2 =2 c2 ≔h
q0 ≔M1 ½p0 Fðu_ 0 , u0 Þ :
C. Compute solution
for n≔1 to ntot solve for fqn u_ n un gT the system of the nonlinear algebraic equations:
Mqn + Fðu_ n , un Þ ¼ pn 2 c 3 2 c 3
1
I
1
I
hI I u_ n 0 I u_ n1 6 7 6 7
¼ + 4 c2 5qn + 4 c2 5qn1
I 0 un I 0 un1 2
I
2
I
2 2
end
The solution procedure is similar to that for the linear systems. Thus,
Eq. (14.11.2a) for t ¼ 0 gives the initial acceleration vector
q0 ¼ M1 ½p0 Fðu_ 0 , u0 Þ, q0 ¼ u
€ (14.11.3)
Subsequently, we apply Eq. (14.11.2a) for t ¼ tn
Mqn + Fðu_ n , un Þ ¼ pn (14.11.4)
Apparently, the second and third of Eqs. (14.10.19a), (14.10.19b) are valid
in this case too, and can be written as
2 c 3 2 c 3
1
I I
1
hI I u_ n 0 I u_ n1 6 2 7 6 2 7
¼ + 4 c 5qn + 4 c 5qn1
I 0 un I 0 un1 I
2 2
I
2 2
(14.11.5)
@CivilMethod
Eqs. (14.11.4), (14.11.5) are combined and solved for qn , u_ n , un with

n ¼ 1, 2, …. Note that Eq. (14.11.5) are linear and can be solved for u_ n , un . Then
substitution into Eq. (14.11.4) results in a nonlinear algebraic equation, which
can be solved to yield qn . In our examples, the function fsolve of MATLAB has
been employed to obtain the numerical results. The steps of the solution proce-
dure are presented in Table 14.11.3. Following these steps, a computer program
called aem_nlin_MDOF.m has been written in MATLAB. The program list is
available on this book’s companion website.
Example 14.11.1 The structure in Fig. E14.11a consists of two extensible
cables with length l ¼ l0 + d0 supporting a concentrated mass m. Study the
dynamic response of the system when the mass undergoes the displacements
u0 ¼ 20 cm, v0 ¼ 15 cm from the initial position. The cables of cross-sectional
area A are assumed massless and are prestressed by the force S0 ¼ EAd 0 =l0 , so
that their stress is tensile during the response. Compute the minimum value of d0
that ensures tension. Assume: A ¼ 3:14 cm2 , l0 ¼ 3:00 m, m ¼ 10 kN m1 s2 ,
g ¼ 9:81 m=s2 , E ¼ 2 107 kN=m2 , and px ¼ py ¼ 0.
(a) (b) (c)

FIG. E14.11 Structure in Example 14.11.1 (a); degrees of freedom (b); forces acting on the
mass m (c).
Solution
The system has two degrees of freedom, Fig. E14.11b. The equations of
motion are formulated using the method of equilibrium of forces. The forces
acting on the mass m are shown in Fig. E14.11c. The equilibrium of forces
gives
m u€ + S1 sin q1 + S2 sin q2 ¼ px (1a)
m v€ S1 cos q1 + S2 cos q2 ¼ py mg (1b)
The elastic forces S1 and S2 are obtained from the relations
l 1 l0
S1 ¼ EA (2a)
l0
l 2 l0
S2 ¼ EA (2b)
l0
@CivilMethod
Referring to Fig. E14.11b, we have

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l1 ¼ ðl v Þ2 + u 2 ¼ l0 ð1 + d 0 y Þ2 + x 2 (3a)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l2 ¼ ðl + v Þ + u ¼ l0 ð1 + d 0 + y Þ2 + x 2
2 2 (3b)
where
u v d0
x¼ , y¼ , d0 ¼
l0 l0 l0
Substituting Eqs. (3a), (3b) into Eqs. (2a), (2b) gives
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
S1 ¼ EA ð1 + d 0 y Þ + x 1 2 (4a)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
S2 ¼ EA ð1 + d 0 + y Þ + x 1 2 (4b)
Moreover, we have
l v 1 + d0 y
cos q1 ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5a)
l1
ð1 + d 0 y Þ 2 + x 2
u x
sin q1 ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5b)
l1
ð1 + d 0 y Þ2 + x 2
l v 1 + d0 + y
cos q2 ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5c)
l2
ð1 + d 0 + y Þ2 + x 2
u x
sin q2 ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5d)
l2
ð1 + d 0 + y Þ2 + x 2
Finally, substituting Eqs. (4a) and (4b), (5a)–(5d) into Eqs. (1a), (1b) yields
the equations of motion
2 3
6 1 1 7
ml 0 x€ + EAx 42 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi5 ¼ px (6a)
2 2
x 2 + ð1 + d 0 y Þ x 2 + ð1 + d 0 + y Þ
2 3
6 1 + d0 y 1 + d0 + y 7
ml 0 y€ + EA42y + qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi5 ¼ py mg
x 2 + ð1 + d 0 y Þ2 x 2 + ð1 + d0 + y Þ2
(6b)
@CivilMethod
Eqs. (6a), (6b) are solved numerically using the AAM and AEM methods.
Fig. E14.12 shows the graphs of u ðt Þ and v ðt Þ while Fig. E14.13 shows
the graphs of S1 ðt Þ and S2 ðt Þ. The results are graphically identical. The mini-
mum value of d0 is obtained iteratively until min ðS1 , S2 Þ ¼ 0. This gives
min d0 ¼ 16:5 cm.
FIG. E14.12 Time history of the displacements u(t) and v(t) in Example 14.11.1.
FIG. E14.13 Time history of the elastic forces S1(t) and S2(t) in Example 14.11.1.
@CivilMethod
14.12 Problems
FIG. P14.1 Dynamic model of the television tower in problem P14.1.
Problem P14.1 The television tower modeled by the system in Fig. P14.1 is
subjected (i) to the wind pressure of Fig. E14.6 with p0 ¼ 2 kN=m2 ,
t1 ¼ 0:5 s and (ii) to the seismic ground motion ug ðt Þ ¼ 2 sin 5t. The column
is approximated by three constant elements of length l ¼ h=3. Adopting the
lumped assumption, derive the equation of motion of the structure and analyze
its response using the mode superposition method. Give the graphs of the time
history of the displacements, the base shear force Qb and overturning moment
Mb . Compute the peak value of the shear force and the bending moment at the
base cross-section of the column as well as at the cross-sections above and
beneath the body B representing a rotating restaurant. The ground is elastic.
The reaction moment of the elastic ground is represented by the expression
MR ¼ CR f, where CR ¼ KI f ; If is the moment of inertia of the planform of
the fundament and K ¼ E=10h the foundation modulus with E being the mod-
ulus of elasticity of the material of the structure. The cross-sections of the flex-
ible column, the planform of the fundament, and the body B are circular with
diameters D, Df ¼ 8D, and DB ¼ 5D, respectively. The specific weight of the
material is g. The fundament and the body B are assumed rigid. Data h ¼ 60 m,
D ¼ 2 m, E ¼ 2:1 107 kN=m2 , g ¼ 24 kN=m3 .
Problem P14.2 The shear frame in Fig. P14.2a is modeled as shown in
Fig P14.2b. Compute the modal contribution factor of (i) the base shear
force Qb , (ii) the overturning moment Mb , and (iii) the top displacement u4
when the excitation is due to ground motion ug ðt Þ ¼ uo sin w t. Assume:
m ¼ 2:0 kN m1 s2 =m; a ¼ 3:0 m; columns cross-sections: first and second
floor 0:30 0:30 m2 , third and fourth floor 0:25 0:25 m2 ; modulus of elastic-
ity E ¼ 2:1 107 kN=m2 .
@CivilMethod
(a) (b)
FIG. P14.2 Shear frame in problem P14.2 (a); dynamic model (b).
Problem P14.3 For the shear frame of Problem P14.2, derive Ritz vectors using
(i) the method of the natural mode shapes and (ii) the method of load dependent
Ritz vectors, and use them to approximate the eigenfrequencies. Then reduce
the number of equations of motion to two.
Problem P14.4 For the shear frame of Problem P14.2, compute the modal
magnification factors Dn and examine whether the static correction method
is recommended for application.
Problem P14.5 For the shear frame of Problem P14.2, compute the peak modal
values Qbn0 , Mbn0 , and u4n0 of the base shear force, overturning moment, and
top displacement, respectively. Then use the ABSSUM, CQM, and SRSS
methods to compute the peak values of these quantities. Compare them with
those resulting from the RHA (Response History Analysis).
Problem P14.6 The two one-story buildings of Fig. P14.6 are connected by a
beam as shown in Fig. P14.6. Formulate the equations of motion. Then compute
the truncation error as a function of the number of employed eigenmodes.
Use both the natural mode shapes and derived Ritz vectors and compare the
computed error. Data: column cross-sections 0:30 0:30 m2 , beam cross-
section 0:20 0:30 m2 , h ¼ 3:5 m, a ¼ 4:0 m, b ¼ 4:0 m, q ¼ 20 kN=m2 ,
b ¼ p=6, and F ðt Þ ¼ 2 cos 5t.
FIG. P14.6 Two one-story buildings in problem P16.6.

@CivilMethod
Problem P14.7 Consider the two-bar system of Fig. P14.7. While the system is
in static equilibrium under the gravity load W ¼ mg, the mass m is given the
additional displacements u0 , v0 . Compute the time history of the displacements
u ðt Þ,v ðt Þ in the horizontal and vertical directions, respectively, as well as the
axial forces S1 ðt Þ, S2 ðt Þ of the bars and give their graphs. Use the numerical
methods CDF, AAM, and AEM to solve the nonlinear equations of motion.
Assume: A ¼ 3:14 cm2 , l1 ¼ 4:0 m, l2 ¼ 3:0 m, m ¼ 10 kN m1 s2 ,
g ¼ 9:81 m=s2 , and E ¼ 2 108 kN=m2 . The bars are assumed massless.
FIG. P14.7 Two-bar system in problem P14.7.
Problem P14.8 The two-story building of Fig P14.8 is modeled as a space

frame. Reduce the degrees of freedom of the structure by considering the axial
restraints of columns as well the diaphragmatic function of the slabs. Adopt the
lumped mass assumption for the columns.
Problem P14.9 The shear frame of Fig. P14.9 is supported on an elastic foun-
dation that is modeled by two nonlinear springs producing the force
FT ¼ CT ðu + u 2 =4a Þ and the moment MR ¼ CR f + f2 =4 , where
CT ¼ EI =10a 3 and CR ¼ EI =5a. Formulate the equations of motion of the
structure and determine the time history of the shear forces and bending
moments of the columns. Then compute their peak values. Data:
¼ 2 kN m1 s2 =m, a ¼ 10 m, h1 ¼ 5h2 =3, h2 ¼ 3:0 m; g ¼ 9:81 m=s2 ,
m
E ¼ 2 107 kN=m2 ; cross-section area of columns k1 0:30 0:30 m2 and k2
0:25 0:25 m2 ; p1 ðt Þ ¼ 5H ðt Þ, and p2 ðt Þ ¼ 8H ðt Þ.
@CivilMethod
FIG. P14.9 Shear frame in problem P14.9.
Problem P14.10 The steel mast of Fig. P14.10a is spherically hinged at point O
and is supported by four massless cables of cross-section A. At the top of the
mast is a steel box of side a and thickness d1 . The mast consists of a circular
steel tube with thickness d2 . Its cross-section varies linearly as shown in the fig-
ure. The external diameters at the cross-sections are D1 at the top C, D2 at B, and
D3 at the base O. The planform of the structure is shown in Fig. P14.10b. The
box is subjected to a wind blast of intensity pðt Þ in the direction b with respect to
the x axis, whose time variation is given in Fig. E14.6. The cables are pre-
stressed to undertake compressive forces. Considering small displacements,
determine the response of the structure. Data: L ¼ 60:0 m, a ¼ 5:0 m,
d1 ¼ d2 ¼ 2 cm, D1 ¼ D3 ¼ 1:0 m, D2 ¼ 1:5 m, p0 ¼ 2:0 kN=m2 , b ¼ p=6,
A ¼ 3:14 cm2 , E ¼ 2 108 kN=m2 , and g ¼ 9:81 m=s2 .
(a) (b)
FIG. P14.10 Steel mast in problem P14.10 (a); planform (b).
@CivilMethod

[1] A.K. Chopra, Modal analysis of linear dynamic systems: physical interpretation, ASCE J.
Struct. Eng. 122 (1996) 517–527.
[2] J.T. Katsikadelis, A generalized Ritz method for partial differential equations in domains of
arbitrary geometry using global shape functions, Eng. Anal. Bound. Elem. 32 (5) (2008)
353–367, https://doi.org/10.1016/j.enganabound.2007.09.001.
[3] E.L. Wilson, A. Der Kiureghian, E.P. Bayo, A replacement for the SRSS method in seismic
analysis, Int. J. Earthquake Eng. Struct. Dyn. 9 (1981) 187–194.
[4] N.M. Newmark, E. Rosenblueth, Fundamentals of Earthquake Engineering, Prentice-Hall,
[5] J.T. Katsikadelis, N. Babouskos, Nonlinear flutter instability of thin damped plates. An AEM
solution, J. Mech. Mater. Struct. 4 (7–8) (2009) 1394–1414.
10.1002/zamm.20120024.
[7] M.A. Dokainish, K. Subbaraj, A survey of direct time-integration method in computational
structural dynamics—I. Explicit methods, Comput. Struct. 32 (1989) 1371–1386.
[8] K. Subbaraj, M.A. Dokainish, A survey of direct time-integration methods in computational
structural dynamics—II. Implicit methods, Comput. Struct. 32 (1989) 1387–1401.
[9] W. Weaver Jr., P.R. Johnston, Structural Dynamics by Finite Elements, Prentice Hall,
[10] K.J. Bathe, E.L. Wilson, Numerical Methods in Finite Element Analysis, Prentice-Hall, Inc.,
@CivilMethod
Chapter 15
Dynamic analysis of multistory

buildings
Chapter outline
15.1 Introduction 693 15.2.4. Equation of motion
15.2 The multistory building 695 of the multistory
15.2.1. The concept of the building 702
multistory element 695 15.3 Dynamic response of
15.2.2. Nodal displacement multistory buildings due to
matrix, nodal force ground motion 715
matrix, transformation 15.4 Problems 723
matrix, and stiffness References and further reading 724
matrix of the MSE 696
15.2.3. Mass matrix of the
MSE and multistory
building 701
15.1 Introduction
The load-bearing systems of modern multistory buildings are skeletal structures
made of steel, RC (reinforced concrete), or a combination of them (composite struc-
tures). The load transfer path is from slabs (plates) to beams, from beams to col-
umns, and from columns to the foundation, which may consist of individual
footings, strip footings, and raft foundations. The first modeling of a building is
the shear building. It is a simple model that is used to approximate the dynamic
response of a building. According to this model, the beam-reinforced slabs are
encountered as plane rigid bodies. The obtained response is acceptable when the
beam-slab system is very stiff, which is a usual case for buildings of reinforced con-
crete. However, in the beam-slab system, the flexible slabs are connected with the
beams either by various types of connectors, as in composite structures, or rigidly,
as in concrete structures, and they cooperate in carrying the live and dead loads. The
deformation of the slab within its plane is very small and can be neglected. Thus,
the functioning of the slab in the horizontal motion can be simulated with that of a
plane rigid body, called diaphragm. This functioning constrains the horizontal dis-
placements and the rotations about the vertical axis at the ends of the columns as
well as the axial deformation of the beams. On the other hand, this approximation
enforces the cooperation of slabs and beams in the vertical direction.

@CivilMethod
The aforementioned functioning of the load-bearing system of the building

allows the approximation of its dynamical response by the following three models:
(a) The slab together with the beams behaves as a rib-reinforced plate (see
Fig. 15.1.1) and is treated as a substructure connected to the columns.
Though this model represents the actual response of the building, it exhibits
considerable difficulties in the analysis of the beam-plate system [1–3],
which increase when the directions of the beam principal axes are inclined
with respect to the middle plane of the plate [4]. On the other hand, the
employed simplified methods to approximate rib-reinforced plates [5]
introduce a considerable error, which cancels the advantages of this model
and prevents its use in building analysis.
(b) The building is approximated by a space frame in which the plates partic-
ipate with their effective breadth, that is, a strip of the plate that increases
the stiffness of the beam, and the beams perform as beams with flanges, that
is, T-beams or G-beams (Fig. 15.1.2). The effective breadth assumption is
adopted as a recommendation by several specification codes. Nevertheless,
the resulting error is not controlled because the effective breadth cannot be
estimated theoretically [6].
(c) A third way to approximate the response of a building is the finite element
method. In this method, the columns are modeled as beam elements and the
beams as beam structures interconnected with the plate, which is treated as
a surface structure. Then the degrees of freedom are reduced by introducing
nodal constraints, which ensure the diaphragmatic functioning of the plates
and if allowed, the omitting of the axial deformation of columns. The finite
element method describes the actual response of the structure reliably and
has been very popular recently because of the increased cheap computa-
tional power. Therefore, it is used in most available professional codes
for the dynamic analysis of buildings. Nevertheless, the method exhibits
a great shortcoming as the number of the eigenmodes is very large and a
procedure should be developed to determine the least number of eigen-
modes that dominate the dynamic response.
In all three models, the slab functions as a diaphragm. In the second model, the
equation of motion of the building can be derived from the equation of motion of
the space frame, as formulated in Section 11.9. Then, the lumped mass assumption
for the columns eliminates the end rotations about the horizontal axes while their
horizontal displacements and rotations about the vertical axis are constrained by
the diaphragm, as described in Section 11.11. If the height of the building is small
compared with its other dimensions, the axial displacements of the columns may
be neglected. The masses of the building are concentrated at the levels of the slabs.
This process reduces the degrees of freedom to N ¼ 3n, where n is the number of
floors. We have illustrated it in Section 14.4.2, where the degrees of freedom of a
three-story building are reduced to 9.
@CivilMethod
Dynamic analysis of multistory buildings Chapter 15 695
Several professional codes have been developed for the analysis of buildings
and in general of structures, such as ETABS, SAP2000, CSiBridge, STAAD,
MIDAS, etc. These codes can be used by professional engineers to efficiently
analyze structures made from various materials. However, the engineers must
be aware of the assumptions adopted by the code. It is highly essential that
the user of the code can check the correctness of the obtained results. This
presupposes a deep understanding of the static and dynamic response of the
structures. Therefore, it is recommended that these codes are employed only
by experienced engineers who master the static and dynamic structural analysis.
FIG. 15.1.1 Plate reinforced with beams.
Effective breadth
FIG. 15.1.2 Plate modeling with a grid of T- and G-beams (cross-section).
In this chapter, we will study the dynamic response of multistory buildings

on the basis of a simpler building model, which takes into account the framing
of horizontal beams and vertical columns. This functioning is not limited indi-
vidually to the floors but extends to the entire height of the building, including
the influence of the elastic support on the ground. The basic concept introduced
by this model is that of the multistory element (MSE). This element pertains
along the whole height of the building and it may be an individual column, a
wall, a frame, an elevator core, a staircase core, or elements that result from their
coupling. The slabs function as diaphragms and the axial deformation of the
multistory elements may be neglected. The masses are considered lumped at
the levels of the floors. Thus, the active degrees of freedom are three for each
floor, two translations and one rotation. This implies that in a building with n
floors, the motion is governed by 3n equations. The method is illustrated with
two buildings, a one-story building and a two-story building.
15.2 The multistory building

15.2.1 The concept of the multistory element
We consider the multistory building of Fig. 15.2.1, which consists of n horizon-
tal plates connected to each other by K vertical elastic elements. The plates
are assumed undeformable in their plane, that is, they function as diaphragms.
@CivilMethod
The vertical elements are rigidly connected with the plates so that the actions
(forces and moments) can be transferred from the plates to the vertical elements
and vice versa.
FIG. 15.2.1 Multistory building.
The vertical elements may connect two or more, or even all, slabs of the
structure. They may be fixed, hinged, or elastically supported on the ground.
These elements will be referred to as multistory elements (MSE). The MSEs
may be columns, frames, walls, isolated or framed with beams, closed sections,
staircase cores, etc. The directions of the principal axes of the MSE may vary
from floor to floor. Nevertheless, their treatment is easier if their principal
directions are the same along the height.
15.2.2 Nodal displacement matrix, nodal force matrix,

transformation matrix, and stiffness matrix of the MSE
We consider an MSE whose axis coincides with the vertical z axis. Its hor-
izontal axes x, y coincide with the directions of the common principal axes
when they do not change from floor to floor or with the predominant prin-
cipal axes. The z axis of all MSEs is parallel to the z axis of the global sys-
tem O xyz of the structure and has the same direction with that. The x, y
axes are generally different for each MSE and are rotated with respect to
the global system of axes (see Fig. 15.2.2). The system of axes i xyz of i
MSE is its local system of axes. The MSE is idealized with its axis. The
points 1, 2, …,n at which the z axis of the MSE intersects the horizontal
slabs are referred to as the nodes of the i MSE.
@CivilMethod
Node
Node
Node
Node
Node
FIG. 15.2.2 Displacements and elastic forces at the node j of the i MSE.
The MSE is deformed due to the deformation of the structure. Its deforma-
tion can be determined by the nodal displacements, which are three for each
node, two translational in the directions of the x,y axes and one rotational about
the z axis. The displacements and elastic forces at the j node of i MSE will be
denoted by uji , vji , wij and Xji , Yji , Mji in the local axes and by u ij , v ij , w
ij and
i i i
X j , Y j , M
in the global axes, respectively.
j
Thus we define the following matrices for the displacements and elastic
forces at the j node of i MSE:
In the local axes n oT
Dij ¼ uji vji wij (15.2.1a)
T
FiSj ¼ Xji Yji Mji (15.2.1b)
In global axes n oT
i ¼ u i v i w
D i
(15.2.2a)
j j j j
n oT
i ¼ X i
F
i
Y j i
M (15.2.2b)
Sj j j
We assume that the local axes of the MSE are the same on all floors. The
transformation matrix from the global system of axes to the local one is
2 3
cos fi sin fi 0
Rij ¼ Ri ¼ 4 sin fi cos fi 0 5 (15.2.3)
0 0 1
@CivilMethod
Hence, the known transformation relations hold

i
Dij ¼ Ri D j (15.2.4a)
i i T i
¼ R D
D (15.2.4b)
j j
i
FiSj ¼ Ri F (15.2.5a)
Sj

i T
i
F ¼ R FiSj (15.2.5b)
Sj
Subsequently, we formulate the total vector of displacements of the MSE
in the local and global systems of axes. To be consistent with Fig. 15.2.2,
we place the displacements in the corresponding vectors starting with the n
node and ending with the first, namely
8 i9
>
> D1 > >
< Di >
> =
Di ¼ 2
, (15.2.6a)
> >
>⋮ >
>
: >
;
Din
8 i9
>
> D >
> 1>>

> =
i
¼ D
D 2 (15.2.6b)
>
> >
>⋮ >
> >
>
: i ;
Dn
Obviously, these vectors are associated by the relations
Di ¼ R i
^ iD (15.2.7a)
i T i
i ¼ R
D ^ D (15.2.7b)
where
2 3
Ri 0 ⋯ 0
60 Ri ⋯ 0 7
î ¼ 6
R 6
7
7 (15.2.8)
4⋮ ⋮ ⋱ ⋮ 5
0 0 0 Ri
By the same reasoning, we define the total vectors of the nodal elastic forces
8 i 9
>
> FS1 >
>
< Fi >
> =
i S2
FS ¼ , (15.2.9a)
>⋮ >
> >
>
: i ; >
FSn
@CivilMethod
8 i 9
>
> >
F
> S1 >
> i >
< >
=

i ¼ FS2
F (15.2.9b)
S
>⋮ >
> >
>
> >
: i > ;
F Sn
which are associated with the relations

FiS ¼ R i
^ iF (15.2.10a)
S
T
i ¼ R
F ^ i Fi (15.2.10b)
S S
Among the components of the nodal elastic forces and the nodal displace-
ments of the MSE, a relation of the following form can be established
ð15:2:11Þ
or
FiS ¼ ki Di (15.2.12)
i
The matrix k with dimensions N N ðN ¼ 3n Þ is the stiffness matrix of the
i MSE with respect to its nodal displacements. In global axes, this matrix is
transformed as
i ¼ k
F i D
i (15.2.13)
S
where
i T i i
i ¼ R
k ^ kR ^ (15.2.14)
The plates undergo translational displacements in the direction of the
y axes and rotations about the z axis. We denote the displacement vector
x,
of the j plate with respect to the O xyz with
8 9
 U >
< 1 >
> =
¼ U
U 2
(15.2.16)
>
> ⋮ >
: > ;
Un
The origin Oj of the axes to which the motion of the plate is referred may
i with U
differ from plate to plate. The relation that links the displacements D j is
j
T
D i ¼ Tj U j , j ¼ 1, 2,…,n (15.2.17)
j i
where
2 3
1 0 0
Tji ¼ 4 0 1 0 5 (15.2.18)
y ij xij 1

is the transformation matrix, in which xij , yij represent the coordinates of the
point i referred to the system of axes with origin the point Oj (see
Section 11.11).
Eqs. (15.2.17) are combined as
i ¼ TT U
D (15.2.19)
i
where
2 3
T1i 0 ⋯ 0
60 T2i ⋯ 0 7
Ti ¼ 6
4⋮
7 (15.2.20)
⋮ ⋱ ⋮ 5
0 0 ⋯ Tni
is the transformation matrix of the i MSE due to the transfer.
Similarly, the elastic forces are transformed to point Oj according to
Eq. (15.2.17)
oi ¼ Tj F
F i , j ¼ 1, 2, …,n
Sj i Sj

8 oi 9 8 i 9
> >
F 2 1 3> F >
>
> S1 >
> T 0 ⋯ 0 > > S1 >
>
< oi >
> = 6 i >
=
FS2Þ 0 T2i ⋯ 7
0 7 FS2
¼6
4⋮ (15.2.21)
>⋮ >
> > ⋮ ⋱ ⋮ 5> > >
>
> >
> >⋮ > >
: oi ; 0 0 ⋯ Tni >: i > ;
F Sn FSn
or
oi ¼ Ti F
F i (15.2.22)
S S
Using Eqs. (15.2.13), (15.2.19), we write Eq. (15.2.22) as
F oi U
oi ¼ k (15.2.23)
S
@CivilMethod
where
oi ¼ Ti k
k i TT (15.2.24)
i
is the global stiffness matrix of the i MSE with respect to points Oj .
15.2.3 Mass matrix of the MSE and multistory building

In order to formulate the mass matrix of the i MSE, we assume that it consists
of successive single-story elements. Thus, the mass of the MSE is lumped at its
nodes. If mji is the mass at the j node and Iji the moment of inertia about the local
vertical axis, then the inertial force at the level of the j plate is
i
i ¼ m
F €
,
ij D j ¼ 1, 2, …,n (15.2.25)
Ij j
where
2 3
mji 0 0
6 mji 0 7
ij ¼ 4 0
m 5, j ¼ 1, 2, …,n (15.2.26)
0 0 Iji
Eqs. (15.2.25) are combined to
i
i ¼ m
F €

iD (15.2.27)
I
where
8 i 9 2 3
> >
F mi1 0 ⋯ 0
> Ii 1 >
> >
< = 60 mi2 ⋯ 0 7
i ¼ F 6 7
F I
I2 , i ¼6
m 7 (15.2.28)
>⋮ >
> > 4⋮ ⋮ ⋱ ⋮ 5
: i >
> ;

F In
0 0 ⋯ min
Eq. (15.2.27) is transformed to point Oj as

oi ¼ m
F €

oi U (15.2.29)
I
where
oi ¼ Ti m
m i TTi (15.2.30)
If mj is the mass of the j plate and Ijc its moment of inertia about the vertical
axis with respect to the center of mass Cj , then the inertial force of the j plate is
c ¼ M
F €
cU
c
Ij j j , j ¼ 1, 2, …,n (15.2.31)
where
2 3
mj 0 0
c ¼6
M
7
4 0 mj 0 5, j ¼ 1, 2, …,n (15.2.32)
j
0 0 Ijc
@CivilMethod
Eq. (15.2.31) is combined to
c ¼ M
F €
cU
c (15.2.33)
I
where
8 c 9 2 c 3 8 c 9
>
> FI 1 >
> M1 0 ⋯ 0 >
> U1 >>
>
<F c => 60 c ⋯ 0 7 >
<Uc >=
c ¼ I2 c ¼6 M 7 c
F I , M 6 2
7, U ¼ 2
(15.2.34)
>
> ⋮ >
> 4⋮ ⋮ ⋱ ⋮ 5 >
> ⋮ > >
>
: c ; > : c>
> ;

FIn 0 0 c
⋯ M
U
n n
Eq. (15.2.33) is transformed to point Oj as

o ¼ M
F €

oU (15.2.35)
I
which
2 3
T1c 0 ⋯ 0 2 3
6 1 0 0
c TT , Tc ¼ 6 0
o ¼ Tc M T2c ⋯ 0 7 7 , Tj ¼ 4 0 1 0 5
M 4⋮ (15.2.36)
c
⋮ ⋱ ⋮ 5 c
ycj xcj 1
n
0 0 ⋯ Tc
15.2.4 Equation of motion of the multistory building

If PA A A A T
j ¼ fPjx Pjy Mjz g denotes the external force that is applied at the point
Aj of the j plate, then it is transformed to point Oj as
Poj ¼ TjA PA
j , j ¼ 1, 2, …,n (15.2.37)
P o ¼ T A PA (15.2.38)
where
8 o9 2 1 3
> P1 > TA 0 ⋯ 0 2 3
>
> >
> 1 0 0
< Po = 60 T2A ⋯ 0 7 7
6 j 6 7
Po ¼ 2
, TA ¼ 6 7, TA ¼ 4 0 1 0 5 (15.2.39)
>
> ⋮ >> 4⋮ ⋮ ⋱ ⋮ 5
: o>
> ; yA
A
j x j 1
Pn 0 0 ⋯ TnA
The dynamic equilibrium of the j plate yields
X
K X
K
o +
F oi +
F oi ¼ Po , j ¼ 1, 2,…,n
F (15.2.40)
Ij Ij Sj j
i¼1 i¼1
@CivilMethod

X
K X
K
o +
F oi +
F oi ¼ Po
F (15.2.41)
I I S
i¼1 i¼1
Further, using Eqs. (15.2.23), (15.2.29), and (15.2.35), we obtain
M € +K
U
U ¼P
(15.2.42)
where
XK
¼M
M o+ oi
m (15.2.43a)
i¼1
X
K
¼
K oi
k (15.2.43b)
i¼1
¼ Po
P (15.2.43c)
The modelling of buildings using the concept of the MSE is not recommended
for the engineering praxis where the analysis is performed using commercial
computer codes. However, it can be used for educational purposes to analyze sim-
ple buildings aiming at understanding their dynamic response. From the examples
presented below we conclude that method of MSEs provides good results.
Example 15.2.1 Formulate the equation of motion of the one-story building of
Fig. E15.1 with respect to the center of mass O of the slab for the ground motion
ug ðt Þ in the direction b. Assume:
(a) The columns k1 , k2 , k3 , k4 have a square cross-section ða a Þ with
moments of inertia Ix ¼ Iy ¼ I .
(b) The beams b1 , b2 have a rectangular cross-section ða2a Þ, hence a
moment of inertia Ib ¼ 8I . The effective breadth of the beams as well as
their torsional stiffness are neglected.
(c) Shear modulus G ¼ 0:40E and mass of the slab r per square meter.
FIG. E15.1 One-story building in Example 15.2.1.
@CivilMethod
Remarks and recommendations:

(i) The theoretical height of the columns is L.
(ii) The theoretical span of the frame k1 b1 k4 is 2L and that of the frame
k2 b2 k1 is L.
(iii) The mass of the columns as well as of the beams below the slab is
neglected. Hence the center of mass of the slab coincides with its geomet-
rical center of mass.
(iv) The slab is assumed undeformable in its plane. Therefore, the axial
deformation of the beams is neglected.
(v) The axial deformation of the columns is neglected. When formulating the
stiffness of the structure, the beams and columns function as frames.
(vi) The columns are assumed fixed on the ground.
(vii) Due to the small stiffness of the slab, the connection between the column
k3 and the slab is a hinge about the horizontal axes.
Solution using the concept of the MSE
Step 1
We determine the MSEs of the structure. From Fig. E15.1 we distinguish the
following MSEs, which are actually one-story elements. Below, they are shown
in their local element axes
2 3
It contributes to the stiffness of the kx 0 0
structure with its translational k ¼ 40 0 05
1
stiffness kx along the local axis x. 0 0 0
2 3
structure with its translational k ¼ 0 0 05
2 4
stiffness kx along the local axis x. 0 0 0
2 3
It contributes to the stiffness of the 0 0 0
structure only with its torsional k ¼ 40 0 0 5
3
stiffness kw because its translation 0 0 kw

stiffnesses along local axes x,y have
been considered in MSE 1 and MSE 2.
2 3
structure with its translational k4 ¼ 4 0 0 0 5
stiffness kx along the local axis x and 0 0 kw
with the torsional stiffness kw . The
translational stiffness along the local
axis y has already been considered in
MSE 2.
@CivilMethod
2 3
structure with its translational k5 ¼ 4 0 ky 0 5
stiffnesses kx and ky along the local 0 0 kw
axes x and y as well as with the
torsional stiffness kw .
2 3
It contributes to the stiffness of the 0 0 0
structure with its translational k ¼ 0 ky 0 5
6 4
stiffnesses ky along the local axes y 0 0 kw
and with the torsional stiffness kw .
The translational stiffness along the
local axis x has been considered in
MSE 1.
Step 2
Formulation of the element transformation matrices Ri due to rotation of the local
axes of the MSE. From Fig. E15.1, we have f1 ¼ f3 ¼ f4 ¼ f5 ¼ f6 ¼ 0,
f2 ¼ p=2. Hence
2 3 2 3
1 0 0 0 1 0
R ¼ R ¼ R ¼ R ¼ R ¼ 4 0 1 0 5, R ¼ 4 1 0 0 5
1 3 4 5 6 2
0 0 1 0 0 1
Step 3
i of the ith element depend on the
Because the plate is rigid, the displacements D

displacements U of the point O. Hence the relation that connects them results
from Eq. (11.11.13) by setting uJ ¼ U, uj ¼ D
i , that is
i ¼ TT U
D (1)
i
where
2 3
1 0 0
Ti ¼ 4 0 1 0 5 (2)
y i xi 1

The transformation matrix is identical to that of the plane frame
(see Table 11.11.1).
Referring to Fig. E15.1 and using Eq. (2), we obtain
2 3
1 0 0
x1 ¼ 0, y1 ¼ L=2, T1 ¼ 4 0 1 05
0:5L 0 1
2 3
1 0 0
x2 ¼ L, y2 ¼ 0, T2 ¼ 4 0 1 0 5
0 L 1
@CivilMethod
2 3
1 0 0
x3 ¼ L, y3 ¼ L=2, T3 ¼ 4 0 1 05
0:5L L 1
2 3
1 0 0
x4 ¼ L, y4 ¼ L=2, T4 ¼ 4 0 1 05
0:5L L 1
2 3
1 0 0
x5 ¼ L, y5 ¼ L=2, T5 ¼ 4 0 1 05
0:5L L 1
2 3
1 0 0
x6 ¼ L, y6 ¼ L=2, T6 ¼ 4 0 1 05
0:5L L 1
Step 4
Formulation of the local stiffness matrices ki of the elements and total stiffness
of the building.
matrix K
MSE 1 (frame k1 b1 k4 )
kx is the translational stiffness of the frame of Fig. a. Assuming flexural

vibrations and working as in Example 11.5.2, the stiffness matrix of the frame
with respect to its three degrees of freedom is
2 3
24 6L 6L
EI
k1 ¼ 3 4 6L 20L2 8L2 5
L
6L 8L2 20L2
and after the static condensation of the rotational degrees of freedom u2 ,u3
2 3
21:43 0 0
EI
k1 ¼ 3 4 0 0 05
L
0 0 0
MSE 2 (frame k1 b2 k2 )
@CivilMethod
kx is the translational stiffness of the frame of Fig. b. The stiffness matrix of the
frame with respect to its three degrees of freedom is
2 3
24 6L 6L
EI
k2 ¼ 3 4 6L 36L2 16L2 5
L
6L 16L2 36L2
and after the static condensation of the rotational degrees of freedom u2 , u3
2 3
22:61 0 0
EI
k2 ¼ 3 4 0 0 05
L
0 0 0
MSE 3, MSE 4, MSE 5, MSE 6
The stiffnesses kx , ky , kw of elements are computed from the relations
kx ¼ 3EI y =L3 , ky ¼ 3EI x =L3 , kw ¼ GI t =L
For a square cross-section a a it is: Ix ¼ Iy ¼ a 4 =12, It ¼ 0:140a 4 ¼ 1:68I .
Hence
2 3 2 3
0 0 0 3 0 0
EI EI
k3 ¼ 3 4 0 0 0 5, k4 ¼ 4 0 0 0 5
L 2 L3 2
0 0 0:672L 0 0 0:672L
2 3 2 3
3 0 0 0 0 0
EI 4 EI
5, k ¼ 4 0 3 0 5
k ¼ 3 0 3 0
5 6
L 2 L3 2
0 0 0:672L 0 0 0:672L
of the building with respect to point O is computed
The stiffness matrix K
from the relation
X
6 T
o ¼
K T i Ri k i Ri ð T i Þ T
i¼1
2 3
27:43 0 7:71L
EI 6 7
¼ 4 0 28:61 16:61L 5
L3
7:71L 16:61L 38:16L2
Step 5
Formulation of the mass matrix M of the building with respect to point O.
Because it is assumed that only the mass of the plate is taken into account,
the mass matrix will result from the inertial properties of the plate. Given that
point O coincides with the mass center of the plate, we have
m ¼ 3:36L2 r
ð2:4LÞ ð1:4LÞ3 ð2:4LÞ3 ð1:4LÞ
Io ¼ r + r ¼ 2:16L4 r
12 12
@CivilMethod
Hence
2 3
3:36 0 0
o ¼ rL2 4 0
M 3:36 0 5
2
0 0 2:16L
Step 6
The equation of motion reads
U
M €
+K
U ¼ Mb
u€g ðt Þ
where U is the vector of the relative displacement with respect to the ground
and b ¼ f cos b sin b 0 gT .
For a ¼ 0:30 m, L ¼ 3 m, E ¼ 2:1 107 kN=m2 , r ¼ 0:4 kN m1 s2 =m2
we obtain
2 3
14400:75 0 12143:25
K ¼4 0 15020:25 26160:75 5
12143:25 26160:75 180306:0
2 3
12:09 0 0
M ¼4 0 12:09 0 5
0 0 69:98
Solution using model (b)

The building is approximated by the space frame of Fig. E15.2. The top
of element 2 is assumed hinged to the slab permitting the relative rotation about
the horizontal axes. The slab behaves as a plane rigid body, which restrains the
horizontal displacements and the rotations about the z axis.
This procedure yields a stiffness matrix, which is practically identical to that
obtained using the concept of the MSE.
Another approach that facilitates the solution is to assume two additional
beam elements (elements 5 and 6) with a very small stiffness.
FIG. E15.2 Space frame modeling of the one-story building in Example 15.2.1.
@CivilMethod
Table E15.1 shows the eigenfrequencies obtained using the MSE modeling
and SAP2000. In both models the torsional stiffness of the beams has been
neglected.
TABLE E15.1 Eigenfrequencies with MSE modeling and

SAP2000.
wn MSE modeling SAP2000 modeling

w1 26.5784 26.5776
w2 34.6436 34.6437
w3 55.6964 55.6911
Example 15.2.2 Formulate the equation of motion of the two-story building shown
in Fig. E15.3 with respect to the origin of the global axes O (center of column 1)
when it is subjected to the ground motion ug ðt Þ in the x direction and evaluate
the eigenfrequencies and eigenmodes. The building is made from reinforced con-
crete with material constants E ¼ 2:1 107 kN=m2 , n ¼ 0:25. The thickness of the
slabs is 15 cm, the cross-sectional area of the columns is 40 40 cm2 , and of the
beams 30 50 cm2 . The columns are assumed fixed at the ground. The loads of the
slabs are: 5 kN=m2 live load and 1 kN=m2 covering.
Planform of the stories Section

FIG. E15.3 Two-story building in Example 15.2.2.
Solution using the concept of the MSE

Cross-sectional moment of inertia and torsional constant of the columns:
0:404
Ix ¼ Iy ¼ ¼ 21:33 104 m4 , It ¼ 0:141 0:404 ¼ 36:01 104 m4
12
Cross-sectional moment of inertia of the beams:
To simplify the solution and limit the numerical computations, the effective
breadth of the beams is neglected. Thus, we have
@CivilMethod
0:30 0:503
Ib ¼ ¼ 31:25 104 m4
12
a. Equation of motion
Step 1
Determination of the MSEs of the structure. Referring to Fig. E15.3, we distin-
guish six MSEs of the frame type, three in the direction of the x axis and three in
the direction of the y axis. The columns are treated as additional MSEs contrib-
uting with their torsional stiffness because their translational stiffnesses are
included in the frames. The stiffness matrices result using the method presented
in Example 11.5.2. The stiffness matrices of all MSEs are given below after the
static condensation of the rotational degrees of freedom.
2 3
29237:03 0 0 16737:43 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
6
k ¼6
1 7
6 16737:43 0 0 13897:72 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
44851:12 0 0 26194:78 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
6
k ¼6
2 7
6 26194:78 0 0 22307:63 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
44851:12 0 0 26194:78 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
6
k ¼6
3 7
6 26194:78 0 0 22307:63 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
41960:02 0 0 23362:92 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
6
k ¼6
4 7
6 23362:92 0 0 18828:40 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
@CivilMethod
2 3
41960:02 0 0 23362:92 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
k ¼6
5
6 23362:92
7
6 0 0 18828:40 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
27145:69 0 0 14736:02 0 0
6 0 0 0 0 0 07
6 7
6 0 0 0 0 0 07
6
k ¼6
6 7
6 14736:02 0 0 11490:06 0 0 77
4 0 0 0 0 0 05
0 0 0 0 0 0
2 3
0 0 0 0 0 0
60 0 0 0 0 0 7
6 7
60 0 15364:52 0 0 8642:40 7
k7, 8, …, 14 ¼ 6
60
7
6 0 0 0 0 0 7 7
40 0 0 0 0 0 5
0 0 8642:40 0 0 8642:40
Step 2
Formulation of the transformation matrices Ri due to the rotation of the local
axes. From Fig. E15.3, we have f1 ¼ f2 ¼ f3 ¼ f7 ¼ f8 ¼ ⋯ ¼ f14 ¼ 0,
f4 ¼ f5 ¼ f6 ¼ п=2. Hence
2 3 2 3
1 0 0 0 0 0 0 1 0 0 0 0
60 1 0 0 0 07 6 1 0 0 0 0 0 7
6 7 6 7
6 0 0 1 0 0 0 7 6 0 0 1 0 0 07
^ ^
R ¼R ¼R ¼6 ^ 6 7 ^ ^ ^ 6 7
7, R ¼ R ¼ R ¼ 6 0 0 0 0 1 0 7
1 2 3 4 5 6
60 0 0 1 0 07 6 7
40 0 0 0 1 05 4 0 0 0 1 0 0 5
0 0 0 0 0 1 0 0 0 0 0 1
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
6 07
^7 ¼ R
R ^ 14 ¼ 6 0
^8 ¼ ⋯ ¼ R 0 1 0 0 7
60 0 0 1 0 07
6 7
40 0 0 0 1 05
0 0 0 0 0 1
Step 3
Formulation of the transformation matrices Ti due to translation of the local
axes. Apparently, in this case, any point of the local x axis can be taken as point i.
Referring to Fig. E15.3 and applying Eq. (15.2.20), we obtain
@CivilMethod
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
60 0 1 0 0 07
MSE 1 : x1 ¼ x2 ¼ 0, y1 ¼ y2 ¼ 0, T1 ¼ 6
1 1 1 1 6 7
7
60 0 0 1 0 07
40 0 0 0 1 05
0 0 0 0 0 1
2 3
1 0 0 0 0 0
6 0 1 0 0 0 07
6 7
6 8 0 1 0 0 0 7
MSE 2 : x21 ¼ x22 ¼ 0, y21 ¼ y22 ¼ 8:0, T2 ¼ 6
6 0 0 0 1 0 07
7
6 7
4 0 0 0 0 1 05
0 0 0 8 0 1
2 3
1 0 0 0 0 0
6 0 1 0 0 0 07
6 7
6 16 0 1 0 0 0 7
MSE 3 : x1 ¼ x2 ¼ 0, y1 ¼ y2 ¼ 16:0, T3 ¼ 6
3 3 3 3 6 7
7
6 0 0 0 1 0 07
4 0 0 0 0 1 05
0 0 0 16 0 1
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
60 0 1 0 0 07
6 7
MSE 4 : x41 ¼ x42 ¼ 0, y41 ¼ y42 ¼ 0, T4 ¼ 6 7
60 0 0 1 0 07
6 7
40 0 0 0 1 05
0 0 0 0 0 1
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
60 5 1 0 0 07
MSE 5 : x1 ¼ x2 ¼ 5:0, y1 ¼ y2 ¼ 0, T5 ¼ 6
5 5 5 5 6 7
7
60 0 0 1 0 07
40 0 0 0 1 05
0 0 0 0 5 1
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
6 0 10 1 0 0 0 7
MSE 6 : x61 ¼ x62 ¼ 10:0, y61 ¼ y62 ¼ 0, T6 ¼ 6
60 0 0 1 0 07
7
6 7
40 0 0 0 1 05
0 0 0 0 10 1
The reaming MSEs are columns, which contribute only with their rotational
stiffness. Thus, we have independently from the directions of their local axes
@CivilMethod
2 3
1 0 0 0 0 0
60 1 0 0 0 07
6 7
60 0 1 0 0 07
T7 ¼ T8 ¼ ⋯ ¼ T14 ¼ 6
60
7
6 0 0 1 0 077
40 0 0 0 1 05
0 0 0 0 0 1
Step 4
Formulation of the stiffness matrix of the structure. Applying Eq. (15.2.43b)
yields the stiffness matrix of the structure with respect to the origin O of
the axes.
X
14 T i
Ko ¼ î k R
Ti R ^ i TT
i
2 i¼1 3
118939:27 0 1076426:90 69126:99 0 628674:79
6 61461:87 264174:87 77
6 0 111065:74 481257:02 0
6 7
6 1076426:90 481257:02 18238842:00 7
628674:79 264174:87 10509145:00 7
¼6
6 69126:99
6 0 628674:79 58512:99 0 535383:28 77
6 7
4 0 61461:87 264174:87 0 49146:87 209042:62 5
628674:79 264174:87 10509145:00 535383:28 209042:62 8827299:10
Step 5
Formulation of the mass matrix of the structure. The surface of the plates is
120:0 m2 . Neglecting the mass of the MSEs, we have

24 6
m1 ¼ m2 ¼ ð8:40 10:40 + 8:0 5:40Þ 0:15 +
9:81 9:81
1 2
¼ 127:76 kN m s
xc1 ¼ xc2 ¼ 4:17, yc1 ¼ yc2 ¼ 9:29, I1c ¼ I2c ¼ 3680:24
Hence the mass matrix with respect to the center of mass is
2 3
127:76 0 0 0 0 0
6 0 127:76 0 0 0 0 7
6 7
6 0 0 3680:24 0 0 0 7
c
M ¼6 6 7
6 0 0 0 127:76 0 0 77
4 0 0 0 0 127:76 0 5
0 0 0 0 0 3680:24
Moreover, the transformation matrix from C to O is
@CivilMethod
2 3
1 0 0 0 0 0
6 0 1 0 0 0 07
6 7
6 9:29 4:17 1 0 0 07
Tc ¼ 6
6 0
7
6 0 0 1 0 077
4 0 0 0 0 1 05
0 0 0 9:29 4:17 1
and applying Eq. (15.2.36) yields the mass matrix with respect to point O
Mo ¼ Tc Mc TTc
2 3
127:76 0 1186:89 0 0 0
6 0 127:76 532:75 0 0 0 7
6 7
6 1186:89 532:75 16928:06 0 0 0 7
¼66
7
6 0 0 0 127:76 0 1186:89 7
7
4 0 0 0 0 127:76 532:75 5
0 0 0 1186:89 532:75 16928:06
Therefore the equation of motion is
€
+ Ko U
Mo U ¼ Mo bug ðt Þ, b ¼ f 1 0 0 1 0 0 gT
b. Eigenfrequencies and eigenmodes

They are computed using the function [PHI,D]¼eig(K,M) of MATLAB.
Table E15.2 shows the eigenfrequencies obtained using the MSE modeling
and SAP2000. In both models the torsional stiffness of the beams has been
neglected.
TABLE E15.2 Eigenfrequencies MSE modeling and SAP2000.
wn MSE modeling SAP2000 modeling

w1 9.390 9.377
w2 10.192 10.085
w3 13.794 13.688
w4 34.129 33.780
w5 35.814 35.299
w6 48.829 48.332
@CivilMethod
2 3
0:0014 0:0455 0:0836 0:0025 0:0684 0:1323
6 7
6 0:0471 0:0011 0:0363 0:0768 0:0015 0:0557 7
6 7
6 7
6 0:0002 0:0003 0:0088 0:0003 0:0006 0:0139 7
6
F¼6 7
7
6 0:0018 0:0701 0:1315 0:0019 0:0440 0:0844 7
6 7
6 0:0762 0:0014 0:0561 0:0475 0:0013 0:0356 7
4 5
0:0002 0:0004 0:0139 0:0003 0:0005 0:0088
15.3 Dynamic response of multistory buildings

due to ground motion
We consider a multi-story building under uniform support excitation (Fig.
15.3.1). We first formulate the mass and stiffness matrices of the free structure
using any method described in Section 15.1 or 15.2.
The displacement vector of the j slab is written as
j ¼ Uj Iug
U (15.3.1)
where U j are the relative displacements of the j slab with respect to the
T
ground, ug ¼ f ug vg wg g the vector of the ground displacements, and I
(a)
(b)
FIG. 15.3.1 Multistory building (a) with support excitation (b).
@CivilMethod
the 3 3 unit matrix. Usually, it is wg ¼ 0. Then the total displacement vector

of the building is written
u¼u + rug (15.3.2)
where
r ¼ f I I ⋯ I gT (15.3.3)
Taking into account that the elastic and damping forces are produced only
by the relative displacements, we can write the equation of motion as (see
Section 14.7.2)
€ + Cu
Mu _ + K
u ¼ Mr€
ug (15.3.4)
it is
For ground motion in direction x,
r ¼ f 1 0 0 1 0 0 ⋯ 1 0 0 gT (15.3.5)
Apparently, it is convenient to refer the displacements to the center of mass
of the plates. Then, taking into account that u c ¼ TTc u ¼ TT
, u c uc and
c T
M ¼ Tc M Tc , Eq. (15.3.4) becomes
c c
€ + Tc Cu
Tc M c u _ + Tc K u
c ¼ Tc Mc TTc r€
ug (15.3.6)
Then premultiplying by T1

c gives
c c
€ + Cc u
Mc u _ + Kc u
c ¼ Mc rc u
€g (15.3.7)
where
Cc ¼ T1 T c 1 T
c CTc , K ¼ Tc KTc (15.3.8)
are the transformed damping and stiffness matrices with respect to the center of
mass of the plates and
8 T 9 2 1 3
>
> T1c > > T1c 0 ⋯ 0
> >
< 2 T = 6 2 1 7
6 7
rc ¼ T T
c r¼>
Tc , T1
c ¼ 60 Tc ⋯ 0 7 (15.3.9)
> ⋮ >
> 4 ⋮ ⋮ ⋱ ⋮ 5
> >
: n T ; 1
Tc 0 0 ⋯ Tnc
To avoid the inversion Tc , we can readily show that (see Section 11.11)
2 3
j 1 1 0 0
Tc ¼ 40 1 0 5 j ¼ 1, 2, …,n (15.3.10)
yj xcj 1
c
1 T
We recall the notation TT
c ¼ Tc (see Section 12.5.3).
@CivilMethod
Example 15.3.1 The building in Example 15.2.2 is subjected to the ground

motion in the x direction due to the 1999 Athens earthquake (see Chapter 6).
Determine:
2 at the top of column 1, the base shear force Qbx , the
2, W
(i) The peak values U
overturning moment Mbx , and the base torsion moment Tb .
(ii) The peak values of the shear force Qx , the bending moment Mx , and the
torsion moment at the base of column 1 (cross-section a).
Assume modal damping x ¼ 0:05 for all mode shapes
Solution
(i) Peak values U 2 , Qbx , Mbx , Tb
2, W
The eigenfrequencies and mode shapes were computed in Example 15.2.2.
Adhering to the steps in Example 14.8.1, we have
Computation of the vectors Rn .
It is R ¼ Mo r, where r ¼ f 1 0 0 1 0 0 gT . Then Eq. (14.3.14) gives
G1 ¼ 0:057, G2 ¼ 15:640, G3 ¼ 0:478
G4 ¼ 0:035, G5 ¼ 3:263, G6 ¼ 0:190
and we obtain from Eq. (14.3.13)
R1 R2 R3 R4 R5 R6
0:001 96:773 0:072 0:004 30:823 0:086
0:338 0:367 0:041 0:341 0:404 0:051
1:360 918:659 15:131 1:339 295:177 9:115
0:002 147:830 0:157 0:003 20:176 0:005
0:549 0:639 0:084 0:210 0:243 0:027
2:207 1399:871 23:322 0:811 194:117 5:854
First, we compute the quantities U st , Q st , M st , T st . The
st , W
2n 2n bxn bxn bn
modal displacements are computed from the relation U n ¼ K1 Rn , which
by virtue
of Eq. (14.3.14) and the relation K ¼ w2n M becomes
U n ¼ Gn =w fn . Hence
2
n
U st ¼ Gn f1n , W
st ¼ Gn f
2n
w2n 2n
w2n 3n
and
st st st
Qbxn ¼ R1n + R4n , Mbxn ¼ h1 R1n + ðh1 + h2 ÞR4n , Tbn ¼ R3n + R6n
Their values are given in Table E15.3. The peak modal values were
computed from Eq. (14.8.4), namely
qn0 ¼ qnst Spa ðx n , Tn Þ

where q U 2 , Qbx , Mbx , Tb Þ and Spa ðx , Tn Þ the spectral pseudoacce-
2, W n
leration, which was computed numerically using the program response_
spectrum_aem.m (see Section 6.2) with x ¼ 0:05. The computed values
@CivilMethod
TABLE E15.3 Modal quantities.

n st
U 2n 105 2 105
W st
Qbxn st
Mbxn st
Tbn
1 0.119 0.015 0.003 0.021 3.567

2 1055.796 6.248 244.603 1618.120 2318.530
3 33.054 3.489 0.229 1.578 38.454
4 0.006 0.0008 0.001 0.003 0.528

5 11.196 0.120 10.647 22.705 101.060
6 0.673 0.071 0.036 0.012 3.261
are given in Table E15.4 together with the peak values of the modal
quantities.
Applying Eq. (14.8.8) gives the correlation matrix
2 3
1000:00 602:02 61:70 4:29 3:89 2:16
6 602:02 1000:00 2:51 7
6 96:66 5:08 4:59 7
6 7
6 61:70 96:66 1000:00 10:19 9:04 4:52 7
r ¼ 10 6
3
6
7
6 4:29 5:08 10:19 1000:00 811:16 70:45 7
7
6 7
4 3:89 4:59 9:04 811:16 1000:00 92:45 5
2:16 2:51 4:52 70:45 92:45 1000:00
Obviously, the off-diagonal elements are not negligible. That was antic-
ipated because there are eigenfrequencies close to each other. Therefore, the
CQC method should be used to compute the peak values. The resulting
values are given in Table E15.5 as compared with those obtained by direct
solution of the equation of motion (Response History Analysis), namely
€
+ Ko U
Mo U ¼ Mo bug ðt Þ, b ¼ f 1 0 0 1 0 0 gT
which has been solved using the MATLAB program aem_lin_MDOF.m

(see Section 14.10.3).
Apparently, although the results obtained by the CQC method are close
to those obtained by the RHA for the displacements, the error may be appre-
ciable for the forces.
(ii) Peak values of Qx and Mx at cross-section a of column 1.
The degrees of freedom of the frame are shown in Fig. E15.4. The displace-
ments u1 , u2 result from the displacements of the MSE 1 while the rotation
from the MSE 7. From the relations (15.2.7a), (15.2.19), we have
^ i TT U
Di ¼ R
i
@CivilMethod
TABLE E15.4 Peak values of the modal quantities.
Spa ðx n , Tn Þ U 2n0 105 2n0 105

W Qbxn0 Mbxn0 Tbn0
n Tn (m/sec ) 2
(m) (rad) (kN) (kNm) (kNm)
1 0.669 3.7824 0.451 0.055 0.0123 0.081 13.493
2 0.616 4.1419 4373.003 25.866 1012.123 6702.093 9603.121
3 0.455 3.4180 112.98 11.924 0.782 5.395 131.435

4 0.184 5.3077 0.031 0.004 0.007 0.0179 2.803
5 0.175 6.0243 67.447 0.723 64.141 136.782 608.816
6 0.129 4.6813 3.151 0.331 0.169 0.055 15.264
TABLE E15.5 Peak values.

U 2 W
2 jQbx jmax jMbx jmax jTb jmax
max max
(cm) (rad) (kN) (kNm) (kNm)
CQC 4.392 0.00027 1015.55 6703.40 9605.00

RHA 4.440 0.00032 1028.28 6726.98 9806.23
FIG. E15.4 Degrees of freedom of MSE 1.
Hence
8 9 8 9
> U1> > 0 >
>
> > > >
> V >
> >
>
>
>
> 0 >
>
>
>
> 1 >
> >
> >
>
> =
< > >
< >
=
0 W
D ¼R
1 ^ T U
1 T ¼ , D 7
¼ ^ T U
R 7 T ¼ 1
> U 2 >
> > 0 >
1 7
> >
> >
> >
>
>
> >
> >
> >
>
>
> V >
> >
> 0 >
>
>
: ;
2 > : >
> ;
0 W2
which yields
1
u1 ¼ u 1 ¼ U 1 , u2 ¼ u 1 ¼ U 2 , w7 ¼ W (1)
1 2 1
The stiffness matrix of the frame before the static condensation

8 9 is
>
> u 3>

< > =
Kaa Kab ua u2 u4
K¼ , u¼ , ua ¼ and ub ¼
Kba Kbb ub u1 >
> u >
: 5> ;
u6
where 2 3

21942:86 21942:86
36876:73 25077:55 6 21942:86 21942:86 7
Kaa ¼ , Kba ¼ 6
4 8668:78 21942:86 5
7
25077:55 25077:55
8668:78 21942:86
@CivilMethod

21942:86 21942:86 8668:78 8668:78
Kab ¼
21942:85 21942:86 21942:86 21942:86
2 3
103700 26250 25600 0
6 26250 103700 0 25600 7
Kbb ¼ 6
4 25600
7
0 143522:22 26250 5
0 25600 26250 143522:22
The nodal rotations ub are computed from the relation
2 3
0:163 0:148
6 0:163 0:148 7
ub ¼ Kbb Kba ua ¼ 6
1 7 u1 (2)
4 0:026 0:107 5 u2
0:026 0:107
The stress resultants at the cross-section a (see Fig. E15.4) result from the
relations
Qxa ¼ k31 u1 + k32 u2 + k33 u3 + k34 u4

(3a)
¼ k31 u1 + k32 u6
Mxa ¼k41 u1 + k42 u2 + k43 u3 + k44 u4

(3b)
¼k41 u1 + k42 u6
GI t
Mta ¼ W1 (3c)
L
because u1 ¼ u1 , u2 ¼ u6 , u3 ¼ 0, u4 ¼ 0.

The quantities kij in the previous expression are the elements of the stiff-
ness matrix of the beam element (Fig. E15.4). By substituting the displace-
ments from Eqs. (1), (2) in the previous relations, we obtain these quantities
in terms of the displacements U 1, U 1 . Subsequently, the modal
2 , and W
quantities qnst are computed byloading with the modal loads Rn . Their peak
values qn0 , q Qxa , Mxa , Mta are obtained using the values Spa ðxn , Tn Þ,
already computed in (i). These values are shown in Tables E15.6 and
E15.7. Finally, Table E15.8 gives the sought peak as obtained by the
CQC method.
@CivilMethod
TABLE E15.6 Modal values of the stress resultants.

Cross-section a
st st
n Qxn Mxn Mtnst
1 0.0042 0.0105 0.0008

2 27.8302 72.0615 0.3190
3 0.8450 2.1981 0.1504
4 0.0006 0.0012 0.00007

5 1.2467 2.6401 0.0102
6 0.0755 0.1599 0.00740
TABLE E15.7 Modal peak values of the stress resultants.
Spa ðx n , Tn Þ
n Tn (m/sec2) Qxn0 Mxn0 Mtn0
1 0.669 3.7824 0.0159 0.0396 0.0029
2 0.616 4.1419 115.270 298.4715 1.3213
3 0.455 3.4180 2.888 7.5130 0.5139
4 0.184 5.3077 0.003 0.0064 0.0004
5 0.175 6.0243 7.5104 15.9047 0.0614
6 0.129 4.6813 0.3534 0.7487 0.0349
TABLE E15.8 Peak values of the stress resultants
jQx jmax jMx jmax jMt jmax

Cross-section (kN) (kNm) (kNm)
a 115.86 299.77 1.370
Remark The procedure applied to the solution of the previous problem is approx-
imate and is recommended only when the peak values are computed using the
pseudoacceleration spectrum provided by the earthquake codes. However, the
procedure is highly simplified if the equations of motion are solved numerically.
@CivilMethod
Actually, this is feasible only for a specified accelerogram of the ground motion.
Therefore, the earthquake codes should tend to provide a representative design
accelerogram in place of the pseudoacceleration spectrum.
15.4 Problems
Problem P15.1 Formulate the equation of motion of the two-story building in
Fig P15.1 when subjected to ground motion ug ðt Þ in the x direction and compute
the eigenfrequencies and mode shapes. The material of the structure is RC with
E ¼ 2:1 107 kN=m2 , n ¼ 0:25. Assume: The thickness of the slabs 15 cm,
cross-sectional dimensions of beams 30 50 cm2 , of columns 1, 3, 5, 7, 9
30 30 cm2 , columns 2, 8 25 40 cm2 , and columns 4, 640 25 cm2 . All col-
umns are fixed to the wall of the basement except for column 3, which is supported
elastically to the ground via an isolated footing with an elastic foundation parameter
k ¼ 30:000 kN=m3 . The beams connecting the footing to the basement walls have
dimensions 30 50 cm2 . Live load of the roof slab 5 kN=m2 and of the mezzanine
slab 10 kN=m2 .
Beam
Beam
Foundation Mezzanine slab
Beam
Roof slab Cross section a–a

@CivilMethod

[1] E.J. Sapountzakis, J.T. Katsikadelis, Analysis of plates reinforced with beams, Comput. Mech.
2 (2000) 66–74.
[2] E.J. Sapountzakis, J.T. Katsikadelis, Dynamic analysis of elastic plates reinforced with beams
of doubly-symmetrical cross section, Comput. Mech. 23 (1999) 430–439.
[3] E.J. Sapountzakis, J.T. Katsikadelis, A new model for the analysis of composite steel-concrete
slab beam structures with deformable connection, Comput. Mech. 31 (3–4) (2003) 340–349.
[4] J.T. Katsikadelis, N.G. Babouskos, Debonding in beam-reinforced plates, in: Proc. International
Conference on Damage Mechanics, Belgrade, Serbia, 25–27 June, 2012.
[5] R. Szilard, Theory and analysis of plates, in: Classical and Numerical Methods, Prentice Hall,
New Jersey, 1974.
[6] J.T. Katsikadelis, E.J. Sapountzakis, A realistic estimation of the effective breadth of ribbed
plates, Int. J. Solids Struct. 39 (4) (2002) 897–910.
[7] J.T. Katsikadelis, Dynamic Analysis of Structures, Symmetria Publications, Athens, 2012
(in Greek).
@CivilMethod
Chapter 16
Base isolation
Chapter outline
16.1 Introduction 725 16.3.2 Reduction of the DOF
16.2 Analysis of the one-story of the superstructure
building with base isolation 727 using mode shapes 745
16.2.1 Linear response of the 16.3.3 Reduction of the
isolation systems 727 superstructure DOF
16.2.2 Modeling of nonlinear using Ritz vectors 745
response of isolation 16.3.4 Linear response of
systems 735 the isolation system 746
16.3 The multistory building with 16.3.5 Nonlinear response
base isolation 741 of the isolation
16.3.1 The equation of motion system 747
of the multistory building 16.4 Problems 753
with base isolation 741 References and further reading 753
16.1 Introduction
Base isolation, also known as seismic isolation, has always been a matter of
great interest to engineers, who have sought to isolate the structure from ground
motion in order to mitigate as far as possible the forces and deformations due to
the earthquake motion of the ground. This would ensure the integrity of the
structure and its contents as well as the protection of its users from injury or
death. Although the concept of base isolation has its roots in the distant past,
more than a century ago, it has only nowadays been a means of designing civil
engineering structures, mainly buildings and bridges, or even protecting old
works of historical value, thanks to the development of reliable low-cost base
isolation systems that guarantee the desirable performance characteristics prac-
tically unlimited. The related literature is extensive [1–4], and the research on
the subject is ongoing due to great interest.
Base isolation uses a collection of structural elements that substantially
decouple a superstructure from its substructure that is resting on a shaking
ground. All in-use isolation systems, despite their wide variation in detail,
can be categorized into two main types as they follow two basic approaches with
certain common features [3]. The first type includes the elastomeric isolators
and the second the sliding systems. The elastomeric isolators perform as a layer

@CivilMethod
with small horizontal stiffness, which is interposed between ground and super-
structure. They allow a relative displacement that can reach their height. Their
vertical stiffness is very large so that they can be considered practically unde-
formed when subjected to large vertical loads. They usually have a cylindrical
shape and are made of hard natural rubber or neoprene layers alternating with
steel plates. Inasmuch as the damping of the rubber is small, mechanical
dampers, hydraulic dampers, or lead cores are used in the elastomeric bearings
(Fig. 16.1.1) to increase damping. The latter bearings are known by the name
LRB (lead rubber bearings).
Lead core
Top steel plate

Steel plates
Rubber layer
Bottom steel plate
Section
FIG. 16.1.1 Lead rubber bearing.
The second type uses sliders between the ground and superstructure. The
shear force transmitted through the interface is significantly reduced by
decreasing the friction coefficient as much as practically possible. However,
friction must not be completely eliminated to prevent movements due to wind
or minor earthquakes. In this type of bearing, the displacements are controlled
by springs, elastomeric devices, or even by curved sliding surfaces. The sliding
interface of these bearings mainly consists of stainless steel and Teflon. The
friction pendulum is a device of this type, Fig. 16.1.2.
Curved surface Teflon coating

from stainless steel
R Articulated slider
Section
FIG. 16.1.2 Friction pendulum bearing.
The curvature of the surface provides resisting forces, which bring the struc-
ture back to its original equilibrium position after displacement. The study of
the dynamic response of structures with sliders is quite complex due to the non-
linearity of the sliding process. The isolation devices reduce stiffness and
increase damping between the ground and structure. Reducing stiffness aims
at shifting the fundamental period of the structure away from the characteristic
@CivilMethod
Base isolation Chapter 16 727
period of the ground motion. Isolated structures have large fundamental vibra-
tion periods compared with the predominant periods of most seismic motions.
Increasing the damping reduces the displacements on the isolators. However,
excessive damping may introduce forces in the structure that increase deforma-
tions in the higher modes. Fig. 16.1.3 shows the pseudoacceleration spectrum of
the Athens earthquake (see Chapter 6). As was pointed out in Section 6.2, the
maximum elastic base shear force of the building is calculated from the
relationship
max Q0 ¼ mS pa ðT , xÞ (16.1.1)
It becomes obvious from Fig. 16.1.3, that for a building with period
TF ¼ 0:6s it is Spa 425cm=s2 , while for an isolated building with TI ¼ 2s
it is Spa 50cm=s2 (see Section 16.2.1). Consequently, the peak base shear
force is considerably reduced, by 8.5 times of that before isolation.
Design spectrum
T (s)
FIG. 16.1.3 Pseudoacceleration spectrum of the Athens earthquake Spa ðT , x Þ, (x ¼ 0:05).
The literature on the dynamic response of base-isolated structures is exten-

sive, and research in this area continues intensively due to the importance of the
subject. Many interesting research programs are under way. This chapter is
introductory and is mainly limited to understanding the dynamic response of
base-isolated structures.
16.2 Analysis of the one-story building with base isolation

16.2.1 Linear response of the isolation systems
The study of the dynamic response of the isolated one-story building allows us
to identify the effectiveness of base isolation in reducing the earthquake-
induced forces. Fig. 16.2.1 shows the model of a one-story shear building to
@CivilMethod
be isolated. More specifically, Fig. 16.2.1a shows the building resting directly
on the ground while Fig. 16.2.1b shows it mounted on a rigid isolated slab. From
now on, we will call the first structure a fixed-base structure while the second
one is an isolated structure. Although the behavior of the isolation system is
nonlinear, the assumption of linear behavior facilitates our purpose in the
first stage.
Isolation
system
Gap Base slab
(a) (b)
FIG. 16.2.1 One-story building: (a) fixed-base structure and (b) isolated structure.
Quantities referred to the fixed-base structure are provided with the subscript
F while those referred to the isolated structure with the subscript I . Moreover,
quantities referred to the isolation system are denoted with the subscript b.
The fixed-base structure is modeled as an SDOF system while the isolated
structure is modeled as a two-degree-of-freedom system (Fig. 16.2.2). Thus,
we have
rffiffiffiffiffi
k
wF ¼ (16.2.1a)
m
2p
TF ¼ (16.2.1b)
wF
c
xF ¼ (16.2.1c)
2mwF
FIG. 16.2.2 Displacements of the isolated structure.
The equations of motion of the isolated structure with respect to the relative
displacements ub and u for ground motion ug ðt Þ are obtained using the method
of equilibrium of forces
@CivilMethod

m 0 u€ c c u_ k k u
+ +
0 mb u€b c c + cb u_ b k k + kb ub
m
¼ u€g (16.2.2)
mb
or setting mb ¼ am, cb ¼ bc, kb ¼ gk

1 0 u€ 1 1 u_ 1 1 u
+ 2xF wF + wF
2
0 a u €b 1 1 + b u_ b 1 1 + g ub
1
¼ u€ (16.2.3)
a g
The determinant of the eigenvalue problem for the undamped vibrations is

1l 1
¼0 (16.2.4)
1 ð1 + g Þ al
where
w2
l¼ (16.2.5)
w2F
The characteristic equation of Eq. (16.2.4) is
f ðlÞ ¼ al2 ð1 + a + g Þl + g ¼ 0 (16.2.6)
which yields
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1+a+g ð1 + a + g Þ2 4ag
l1, 2 ¼ (16.2.7)
2a 2a

1 1
F¼ (16.2.8)
1 l1 1 l2
The stiffness kb of the isolation system is very small compared to the stiff-
ness k of the fixed-base structure while mb is of the same order of magnitude as
m or smaller. Thus, we may limit our study to the following two cases:
(a) g ¼ 0:1 and a ¼ 1
9
l1 ¼ 0:0487, l2 ¼ 2:0514 >
2 3>>
>
1:000 0:951 >=
w1 ¼ 0:220wF , w2 ¼ 1:432wF , F1 ¼ 4 5 (16.2.9)
>
0:951 1:000 > >
>
>
;
T1 ¼ 4:529TF , T2 ¼ 0:698TF
@CivilMethod
(b) g ¼ 0:1 and a ¼ 0:1

9
l1 ¼ 0:084, l2 ¼ 11:916 >
=
1:000 0:0916
w1 ¼ 0:289wF , w2 ¼ 3:452wF , F2 ¼ (16.2.10)
0:916 1:000 > ;
T1 ¼ 3:452TF , T2 ¼ 0:289TF
(a)
(b)
FIG. 16.2.3 Mode shapes and periods of the isolated structure.
The mode shapes together with the corresponding natural periods for the
examined two cases are shown in Fig. 16.2.3. The first mode has a large natural
period, which differs slightly from the mode of a rigid isolated structure,
f1 0:916gT f 1 1 gT . Obviously, only the isolation system is deformed in this
mode, whereas the structure behaves as a rigid body. Therefore, this mode is
called the isolation mode. The second mode has a small natural period and
expresses the deformation of the structure. Therefore, it is called the structural
mode. This mode, however, contributes little to the earthquake-induced forces
in the structure (see Example 16.2.1).
At this point, we make a remarkable observation. If we set l1 ¼ g=ð1 + aÞ in
Eq. (16.2.6) and take into account that g is small, we have
f ðl1 Þ ¼ g 2 0 (16.2.11)
that is, this value of l1 is the first eigenvalue as it satisfies Eq. (16.2.4). If this
value of l1 is introduced in Eq. (16.2.5), we obtain
@CivilMethod
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kb
w1 (16.2.12)
m + mb
This suggests defining the following nominal dynamic parameters

2p
Tb (16.2.13a)
wb
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kb
wb (16.2.13b)
m + mb
cb
xb (16.2.13c)
2ðm + mb Þwb
The previously defined quantities Tb and x b can be interpreted as the natural
period and damping ratio, respectively, (i) of the isolated structure considered as
a rigid body or (ii) of the isolation system where the structure is also considered
as a rigid body.
Eqs. (16.2.13a)–(16.2.13c) are very useful because they allow estimating the
desired fundamental natural period and the damping of the isolated structure
with sufficient accuracy, or in other words the characteristic parameters of
the isolation system so that the structure has a specified natural period and
damping, which is the objective of seismic isolation. For example, we assume
that a fixed-base structure has a mass m ¼ 25kNm1 s2 , a base mass
mb ¼ 15kNm1 s2 , and we want to have Tb ¼ 2s and x b ¼ 15%. From
Eqs. (16.2.13a)–(16.2.13c), we obtain
2p
wb ¼ ¼ 3:141, kb ¼ 394:63kNm1 , cb ¼ 37:68kNm1 s1
Tb
Further, Eqs. (16.2.1c), (16.2.13c) yield
c b m + m b wb x b
b¼ ¼
c m wF x F
or
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x b
b¼ ð1 + a Þg (16.2.14)
xF
For the previously selected values of a,g and taking xb =xF ¼ 7:5, we obtain
the damping matrices

1 1
g ¼ 0:1, a ¼ 1 ) b ¼ 3:35 C1 ¼ c
1 4:35

1 1
g ¼ 0:1, a ¼ 0:1 ) b ¼ 2:48 C2 ¼ c
1 3:48
In order to examine the orthogonality of C1 and C2 with respect to the mode
shapes of the undamped free vibrations, we compute the matrices
@CivilMethod

3:032 3:0902
FT1 C1 F1¼c (16.2.15a)
3:0902 7:1564

2:0879 2:1799
F2 C2 F2 ¼ c
T
(16.2.15b)
2:1799 3:6716
We observe that the orthogonality condition is not satisfied. Moreover, the
off-diagonal elements are of the same order of magnitude as the diagonal ones,
hence they cannot be neglected (see Example 12.10.3). This was anticipated
because the difference in the damping between the structure and the isolation
is large. Therefore, the method of mode superposition presented in
Chapter 14 using real mode shapes cannot be applied. On the other hand, the
uncoupling of the equations of motion using complex modes as presented in
Section 12.10 is not recommended because this procedure, though effective,
is not convenient to understand the physical response of the structure. The
numerical solution of the equations of motion offers a simple and efficient alter-
native means to overcome this problem. Nevertheless, the use of the mode
superposition method with the real modes of the undamped vibrations allows
us to draw useful conclusions for the dynamic response of the isolated
single-story building, as will be shown in the next example.
Example 16.2.1 Compute the peak values of the base shear force Qb , the over-
turning moment Mb , the displacement ub , and the relative displacement u ub
as well as the peak value of the shear force at the base of the columns of the one-
story isolated building of Fig. 16.2.1 when it is subjected to the Athens 1999
earthquake using the response spectrum method. Compare the computed quan-
tities with those of the fixed-base structure. Assume: m ¼ 50kNm1 s2 ,
k ¼ 15, 000kNm1 , mb ¼ 0:7m, Tb ¼ 2s, xF ¼ 3%, xb ¼ 15%.
Solution
(i) Characteristic parameters
wF ¼ 17:32s1 , TF ¼ 0:362s
wb ¼ 2p=Tb ¼ 3:141s1 , kb ¼ w2b ðm + mb Þ ¼ 838:59kNm1 m
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x
a ¼ mb =m ¼ 0:7, g ¼ kb =k ¼ 0:056, b ¼ ð1 + aÞg b ¼ 1:541
xF
c ¼ 51:961kNm1 s1
(ii) Eigenfrequencies, mode shapes, and modal quantities.
Eqs. (16.2.7), (16.2.8) give
l1 ¼0:0323, w1 ¼ 3:113, l2 ¼ 2:476, w2 ¼ 27:255

T1 ¼2:018s, T2 ¼ 0:230s

1 0:677
F¼
0:968 1
@CivilMethod
Modal masses and modal dampings

1 0 1:655 0 M1 0
M ¼m , FT MF ¼ m ¼
0 0:7 0 1:158 0 M2

1 1 1:444 1:438
C ¼c , FT CF ¼ c
1 2:541 1:438 4:354
For the uncoupling of modes, we take the approximate damping matrix

1:444 0 C1 0
FT CF c ¼
0 4:354 0 C2
x1 ¼C1 =2M1 w1 ¼ 0:145, x 2 ¼ C2 =2M2 w2 ¼ 0:0716
We observe that the first modal damping differs little from the isolation
damping x b .
st
(iii) Computation of the modal quantities Qbn st
, Mbn st
, ubn , ðu ub Þst
n.
The loading due to ground motion is pðt Þ ¼ M1u€g ðt Þ with 1 ¼ f1 1gT .

Namely, it has the form pðt Þ ¼ Rf ðt Þ, where RT ¼ m f 1 0:7 g ¼ f 50 35 g.
Following the procedure described in Section 14.3 (Example 14.3.1), we have
fT1 R fT R
G1 ¼ ¼ 1:0132, G2 ¼ 2 ¼ 0:0195
M1 M2

50:660 0:661
R1 ¼G1 Mf1 ¼ , R2 ¼ G2 Mf2 ¼
34:310 0:683
X2
st
Qbn ¼ Rkn ¼ 1T Rn
k¼1
X
2
st
Mbn ¼ hk Rkn ¼ hT Rn , hT ¼ f 3:30 0:30 g
k¼1
The vector ust

n is computed from the relation
st
ubn Gn
ust ¼ ¼ K1 Rn ¼ 2 fn
n st
un wn
hence

0:1045 0:1776
ust
1 ¼ , u2 ¼
st
104
0:1012 0:2627
Assuming that the columns of the building are identical, the base shear force is
k
Qnst ¼ ðu ub Þst
n
2
The computed modal values are given in Table E16.1.
@CivilMethod
TABLE E16.1 Modal contribution values.

st st st
n Qbn Mbn ubn (cm) ðu ub Þst
n (cm)
Qnst
1 84.97 177.47 10.13 0.338 25.330

4 4
2 0.022 1.97 0.26 10 0.44 10 0.330
(iv) Computation of the peak values jQb jmax , jMb jmax , jub jmax , ju ub jmax ,
jQ jmax .
They are computed using the response spectrum method. First the quantities Qbn0 ,
Mbn0 , ubn0 , ðu ub Þn0 , Qn0 are computed using the relation (14.8.4), that is,
qn0 ¼ qnst Spa ðTn , xn Þ
The quantities Spa ðxn , Tn Þ are computed using the program response_
spectrum_aem.m given in Chapter 6 for the Athens earthquake and are pre-
sented together with the peak modal values in Table E16.2. For the fixed-base
structure, it is Spa ðT ¼ 0:362, x ¼ 0:03Þ ¼ 4:116cm=s2 .
TABLE E16.2 Peak modal values Qbn0 , Mbn0 , ubn0 , ðu ub Þn0 , Qn0 .
Spa ðTn , x n Þ Qbn0 Mbn0 ðu ub Þn0 Qn0
n Tn (s) xn (cm/s2) (kN) (kN m1) ubn0 (cm) (cm) (kN)
1 2.018 0.146 37.05 31.48 65.75 3.753 0.125 9.38
2 0.230 0.0716 621.01 0.14 12.23 0.161e 3 0.023 2.05
The matrix of the correlation coefficients obtained from Eq. (14.8.8) is

1 0:00495
r¼
0:00495 1
TABLE E16.3 Peak values obtained by CQC.
jQb jmax jMb jmax jub jmax ju ub jmax jQ jmax

Building (kN) (kN m1) (cm) (cm) (kN)
Isolated 31.49 66.83 3.748 0.126 9.59
Fixed-base 205.80 617.40 0 1.372 102.90

Reduction % 84.7 89.2 90.8 90.7
@CivilMethod
Table E16.3 shows the peak values for the isolated building obtained by the
CQC method together with the corresponding values obtained for the fixed-base
structure. The latter values are computed by treating the structure as an SDOF
system with T ¼ 0:362s and x ¼ 0:03. Obviously, the reduction of all quantities
is considerable in the isolated structure.
In obtaining the previous solution, an approximate damping matrix was used
by omitting the off-diagonal elements in order to avoid the modal superposition
method with complex eigenfrequencies and eigenmodes (see Section 12.10).
This, however, can be surpassed by solving Eq. (16.2.2) numerically. For the
data of the problem, the computed peak values are given in Table E16.4.
TABLE E16.4 Peak values obtained using a numerical solution of Eq. (16.2.2).
Isolated jQb jmax jMb jmax jub jmax ju ub jmax jQ jmax
building (kN) (kN m1) (cm) (cm) (kN)
Approximate 31.48 66.82 3.753 0.126 9.59
Numerical 31.28 95.83 3.724 0.193 14.46
16.2.2 Modeling of nonlinear response of isolation systems

16.2.2.1 Linear springs or laminated rubber bearings
with flat sliders
In this case, the nonlinear equations of motion of the isolated one-story building
are written as
m u€ + cðu_ u_ b Þ + k ðu ub Þ ¼ m u€g (16.2.16a)
mb u€b cðu_ u_ b Þ + fD k ðu ub Þ + fS ¼ mb u€g (16.2.16b)
where fD is the damping force of the isolation system. The restoring force fS of
the bearing represents the force that brings the system back to the equilibrium
position. It may be linear, given by
fS ¼ kb ub (16.2.17)
or nonlinear.
The linear springs and the laminated rubber bearing provide the restoring in
the isolation system. The damping in these systems is due to the friction
between the sliding interfaces. The motion of the sliding bearings can be
resolved in two modes:
(a) Stick mode. This occurs when the shear force at the sliding interface of the
bearing fails to overcome the maximum friction force. In such an occur-
rence, the relative velocity between the interfaces is zero.
@CivilMethod
(b) Slip mode. When the shear force at the interface of the bearing reaches the
maximum friction force, the bearing is forced to slide.
The damping force fD due to friction at the sliding interfaces satisfies the
condition
jfD j mW (16.2.18)
where m is the friction coefficient and W the weight applied to the bearing.
The friction coefficient can be either constant as considered in Coulomb’s
model (see Chapter 7), or depending on the sliding velocity and the bearing
pressure.
For Teflon-steel interfaces, it has been proposed as [5,6],
m ¼ mmax ðmmax mmin Þea|u_ | (16.2.19)
where mmax and mmin are the maximum and minimum values of the friction
coefficient, respectively, u_ the sliding velocity, and a a coefficient controlling
the transition between mmax and mmin . Obviously, it is m ¼ mmin if u_ ¼ 0 and
m ! mmax if u_ ! 1.
Hence the condition for nonsliding is
jfD j < mW and u_ ¼ 0 (16.2.20a)
while for sliding
fD ¼ mW sgnðu_ Þ (16.2.20b)
16.2.2.2 Linear springs or rubber bearings and nonlinear dampers

In another model, the nonlinear damping force is given by
fD ¼ cb sgnðu_ Þju_ ja (16.2.21)
where sgn denotes the signum function. The exponent a describes the
nonlinearity of the damping force and cb is the damping coefficient.
16.2.2.3 Friction pendulum bearing

In these devices, the restoring force is provided by the curvature and the
damping force is due to friction (see Section 16.2.2.2).
Fig. 16.2.4 shows the friction pendulum bearing in operation. The position
shown is not the stable equilibrium position. In order to determine the restoring
force fS , we consider the lift h of the structure due to its horizontal displacement
u from the equilibrium position. Referring to this figure, we have
h ¼ Rðr
1 cos fÞ ffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2 (16.2.22)
¼R 1 1
R
@CivilMethod
where R is the radius of curvature of the sliding surface and f is the

sliding angle.
The potential energy A due to gravity accumulated during this lift is
" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #
u 2ffi
A ¼ Wh ¼ WR 1 1 (16.2.23)
R
where W ¼ mg is the weight supported by the bearing.

The restoring force is
dA W u
fS ¼ ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2ffi (16.2.24)
du R
1
R
FIG. 16.2.4 Friction pendulum bearing in operation.
and it represents the stiffness of the isolation system. Obviously, this force
depends nonlinearly on the displacement u. Expansion of fS in Taylor’s series
with respect to u gives
W W 3W 5

fS ¼ u + 3 u3 + 5
u + O u7
R 2R 8R
1 u 2 3 u 4
6 (16.2.25)
W
¼ u 1+ + +O u
R 2 R 8 R
The approximation by a finite number of terms depends on the ratio umax =R.
For example, if it is umax ¼ 0:25m and R ¼ 1:00m, the contribution of the
second term is 3:12% and that of the third 0:14%. Hence, the response can
be reasonably approximated as linear
W
fS ¼ u (16.2.26)
R
Thus the stiffness of the bearing is
W
kb ¼ (16.2.27)
R
@CivilMethod
yielding
wb ¼ g=R (16.2.28a)
Tb ¼ 2p R=g (16.2.28b)
We observe that the natural period of the isolated structure does not depend
on the supported weight, hence on the mass of the structure. Eq. (16.2.28b) can
be employed to establish the radius of curvature R of the bearing for a specified
natural period. For example, if the desired natural period Tb ¼ 2s, the radius of
curvature should be
g
R¼ Tb2 1:00m (16.2.29)
ð2pÞ2
The Taylor’s series expansion of the restoring force, though not necessary
from the computational point of view, gives an insight into the response of the
friction pendulum bearing. Nevertheless, linearization must be done with great
caution because the nonlinear behavior of the system may be completely differ-
ent. When, for example, we keep the cubic term in the expression (16.2.25), the
equation of motion of the SDOF system takes the form
u€ + w2 u + mu 3 ¼ u€g (16.2.30)
where
w¼ g=R (16.2.31a)
g
m¼ 3>0 (16.2.31b)
2R
Eq. (16.2.30) is known as the Duffing equation with hardening. The eigen-
frequency of free vibrations depends on the amplitude of the vibration. The
is complicated,
response of the system to harmonic excitations (u€g ¼ p0 cos wt)
especially in the region of resonance where instability and jump phenomena
arise. The response becomes even more complicated in the presence of damp-
ing. Modern developments use fractional derivative models to simulate the dis-
sipation forces. However, such analyses are beyond the scope of this book.
Therefore, the interested reader is advised to look in the relevant literature.
16.2.2.4 High damping rubber bearing or lead rubber

bearing—Bilinear model
In this case, the nonlinear equations of motion of the one-story isolated building
become
mb u€b cðu_ u_ b Þ k ðu ub Þ + fS ¼ mb u€g (16.2.32b)
@CivilMethod
The curve representing the actual variation of the restoring force fS versus
the displacement in this model is approximated by the bilinear response
(Fig. 16.2.5), where Ke is the elastic stiffness, Ky the stiffness in the yield phase,
Fy the yield force, and Uy the yield displacements
FIG. 16.2.5 Restoring force in the bilinear hysteretic model.
The solution of the motion equations is accomplished by the step-by-step

incremental numerical integration. This technique presents considerable pro-
gramming difficulties because, at each step, we have to follow the hysteretic
loop (Example 5.4.4).
The solution is greatly simplified by the linearization of the nonlinear
behavior. In this approach, the hysteretic isolator behaves with an equivalent
viscoelastic response as follows:
fS ¼ Keff ub + cb u_ b (16.2.33)
The effective modulus of elasticity Keff (Fig. 16.2.6) is computed from the
relation
FIG. 16.2.6 Hysteresis loop in the bilinear model.
@CivilMethod

Ke Uy + Ky umax Uy
Keff ¼ (16.2.34)
umax
Subsequently, the viscous part cb can be obtained from
cb ¼ 2xeq ðmb + m Þwb (16.2.35)
where the equivalent linear viscous damping ratio xeq is calculated using
Eqs. (7.3.2), (7.3.3), which for r ¼ umax and Keff give
WD
x eq ¼ 2
(16.2.36)
2pKeff umax
The area WD of the hysteresis loop is obtained from Fig. 16.2.6.
1
WD ¼2Fd 2umax 2 2Fd 2Uy

2 (16.2.37)
¼4Fd umax Uy
This approach is permitted by the regulations (design codes) and is
very useful, at least at the stage of preliminary design. This effective modulus
is used in Eqs. (16.2.13a)–(16.2.13c) with kb ¼ Keff to compute the natural
period
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Keff 2p
wb , Tb (16.2.38)
m + mb wb
Eqs. (16.2.36)–(16.2.38) are used in a repetitive process to determine the
characteristics of the seismic isolation system at each step.
16.2.2.5 Hysteretic isolators—Bouc-Wen model

This model, which is used for most isolation bearings, is described by a
differential equation. The restoring force is given by the relation
fS ¼ aKe u + ð1 aÞFy Z (16.2.39)
where a ¼ Ky =Ke , and Ke , Ky are the elastic stiffness and the stiffness after
yielding; Z is a dimensionless quantity determining the hysteretic behavior
of the model and is obtained from the solution of the nonlinear differential
equation.
Z_ Uy ¼ g ju_ jjZ jn1 z bu_ jZ jn + Au_ (16.2.40)
in which g, n,b and A are dimensionless shape parameters determined from the
hysteresis loop and Uy is the yield displacement. For certain materials it is
@CivilMethod
b + g ¼ 1, A ¼ 1, and n ¼ 2. Eq. (16.2.40) can be written in the computationally

more convenient form
Z_ Uy ¼ fA jZ jn ½g sgnðuZ
_ Þ + b gu_ (16.2.41)
In this case, the nonlinear equations of motion of the one-story building
become
mb u€b cðu_ u_ b Þ k ðu ub Þ + fS ¼ mb u€g (16.2.42b)
where fS is obtained from Eq. (16.2.39) with u ¼ ub . Eqs. (16.2.42a), (16.2.42b),
(16.2.41) constitute a system of three nonlinear equations, which can be
efficiently solved using the method presented in Refs. [7–10].
The Bouc-Wen model, like all models whose hysteric behavior is described
by a differential equation, exhibits two remarkable advantages over the bilinear
trace model:
a. It can simulate a variety of hysterical behaviors, ranging from the simple
elastoplastic behavior to the general hysteretic one, by selecting appropri-
ately the shape parameters.
b. It simplifies the numerical solution of the nonlinear equations of motion
because it is not required to follow the hysteresis loop during the
step-by-step integration.
16.3 The multistory building with base isolation

16.3.1 The equation of motion of the multistory building with base
isolation
We consider an isolated multistory building of n floors, with diaphragmatic
behavior of its slabs, Fig. 16.3.1. The building is not symmetric with respect
to the horizontal axes x, y. Therefore, the stiffness matrix of the fixed-base
structure (superstructure) is not diagonal. The mass matrix is not diagonal,
too, in the general case where the centers of mass of the plates do not lie on
the same vertical line. The vertical axis z of the reference of motion of the plates
passes through the center of mass O of the base plate.
The fixed-base structure has N ¼ 3n degrees of freedom. To formulate its
equation of motion due to ground motion, we decompose the total displacement
vector as [10, 11]
utot ¼ u + rub + rug (16.3.1)
where
r ¼ f I I ⋯ I gT , I is the 3 3 unit matrix.
@CivilMethod
FIG. 16.3.1 Multistory building with base isolation.
FIG. 16.3.2 Isolation base plate.
8 9 8 9
< un = < uj =
u¼ ⋮ with uj ¼ vj , j ¼ 1, 2, …,n (16.3.2)
: ; : ;
u1 wj
is the 3n 1 vector of the relative displacements of the fixed-base structure with
respect to the base (Fig. 16.3.2),
8 9
< ub =
u b ¼ vb (16.3.3)
: ;
wb
is the 3 1 vector of the relative displacements of the base with respect to the
ground and
8 9
< ug =
ug ¼ v g (16.3.4)
: ;
wg
is the 3 1 vector of the ground displacements. Usually, it is wg ¼ 0.
@CivilMethod
Hence, the equation of motion of the fixed-base structure is written

M€u + Cu_ + Ku ¼ Mr u €g
€b + u (16.3.5)
in which M,C,K are the 3n 3n mass, damping, and stiffness matrices of the
fixed base structure, respectively. They are formulated using the methods
described in previous sections.
The equation of motion of the base results from its dynamic equilibrium with
respect to the point of reference of the motion O, which is, as we have assumed,
its center of mass. The forces exerted at the isolation base are shown in
Fig. 16.3.3 and are:
FIG. 16.3.3 Forces exerted on the isolation base plate.
1. The vector of the inertia forces due to the base motion is

8 b9
< fIx >
> =
f I ¼ fIyb
b
(16.3.6)
: b >
> ;
fI w
which is expressed in terms of the vector of the total acceleration

f bI ¼ Mb u €g
€b + u (16.3.7)
where
2 3
mb 0 0
Mb ¼ 4 0 m b 0 5 (16.3.8)
0 0 Ib
is the mass matrix of the base with respect to its center of mass.
2. The vector of the forces exerted on the base by the superstructure is
8 a9
< Fx >
> =
F ¼ Fya
a
(16.3.9)
: a>
> ;
Fw
which is expressed as the sum of the elastic forces f Sj and damping forces
f Dj of all the floors above the base. Namely,
@CivilMethod
X
n X
n
Fa ¼ f Dj + f Sj (16.3.10)
j¼1 j¼1
which we may write as

Fa ¼ rT ðf D + f S Þ (16.3.11)
But it is
fD + fS ¼ fI

¼M u €+r u€b + u
€g
Therefore

Fa ¼ rT M u€+r u€b + u
€g (16.3.12)
3. The restoring forces exerted by the isolation system are

8 9
< Rx =
Rðu_ b , ub Þ ¼ Ry (16.3.13)
: ;
Rw
The expression for Rðu_ b , ub Þ depends on the response of the bearings. It may
be linear, nonlinear, or a combination of both. The latter case is the general and
has the form
Rðub , u_ b Þ ¼ Cb u_ b + Kb ub + f nl
b (16.3.14)
where Cb is the matrix of the damping coefficients in case of the presence of
linear viscous damping devices, Kb the stiffness matrix of the isolators with
a linear response, and f nl
b the vector of nonlinear restoring forces due to the
hysteretic or friction isolators.
On the basis of the above, the equation of the dynamic equilibrium of the
base gives
f bI Fa + Rðub , u_ b Þ ¼ 0 (16.3.15)
or taking into account Eqs. (16.3.7), (16.3.12), (16.3.14), the equation of motion
of the isolation base becomes

Mb u €g + rT M u
€b + u €+r u €b + u€g + Cb u_ b + Kb ub + f nl
b ¼0 (16.3.16)
Eqs. (16.3.5), (16.3.16) represent the equations of motion of the isolated
structure, which in terms of the unknown displacements u and ub read
M€ ub + Cu_ + Ku ¼ Mr€
u + Mr€ ug (16.3.17a)

rT M€ €b + Cb u_ b + Kb ub + f nl
u + rT Mr + Mb u €g (16.3.17b)
b ¼ r Mr + Mb u
T
@CivilMethod
16.3.2 Reduction of the DOF of the superstructure

using mode shapes
The above equations can be further simplified by transforming Eq. (16.3.17a)
using modal coordinates, namely
u ¼ FY (16.3.18)
where F is the matrix of the mass-normalized eigenmodes.
Substituting Eq. (16.3.18) into Eqs. (16.3.17a), (16.3.17b) and premultiply-
ing the first of the equations by FT yields
"h i #( ) " h i #
I FT Mr Y€ 2x j wj 0 Y_ w2j 0 Y
+ +
r MF rT Mr + Mb
T €b
u 0 Cb u_ b 0 Kb ub
T
0 F Mr
+ nl ¼ T €g
u (16.3.19)
fb r Mr + Mb
h i
where 2xj wj and w2j are diagonal matrices with dimensions 3n 3n. Note
the matrix C is assumed proportional. We observe that the nonlinearity is
limited to the three equations of motion of the isolation base. Eq. (16.3.19)
has the form
uI + C*u_ I + K*uI + f nl
M*€ I ¼ p I ðt Þ (16.3.20)
where

Y
uI ¼ (16.3.21)
ub
is the vector of the generalized displacements of the isolated structure. The
remaining quantities in Eq. (16.3.20) result by comparing it with Eq. (16.3.19).
16.3.3 Reduction of the superstructure DOF using Ritz vectors

The linear response of the superstructure allows the reduction of the number of
equations (16.3.17a) using the Raleigh-Ritz method presented in Section 14.5.
If u ¼ Ψz is the Ritz transformation, then substituting it into Eqs. (16.3.17a),
(16.3.17b) and premultiplying the first equation by ΨT yields
e z + ΨT Mr€
M€ ub + C_e z + Kz
e ¼ ΨT Mr€ ug (16.3.22a)

rT MΨ€z + rT Mr + Mb u €b + Cb u_ b + Kb ub + f nl
b ¼ r Mr + Mb u
T
€g
(16.3.22b)
e e e
where M ¼ Ψ MΨ, C ¼ Ψ CΨ, K ¼ Ψ KΨ are the transformed mass, damp-
T T T
ing, and stiffness matrices, respectively. Their dimension is K K , where K <

N is the dimension of the vector z.
@CivilMethod
If Z ¼ f z1 z2 ⋯ zK gT is the matrix of the eigenmodes and

e1 , w
ðw
e 2 , …, w
eK Þ the eigenfrequencies of the reduced eigenvalue problem
e
Kw 2e
e M z ¼0, then the transformation
z ¼ ZY e ðt Þ (16.3.23)
n oT
e ðt Þ ¼ Y
decouples the linear equations with Y e 1 ðt Þ, Y
e 2 ðt Þ, …, Y
e K ðt Þ being
the generalized coordinates. Thus, we obtain
" # "h i # "h i #( )
e T Mr n o n o 2 e
Y
I F €
eu 2x e
w 0 _
e u_ b + e
w 0
Y €b + j j
Y j
e
r MF r Mr + Mb
T T 0 Cb 0 Kb ub
( ) ( )
T
e Mr
0 F
+ nl ¼ €g
u (16.3.24)
fb rT Mr + Mb
where
e ¼ ΨZ
F (16.3.25)
It is convenient to use the first K eigenmodes of the superstructure as
Ritz vectors.
16.3.4 Linear response of the isolation system

Although the linear modeling of the isolation system is not always realistic, it
is nevertheless useful for assessing the response of the isolated building in a
preliminary analysis and design stage. We assume that the isolation system
consists of m isolators and that kxi , kyi denote the stiffness of the i isolator
in the directions x and y, respectively, and that kwi ¼ 0. Actually, the base-
isolation system reacts with a shear force because the stiffness of the isolators
is small relative to the transverse stiffness of the base. Apparently, we can
readily show that the total stiffness matrix of the isolation system is given
by the relation
2 m 3
X Xm
6
i
kx 0 i
y i kx 7
6 i¼1 7
6 X
i¼1
X 7
6 m m
7
6
Kb ¼ 6 0 i
ky xi kyi 7 (16.3.26)
7
6 i¼1 i¼1 7
6 X m Xm Xm 7
4 5
yi kxi xi kyi yi2 kxi + xi2 kyi
i¼1 i¼1 i¼1
Similarly, if cxi , cyi

represent the damping coefficients of the i isolator in
the directions x and y, respectively, and that cwi ¼ 0, the damping matrix of
the isolation system results from the relation
@CivilMethod
2 X
m X
m 3
6 0 cxi yi cxi
7
6 i¼1 7
6 i¼1
7
6 X X 7
6 m m
7
6
Cb ¼ 6 0 i
cy i
xi cy 7 (16.3.27)
7
6 i¼1 i¼1 7
6 7
6 X m Xm Xm 7
4 2 i 5
i
yi cx i
xi cy 2 i
yi cx + xi cy
i¼1 i¼1 i¼1
Moreover, it is f nl
b ¼ 0 and the total restoring force of the isolation system is
expressed as
Rðub , u_ b Þ ¼ Cb u_ b + Kb ub (16.3.28)
In this case, it is obvious that the equation of motion (16.3.20) becomes lin-
ear, but the matrix Cb is not diagonalized by the eigenmodes of the undamped
free vibrations. The solution of the resulting equation can be obtained numer-
ically using any of the methods presented in Section 14.10.
16.3.5 Nonlinear response of the isolation system

The nonlinear two-dimensional behavior of the isolation system can be
expressed by the biaxial sliding model or the biaxial hysteretic model (bilinear
or Bouc-Wen) [6]. In the following, the presentation will be limited only to the
Bouc-Wen model, mainly because of the mentioned advantages. Regarding the
other models, interested readers are advised to look into the related literature,
which is quite rich.
The restoring forces are described by the equations
fSx ¼ a1 Kxe u + ð1 a1 ÞFxy Z1 (16.3.29a)
fSy ¼ a2 Kye v + ð1 a2 ÞFyy Z2 (16.3.29b)
where a1 ¼ Kxy =Kxe , a2 ¼ Kyy =Kye are stiffness ratios before and after yielding in
the x and y directions, respectively, u, v the respective displacements, and
Z1 ,Z2 two nondimensional parameters obtained from the solution of the differ-
ential equations

Z_ 1 Uxy ¼ A Z12 ½g sgnðuZ
_ 1 Þ + b u_ fZ1 Z2 ½g sgnðvZ
_ 2 Þ + b gv_ (16.3.30a)

Z_ 1 Uyy ¼ A Z12 ½g sgnðuZ
_ 1 Þ + b u_ fZ1 Z2 ½g sgnðvZ
_ 2 Þ + bgv_ (16.3.30b)
in which Uxy , Uyy represent the displacements at the yield point in the x and y
directions, respectively.
The hysteretic isolators cause the nonlinear force f nl
b , which is evaluated
from the expression
@CivilMethod
8 m 9
> X >
>
> i
fSx >
>
>
> >
>
>
> >
>
>X
<
i¼1
m >
=
i
fb ¼
nl
fSy (16.3.31)
>
> >
>
>
>
i¼1
m >
>
>
> X >
>
>
> i
i >
>
: x i f Sy yi fSx ;
i¼1
i i
The forces fSx , fSy are obtained from Eqs. (16.3.29a), (16.3.29b) with
u ¼ u b y i wb (16.3.32a)
v ¼ v b + x i wb (16.3.32b)
Eq. (16.3.20) is now nonlinear and can be solved by the methods presented
in Section 14.11 after appropriate modification to include Eqs. (16.3.30a),
(16.3.30b).
Example 16.3.1 The two-story building of Fig. E16.1 is seismic isolated
using LRB. Study its dynamic response when it is subjected to ground motion
in the x direction due to the Athens 1999 earthquake and compute:
2 at the top of column 1, the shear force Qx , the bend-
(i) The peak values U 2 , W
ing moment Mx , and the torsion moment at the base (foot) of the same
column.
(ii) The peak values of the base shear force Qbx , overturning momentMbx , the
base torsion momentTb .
Data:
a. Modal damping of the superstructure x ¼ 0:05 for all eigenmodes.
b. The eight isolators will be of the same type with Keff ¼ 500kNnm1 and
xeq ¼ 0:113, and will be placed under the eight columns.
c. The base, which is considered undeformed, has a thickness 0:15m and a
total load 2kN=m2 .
Solution
The analysis will be done by considering a linear behavior of the damping sys-
tem. We will formulate the equation of motion (16.3.19) with respect to the cen-
ters of mass of the plates, which are located on the same vertical axis through the
point xc ¼ 4:17m, yc ¼ 9:29m.
The mass matrix of the superstructure is
2 3
127:76 0 0 0 0 0
6 7
6 0 127:76 0 0 0 0 7
6 7
6 0 0 3680:24 0 0 0 7
6 7
M ¼6
c
7
6 0 0 0 127:76 0 0 7
6 7
6 7
4 0 0 0 0 127:76 0 5
0 0 0 0 0 3680:24
@CivilMethod
FIG. E16.1 Two-story building with base isolation.
and that of the base

2 3
74:52 0 0
Mcb ¼ 4 0 74:52 0 5
0 0 2152:54
The stiffness matrix is transformed with respect to the center of mass as
Kc¼ T1 KTT
2 c c
3
118939:27 0 28518:93 69127:0 0 13515:0
6 7
6 0 111065:74 18112:89 0 61461:87 7878:86 7
6 7
6 7
6 28518:93 18112:89 6421688:10 13515:0 7878:86 3660006:20 7
¼6
6
7
7
6 69127:0 0 13515:0 58212:99 0 8202:46 7
6 7
6 61461:87 7878:86 4100:17 7
4 0 0 49146:87 5
13515:0 7878:86 3660006:20 8202:46 4100:17 3041143:10
where
2 3
1 0 0 0 0 0
6 0 1 0 0 0 07
6 7
6 9:29 4:17 1 0 0 07
Tc ¼ 6
6 0
7
6 0 0 1 0 077
4 0 0 0 0 1 05
0 0 0 9:29 4:17 1
@CivilMethod
The obtained eigenfrequencies and eigenmodes of the superstructure are

w1 ¼ 9:390 w4 ¼ 34:129
w2 ¼ 10:192 , w5 ¼ 35:814
w3 ¼ 13:794 w6 ¼ 48:829
2 3
0:00018 0:04840 0:00118 0:00085 0:07398 0:00354
6 0:04639 0:00018 0:00067 0:07529 0:00097 0:00210 7
6 7
6 0:00018 0:00031 0:00888 0:00037 0:00059 0:01387 7
F¼6
6 0:00026
7
6 0:07398 0:00256 0:00057 0:04839 0:00205 77
4 0:07530 0:00033 0:00179 0:04637 0:00059 0:00139 5
0:00022 0:00041 0:01387 0:00027 0:00047 0:00887
The matrix Kb is computed using Eq. (16.3.26)p with kx ¼ ky ¼ Keff and Cb
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
using Eq. (16.3.27) with cxi ¼ cyi ¼ 2m i wb x eq ¼ 2x eq m i Keff , where m i ¼ Ni =g
is the mass of the load of the i column. The loads Ni obtained from the static
analysis of the building are N = {281.23 281.23 535.68 775.25 282.12 281.23
520.80 281.23}. Using these data, Eqs. (16.3.26), (16.3.27) give
2 3 2 3
4000 0 1160 254:42 0 33:63
Kb ¼ 4 0 4000 820 5, Cb ¼ 4 0 254:42 24:66 5, f nl
b ¼0
1160 820 217442 33:63 24:66 12452:77
Thus, Eq. (16.3.20) becomes
uI + C*u_ I + K*uI ¼ pI ðt Þ
M*€ (1)
where
2 3
1 0 0 0 0 0 0:06 15:55 1:77
6 0 1 0 0 0 0 15:64 0:06 2:68 7
6 7
6 0 83:73 7
6 0 1 0 0 0 0:48 0:26 7
6 0 3:69 0:18 7
6 0 0 1 0 0 0:035 7
6 7
M* ¼ 6 0 0 0 0 1 0 3:26 0:05 0:45 7
6 7
6 0 0 0 0 0 1 0:19 0:12 18:4 7
6 7
6 0:06 15:64 0:48 0:03 3:26 0:19 330:04 0 0 7
6 7
4 15:55 0:06 0:26 3:69 0:05 0:12 0 330:04 3:83 5
1:77 2:68 83:73 0:18 0:45 18:4 0 3:83 9513:13
2 3
0:940 0 0 0 0 0 0 0 0
60 1:019 0 0 0 0 0 0 0 7
6 7
60 0 7
6 0 1:379 0 0 0 0 0 7
60 0 7
6 0 0 3:413 0 0 0 0 7
6 7
C* ¼ 6 0 0 0 0 3:581 0 0 0 0 7
6 7
60 0 0 0 0 4:883 0 0 0 7
6 7
60 0 0 0 0 0 254:42 0 33:63 7
6 7
40 0 0 0 0 0 0 254:42 24:66 5
0 0 0 0 0 0 33:63 24:66 12452:77
@CivilMethod
2 3
88:311 0 0 0 0 0 0 0 0
6 7
6 0 103:88 0 0 0 0 0 0 0 7
6 7
6 7
6 0 7
6 0 0 190:28 0 0 0 0 0 7
6 7
6 0 7
6 0 0 0 1164:76 0 0 0 0 7
6 7
6 7
K* ¼ 6 0 0 0 0 1282:66 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 2384:40 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 4000 0 1160 7
6 7
6 7
6 0 0 0 0 0 0 0 4000 820 7
4 5
0 0 0 0 0 0 1160 820 217442
8 9
>
>
0:057 >
>
>
> >
>
>
> 15:639 >
>
>
> >
>
>
> >
>
>
> >
>
>
> 0:478 >
>
>
> >
>
>
> >
> 0:035 >
> >
> ( )
>
< >
= Y
PI ¼ 3:263 u€g , uI ¼
>
> >
> ub
>
> >
> 0:19 >
> >
>
>
> >
>
>
> >
>
>
>330:04 >
>
>
> >
>
>
> >
>
>
> >
>
>
> 0 >
>
>
: >
;
0
Eq. (1) is solved numerically using the program aem_lin_MDOF.m

developed in Section 14.10 with Dt ¼ 0:005. Table E16.5 shows the peak values
2 at the top of column 1 and the stress resultants at the base of the same
of U 2 , W
column. Table E16. 6 shows the peak values at the base of the building. Finally,
Figs. E16.2 and E16.3 show the time history of U 2 , W 2 (relative to the ground and
the isolation base) at the top of column 1.
TABLE E16.5 Peak values at column 1.

U 2 2
W jQx jmax jMx jmax jMt jmax
max max
Building (cm) (rad) (kN) (kN m) (kN m)

With Relative to the 4.08 0.000315
isolation ground
Relative to the 0.60 0.000034 16.39 42.29 0.16
base
Without isolation 4.44 0.00032 115.86 299.77 1.37
Reduction % 86.5 89.4 88.8 85.9 88.3
@CivilMethod
TABLE E16.6 Peak values at the base of the building.
jQbx jmax jMbx jmax jTb jmax

Building (kN) (kNm) (kNm)
With isolation 137.58 888.06 43.95
Without isolation 1028.28 6726.98 258.53
Reduction % 86.6 86.8 83
FIG. E16.2 Displacement U 2 at the top of column 1.
2 at the top of column 1.

FIG. E16.3 Rotation W
@CivilMethod
16.4 Problems
Problem P16.1 The two-story building of Fig. P16.1 is seismic isolated. Com-
pute the reduction of the peak values of the displacements at the top of column 1
as well as of the stress resultants Qx , Mx , Mt at the foot of the same column
when it is subjected to the Athens 1999 earthquake in the x direction. Consider
two types of isolators: (i) bilinear and (ii) Bouc-Wen. The building is made from
reinforced concrete. Assume the following data:
Modulus of elasticity E ¼ 2:1 107 kNm2 ; Poisson ratio n ¼ 0:20; specific
weight g ¼ 25kNm3 ; thickness of the slabs d ¼ 0:15m; and the isolation base
d ¼ 0:20m; columns of the first floor 0:40 0:40m2 and second floor
0:35 0:35m2 ; beams 0:30 0:50m2 ; live load of slabs p ¼ 5kNm2 ;
Bilinear isolators Kx = Ky = Ke = 900kN / m, a = 0.27, Fy = 9.6kN, Fd = 7kN,
Uy = 0.01m;
Bouc-Wen isolators Kx = Ky = Ke = 900kN / m, Fy = 9.6kN, Uy = 0.01m,
a1 = a2 = a = 0.27
FIG. P16.1 Two-story building.

[1] F. Naeim, J.M. Kelly, Design of Seismic Isolated Structures. From Theory to Practice, John
Wiley & Sons, Inc., New York, 1999.
[2] S.I. Skinner, W.H. Robinson, G.H. McVerry, An Introduction to Seismic Isolation, John Willy
& Sons, New York, 1993.
[3] T.T. Soong, M.C. Constantinou, Passive and Active Structural Vibration Control in Civil Engi-
neering, Springer-Verlag, Wien, 1994.
[4] A. Kamrava, Seismic isolators and their types, Curr. World Environ. 10 (Spec. Issue 1) (2015)
27–32.
[5] A. Mokha, M.C. Constantinou, A.M. Reinhorn, Teflon Bearings in Asseismic Base Isolation:
Experimental Studies and Mathematical Modeling, 1998. Technical Report NCEER-88-038.
@CivilMethod
[6] P. Tsopelas, M.C. Constantinou, A.M. Reinhorn, 3D-BASIS-ME: Computer Program for
Nonlinear Dynamic Analysis of Seismically Isolated Single and Multiple Structures and
Liquid Storage Tanks, Report No. NCEER 94-0010. Nat. Ctr. for Earthquake Engrg. Res,
State Univ. of New York, Buffalo, NY, 1994.
10.1002/zamm.20120024.
[8] G.C. Tsiatas, A.E. Charalampakis, A new hysteretic nonlinear energy sink (HNES), Commun.
Nonlinear Sci. Numer. Simul. 60 (2018) 1–11.
[9] Katsikadelis, J.T. A new numerical integration method for higher order ordinary differential
equations (to be published).
[10] S. Nagarajaiah, A.M. Rheinhorn, M.C. Constantinou, Nonlinear Dynamic Analysis of Three-
Dimensional Isolated Structures, 1989. Technical Report NCEER-89-0019.
[11] P. Tsopelas, S. Nagarajaiah, M.C. Constantinou, A.M. Reinhorn, Nonlinear dynamic analysis
of multiple building base isolated structures, Comput. Struct. 50 (1) (1994) 47–57.
@CivilMethod

Dynamic Analysis of Structures PDF

Uploaded by

Copyright:

Available Formats

Dynamic Analysis of Structures PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dynamic Analysis of Structures PDF

Uploaded by

Copyright:

Available Formats

Dynamic Analysis of Structures

Library of Congress Cataloging-in-Publication Data

For information on all Academic Press publications

Publisher: Matthew Deans

This work is also dedicated to my numerous students

fractional derivative, contrary to the integer-order derivative, has no direct

Knowledge of the dynamics of structures is particularly necessary for engi-

It is a pleasure to make grateful acknowledgment of many helpful sugges-

General concepts and principles

Dynamic Analysis of Structures. https://doi.org/10.1016/B978-0-12-818643-5.00001-7

varying position-dependent distributed loading pðx, t Þ (Fig. 1.1.1a). We can

From the Euler-Bernoulli beam theory we have

Substituting Eq. (1.1.2) into Eq. (1.1.1) yields

1.2 Types of dynamic loads

FIG. 1.2.1 Periodic load.

From the analytical point of view, it is convenient to divide the deterministic

FIG. 1.2.3 Nonperiodic load due to explosion.

could be considered as a special case of the dynamic problem with negligible

1.3 Dynamic degrees of freedom

a. The term displacent denotes both translation and rotation.

two differential equations of motion are necessary to determine these displace-

FIG. 1.3.3 Multi-degree-of-freedom (MDOF) system.

FIG. 1.3.4 System with infinite degrees of freedom. Continuous system.

1.4 Dynamic model and formulation of the equation

Fig. 1.4.1 shows a simple dynamic model of a SDOF system. It consists of a

FIG. 1.4.2 Forces applied to the free body.

1.5 Derivation of the equations of motion using

If R ¼ X ðt Þi + Y ðt Þj + Z ðt Þk is the position vector of a particle A of

in which r ¼ rðx, y, z Þ is the mass density of the body.

It is often convenient to study the motion with reference to an arbitrary point

When small displacements are considered, as in the theory of linear vibra-

Similarly, Eqs. (1.5.6a)–(1.5.6c) are written

u ðt Þ ¼ a tan fðt Þ afðt Þ, u ðt Þ ¼ BB 0 cos f BB 0 (1)

because we have assumed small displacements. Moreover, AA0 ¼ BB 0 =2 ¼ u=2

The inertial force at the center of mass along y

The inertial moment about the center of mass

The external force

which by virtue of Eq. (2) becomes

Hence, Eq. (5) becomes

Example 1.5.2 Equation of motion of a frame with a rigid column

The elastic shear force at the beam end C QS ¼ ð1:5L

MIA + MS + MR + QS ð3 4Þ + 2fD ð5 6Þ pðt Þ ð1 2Þ ¼ 0

which after substituting their exressions becomes

Further, linearizing (f2 0) gives

If the displacement u ¼ Lf at the level of the beam is taken as the parameter

Example 1.5.3 Equation of motion of a system of rigid bodies

or taking into account that

The external load is 2Lpðt Þ.

fS ð2LÞ fD ð3:25LÞ fIK ð1:5LÞ fIG ð3:25LÞ

or inserting Eqs. (1)–(6) into Eq. (7) we obtain

RA + pðt Þ2L fIK fIG fS fD ¼ 0

or using Eqs. (1)–(3), (5) we obtain

RA ¼ ku + 1:625cu_ + 2:375m u€ 2L

Example 1.5.4 Equation of motion of a single-story shear building

5:21u€ + 708:75u ¼ sin 13t

Example 1.5.5 Equation of motion of a two-story shear frame

(a) (b) (c)

The equations of motion result from the dynamic equilibrium of forces

m1 u€1 + k1 ðu1 u2 Þ ¼ p1 ðt Þ (1)