Response Spectrum Method Gupta
Response Spectrum Method Gupta
Response Spectrum Method Gupta
AJAYA K U M A R GUPTA.
Prokssor of Civil Engineering
North Carolina State University
F O R E W O R D BY
WILLIAM J. H A L L
Professor and Head. Civil Engineering
-
Universit.~of lllinois at Urbana Champaign
PO Box 447
Osney Mead, Oxford OX2 OEL, England Brookline Village
25 John Street, London WClN 2BL, England Massachusetts 02 147
23 Ainslie Place, Edinburgh EH3 6M, Scotland (Orders: Tel: 61 7 524-7678)
107 Bany Street. Carlton '
Victoria 3053, Australia I .
Canada
Oxford University Press
All rights reserved. No part of this book may be 70 Wynford Drive
reproduced in any form or by any electronic or Don Mills
mechanical means, including information Ontario M3C 159
storage and retrieval systems, without permission (Orders: Tel: 416 441-2941)
in writing from the publisher, except by a Australia
reviewer who may quote brief passages in a review Blackwell Scientific Publications
(Australia) Pty Ltd
First published 1990 107 Barry Street
Carlton, Victoria 3053
Set by Tima Graphics, Singapore (Orders: Tek 03 347-0300)
Printed and bound at the University Press,
Cambridge. England Outside North America and Australia
90 91 92 93 4 3 2 1 Marston Book S e ~ c e Ltd
s
PO Box 87
Oxford OX2 ODT
(Orders: Tel: 01 1 44 865 79 1155)
Library of Congress
Cataloging-in-Publication Data
Gupta Ajaya K.
Response spectrum method in seismic
analysis and design of structures / Ajaya
Kumar Gupta; foreword by W.J. Hall.
P- cm. - (New directions in
civil engineering)
ISBN 0-86542-1 15-3
1. Earthquake engineering. 2. Structural
engineering. 3. Seismic waves.
1. Title. 11. Series.
TA654,6.G87 1990
624.1'7626~20
British Library
Cataloguing-in-Publication Data
Gupta, Ajaya Kumar
Response spectrum method in seismic
analysis and design of structures,
I. Structure. Analysis - A
.
Foreword, ix
Preface, xi
Acknowledgments, xv
1 Structural dynamics and response spectrum, 1
I. 1 Single-degree-of-freedom system, I
1.2 Response spectrum, 2
1.3 Characteristics of the earthquake response spectrum, 6
1.4 Multi-degree-of-freedom systems. 7
References, 10
2 Design spectrum, 11
2.1 Introduction, I I
2.2 'Average' elastic spectra, 12
2.3 Site-dependent spectra. 16
2.4 Design spectrum for inelastic systems, 23
2.5 Comments. 27
References, 28
This book devoted to the Response Spectrum Method contains concise sections
on a number of the major topics associated with the application of spectrum tech-
niques in analysis and design. Although the theory of spectra has been understood
for some extended period of time, it was only in the past twenty years that the
approach was adopted in a major way by the profession for use in engineering
practice. This development came about as a result of three major factors, namely
that the theory and background of spectra was more fully understood, that the
theory was relatively simple to understand and use, and because there was a need
for such a simple approach by the building codes and by the advanced analysis
techniques needed in the design of nuclear power plants and lifeline systems.
The author rather directly presents his interesting and informative interpreta-
tions of various spectrum techniques in the topical chapters. He correctly points
out that much work remains to be accomplished, which is accurate, for spectra in
general only depict maxima of various effects, and in many cases, especially
where nonlinear effects are to be treated, it is often desirable to know more about
the response than just a maximum value. Research on such topics presently goes
forward on such matters at a number of institutions, and in time will lead to even
greater understanding of the theory, and to new approaches of application. In this
connection one can cite subtle yet important differences in use and interpretation
of spectra. For example, the term 'response spectrum' normally is used to refer to
a plot of maximum response parameters as a function of frequency or period, for
a given excitation of the base of a single-degree-of-freedomdamped oscillator, as
for acceleration time history of excitation associated with a specific earthquake.
On the other hand a design spectrum is a similar shaped plot selected as being
representative of some set of such possible or plausible excitations for use in
design; as such it is a characterization of effects that might be expected as a result
of some possible range of excitation inputs, and possibly adjusted to reflect risk or
uncertainty considerations, personal safety requirements, economic considera-
tions, nonlinear effects, etc. One can immediately discern the differences, directly
or subtly as may be the case.
It is believed that the reader will find the interesting presentation by Dr Ajaya
Gupta to be educational and informative, and hopefully such as to promote
additional effort to improve even further our understanding of the theory and
applications thereof.
W. J. HALL
Professor and Head, Civil Engineering
University of Illinois at Urbana-Champaign
In modern earthquake engineering the response spectrum method has emerged as
the most commonly used method of analysis. The primary reason for this
popularity is the fact that it provides the designer with a rational and simple basis
for specifying the earthquake loading. Another reason often cited is that the
method is computationally economical. If a comparison is made between the
computational effort required in, say, a modal superposition analysis of a multi-
degree-of-freedom structure subjected to a specified ground motion history, and
that in a response spectrum analysis including the evaluation of the response
spectrum from the same motion history, it is not clear whether the response
spectrum method would do much better. A major part of the effort, which is com-
mon in both the methods, is the solution of the eigenvalue problem. In fact, if the
objective is to evaluate the response of a structure subjected to a known
earthquake ground motion, there should not be any question about using a
standard time-domain analysis, or alternatively, an equivalent frequency-
domain analysis. It is when we are designing a structure for a potential future
earthquake that the response spectrum method is much more relevant.
Criticisms of the response spectrum method arise from the fact that the-
temporal information is lost in the process of evaluating the spectrum. In the
words of Robert Scanlan:' 'Multi-degree-of-freedom cases are thus improperly
served, intermodal phasings, in particular, being unaccounted for.' Further he
points out: 'The needs arising in the design of secondary responding equipment
(piping, machinery, etc., on upper floors of a structure) are not adequately met by
the given design response spectra. That is, the given primary shock spectra do not
lead directly and simply to definition of corresponding secondary shock spectra.'
Similar difficulties arise in combining the responses from three components of
the earthquake.
Much progress has been made in the last decade. Lack of temporal
information in the response spectrum method no longer appears to be a
handicap. Rational rules are now available to combine responses from various
modes, and from three components of earthquake motion. These rules account
for the physics of the problem, and can be further justified in the same spirit as
the design spectrum itself, as a representation of expected response values in an
uncertain world. Response of secondary systems can now be evaluated using
efficient modal synthesis techniques in conjunction with the response spectrum
method. Alternatively, the secondary spectrum, or the instructure response
'R. H. Scanlan. On Earthquake Loadings for Structural Design, Earthquake Engineering and
Sfrucfural Dynamics, Vol. 5, 1977, pp. 203-205
spectrum can be evaluated by applying similar modal synthesis techniques to the
secondary single-degree-of-freedom oscillator coupled to the primary system.
These new techniques directly use the design response spectrum at the base of the
primary structure as seismic input and account for the effects of mass interaction
(between the equipment and the structure) and of multiple support input into the
secondary system. In doing so it is no longer necessary to convert the design res-
ponse spectrum into a 'compatible' motion history or a power spectral density
function. The question of noncIassica1 damping introduced in the coupled
primary-secondary system, which had not even been specifically raised 10 or 15
years ago, is now adequately addressed.
This brings us to the objective of this book. It is intended to bring together in
one volume the wealth of information on the response spectrum method that has
been generated in recent years. Needless to say that this information has reached
a critical mass suitable for a book. This book can be used as a text or as reference
material for a graduate level course. Although Chapter 1 begins with the
introductory information about the single-degree-of-freedom systems that leads
into the definition of the response spectrum, 1feel that most students will be more
comfortable with the material in subsequent chapters if they already have had an
introductory structural dynamics course. This book should also serve as a useful
reference for prac~icingengineers. It should help them appreciate the analytical
techniques they are already using. In many cases the book may also help them
improve those techniques, especially when the improvement would lead to
enhanced accuracy, often resulting in significantly lower response values.
It is assumed throughout the book that we are dealing with linear systems.
There are two exceptions. In Chapter 2 a brief treatment is given to inelastic res-
ponse spectra. Chapter 8 deals with conventional buildings which are customarily
designed to undergo significant inelastic deformation under the worst loading
conditions. Inelastic behavior has always been a part of seismic design of
buildings, unintentionally in the beginning, and later with full knowledge and
intention. Yet, our knowledge of the topic is relatively limited. Inelastic seismic
behavior and design continue to be a topic of active research. Detailed coverage
of current research on the topic goes beyond the realm of the response spectrum
method, and is beyond the scope of this book. Brief treatments in Chapter 2 and
Chapter 8 are intended to provide a useful link between the response spectrum
method and the design of conventional buildings. It should be of particular
interest to the students to see the link established and, at the same time, recognize
the limitations of the link.
I have emphasized deterministic modeling of the earthquake response
phenomenon. For a given earthquake ground motion, the maximum response
values for a single-degree-of-freedom system-which are the basis of the
definition of response spectrum-are deterministic quantities. For a multi-
degree-of-freedom system, therefore, the maximum response values in individual
modes are also deterministic quantities. The modal combination rules are based
partly on the physics of the problem, that is on deterministic concepts, and partly
on the random vibration modeling of the phenomenon. Strictly speaking, then,
these rules do not apply to responses from individual earthquakes. On the other
hand, we can look upon the modal combination rules as tools for giving
approximate values of the deterministic maximum response values. It is in this
spirit that the response spectrum analysis results have been repeatedly compared
with the corresponding time-history maxima for individual earthquakes, treating
the latter as the standard. This concept is especially powerful when judging two or
more modal combination rules within the response spectrum method. A rule
which models the physics well is likely to give results which are reasonably close
to those obtained using the time-history analysis.
Probabilistic concepts play an important role in the definition of the design
spectrum, as they do in defining other kinds of loads too. These concepts are most
useful when all the available deterministic tools have been carefully employed.
One should not replace the other. Great strides have taken place in recent years in
the development and application of random vibration techniques to the
earthquake response problems. Important contributions have been made to the
response spectrum method using the random vibration concepts. This book has
not covered those techniques and concepts for most part.
My interest in the response spectrum method has been the primary motiva-
tion for writing this book. This interest has been sustained through many years of
research on related topics in collaboration with coworkers and students. Such
personal involvement in the topic has its advantages and disadvantages in writing
a book. The advantages are obvious. The main disadvantage is that I may not be
able to do full justice in presenting the works of other researchers. To that end, I
shall welcome criticism and suggestions from the readers, which I hope will
improve the future editions of this book.
A. K. GUPTA
Acknowledgments
where o is circular frequency of the structure in radians per second and [ is the
damping ratio. For free response to be vibratory, ( < 1. For most structures l; is
small, say < 0.1, or 10%. We note that the frequency in Hertz (Hz) or in cycles per
second (cps)f = 0/2n, and that the period of vibration T = I /f= 2x10, which is
in seconds.
Equation 1.4 can be solved using standard numerical techniques. As a result
we can obtain the time histories of displacement, velocity and acceleration, of the
spring and the damping forces. and any other related response time history. See
the Appendix.
. .
Mass M
Lateral
Stiffness
C--
(-4
(a) One story model (b) Model subjected to
ground motion
-f.
Time, sec
Fig. 1.2 Ground acceleration history of El Centro earthquake (SOOE,1940).
For designing a structure, we are most interested in the maximum spring force
f,,which can be evaluated if the maximum relative displacement u is known. A
plot between maximum relative displacement and structural frequency is called
the displacement response spectrum. Its ordinates are called spectral displace-
ments, and are denoted by S,(f; (). Depending upon the context, they can also be
denoted by SD(o,0, SD(f ), SD(o),or simply by SD.Let us write
Figure 1.2 shows the ground acceleration time history of the El Centro (SOOE,
1940) earthquake. The corresponding displacement response spectrum is shown
in Figure 1.3(a).
Let us consider the spring force-displacement relationship f, = ku. We have
indicated earlier that if the relative displacement u is known, we can find the
spring forcef,. Alternatively, if the spring force is known, we can determine the
corresponding relative displacement. We can visualize this as a pseudo-static
problem shown in Figure 1.4. Now let us think of-&as a pseudo-inertia force,
which can be written in terms of the pseudo acceleration a as ma. The
relationships, ma = f, = ku, give a = (k/m)u = 0 2 u . The absolute maximum
value of a is called spectral acceleration SA.We can easily see
Having defined the response spectra for relative displacement and for pseudo
acceleration, we wish to define a response spectrum for velocity. It can be done in
more than one way. First, let us define a spectral velocity Svsuch that the kinetic
energy associated with it is equal to the maximum strain energy of the spring,
(1/2)mS: = ( 1 / 2 ) k ~ iThis
. gives
Because of this relationship it is possible to read S,, Svand S, from the same
logarithmic chart shown in Figure 1.5. This chart is known as the tripartite chart
because, for any frequencyJ there are three scales, one each for SD,Svand SA.
Now consider the second way of defining a velocity spectrum. We shall denote
the new quantity by S;. It is defined as the absolute maximum relative velocity
S; I
= max u ( t ) 1. (1.10)
The relative velocity spectrum is shown in Figure 1.3(b) with the dashed lines.
The two spectra in the figure are close in the intermediate frequency range; the
pseudo velocity spectrum is higher in the high frequency range, and the relative
velocity spectrum is higher in the low frequency range. Thus, as a rule, we cannot
substitute one spectrum for the other.
For the SDOF structure, the response spectrum quantity of interest is any one
of S,, S, or S,. Also, for the classically damped r n u ~ ~ ~ ~ f - f r e e d o m
(MDOF) systems defined in Section 1.4, we need only one o&*drce spectra.
We shall see in Chapter 5, that we also need S; for nonclassically?
systems.
4=Mlij+og;= ['a
by ur = [u, u2 u,]. The inertia force vector is
0
in2
0 in3
ii, + ii,
[ii2+iig],
ii, + ii,
(1.1 1)
where U, is a vector of ground acceleration ii,. The vector of spring forces is given
by
In the above equation M is the mass matrix of the structure, K the stiffness
matrix, and the vector 1 consists of unit elements. For the 3-DOF structure these
matrices are explicitly defined above. For other MDOF structure these matrices
(a) Simple 3-DOF System (b) Deformed Shape
can be obtained using standard procedures [2]. A more general form of the
undamped equation of motion is
The vector U,,defines static structural displacements when the support undergoes
a unit displacement in the direction of the earthquake. For the simple structure at
I
hand, it is easy to see that U,becomes 1 as in Equation 1.13.
The mode shapes and the frequencies of the structure are obtained by solving
the following eigenvalue problem
Mode 1
f = o.0708 Hz f
Mode 2
= 0 . 1 9 8 4 HZ f - Mode 3
0.2a7fi HZ
-
Fig. 1.7 Unnormalized mode shapes and frequencies of a
m, = m, m , , k , = k 2 = k,.
3-DOFsystem,
where y, terms are called normal coordinates, and are functions of the time
variable t. Substitution of Equation 1.17 in Equation 1.14, premultiplication by
i$jT, and the application of the orthogonality conditions from Equation 1.16 gives
Equation 1.18 is similar to Equation 1.4 for the SDOF structure for the
undamped case.
It is difficult accurately to define the damping matrix for a MDOF structure.
Often it is assumed that the damping matrix C has orthogonality properties
similar to those of M and K,and that we can define the damping ratio for each
mode just as we did for a SDOF structure
Structures that have the idealized damping matrix property given by the above
equations are called classically damped. Equation 1.18 is replaced by
In the modal superposition method Equation 1.21 is solved to obtain the time
histories of the normal coordinates y,, which with Equation 1.17 give the history
of the relative displacement vector U,etc.
We shall now use the above concept to apply the response spectrum method to
the MDOF structure. The comparison of Equations 1.4 and 1.21 shows,
r,)
yi(f) = y,u(t), when o = w, and & = 6,. Hence, yimar= y, SD(w,. = y,SD;.
Thus, the maximum displacement vector in the ith mode can be written as
Given the displacement vector U,,,,,we can determine the maximum value of
any response of interest. Methods of combining maximum response values from
various modes, and from three components of earthquake are presented in
Chapters 3 and 4, respectively.
References
I . A.K. Chopra, Dwamics of Struc~ures- A Prirtier, Earthquake Engineering Research
Institute, Berkley, California, I98 1 .
2. R.W. Clough and J. Penzien, D7vnarnicsofStructures, McGraw Hill, New York, 1975.
Chapter Z/Design spectrum
2.1 Introduction
We have reviewed the basic concepts of dynamic structural analysis in Chapter 1.
Only linear elastic behavior has been considered. The purpose was to set the stage
for other topics. Indeed, theory and techniques of structural dynamics have
reached a stage of advancement such that it is fair to say that any structure that
can be mathematically modeled can be analyzed subject to any given transient
forcing function, e.g. an earthquake ground motion. The structure may have any
given linear or nonlinear constitutive properties. Large displacement effects can
also be accounted for. For us the key here is the definition of the earthquake
ground motion. If we know the ground motion history, we can analyze the
structure and design it. But the earthquake we are talking about has not yet
occurred.
In many ways the problem of specifying a future earthquake is not very
different from specifying any other load for design. The actual live load on a
building floor varies a great deal during its lifetime, and it is not uniformly
distributed on the floor area. There are at least three idealizations involved in live
load specification. First, we idealize the actual floor distribution of furniture,
people and other live loads as a uniformly distributed load such that the design
quantities of interest-the slab moments-have approximately the same magni-
tude. Second, we estimate the likely maximum magnitude of this uniformly
distributed load during the lifetime of the building. Finally, we design the floor
using appropriate load factors, capacity reduction factors or factors of safety. The
end product is a slab, which, incidentally, has a relatively definite resistance. The
art of specifying the load and the remainder of the design procedure is then a
'recipe', which more than anything else assures a resistance. Therefore, when we
are specifying a load, we are really specifying the resistance or the level of
resistance in a structure or a structural element.
The specification of the earthquake load consists of determining themagni-
tude and intensity of the design ground shock at a given site, and of somehow
converting them into the ground motion parameters. The intensity of the design
earthquake is determined from the seismological and geological data concerning
earthquakes and their occurrence. It is well to note that available data base is far
from adequate and is the major source of uncertainty in earthquake-resistant
design. It is then unrealistic in most cases to expect that we can characterize a
future ground motion in any detail based on the design intensity and any other
limited information available. Approximate procedures have been developed to
11
give the estimate of the peak ground acceleration associated with intensity levels.
In some cases the peak ground displacements and velocity can also be
approximately estimated.
Given the peak ground motion parameters-displacement, velocity and
acceleration-techniques have been developed to define smooth spectrum
curves, which are called design spectra. The main difference between the
response spectrum presented in the previous chapter and the design spectrum we
are discussing now is that the former represents the response to an actual
earthquake and the latter defines the level of seismic resistance we are to design
for. Just as in the example of the live load, the design spectrum idealizes the real
phenomenon to fit it into a design recipe.
Period (set)
Fig. 2.1 (a) Housner velocity design spectra. (b) Housner acceleration design spectra, peak
ground acceleration = 0.125 g [4l.
I
tion was made to the particular ground motion parameter for which the response
spectrum curve was most nearly parallel to the coordinate. In the Newmark
study[8], the spectral ordinates were assumed to have a normal distribution; the
data would also fit the log-normal distribution[9]. The design spectrum can be
obtained from the maximum ground displacement, velocity and acceleration
values, if the respective amplification factors are known. These amplification
factors were obtained for the mean spectrum (50% probability level) and for the
mean plus one standard deviation spectrum (84.1% probability level). Table 2.1
summarizes these amplification factors for the horizontal components of
earthquake for four damping values. Values simiIar to those in Table 2.1 for
84.1% probability level have been adopted by a recent ASCE standard[lO]. In the
Newmark study[8], it is recommended that the transition from amplified
acceleration to ground acceleration begin at 6 Hz for all damping values and end
at 40, 30, 20, 20 Hz, respectively for damping ratios of 0.5%, 2.0%, 5.0% and
10.0%. In the ASCE Standard[lO], this transition occurs between 9 and 33 Hz for
all damping values. Corresponding to a Ig maximum ground acceleration, it was
found[8] that the maximum ground displacement and velocity were approxima-
tely 36 in and 48 in sec-I, respectively, for alluvium soil, and 12 in and 28 in sec-'
for rock. For both types of soiis, we have ad/u2 2: 6, the value recommended by
the ASCE Standard[lO]. Figure 2.3 shows the 'Newmark spectrum' for Ig
maximum ground acceleration on an alluvium soil.
The joint Newmark-Blume recornmendations[l l ] were later adopted in a
modified form by the United States Atomic Energy Commission (now the US
Nuclear Regulatory Commission-USNRC)[12]. Figure 2.4 shows the USNRC
spectra for Ig ground acceleration. As can be seen, there are no major differences
between Figures 2.3 and 2.4. The recommendations for the vertical component of
Fig. 2.3 Newmark design spectra for alluvium soil; maximum ground acceleration = I g.
earthquake vary much more, and a complete discussion is beyond the scope of
this book. The ASCE Standard[lOJ recommends that the design spectra for the
vertical component be obtained by multiplying the corresponding spectra for the
horizontal component by a factor of 2/3.
Frequency. Hz
Fig. 2.4 United States Atomic Energy Commission design spectra [12].
Regulatory Guide
0 .5 1 1.5 2 2.5 3
Period, Seconds
Fig. 2.5 Seed, Ugas and Lysmer site-dependent spectra and Atomic Energy Commission
spectrum; mean plus one standard deviation (84.1 percentile), peak ground acceleration =
1 g, damping ratio = 0.05 [14].
Mohraz had three components for each earthquake. The peak accelerations
for the three components are ordered from the largest to the smallest as follows:
the larger horizontal a,,, the smaller horizontal a,, and the vertical a,. The mean
value of the ratios aSH/aLH and a,/aLH for all the site conditions are 0.83 and 0.48,
respectively. The corresponding 84.1 percentile values are 0.98 and 0.65. That
means that at the 84.1 percentile level the maximum ground acceleration for the
smaller horizontal component is almost equal to the maximum horizontal
acceleration for the larger horizontal acceleration, and that the maximum vertical
acceleration is almost 213 of the latter. This is consistent with the assumption
commonly made for the design of critical facilities[lO].
Mohraz performed a detailed statistical study on the ratios of the ground
motion parameters, u/a and ad/v2. When the ratios are known, we can calculate
the values of u and d; the value of a is commonly given based on seismological
considerations. Table 2.2 lists the mean values of the ratios v/a and ad/v2, along
with the corresponding values of u and d for various site conditions for a 1 g
maximum ground acceleration. For illustration here, the larger horizontal
Table 2.2 Ground motion parameters[ 161
Based on Mohraz[l6]
Rock 27 6.9 27 13.0 31 7.6 31 18.9
Less than 30 ft 37 5.2 37 18.4 37 8.5 37 30.1
alluvium underlain
by rocks
30-200 ft Alluvium 33 5.6 33 15.8 33 9.1 33 25.6
underlain by rocks
Alluvium 51 4.3 51 28.9 51 5.0 51 33.7
Based on Mohraz,
Hall and Newmark(81
Rock 28 6 28 12 - - - -
Alluvium 48 6 48 36 - - - -
component and the vertical component are included in Table 2.2. The table
indicates that the v / a ratios for rock are substantially lower than those for
alluvium. These ratios for the two alluvium layers underlain by rock are between
those for rock and alluvium. The v / a ratios for the larger horizontal components
are the same as that for the vertical component, except in the case of rock, when
they are close. The ad/v2ratios given in Table 2.2 indicate that the ratios for allu-
vium are smaller than those for rock and alluvium layers underlain by rocks.
Since ad/u2 is a measure of the width of the spectra, the ratios indicate that the
average spectrum for a rock site is flatter than for alluvium site or for a site with
alluvium layers underlain by rock. The values of vla and adlv2 ratios, and of v
and suggested earlier by Mohraz, Hall and Newmark(81 for the horizontal
component only are also given in Table 2.2 for comparison. There are minor dif-
ferences between the old and the new v/a values. A relatively greater change
occurs in the value of ad/v2for the alluvium site (6-4.3) and in the corresponding
value of d (36-28.9 in). We note that Mohraz[I 61 also gives the median and 84.1
percentile values, which d o show dispersion in these values. For most practical
applications we believe that using the mean values of the ratios should be
adequate.
Mohraz(l61 showed that his average spectrum for the rock site, 5% damping
ratio compared well with the corresponding spectrum given by Seed, Ugas and
DESIGN S P E C T R U M / 2 1
-Alluvium
4 -
!\ -.- Lesa than 30 ft. Alluvium on Rock
,-,Rock
1 I
0 I
0 .5 1 1.5 2 2.5 3
Period, Seconds
Fig. 2.6 Mohraz average site dependent spectra; peak ground acceleration
ration = 0.02 [16].
- I g, damping
Lysmer[l4]. Mohraz's average spectra normalized to 1 g for the four sites are
shown in Figure 2.6 for 2% damping ratio. We observe from Figure 2.6 that the
acceleration amplification for the alluvium deposits extends over a larger
frequency region than the amplification for other site categories. Further, the
maximum spectral accelerations for the two sites with alluvium layers underlain
by rock is greater than the maximum spectral acceleration for either rock site or
the alluvium site. For periods ~ 0 . 2sec (frequencies > 5 Hz), the spectral
ordinates for alluvium sites are less than those for the other three site types
studied. In the periods >0.5 sec (frequencies < 2 Hz), the spectral ordinates for
the alluvium site are greater than those for the other three. The alluvium spectral
values are approximately 2.5-3 times the rock spectral values in the 1.5-3 sec
period range (frequencies, 0.3-0.7 Hz).
Mohraz [I 61 has presented comprehensive statistics of displacement, velocity
and acceleration amplification factors for all the four site types, damping ratios
0-20%, and the three components of earthquake. Amplification factors are given
for the mean (50% probability level) and the mean plus one standard deviation
(84.1% probability level) spectral values. Amplification factors for the larger
horizontal component for the alluvium site only are reproduced in Table 2.3, for
three damping ratios (2%, 5% and 10%) and for 50% and 84.1% probability levels.
Table 2.3 Amplification factors for larger horizontal component for alluvium site
suggested by Mohraz(l61
Coefficients
Frequency, Hz
Fig. 2.7 Mohraz and Newmark site dependent spectra; peak ground acceleration = 1 g.
damping ratio = 0.02.
of unity, urn= u,, and we have the elastic response spectrum curve. The
inelasticity is introduced when p > 1. Again, for each response spectrum curve a
constant value of p is assumed. For a single-degree-of-freedom system of given
frequency and damping, the solution process consists of assuming a value of u,,
and integrating the nonlinear equation of motion numerically for the earthquake
ground motion history which gives u,. For the assumed value of u,, the
calculated urnis not likely to give the ductility factor urn/uyequal to the desired p.
For each point, therefore, several solutions have to be performed, each with a
different value of u,. When the calculated urn/uyis sufficiently close to the desired
p, the iteration stops.
The elastic response spectrum is based on maximum relative displacement,
which is also a measure of the maximum spring force. We recall, the relationship
S, = oZSD is obtained on the basis that the pseudo-inertia force given by S, is
equal to the spring force given by SD.For an inelastic single-degree-of-freedom
system the measure of spring force is u,. Since u, and urnare so conveniently
related, urn= pu,, we can use either of the two displacements for drawing the
response spectrum, as long as we know which one it is in a given case. The spec-
trum based on urncan be called the maximum displacement spectrum, and that
based on u , , the yield spectrum.
Early studies on the inelastic response spectrum were made by Newmark and
coworkers [5,6,17]. They reached the following conclusions:
1. For low frequency systems, the maximum displacement for the inelastic
system (urn)is the same as for an elastic system having the same frequency.
2. For intermediate frequency systems, the total energy absorbed by the spring
is the same for the inelastic system as for an elastic system having the same
frequency.
3. For high frequency systems, the force in the spring is the same for the
inelastic system as for an elastic system having the same frequency.
Let us denote the elastic spectral values by s E , and the corresponding
maximum displacement and yield spectra values by SMand SY,respectively. The
above conclusions give the following relationships:
1
Intermediate frequency range, SM= P p , SY =
J ( ~ P- 1) J(2p - I)
High frequency range, SM= p s E ,sY
= sE. (2.1)
I
Riddell and Newmark11 81 performed a relatively detailed study on the topic
I of the inelastic spectrum. They found that the factors used to modify an elastic
spectrum into an inelastic spectrum are functions of the damping ratio and of the
1 I
type of the material force-deformation relationship. Overall, however, they
I
confirmed the conclusions of the earlier studies[5,6,17]. They evaluated inelastic
spectra for elastoplastic, bilinear and stiffness degrading systems. They con-
cluded that the ordinates of the average inelastic spectra for the three material
models did not differ significantly. They also found that the spectrum for the
elastoplastic material was always slightly conservative as compared with those for
the other two materials. That leaves the effect of damping values. One may use
the more refined modification factors given by Riddell and Newmark[lS].
DESIGN S P E C T R U M 1 2 7
2.5 Comments
An effort has been made in this Chapter to summarize some of the available
information on design spectra in a simplified form. For many readers the
information presented here may be adequate. For others, especially, those
involved in the design of major facilities and those interested in pursuing the
topic for research, there is a vast growing body of literature available. A
philosophical and historical perspective is given by Newmark and Ha11[20] and
Housner and Jennings[2 11 in two EERI monographs.
Broadly classified spectral shapes presented in this Chapter should serve a
useful purpose when more detailed and precise data f0r.a site can not be obtained.
It is not uncommon in many cases that the only motion parameter that is
specified is the peak ground acceleration. The use of peak ground acceleration in
conjunction with a standard spectral shape for a site in the vicinity of a fault can
grossly overstate the response values in the frequency range of interest. It is much
more desirable somehow to evaluate and use the three major motion parameters:
the peak acceleration, velocity and displacement. For sites like those in Mexico
City, specific knowledge about the local conditions is very important. Housner
and Jennings recommend[2 I]: 'A much better method of describing the ground
motion simply would be to compare it to a known accelerogram, such as recorded
in Taft, California in 1952, or to a synthesized accelerogram. The decription
could thus be phrased as: 1.5 times as intense as Taft 1952, with duration of
strong shaking 1.2 times as long and with similar frequencies of motion.'
References
1. M.A. Biot, A Mechanical Analyzer for the Prediction of Earthquake Stresses, Bulletin of the
-
Seismological Societv of America, Vol. 31, 194 1, pp. 15 1 17 1.
2. M.A. Biot, Analytical and Experimental Methods in Engineering Seismology, Proceedings,
ASCE, Vol. 68, 1942, pp. 49-69.
3. G.W. Housner, An Investigation of the Effects of Earthquakes on Buildings, Ph.D. Thesis,
California Institute of Technology, Pasadena, California, 1941.
4. G.W. Housner, Behavior of Structures During Earthquakes, Journal of Engineering
Mechanics Division, ASCE, Vol. 85, No. EM4, 1959, pp. 109- 129.
5. N.M. Newmark and AS. Veletsos, Design Procedures for Shock Isolation Systems of
Underground Protective Structures, Vol. 111, Response Spectra of Single-Degree-of-
Freedom Elastic and Inelastic Systems, Report for Air Force Weapons Laboratory, by
Newmark, Hansen and Associates, R T D TDR 63-3096, June 1964.
6. A.S. Veletsos, N.M. Newmark and C.V. Chelapati, Deformation Spectra for Elastic and
Elasto-Plastic Systems Subjected to Ground Shock and Earthquake Motions,. Proceedings.
Third U,,'orldConference on Earthquake Engineering, New Zealand, 1965.
7. J.A. Blume, N.M. Newmark and L.H. Corning, Design ofMuhi-Story ReinJorced Concrete
Buildingsfor Earthquake Motions, Portland Cement Association, Chicago. Illinois, 1961.
8. B. Mohraz, W.J. Hall and N.M. Newmark, A Study of Vertical and Horizontal Earthquake
Spectra, Nathan M. Newmark Consulting Engineering Services, Urbana, Illinois; AEC
Report No. WASH-1255, 1972.
9. J.A. Blume, R.L. Sharpe and J.S. Dalal, Recommendations for Shape of Earthquake
Response Spectra, John A. Blume and Associates, San Francisco, California; AEC Report,
No. 1254, 1972.
10. American Society of Civil Engineering, Standard for the Seismic Analysis ofSafey Related
Nuclear Structures, September 1986.
I I . N.M. Newmark, J.A. Blume and K.K. Kapur, Seismic Design Criteria for Nuclear Power
Plants. Journal ofthe Power Division, ASCE, Vol. 99, 1973, pp. 287-303.
12. United States Atomic Energy Commission. Design Response Spectra for Seismic Design of
Nuclear Power Plants, Regulatory Guide, No. 1.60, 1973.
13. S.H. Hayashi. H. Tsuchida and E. Kurata, Average Response Spectra for Various Subsoil
Conditions. Third Joint Meeting. L!S. Japan Panel on Wind and Seismic Efects, UJNR,
Tokyo, May 197 1.
14. H.B. Seed, C. Ugas and J. Lysmer, Site-Dependent Spectra for Earthquake-Resistant
Design, Bulletin ofthe Seismological Society of America, Vol. 66, No. 1. February 1976,
pp. 22 1-243.
15. E. Kuribayashi, T. Iwasaki, Y. lida and K. Tuji, Effects of Seismic and Subsoil Conditions
on Earthquake Response Spectra, Proceedings, International Conferenceon Microzonation,
Seattle, Washington, 1972, pp. 499-5 12.
16. B. Mohraz, A Study of Earthquake Response Spectra for Different Geological Conditions,
Bulletin ofthc Seismological Socie?rl ofAmerica, Vol. 66, No. 3, June 1976, pp. 9 15-935.
17. A.S. Veletsos and N.M. Newmark, Effect of Inelastic Behavior on the Response of Simple
Systems to Earthquake Motions, Proceedings, Second World Conference on Earthquake
Engineering, Vol. 11, 1960.
18. R. Riddell and N.M. Newmark, Statistical Analysis of the Response of Nonlinear Systems
Subjected to Earthquakes, Strrrcrural Rcsearch Series, No. 468, Department of Civil
Engineering. University of Illinois at Urbana-Champaign, Urbana, Illinois. August 1974.
19. F.E. Elghadamsi and B. Mohraz, Site-Dependent Inelastic Earthquake Spectra, Technical
Report, Civil and Mechanical Engineering Department, Southern Methodist University,
Dallas, Texas. June 1983.
DESIGN SPECTRUM/29
20. N.M. Newrnark and W.J. Hall, Earthquake Spectra and Design (Engineering Monograph on
Earthquake Criteria, Structural Design, and Strong Motion Records, M.S. Agbabian,
Coordinating Editor), Earthquake Engineering Research Institute, Berkeley, California,
1982.
2 1. G.W. Housner and P.C. Jennings, Earthquake Design Criteria (Engineering Monograph on
Earthquake Criteria, Structural Design, and Strong Motion Records, MS. Agbabian.
Coordinating Editor,), Earthquake Engineering Research Institute, Berkeley. California,
1982.
Chapter 3/Combination of modal responses
3.1 Introduction
The equation of motion for an N-degree of freedom system was presented in
Chapter I, and is rewritten below
where M , Cand K are mass, damping and stiffness matrices, respectively; U is the
relative displacement vector; Ubis the static displacement vector when the base of
the structure displaces by unity in the direction of the earthquake; u, is the
earthquake ground acceleration; and a super dot (.) represents the time
derivative. The structure has N-orthogonal modal vectors $,, i = 1 - N. For the
present treatment, we assume that the modal vectors have been scaled such that
$, M $, = 1. Also, we recall from Chapter 1, $, K $, = of and
$i ci
C q i = 20,C,,, where o,is the circular frequency in radians per second, and
is'the damping ratio, both for mode i. In the modal superposition method we use
the following transformation (or modal superposition equation)
Equation 3.5 gives the maximum relative displacement for each mode. The
superposition equation, Equation 3.2, applies only when we know the time
histories of all the modal displacement vectors in all the modes. Equation 3.5,
however, does not provide that information. In general, it is unlikely that the
COMBINATION O F MODAL RESPONSES/31
maximum values of U;in different modes would occur at the same time. How
should we then combine these maximum modal vaiues? Given the modal
displacement vector U;, we can evaluate any other response of interest in the
same mode, R,. The vector Ui,,, gives Rim,,. The problem with combining
various modal U;,,,stated above also applies to the response R,,,,. For brevity,
we shall drop the subscript max. From now on, the term R, would denote the
maximum value of the response in mode i.
It is obvious that the upper bound of the combined response is given by the
absolute sum of the modal values
In the first of Equation 3.14 we have E, = 1 for i = j; and in the second one we
take into account the symmetry property, ciJ = E,~.
Equation 3.1 3 or 3.14 is popularly known as the double sum equation [7]. In
the correct form of the equation all the sums are algebraic. The US Nuclear Regu-
latory Commission incorrectly uses an absolute sign in .front of the double
summation 181. Consider two modes with equal frequencies and damping values.
Then, E , = ~ I, and Equation 3.13 gives
C U M B I N A T I O N OF MODAL RESPONSES133
As indicated earlier, the double sum rule correctly gives the combined response as
the algebraic sum of the two modal response values. On the other hand, if the two
modes had sufficiently separated frequencies, E , , 2: 0,and we would get
which is the SRSS rule. In general, the value of E,, varies between zero and unity.
For a given earthquake ground motion the value of E, can be evaluated
numerically, in accordance with Equation 3.12. We should make two comments
here. First, the double sum rule, or any combination rule for the response
spectrum method, is an approximate rule. Even when the value of E,, is evaluated
'exactly' from Equation 3.12, the combined responses would be approximate.
When the calculations are performed on several earthquake ground motions, the
combination rule would, on the average, give a reasonably accurate estimate of
the combined response. The second comment is about the response spectrum
method, and is related to the first comment. If the objective were to obtain the
response values for a known ground motion record, the appropriate technique
would be one of the time-history integration methods, e.g. the modal superposi-
tion method. The most appropriate application of the response spectrum method
is to design problems for which the future earthquake is not known. For this pur-
pose, we not only need 'average' spectral shapes which are presented in Chapter
2, we also need representative values of the modal correlation coefficient E,,,
which on average weuld give sufficiently accurate combined response values.
Rosenblueth and Elorduy [9] assumed the earthquake ground motion to be a
finite segment of white noise, and assumed the response to be damped periodic-
of the form e-C"' sinoDt. Based on their work, the correlation coefficient can be
written as
in which o, and o, are the circular frequencies of the two modes in radians per
second;.. o,,_.grid oD, are the corresponding damped frequencies,
"\
r'i;,, = J(l - Sf) o,/, wDJ= J(l - (j) o,; and (:and 6; are the equivalent damping
r z t o s wTkh account for the reduction in the response due to the finite nature of
the white noise segment.
where s is the effective duration of the white noise segment. The duration s is not
the total duration of the ground motion. It is not clear how exactly to evaluate it.
Villaverde [ lo] obtained numerically values of s for several ground motions by
exploiting its relationship with the expected value of pseudo velocities at
different damping values. However, he did not suggest any method of evaluating
s for a given response spectrum.
We note from Equation 3.16 that the value of effective duration has a
significant contribution in the lower frequency range only. In the higher
frequency range (;fi c,.
For the lower frequency range, the effective duration of
the earthquake may be represented by the duration of the strong motion. A
measure of the effective duration of the strong motion can be obtained from the
Husid plot[I 11, which is the graphical representation of the following function
H(t) =
I,' iii(f) dl
[iii(t) dl '
in which t, is the final value oft. By definition, 0 S H ( t ) I:1. The Husid plot for El
Centro (SOOE, 1940) ground motion is shown in Figure 3.1. The function H(t)
builds slowly initially because of the weak motion at the early phase of ground
shaking. In the intermediate duration, the H ( t ) builds rapidly. In the final phase,
very little seismic input is developed. It is clear from Figure 3.1 that the
intermediate portion of the Husid plot comprises the significant strong motion
contribution. For definitiveness, but arbitrarily, the first 5% and the last 5% are
95%
1, = 1.67 sec 1,126.15 sec
Duration
5%
t
0 10 20 30 40 50 60
Time. sec
I Fig. 3.1 Husid plot for El Centro earthquake (SOOE, 1940). (Based on Nau and Hall[l I]).
COMBINATION O F M O D A L RESPONSES135
deleted from the plot. The remaining 90% is defined as the significant strong
motion portion as shown in Figure 3.1. This duration for El Centro record is
24.5 sec. In using Equations 3.15,3.16 in conjunction with a design spectrum, the
value of the duration s should be specified.
r; r'
By substituting the expressions for and in Equation 3.15 from Equation
3.16, we obtain
The coefficient c,, was evaluated numerically for ten strong ground motion
records. On the basis of their study, an expression of the type given below was
suggested [ 12]
r,,
in which is the average damping value. Figure 3.2 shows a comparison of the
value of E,, obtained from Equations 3.18 and 3.19, with the average of
numerically obtained values from ten earthquake records[ 121. Since Equation
3.19 is based on the average of c,,values obtained from several records, it is more
appropriate to use it for a broad band earthquake input.
fi=O.l H Z , ~ = $ = O . O I
.c
9)
0
- Formula
Numerical
0
The double sum equation in which Der Kiureghian's expression is used has been
called the complete quadratic combination (CQC)[15]. When the two modes
have equal damping values, it can be shown that the E , values obtained from the
Singh-Chu equation[l3] and those from Equation 3.20 (Der Kiureghian[l4]) are
about the same. Both can be significantly lower than those given by Equation 3.15
(Rosenblueth and Elorduy[9]) and 3.18 (Gupta and Cordero[l2]) within the
frequency ranges of interest. As will be shown later, these differences in the 8,
values result in response variations that are not negligible.
There is another important element in the expressions of the correlation
coefficient which has not been explicity recognized in the published studies so
far. Equations 3.15 and 3.18 are likely to overestimate the values of E,, when the
damping ratios of two modes are sufficiently different. Consider a situation when
w, and wJare large enough that the effect of the finite duration on the values of E,,
in these equations can be neglected. Equations 3.15 and 3.18 can be approxi-
mated as follows:
Equation 3.22 includes a coefficient, J(c,l;i)/c,,,that Equation 3.21 does not. The
variation in the value of the coefficient with c,/c,ratio is tabulated below.
COMBINATION OF M O D A L RESPONSES/37
The coefficient is approximately unity when l,, and 6, are not very different.
On the other hand it is much less than unity when 4, and 6, are sufficiently apart.
This would have serious influence on the correlation coefficients for modes with
close frequencies. Consider an example, w, = o,and l,,/l,, = 0.2. Equation 3.21
would give E,, = 1.0, and Equation 3.22, E, = 0.745. Further, consider the
response of a secondary oscillator in resonant modes. Assume, R, = 1.01,
R, = -0.99. The first value of E,,, Equation 3.21, would give R = 0.02, the
second value, Equation 3.22 would yield R = 0.7 14.
Our recent numerical experimentation using the actual earthquake motion
data at North Carolina State University indicates that the correlation coefficient
values and the resulting combined response values are relatively more realistic
when the coefficient ,/(<,l;,)/l,,,is included than when it is not. Consequently,
Equation 3.18 should be modified as follows
% Error in results
Orthogonal (N-S)
response
Deflection
Average error 18 32
Maximum error 33 67
Shear
Average error 17 24
Maximum error 31 55
Overturning
moment
Average error 16 25 218 520
Maximum error 25 51 299 658
Torsional response
Torques
Average error 9 7
Maximum error 27 26
have relatively higher errors; the errors from the absolute sum calculations are the
highest. Similar conclusions can be made about the torsional response, except
that the absolute sum values now have much higher errors. All the combination
rules have the highest errors in the orthogonal north-south response. The double
sum method using the Rosenblueth-Elorduy[9] modal correlation coefficient
gives the best results. The results from the SRSS and the absolute sum
combination rules are unacceptable. The orthogonal north-south response values
from the San Fernando-Pacoima Dam excitation are shown in Figure 3.3. The
COMBINATION O F MODAL R E S P O N S E S / 3 9
ROOF
1 I c I I
1 2 3 4 1000 2000. 3000 4000 5000
Fig. 3.3 Comparison of modal combination rules: (a)story deflections, (b) story shears,
(c) story overturning moments(l61. (Reprinted by permission of John Wiley & Sons Ltd.)
order of accuracy between different combination rules observed from the figure is
the same as that concluded on the basis of Table 3.1.
of Lindley and Yow[l8]. Singh and Mehta[20] formulate the problem using the
modal acceleration approach and suggest making use of relative velocity and
relative acceleration spectra. The Singh-Mehta method is theoretically rigorous,
and gives accurate results. The availability of the relative velocity and relative
acceleration spectra may be a problem. Among the methods proposed by
Kennedy [ 171, Lindley and Yow [ 181 and Hadjian [ 191, the Lindley-Yow method
appears to be most rational and is likely to give most accurate results for a struc-
ture having frequencies in the neighborhood of or greater than the rigid
frequency. The method is likely to run into trouble for modes having frequencies
significantly lower than the rigid frequency. Even for frequencies immediately
below the rigid frequency, the method of calculating the relative spectral
acceleration is somewhat arbitrary.
We will now present a method developed by Gupta and coworkers[12,2 1-25].
It has been pointed out above that the responses from the modes having
frequencies greater than the rigid frequency are perfectly correlated with the
input acceleration history. What happens at a frequency immediately below the
rigid frequency? Let us call the correlation between a modal response and the
input acceleration as the rigid response coefficient, because at frequencies equal
to or higher than the rigid frequency the response is rigid and the correlation is
unity. Figure 3.4 shows the variation of rigid response coefficient for the San Fer-
nando earthquake (Hollywood Storage, EW, 197 1). The rigid response coefficient
becomes almost unity at about 20 Hz, which is much less than the rigid
frequency, which is about 30 Hz. Below 20 Hz the coefficient shows a gradual
diminishing trend, and becomes zero at about 2.5 Hz. This means that even the
modes having frequencies below the rigid frequency have a 'rigid content.' This is
the natural consequence of gradually changing proportions of the contributions
of the transient response (damped periodic response) and the steady state
response (rigid response).
Based on the above discussion, we can divide a modal response. R,, at a fre-
quency w, c or,into two parts: the rigid part, R,', and the damped periodic part,
RP. Denoting the rigid response coefficient by a,,the rigid part is defined by
It is assumed that the rigid part and the damped periodic part are statistically
independent
R: = +
( ~ j ) (R,?)',
~ R,? = J(l - a,?)R, . (3.25)
0.8- -
C
C
.g -
Ii 06-
1
K
04-
-
S
.-0
a 02- -
0 - I I l l
Fig. 3.4 Variation of rigid response coefficient with frequency, San Fernando earthquake
(Hollywood Storage, EW, 1971)[23].
The damped periodic pans are combined using the standard double sum
equation:
I 1 />I
in which
B,, = a,., + ,/[(I - a f ) ( l - a;)] E,,.
Equations 3.31 and 3.32 include the effec:t of rigid resplonse in the modified
correlation coefficient T,,. A comparison of the average numerical values of 8,
obtained from ten earthquake records and the values calculated from Equation
3.32 is shown in Figure 3.5. The agreement between the numerical values and
Equation 3.32 is good. Singh and Mehta[20] have also proposed an expression for
8, which includes the effect of rigid response.
The time history and the response spectrum analyses were performed on five
3-degree-of-freedom systems, each subjected to three actual earthquake ground
motion records, and to three instructure calculated motions. The five buildings
had fundamental frequencies from 2-64 Hz. The building model and the
unnormalized mode shapes (which are same for all the five buildings) are shown
in Figure 3.6. In the response spectrum method, the modal responses were
combined using four methods: SRSS, Kennedy [l7], Hadjian [I 91, and Gupta and
Chen[21]. The combined response spectrum values were compared with the
corresponding maximum values from the time-history method. The following
responses were considered: story displacements, story inertia forces, story shears
and story moments. It was found that the story displacements were dominated by
the fundamental mode and their response spectrum values were insensitive to the
method of combination. Therefore, displacements were eliminated from the
comparison. A statistical summary of the errors in the four methods of modal
combination is given in Table 3.2. The mean error in all the methods is relatively
small. The most important error parameter for consideration is the standard
deviation. Relative to the value for Gupta's method, the standard deviation for
Hadjian's method is 2.4, that for Kennedy's method is 2.9, and for the SRSS
method is 4.6. A comparison of the time-history response results with those from
the response spectrum method using Gupta's method of modal combination is
shown in Figure 3.7 for the San Fernando (Hollywood Storage, EW, 1971)
earthquake. The dotted lines in Figure 3.7 show the response spectrum results
only when they are not superimposed by the time-history results shown by the
solid lines.
lido Hz
C
- Formula
Nwnerlcal
o
- Formula
1
0
0
o Numerical
Fig. 3.5 Comparison of modal correlation coefficients including the effect of rigid
response [ 121.
Fig. 3.6 Building model and unnormalized mode shapes (same for all five building)[21].
Method
Descri~tion SRSS Kennedv Hadiian Gu~ta
Fig. 3.7 Comparison of maximum building response from time-history analysis and from
response spectrum method using Gupta's modal combination method for San Fernando
earthquake (Hollywood Storage, EW, 1971)[21].
The summations in Equation 3.37 are in the time-domain. Equations 35-37 give
The displacement vector U, gives the residual rigid response. In the modal
superposition method, the response history in modes having frequencies up to
the rigid frequency is obtained by the usual algorithms. The response in all the
remaining modes is given by U,, in Equation 3.39. Although the vector Uois time-
dependent, we need not solve the simultaneous equation at each time step. Since
the time dependence is introduced by the term ii,, which in the present case is
a scalar, we solve the simultaneous equation once with - M U,,, on the right hand
side, and multiply the solution by ii, at each time step. The complete displace-
ment vector is given by Equation 3.37. The above procedure is not only
economical compared with doing calculations for all the modes, but in most cases
it will also be more accurate. It is usually difficult to evaluate the frequencies and
the mode shapes of the higher modes very accurately.
In the response spectrum method, the responses in modes having frequencies
I up to the rigid frequency are calculated by Equation 3.5 as before. For the
, /
. residual rigid response, Equation 3.39 is replaced by
i 4
K Uoma,
= M UbO
(ZPA). (3.40)
I
$ ,
The residual value of any response, Ro, is then calculated from U
,
,, . The modal
1: I
I combination procedure described in the previous section, Equations 3.24-3.28,
is applied with one change. Equation 3.28 is modified as follows:
This method of residual rigid response calculation was applied to a piping sys-
tem shown in Figure 3.8 by Gupta and Jaw[24]. The piping was subjected to Taft
(N2 1E, 1952) ground motion which has rigid frequency, f r -
20 Hz. There were
nine modes having frequencies <20 Hz, n = 9. We are presenting here some of
the results from Reference[24], comparing the time-history results with the
response spectrum results with or without residual rigid response. Table 3.3 gives
the significant element forces and moments in the vicinity of the supports
obtained from the time-history analysis along with the errors in the response
spectrum calculations. Ignoring the residual response gives error as high as
-98.6%, which means the calculated value is 1.4% of what it should be, clearly an
unacceptable situation. Excluding the residual rigid response introduces a mean
error of -43% as opposed to mean error of only 4% when the residual rigid re-
sponse is included. This shows a definite bias in the response results towards
under-estimation when the residual rigid response is not included. The exclusion
of residual rigid response also gives much larger error standard deviation and
RMS error than when it is included.
References
1. L.E. Goodman, E. Rosenblueth and N.M. Newmark. Aseismic Design of Elastic
Structures Founded on Firm Ground. Proceedings, ASCE, November 1953, pp. 349-1,
349-27.
2. R.L. Jennings and N.M. Newmark, Elastic Response of Multi-Story Shear-Beam-Type
Structures Subjected to Strong Ground Motion, Proceedings, Second World Conference on
Earthquake Engineering, Vol. 11, Tokyo, 1960.
3. H.D. Merchant and D.E. Hudson, Mode Superposition in Multi-Degree-of-Freedom
Systems Using Earthquake Response Spectrum Data. Bulletin of the Seismological Society
ofAmerica, Vol. 52, No. 2, 1962, pp. 405-416.
4. R.W. Clough, Earthquake Analysis by Response Spectrum Superposition, Bulletin of the
Seismological Society of America, Vol. 52, No. 3, 1962, pp. 647-680.
5. N.M. Newmark, W.H. Walker. A.S. Veletsos and R.J. Mosborg, Design Procedures for
Shock Isolation Systems of Underground Protective Structures, Report for Alr Force
Weapons Laboratory, by Newmark, Hansen and Associates, RTD TDR 63-3096,
December 1965.
6. A.G. Davenport, Note on the Distribution of the Largest Value of a Random Function with
Application to Gust Loading, Proceedings. lnstltution of Civd Engrneers, Vol. 28. 1963,
pp. 187- 196.
7. A.K. Singh, S.L. Chu and S. Singh, Influence of Closely Spaced Modes in Response.
Spectrum Method of Analysis, Proceedings, Speciality Conference on Structural Design of
Nuclear Power Plant Facilities, ASCE, Chicago, Illinois, 1973.
8. United States Nuclear Regulatory Commission, Design Response Spectra for Nuclear
Power Plants, Nuciear Regulatory Guide, No. 1.60, Washington, DC, 1975.
9. E. Rosenblueth and J. Elorduy, Response of Linear Systems in Certain Transient
Disturbances, Proceedings, Fourth World Conferenceon Earthquake Engineering, Santiago,
Chile, 1969, A-1, pp. 185-196.
10. R. Villaverde, On Rosenblueth's Rule to Combine the Modes of Systems with Closely
Spaced Natural Frequencies, Bulletin of the Seismological Society of America, Vol. 74,
No. 1, February 1984, pp. 325-338.
I I. J.M. Nau and W.J. Hall, An Evaluation of Scaling Methods for Earthquake Response
Spectra, Structural Research Series, No. 499, Department of Civil Engineering, University
of Illinois at Urbana-Champaign, Urbana, Illinois, May 1982.
I 12. A.K. Gupta and K. Cordero, Combination of Modal Responses, Transaction. Srxth
I Internatronal Confirence on Structural Mechanrcs in Reactor Technology,Paper No. K7115,
1 Paris, August 198 1.
r
i
13. M.P. Singh and S.L. Chu, Stochastic Considerations in Seismic Analysis of Structures,
Earthquake Engineering and Structural Dynamics, Vol. 4, 1976, pp. 295-307.
I 14. A. Der Kiureghian, A Response Spectrum Method for Random Vibrations, Report No.
UCBIEERC-8O//I5, Earthquake Engineering Research Center, University of California,
Berkeley, California, 1980.
15. E.L. Wilson, A. Der Kiureghian and E.P. Bayo, A Replacement for the SRSS Method in
Seismic Analysis, Short Communrcation. Earthquake Engineering and Structural Dyna-
mics, Vol. 9, 198 1, pp. 187-1 94.
16. B.F. Maison, C.F. Neuss and K. Kasai, The Comparative Performance of Seismic Response
Spectrum Combination Rules in Building Analysis, Earthquake Engineering and Structural
Dynamrcs, Vol. 11, 1983, pp. 623-647.
17. R.P. Kennedy, Recommendations for Changes and Additions to Standard Review Plans
and Regulatory Guides Dealing with Seismic Design Requirements for Structures, Report
prepared for Lawrence Livermore Laboratory, Published in NUREGICR-I 161, June
1979.
18. D.W. Lindley and J.R. Yow, Modal Response Summation for Seismic Qualification,
Second ASCE Specralitj~Confirence on Civil Engineering and Nuclear Power, Vol. VI,
Paper 8-2. Knoxville, TN, September 1980.
19. A.H. Hadjian, Seismic Response of Structures by Response Spectrum Method, Nuclear
Engineer~ngand Design, Vol. 66, No. 2, August 198 1, pp. 179-201.
20. h . ~Singh
. and K.B. Mehta, Seismic Design Response by an Alternative SRSS Rule,
Earthquake Engrneering and Structural Dynamics, Vol. 11, 1983, pp. 77 1-783.
21. A.K. Gupta and D.C. Chen, Combination of Modal Responses: A Follow Up, Report,
Department of Civil Engineering, North Carolina State University, Raleigh, November
1982.
22. A.K. Gupta and D.C. Chen, A Simple Method of Combining Modal Responses,
Transactions, Sewnth International Conference on Structural Mechanics in Reactor
Technology, Paper No. K3/ 10, Chicago, August 1983.
23. A.K. Gupta and D.C. Chen, Comparison of Modal Combination Methods, Nuclear
Engineenng and Design, Vol. 78, March 1984, pp. 53-68.
24. A.K. Gupta and J.W. Jaw, Modal Combination in Response Spectrum Analysis of Piping
Systems, in Seismic Eflects in PVP Components, PVP. Vol. 88, ASME, 1984, pp. 1-12;
presented at Pressure Vessel and Piping Conference, San Antonio, Texas, June 1984.
25. A.K. Gupta, Modal Combination in Response Spectrum Method, Proceedings, Eighth
World Conference on Earthquake Engineering, San Francisco, 1984.
26. American Society of Civil Engineers, Standard for the Sersmic Analysis of Safety-Related
Nuclear Structures, September 1986.
27. K.M. Vashi, Computation of Seismic Response from Higher Frequency Modes, Journal of
Pressure Vessel Technology, ASME, Vol. 103, February 198 1, pp. 16- 19.
28. G.H. Powell, Missing Mass Correction in Modal Analysis of Piping Systems, Transactions,
Fifih International Conference on Structural Mechanics in Reactor Technology, Paper No.
K1013, 1979.
29. A.J. Salmonte, Consideration on the Residual Contribution in Modal Analysis, Earthquake
Engineering and Structural Dynamics, Vol. 10, 1982, pp. 295-304.
Chapter 41Response to multicomponents
of earthquake
4.1 Introduction
The earthquake motion at any point can be resolved into three orthogonal
components, two horizontal and one vertical. Penzien and Watabe[l] have
shown that two horizontal components, which are approximately radial and
tangential with respect to the epicenter, are uncorrelated. Any other orientation
of horizontal axes leads to partially correlated horizontal components. This
would normally be the case because the buildings or the structures are, in general,
unlikely to be placed along the radial and tangential directions. The vertical
component of the earthquake motion always has some correlation with the
horizontal components. Because of the wave motion, a building is also subjected
to three rotational components[2-71. These rotational components are mutually
correlated, and have a strong correlation with the translational components. The
structures supported on multiple supports, such as bridges can be considered to
be subjected to multiple 'components.' These components would have variable
degrees of correlation.
Let us denote the earthquake component by the capital letter subscripts, and
modes by the small letter subscripts. For example, R,, represents the response
spectrum value of a response in the ith mode of vibration due to Ith earthquake
component. Following a procedure similar to that used for the derivation of
Equation 3.13, we can write the following equation for component and modal
response equation [8]:
in which ci,includes the effect of modes with close frequencies and that of the
rigid response at higher frequencies. In Chapter 3, we denote the corresponding
correlation coefficient by BO;we are dropping the bar in Equation 4.2 here for
brevity without any loss of meaning. Equation 4.2 can be rewritten as
Equations 4.1-4.3 yield the maximum probable value of any response. Often
the design of a structural element is based upon more than one response, for
example, a column subjected to axial force P and bending moment M, or a metal
element subjected to stresses ox, o,, o,, T,,, T,,, T ~ .It is unlikely that the
maximum values of these responses would occur simultaneously. In most
conventional design procedures, it is implicitly assumed that the maximum
responses do occur simultaneously. This assumption introduces error on the safe
side which may be significant. Various aspects of this problem are discussed in
subsequent sections.
Before we proceed, let us consider the nature of Equation 4.3. It can be viewed
as the definition of the length R of a vector Rli in' a 3N-Riemannian space [ 141
having a metric tensor E,,, where N is the number of modes. Now consider the
instance of time when a particular response U reaches its maximum value Urnax.
Equation 4.4 gives
In Equation 4.6, u,, is a unit vector in the Riemannian space. Equations 4.3 and
4.6 give
At the instant, Kli has a value which gives U,,, we will have
Since, v, is an arbitrary unit vector, without any loss of generality we can write
J C l s1.
l i j
Let us now substitute Equation 4.10 into Equation 4.6, which gives
C 111
l i j
E,, uIiv I = 1. (4.12)
In the Riemannian space, the cosine of the angle between unit vectors uli and uIJ is
given by
COS (u, V) =
l
1
i j
E;, MI; U1,.
which is possible when the unit vectors ull and v, are parallel. If we assume that
the positive direction of both the vectors is same, then we conclude
Equation 4.17 gives the value of K,,, at the instant when U,,, is attained. We
shall now proceed to find a relationship for K,, which will be applicable to any
arbitrary response. The coefficient E,, is an element of the N X N matrix e, where
N is the number of modes. The matrix e is positive-definite by definition, see
Chapter 3. Let us define the inverse of the matrix E by e-. The solution of
Equation 4.17 gives uIj = 16, K,, which with Equation 4.6 yields
i
In Equation 4.1 8, KIl (or K,,) values are unknown. Any set of values of Kit,
which satisfy Equation 4.18, will yield the maximum value of a response Rr
where r is the response number. The value of other responses at the time when Rr
is maximum, can be obtained by another equation similar to Equation 4.19,
RESPONSE T O MULTICOMPONENTS O F EARTHQUAKE155
where s # r, the K,, values are the same as those in Equation 4.19. Thus, all the
R'and I? values occurring simultaneously are evaluated. In practical applica-
tions, one may not be able to identify the response Rr, which is maximum for a
given set of K, satisfying Equation 4.18. However, that is not necessary. In
practice, Equations 4.18 and 4.19 can be used as follows. Evaluate sufficient sets
of K , values which satisfy Equation 4.18. Then for each set of K, values evaluate
Rr from Equation 4.19, for all responses R' of interest. For example, in case of a
beam column, we have two responses of interest: axial force P and bending
moment M. Then, R ' = P, R * = M. Each set of Rr values (example, P, M) gives
the simultaneously occurring responses which should be considered in the design.
The design function @ in Equation 4.20 is a linear function of the responses Rr,
provided the influence coefficients A' are constants. It is not unusual for a design
function to be a nonlinear function of the response values, in which case the in-
fluence coefficients, A', are functions of the response values also. For the purpose
of Equation 4.20, however, A' is assumed to be locally constant. This assumption
is justifiable in view of the fact that the response analysis which gives the Rr
values itself is linear. For the ith mode, Ith component of earthquake, we have
Equations 4.3 and 4.20 give the maximum response value of the design function
as
$2 = C 711& A ~ A S R ; R ; ,
r s l i j
It is obvious from Equation 4.22 that for design purposes, it is sufficient to know
G" values, and it is not necessary to know R;, values. At least, it is not necessary to
work with R;, values.
We shall now introduce the 'equivalent modal responses' denoted here by a:.
The subscript in a:denotes the equivalent mode number. The equivalent modes
are defined such that
The subscripts i, j in the triple summation in Equation 4.25 refer to the actual
mode numbers, and the single summation refers to the equivalent mode
numbers. Whereas, the modal response values R, constitute a 3N-Reimannian
a:
space with E,,as the metric tensor, Equation 4.25 indicates that values are part
RESPONSE TO MULTICOMpONENTS OF EARTHQUAKE157
of an M-Cartesian space whose metric tensor is given by an identity matrix. Since
the Cartesian space is a special form of Reimannian space, we can also apply the
conclusions of the previous section here.
Equations 4.18 and 4.19 are now replaced by
It is clear that the above equations represent an ellipse, which we shall call
'interaction ellipse.' In general, when M > 2, Equation 4.26 represents an
ellipsoid in an M-space. R , , E2are like direction cosines. For design purposes,
+
several sets of i?,, RZvalues which satisfy the equation R: K: = I should be
used, as shown in Table 4.2. The calculated values of P and M are plotted in
Figure 4.1, the points are joined to form a polygon. Also shown in Figure 4.1 are
the exact ellipse, and a rectangle which will be obtained when we assume that the
maximum values of + P and -+ M occur simultaneously.
No. K, P M No. K, P M
RESPONSE TO MULTICOMPONENTS OF EARTHQUAKE159
Interaction t 12.0
,r----
I
Rectangle for
Conventional Method
Equation 4.30 is the desired equation of the interaction ellipsoid. This derivation
is based on Gupta and Chu[l2]. An alternative derivation is given by Gupta and
Singh [ 1 71.
the corresponding errors are given in Table 4.3. Rosenblueth and Contreras[l I]
give values of C, when the maximum errors on the safe and unsafe side are equal.
Many problems of practical interest have M = 2 or 3. As can be seen from
Table 4.3, the maximum error on the conservative side will be only 8% or 13% in
the two cases, respectively. The highest possible error is 30%. These are
maximum possible errors. In most cases the errors will be somewhere between
zero and the maximum value. It is shown by Gupta, Fang and Chu[19] that the
RESPONSE TO MULTICOMPONENTS OF EARTHQUAKE161
Rp x l m , = D l , Dl = Permutations (tC,).
I
Equation 4.33 will give 2 M ~sets ! of R~pp,,,m, values. Let us apply this
procedure to the beam-column problem of Section 1.2. The values of D, and the
corresponding P, M points are given in Table 4.4. These points are plotted in
Figure 4.2, to give the polyhedron enveloping the interaction ellipse.
When M is relatively large, the number of permutations in Equation 4.33 will
become very large. For example, for M = 6, 2MM ! = 46 080. Although such
No. D, D, P M
Polygon for
approximate method t "SO
Rectangle for
conventional method
Fio. 4.2 Interaction ellipse and the approximate polyhedron for the column problem[l6].
r Conventional
Fig. 4.3 Interaction diagrams for combined static and seismic loads[l6].
must be resisted simultaneously with the seismic loads. Thus, for design
purposes,
R;,I RLc + Rkismic . (4.35)
Equation 4.35 amounts to shifting the center of ellipsoid to the point R&,,,. For a
safe design, all the possible RW,, points should be within the resisting capacity of
the structural element under consideration.
The design problem of a reinforced concrete beam-column is illustrated in
Figure 4.3. It is shown that the interaction ellipse representing the seismic loading
with origin shifted for the static loading is completely inscribed by the capacity
interaction diagram of the beamcolumn section. Also shown in the figure is the
rectangle given by the conventional method in which the maximum seismic Pand
M values are assumed to occur simultaneously. The conventional method would
have required a stronger, and therefore, less economical section.
Methods similar to those presented here have been applied to the design of
building cross-sections, base slabs and reinforced concrete columns, and to the
analysis of base slabs with local uplift by Gupta and Chu[12]. The design of steel
beam-columns have been studied by Gupta, Fang and Chu[l9]. Gupta has
applied these methods to a reinforced concrete nuclear pressure vessel, shear-
walls[l6] and piping[l6,20] systems.
References
1. J. Penzien and M. Watabe, Simulation of 3-Dimensional Earthquake Ground Motion,
Bulletin of International Institute of Seismology and Earthquake Engineering, Vol. 12,
1974, pp. 103-1 15.
2. J.L. Bogdanoff, J.E. Goldberg and A.J. Schiff, The Effect of Ground Transmission Time on
the Response of Long Structures, Bulletin o f the Seismological Society ofAmerica, Vol. 55,
No. 3, June 1965, pp. 627-640.
3. N.M. Newmark, Torsion in Symmetric Buildings. Proceedings. Fourth U'orld Conference
on Earthquake Engineering, Vol. 2, Paper A-3, Santiago, Chile, 1968, pp. 19-32.
4. E. Rosenbueth, The Six Components of Earthquake, Proceedings, Twelfih Regional
Conferenceon Planning and Design o f Tall Buildings, Sydney, Australia, 1973, pp. 63-81.
5. N.D. Nathan and J.R. MacKenzie, Rotational Components of Earthquake Motion,
Canadian Journal ofCivil Engineering, Vol. 2, 1975, pp. 430-436.
6. E. Rosenblueth, Tall Building Under Five-Components Earthquake, Journal of the
Structural Division, ASCE, Vol. 102, No. 2, February 1976, pp. 453-459.
7. W.K. Tso and T.I. Hsu, Torisonal Spectrum for Earthquake Motion, JournalofEarthquake
Engineering and Structural Dynamics, Vol. 6, 1972, pp. 375-382.
8. A.K. Gupta, Multicomponent Seismic Design, Proceedings, Seventh U'orld Conference on
Earthquake Engineering, Istanbul, Turkey, September 1980.
9. M. Amin and A.H.S. An& Nonstationary Stochastic Model of Earthquake Motion, Journal
o f Engineering Mechanics Division, ASCE, Vol. 94, No. EM2, April 1968.
10. M. Ghafory-Ashtiany and M.P. Singh, Seismic Response for Multicomponent Earth-
quakes, Technical Report, NSF Grant No. CEE-8214070, Department of Engineering
Science and Mechanics, Virginia Polytechnic Institute and State University, Blacksburg,
April 1984.
I I . E. Rosenblueth and H. Contreras, Approximate Design for Multicomponent Earthquakes,
Journal o f Engineering Mechanics Division, ASCE, Vol. 103 No. EMS, 1977, pp. 88 1-893.
12. A.K. Gupta and S.L. Chu, A Unified Approach to Designing Structures for Three
Components of Earthquake, Proceedings, International Symposium on Earthquake Struc-
tural Engineering, St. Louis, Missouri, August 1976, pp. 58 1-596.
13. A.K. Gupta and S.L.Chu, Probable Simultaneous Response by the Response Spectrum
Method of Analysis, Nuclear Engineering and Design, Vol. 44, 1977. pp. 93-95.
14. J.L. Synge and A. Schild, Tensor Calculus, University of Toronto Press, Toronto, 1949.
15. A.K. Gupta and S.L. Chu, Equivalent Modal Response Method for Seismic Design of
Structures, Nuclear Engineering and Design, Vol. 44, 1977, pp. 87-9 1.
16. A.K. Gupta, Design of Nuclear Power Plant Structures Subjected to Three Earthquake
Components, Proceedings. ASCE Speciality Conjerence on Civil Engineering and Nuclear
Power, Knoxville, Tennessee, September 1980.
RESPONSE TO MULTICOMPONENTS O F EARTHQUAKE/^^
17. A.K. Gupta and M.P.Singh, Design of Column Sections Subjected to Three Components of
Earthquake, Nuclear Engineering and Design, Vol. 41, 1977, pp. 129-1 33.
18. A.K. Gupta, Approximate Design for Three Earthquake Components, Journal of Engin-
eering Mechanics Division, ASCE, Vol. 104, No. EM6, December 1978, pp. 1453- 1456.
19. A.K. Gupta, S.J. Fang and S.L. Chu, A Rational and Economical Seismic Design of Beam-
- Columns in Steel Frames, Transactions. Fourth International Conference on Structural
Mechanics in Reactor Technology, Paper No. K9/7, San Francisco, 1977.
20. A.K. Gupta, Rational and Economic Multicomponent Seismic Design of Piping Systems,
Journal of Pressure Vessel Technology, ASME, Vol. 100, November 1978.
Chapter 5/Nonclassically damped systems
5.1 Introduction
Equations of motion of the classically damped multi-degree-of-freedom (MDOF)
systems can be transformed into a set of independent modal equations using the
real-valued eigenvectors and eigenvalues of the undamped systems as was done
in Chapter 1. However, in many real systems the modal equations are coupled by
the nonclassical damping maxtrix[l]. In many cases, nonclassically damped
systems can be approximated by a classically damped system without a
significant loss of accuracy. On the other hand, there are important practical
situations when the nonclassical nature of the damping matrix cannot be ignored.
Such is the case when a structure is made up of materials with different damping
characteristics in different parts. For example, a combined analytical model of a
soil-structure system is nonclassically damped. Another example is a coupled
structure-equipment (primary-secondary) system.
Whether damped classically or nonclassically, one can always evaluate the
response of a MDOF system using the direct time-history integration analysis. In
classically damped systems, the advantage of modal analysis is that often the
response can be represented by a few modes only. For nonclassically damped sys-
tems, one may still use the undamped mode shapes to obtain the transformed
coupled modal equations as was done by Clough and MojtahediI2J. Because the
resulting equations are coupled, the complete transformed system must be
integrated simultaneously, as in the direct integration method. The degree of
efficiency thus achieved is much less than when the equations are uncoupled. On
the other hand, this procedure can be used to reduce the number of degrees-of-
freedom by considering only a few of the undamped modes.
Frequencies and mode shapes of a nonclassically damped system are complex
and can be calculated using Foss' method[3]. The complex mode shapes and
frequencies can be used to obtain first order uncoupled modal differential
equations[4]. Itoh[5] solved these equations in conjunction with the FFT (fast
Fourier transform) procedure.
SinghI61, using the random vibration approach, developed a response
spectrum method for nonclassically damped systems. Implicit in the method is
the dependence of response on two spectra, one based on the maximum relative
displacement, and another one based on the maximum relative velocity. Singh[6]
assumed that the relative velocity spectrum is the same as the relative
displacement spectrum (when expressed in same units) in the low and interme-
diate frequency ranges, that the relative velocity spectrum has zero ordinates in
66
NONCLASSICALLY D A M P E D SYSTEMS/67
the higher frequency range, and that there is a linear transition on the logarithmic
scale between the frequencies the relative velocity spectrum diminishes from
relative displacement equivalence to zero. As will be shown later, these
assumptions are not entirely true. Although these assumptions may introduce
errors in certain special cases, in most cases, the method should yield satisfactory
results. Singh [6] also presented a comprehensive modal response combination
procedure. The main criticism of Singh's method is that it is relatively tedious.
Villaverde and Newmark[7] performed a deterministic formulation starting
with the complex frequencies and mode shapes. For each complex mode shape
and its conjugate, they explicitly showed that the response can be represented in
two parts, one based on the relative displacement spectrum and another one
based on the relative velocity spectrum. They assumed that the two spectra are
equivalent when expressed in the same units. This assumption does not hold true
in the high frequency range.
Igusa and Der Kiureghian[8], and later Gupta and Jaw[9], and Veletsos and
Ventura[lO] showed that the displacement vector of a nonclassically damped
MDOF system may be expressed as a linear combination of the displacements
and velocities of equivalent single-degree-of-freedom (SDOF) systems. Igusa and
Der Kiureghian[8] treated the earthquake as white noise and, thus, were able to
simplify the final modal response combination equation greatly. Gupta and
Jaw[l 11 extended their earlier formulation[9] to the response spectrum method.
They also proposed a method for estimating the relative velocity spectrum that
would be needed for the analysis.
where M, Cand Kdenote mass, damping and stiffness matrices, respectively; Uis
the relative displacement vector; Ub is a displacement vector obtained by
statically displacing the support by unity in the direction of the input motion; 14, is
the ground (or support) displacement; the super dot (.) represents a derivative
with respect to the time variable. In the Foss approach[3], Equation 5.1 is cast
into a 2N-dimensional matrix equation as follows:
where
The free vibration equivalent of Equation 5.2 yields N complex eigenvectors and
eigenvalues, along with their conjugates. These eigenvectors satisfy various
orthogonality conditions[4]. The complex eigenvectors can be written in terms of
N- dimensional vectors as
in which zi and 2, can be called complex normal coordinates. The bar (-) in Equa-
tion 5.5 and elsewhere denotes the complex conjugate. Equation 5.5 with
standard operations yields the following uncoupled equations.
in which
where
Let us write various parameters in Equations 5.6 and 5.9 in terms of their real and
imaginary parts.
For a given mode, 6 and q are functions of the time variable only. The parameters
(a, p) define the mode shape and differ among the various DOF as they should. In
a classically damped system, the variation in (a,p) can be represented in terms of
a real vector, which, in fact, represents the mode shape. In a nonclassically
damped system, on the other hand, both a, P vary independently. Thus, each
DOF has its own relative phase. In a classically damped system, we can write one
differential equation in terms of the real normal coordinate and real coefficients.
In a nonclassically damped system, if we wish to deal with real values, we must
write a differential equation for each DOF.
Substituting Equation 5.10 into Equation 5.6 and separating real and
imaginary parts, and then solving for 6 and q, we get
ij + 2 c , o i 5 + of6 - -(<,wig - wDih)ii, - gu,,
In terms of the solution of Equation 5.1 5, the solution of Equation 5.13 can be
written as
u,, = r (x, cose - i,sin 8/o,). (5.16)
Equation 5.16 can be used to obtain the history of the displacement u,,. It can be
written as
u:'.=~~x,, y(,=rcos8,
= ,, tyy' = r sin 8/61,. (5.17)
We would like to write Equation 5.17 in matrix notations. We note from
Equations 5.10
The nature of these spectral properties is discussed in more detail in the next sec-
tion. Equations 5.19, 5.22 and 5.23 give
d
Uimax
= y : ~ & , UP,, = yPS& (5.24)
NONCLASSICALLY DAMPED SYSTEMS171
For each complex mode i and its conjugate, Equation 5.24 gives two response
displacement vectors. These vectors, in turn, yield two modal values of any
response, R: and R,",in which the suffix 'max' has been dropped for brevity. In a
classically damped case the vector yrP becomes null, so the response Ry is zero. In
those cases the problem reduces to combining the maximum modal responses
from various modes, R:'. Now, we have two maximum responses from each
mode. The method of combining these responses is presented subsequently.
in which all the quantities are based on the maximum relative displacement. We
can write a similar relationship for the other spectral value:
where, now all the quantities are based on the maximum relative velocity. T o
distinguish the quantities in Equations 5.25 and 5.26 we shall call all the sJquan-
tities in Equation 5.25 the displacement spectrum values, and the S"quantities in
Equation 5.26, the velocity spectrum values. Again, in both cases the units are de-
fined by the subscripts D, V o r A.
For a given earthquake motion history, the procedure for obtaining the
velocity response spectrum is straightforward. For design purposes, the displace-
ment spectrum is specified, the velocity spectrum is not. Theoretically, it should
be possible to specify the velocity spectrum also. We shall investigate here the
characteristics of the velocity spectrum and its relationship with the displace-
ment spectrum. The objective is to estimate the velocity spectrum from a given
displacement spectrum.
Figure 5.1 shows the displacement and the velocity spectra for El Centro
(SOOE, 1940) record. A 2-sec initial pulse was added to correct for the initial
ground motion values[13,14]. Both spectra are in acceleration(g) units. As such
we are dealing with the ~ dand
, S: values. We observe from Figure 5.1 that the two
spectra are approximately equal in the intermediate frequency range. The
displacement spectrum is higher in the higher frequency range, and the velocity
spectrum is higher in the lower frequency range. Similar observations have been
made by others in the past; see for example Nau and Ha11[14].-
Frequency. Hz
Fig. 5.1 Displacement and velocity spectra for El Centro Earthquake (SOOE, 1940);
damping ratio = 0.05[ l 11.
...
,,)L = -
maa
(,,"= -
..
Lg' mar
Ug max U g max
Also
wL OH
f = -(Hz),f = -(HZ).
2a 2a
If we knowf and f ",we can accurately determine S: from S; in the high and
low frequency ranges. Further, we know that in the intermediate frequency range,
S ; - ~ d , This,
. for most part, completes the estimation of S ; . The exception is a
band of frequencies between 'intermediate' and 'high' frequencies. This is the
band of frequencies in which the displacement spectrum transitions from almost
totally non-rigid (damped periodic) to almost totally rigid, see Chapter 3. This
transition is characterized by a change of correlation between the response
history and the input ground acceleration history from 0 to 1. This correlation
coefficient is called the rigid response coefficient, to be denoted here by a d . ~ h e
coefficient ad varies from 0 to 1 between two frequencies f' and f *.These
frequencies are estimated from the following equations
d
W'
f l = % ,W 1 =
S A maa
- , fZ = (f' + 2f ')/3, (5.30)
S L a a
wheref 'is the minimum frequency at which S: z ii,,,,, and is commonly known
as the rigid frequency.
Now, for fs f ', Equation 5.27 holds; for/" rf 2f L,S" = sd,and for f 2 f'
Equation 5.28 holds. In the range f z rf 2 f ', we need to account for the
transition from the damped periodic to rigid response. As in Chapter 3, we
separate Sd into two parts, the damped periodic and the rigid as follows:
Further,
a t f = f 2 , a v = a d = 1,and
at f = f H , a V =ad.
and the rigid frequency f ', defined earlier. For twelve earthquake records, these
frequencies (f ,' f and f ') along with the two key frequencies of interest
( f L and f H, are listed in Table 5.1. We observe that f0 and f are of approxim-
ately similar magnitude and that f O <f ' <f H c f '. This observation is important
in that it defines the neighborhood of the desired key frequencies f Landf H. Also
given in Table 5.1 are the ratios f L/fO,f H/f I and f H/fr. There is quite a bit of
scatter in these ratios among various earthquake records. Such a scatter is not
unexpected. The standard smooth shape of the design spectrum is obtained from
averaging much scattered values. It is reasonable to use the average values of
these ratios to estimate f and f frequencies. Therefore
The two estimates of f H are comparable in most cases. The equation in terms of
f' may be preferable, since the frequency f can be evaluated directly from
Equation 5.36. A certain degree of judgment may be required in defining f r in a
non-smooth spectrum.
Assuming that the earthquake motion is stationary and ergodic, the standard
deviation can be expressed as (Chapter 3)
Table 5.1 Key frequencies for various ground motions[l I]
where td is the duration of response, op is the standard deviation for RP(t), etc.
and
in which a;' is the standard deviation for x,, and o:'for f,.
The maximum value is given by a peak factor times the standard deviation. If
we assume that the peak factor is the same for all quantities, and that the
maximum values are the same as the maximum of the time-history values, we can
write the desired modal combination equation
In the classically damped systems R,"= 0, and Equation 5.42 degenerates into
-
the standard double sum equation. Various authors have given expressions for
the correlation coefficient E; see Chapter 3. For frequencies less than the key
frequency, based on Igusa and Der Kiureghian[8], we observe E; ~ , j=
l E,,.
Further, we can write
As is well known, the correlation coefficients and ~ , linj Equation 5.42 are
E:
unity (or nearly unity) when i==j,or when o , = w, and 6, = 6,. These coefficients
diminish rapidly as the ratio o,/o,departs from unity. On the other hand, the
correlation coefficient p,, is zero when i = j or when o, = o,. This coefficient
first increases in magnitude to a maximum value as o,/o,departs from unity, but
it soon diminishes too by virtue of the diminishing E,, in the expression.
It can be assumed that the two rigid parts are uncorrelated. Therefore,
The damped periodic parts can be combined in accordance with Equation 5.42.
The above equations can be condensed into one equation with modified
correlation coefficients.
Hence,'
We have denoted the response vector in the ith complex conjugate pair of modes
as Ui in Equation 5.21. From Equation 5.59 we infer
The structure has N-DOF and N pairs of conjugate modes. Let us assume that
there are n pairs of modes that either have frequencies less than the rigid
frequenciesf ror have significant values of U:. The higher of the two values of
n given by these conditions should be used. We denote the combined response in
the modes beyond those n pairs by U,. The total response vector is given by
1 The response of the structure in the modes having frequencies greater than the
rigid frequency is pseudo-static. Therefore, u0and & vector terms in Equation
1
5.62 can be ignored and we get
N O N C L A S S I C A L L Y D A M P E D SYSTEMS181
only one static analysis. The other term y:/o: is known from the modal
i- l
analysis.
The history of the residual response U, is evaluated by multiplying the vector
The vector U, gives the value of any residual rigid response R,. The modal
combination equations, Equations 5.45-5.48 are appropriately modified. For
example, Equation 5.45 becomes
5.8 Application
The response spectrum method was applied to nine primary-secondary systems
of the type shown in Figure 5.3 [I 11. The story stiffness and mass of the primary
system were kept constant, and those of the secondary system were varied, Table
5.2, to obtain different coupled systems. The first eight systems were selected
such that the uncoupled analysis of the primary system would introduce
approximately 10% error in the fundamental frequency. The ninth system was
selected to make the fundamental frequencies of the primary and secondary
systems identical, the tuned case.
The uncoupled frequencies and mode shapes of the primary and secondary
systems are shown in Figure 5.4. The damping ratio of the primary system is 796,
and of the secondary system is 2%. A method of obtaining the coupled
nonclassical damping matrix is described in Chapter 6. The coupled frequencies
and damping ratios for all the nine cases under consideration are given in Table
5.3. The coupled modes which have damping ratios close to 7% are dominated by
the primary system modes, those having damping ratios close to 2% have
predominant secondary system modes, and those which have damping ratios in
between have significant participation from both the primary and secondary
system modes. The coupled modal vectors y%nd y " are shown in Figures 5.5
and 5.6.
Primary system Secondary system Coupled system
All nine systems were subjected to the twelve earthquake motions listed in
Table 5.1. Two methods of analysis were used, the direct time-history integration
using the Newmark's f3 method and the proposed response spectrum approach.
The direct time-history integration results were treated as standard and the
response spectrum results were compared against them.
Details of Case 2 subjected to El Centro ground motion are given in Tables 5.4
and 5.5. It is observed that the response spectrum method does yield satisfactory
results. In all, for nine systems, twelve time histories each, there were one
Frequencies (Hz)
(b)
hundred and eight direct time history, response spectrum analyses pairs. In most
cases the errors were sufficiently small. The highest error in displacements was
41% and that in element forces 45%. In a comparative study performed by
Maison, Neuss and Kasai in Reference[l6] of Chapter 3, the maximum error was
found to be 67%when double sum combination was used for a classically damped
building. In other modal combination methods (SRSS, absolute sum) the error
Table 5.4 Comparison of nodal displacements (in), Case 2, El Centro (SOOE, 1940)[1 I]
Mode 6 Mode 6
f = 12.8Hz.1~7.0% f = 16.9Hz. {=?.o%
[:::::l : l ~ . o ~ ~ ~ ~ ~
0.038 0.038 -0.009 -0.009
0.05f'08~.0 -0.001
0.0 19
-0.074 -0.03 1
0.065 0.0 17
Mode 7 Mode 8
f = l9.9Hz,{=7.O% f = 21.9Hz.{=7.0%
- Mode 3 (10-3
f =6.5Hz,(=2.2%
Mode 4 (lo-')
f =8.2~z.(=6.7%
-0265
-0.491
-0.559
-0.438
was much higher. The combined error statistics for all the cases analyzed are sum-
marized in Table 5.6. The mean error is in the range of 1%, and the standard
deviation of the percent error is in the order of ten. This shows that the response
spectrum method for the nonclassically damped system is reasonably accurate.
The degree of accuracy is of the same order as is usually expected in the response
spectrum analysis of classically damped systems.
References
I. G.B. Warburton and S.R. Soni, Errors in Response Calculations of Non-Classically
Damped Structures, Earthquake Engineering and Strzictural Dynamics, Vol. 5. 1977.
pp. 365-376.
2. R.W. Clough and S. Mojtahedi, Earthquake Response Analysis Considering Non-
Proportional Damping, Earthquake Enginrering and Structural Dynamics, Vol. 4 , 1976,
pp. 489-496.
3. K.A. Foss, Coordinates Which Uncouple the Equations of Motion of Damped Linear
Dynamic Systems, Journal ofApplied Mechanics, Vol. 25, September 1958, pp. 361-364.
4. W.C. Hurty and M.F. Rubenstein, Dynamics of Structure, Prentice Hall, Clifton. New
Jersey, 1964.
5. T. Itoh, Damped Vibration Mode Superposition Method for Dynamic Response Analysis,
Earthquake Engineering and Structural Dynamics, Vol. 2, 1973, pp. 47-57.
6. M.P.Singh, Seismic Response by SRSS for Nonproportional Damping, Journal of the
Engineering Mechanics Diuision, ASCE, Vol. 106, No. EM6, December 1980,
pp. 1405-1419.
7. R. Villaverde and N.M. Newmark, Seismic Response of Light Attachments to Buildings,
Structural Research Series, No. 469, University of Illinois at Urbana-Champaign, February
1980, Chapter 5.
8. T. lgusa and A. Der Kiureghian, Dynamic Analysis of Multiply Tuned and Arbitrarily
Supported Secondary Systems, Report No. UCB/EERC-83/07, University of California,
Berkeley, July 1983.
9. A.K. Gupta and J.W. Jaw, Seismic Response of Nonclassically Damped Systems, Nuclear
Engineering and Design, Vol. 91. January 1986, pp. 153-1 59.
10. AS. Veletsos and C.E. Ventura, Modal Analysis of Non-Classically Damped Linear
Systems, Earthquake Engineering and Structural Dj~namics,Vol. 14, 1986, pp. 2 17-243.
I I. A.K. Gupta and J.W. Jaw, Response Spectrum Method for Nonclassically Damped
-
Systems, Nuclear Engineering and Design, Vol. 91, January 1986, pp. 161 169.
12. M.P. Singh, M. Ghatory-Ashtiany, Modal Time History Analysis of Nonclassically
Damped Structures for Seismic Motion, Earthquake Engineeringand Structural Dynamics,
Vol. 14, No. 1, January - February 1986, pp. 1 13-146.
13. D.A. Pecknold and R. Riddell, Effect of Initial Base Motion on Response Spectra, Journal
of the Engineering Mechanics Division, ASCE, Vol. 105, No. EM6. December 1979,
pp. 1057-1 060.
14. J.M. Nau and W.J. Hall, An Evaluation of Scaling Methods for Earthquake Response
Spectra, Structural Research Series, No. 499, University of Illinois at Urbana-Champaign,
May 1982.
15. A.K. Gupta and J.M. Tembulkar, Dynamic Decoupling of Multiply Connected MDOF
Secondary Systems, Nuclear Engineering and Design, Vol. 81, 1984, pp. 375-383.
Chapter 6JResponse of secondary systems
6.1 Introduction
In major industrial buildings, such as nuclear power plants, as with other
common constructions such as high rise buildings, it is impractical to perform a
coupled dynamic analysis of the primary system (building) and the secondary
system (HVAC, piping, equipment, etc.) using the conventional analytical tools.
The stiffness and inertia properties of the two systems may be quite dissimilar,
which is likely to cause numerical problems in a coupled analysis. For other
practical reasons also, it is customary to perform seismic analysis of the two
systems separately. The effect of decoupling on the primary system is presented
in Chapter 7. We shall discuss in this Chapter the problems associated with the
decoupled response of secondary systems, and the techniques by which the
accurate response of a secondary system can be calculated.
A popular method of calculating the response of secondary systems is by using
the floor response spectrum, or more accurately, the instructure response
spectrum (IRS). In the conventional IRS method, interaction between the
primary and secondary systems is ignored, which may have significant effect in
the resonant frequency range. For the multiply supported secondary systems, it is
customary to use a common IRS input, which is obtained by enveloping the IRS
at various connecting degrees of freedom (DOF). The effect of relative motions
between supports is incorporated by performing a separate static analysis. This
procedure may lead to considerable over-estimation of the seismic response of
the secondary system.
Often the earthquake input to the primary system is defined in terms of a
design response spectrum, and the ground motion time history is not known. In
the conventional procedure, a response spectrum compatible ground motion
history is created and is used for generating the IRS. The main criticism of the
procedure is that it is non-unique. The problem can be overcome in part by using
several different ground motion histories compatible with the same input
response spectrum. The time-history solutions are uneconomical to start with,
and the use of several time histories further adds to the cost. In view of all the
problems associated with the conventional method, the last two decades have
been an ongoing search for a direct method. Penzien and Chopra[l] were among
the first (1 965) to explore the topic. They were followed by Biggs and Roesset[2]
(1 97O), and Kapur and Shao[3] ( 1 973).
These early efforts were semi-empirical and heuristic, and were not found to
be generally acceptable. The time-history procedure mentioned above continued
to be the commonly used method, as it still is today. Other rational alternatives,
however, have been presented, and some are being used. Perhaps the first among
these alternatives was a stochastic method developed by Singh[4]. The modal
properties of the primary system are used to obtain the power spectral density
function at any connecting degree of freedom directly from the input response
spectrum. The method assumes the ground motion to be stationary. This
assumption leads to over-estimation of response in the lower frequency range.
Singh[S] later suggested corrective measures to rectify some of the problems.
The stochastic method is computationally efficient. Further, the method avoids
the use of time history for analysis of the primary system. In recent years,
therefore, the method has been gaining in acceptability. Singh's method, as
originally developed[4,5], has some of the same problems as the conventional
IRS method.
Peters, Schmitz and Wagner's[6] presented a method of determining the IRS
by evaluating the mode shapes of the coupled primary and secondary system. As
is usual, the secondary system was assumed to be a massless SDOF oscillator. In
this case, it is assumed that the frequencies and mode shapes of the primary
system do not change. One simply includes the appropriate modal displacement
terms related to the secondary mass. Plus, a new modal vector is added for the
extra DOF. Like most other methods, this method has problems in cases of tuned
and nearly tuned secondary systems.
Peters, Schmitz and Wagner's[6] evaluation of the mode shapes of the
coupled system based on the uncoupled mode shapes may be considered to be a
turning point in the evaluation of the coupled system response calculation. It
opened the way for consideration of more complex secondary systems. Sackman
and Kelly have made much contribution in the area of secondary system response
and have many publications, see for example Reference[7]. They considered the
secondary system with nonzero mass and introduced a rationally derived
expression for mass ratio, r,,,, between the secondary and primary system masses.
They, for the first time, rationally tackled the problem of tuned and nearly-tuned
secondary systems and established the role of the mass ratio in this context.
Ruzicka and Robinson[8] have 'studied the tuned systems in detail and
propose three different approximate methods. However, they conclude, 'know-
ledge of the Fourier amplitude spectrum is essential if the response of a tuned
secondary system is to be estimated accurately.' We have since learned that one
can find the response of a secondary system for tuned and untuned systems
without the knowledge of the Fourier spectrum. Villaverde and Newmark[9]
have developed approximate methods for evaluating secondary system response.
Sackman, Der Kiureghian and Nour-Omid [ 101 used a perturbation technique
to develop modal properties of the coupled system. As in Reference[7], the
secondary system was still a SDOF system. They also accounted for changes in
the frequencies of the primary system and for their effect on the mode shapes.
RESPONSE O F S E C O N D A R Y SYSTEMS/91
They used orthogonality conditions to improve mode shape of the modes with
close frequencies, a situation that arises when the secondary system is tuned or
nearly-tuned. Unlike reference[7], however, this solution is not extended to a
deterministic evaluation of the IRS. Instead. in a companion paper Der
Kiureghian, Sackman and Nour-Omid[l I] used the mode shapes and frequencies
to evaluate response to a stochastic input. The new stochastic method is an
improvement on the old method[4,5] in that it accounts for interaction between
equipment and structure, correlation between closely spaced modes, etc. How-
ever, other problems inherent in the stochastic method remain. Unlike
Singh [4,5], Der Kiureghian, Sackman and Nour-Omid [ 1 1] assumed the earth-
quake to be a white noise. Hernried and Sackman[l2] used the perturbation
technique to develop mode shapes of a coupled MDOF primary and MDOF
secondary system.
Gupta[l3], and Gupta and Jaw[l4] developed an approximate method for
evaluating the complex eigenvalues and eigenvectors of nonclassically damped
primary-secondary systems. The method was applied to evaluate the coupled
response of the secondary system [13,15,16]. An improved IRS method was also
developed which accounts for the interaction effects and the correlation between
responses from various support motions [13,17]. All these methods[ 13,15- 171
used the response spectrum at the base of the building as input without
converting it into a compatible time history, or to a power spectral density
function. Igusa and Der Kiureghian(l81 have proposed a perturbation method
for evaluating the complex eigenvalues and eigenvectors of nonclassically
damped primary-secondary systems. As in Reference[l 11, the emphasis is on the
response to stochastic input. Asfura and Der Kiureghian[l9] have applied a
similar technique to develop an IRS method. The present treatment is based on
References [I 31-[ l 71.
when M, C a n d K are the mass. damping and stiffness matrices, respectively, and
U is the displacement vector. It is assumed that the uncoupled primary and
secondary systems are classically damped. Let us denote the ith mode shape of
the uncoupled primary system by and the ath mode shape of the uncoupled
secondary system by $., The mode shapes are normalized such that
$riT M, $ci= 1 and '$ , M, $ , = 1. The subscript p denotes a primary system
property, and the subscript s a secondary system property. The subscript i and
other lower case letters denote the primary system modes, and the subscript a and
other Greek letters denote the secondary system modes.
In terms of the uncoupled mode shapes we can write
Substituting Equation 6.2 in Equation 6.1 and premultiplying by a T , we get
= 0,a # P, (6.4)
where o ,and 6 , are the circular frequency and the damping ratio, respectively,
for the ith uncoupled mode of the primary system; w, and l,, are the
corresponding values for the ath secondary system mode. In Equation 6.4 and
later in this Chapter, subscript c denotes the primary DOF which are connected
with the secondary system; and subscript s, as before, denotes the secondary
DOF. The matrices K: and C& are the stiffness and damping contributions of the
secondary syFtems.
To evaluate the term with K, in Equation 6.4, let us define the following
matrices[20]
The matrix U, contains one secondary system vector for each connecting DOF.
Each such vector represents the static deformation shape of the secondary system
when the corresponding connecting DOF undergoes a unit displacement. y, is a
RESPONSE O F S E C O N D A R Y SYSTEMS193
row of participation factors for the secondary system, one element for each
connecting DOF. Following the derivation for U,in Chapter 3, we can write
Kc= --CK,O.Y~.
a
Lh= -@:Yea
T
Therefore,
Ria= -of, f)dT ymT. (6.8)
For SDOF primary and secondary system y, $, degenerates into the square root
of the mass ratio J(m,/m,).It has been shown in Reference[19] that the product
$dry,Ty, $eican be viewed as the ratio of the kinetic energies of the secondary
and primary systems. An energy-mass ratio is, therefore, defined as
4a = $cT yeaTyea $ei, r,h'2 = +idTyear = ym h i - (6.9)
Equations 6.8 and 6.9 give
Ria= -o,,rIa
2 1/2
. (6.10)
To evaluate KA, let us subject the secondary system to a rigid body motion,
which gives
We can 2crive the elements of K * ' ~ in ~ ,Equation 6.18, from those of K', in
Equation 6.17, by omitting terms which are of the order of r,,, and retaining the
terms which are of the order r,!Lz.Further, Equation 6.18 also ignores the AW;,
c,,
term. If *,=' is of the same order as and c,, then the assumption[18] to omit r,,
terms is justifiable. Therefore, K * ' can~ ~ be viewed as a simplified form of K'
c,,
when r:' is of the same order as and < , and when the secondary system does
not offer significant static constraint represented by mi,,.
Dynzmic coupling between primary and secondary systems is caused because
of the t e r n r,,, the energy-mass ratio in Equations 6.17 and 6.18. In particular,
when r , = 0. the terms K: = 0, thus, dynamically uncoupling the two systems.
The t e n lo:, is independent of the energy-mass ratio, and accounts for the
static constraint only. It is like introducing a massless spring between the DOF i
and j.
We also have
Equation 6.23 forms the basis of evaluating the complex coupled eigenvalues
corresponding to primary system modes.
Equation 6.23 is solved iteratively. This last equation is a quadratic in 3,. if the
last expression is known. This last expression, however, includes the unknown h,
terms. Therefore, we assume a trial value fork,, evaluate the expression and then
solve for k,. The new value of A, is used as the next trial value and so on, until a
convergence is reached. This algorithm yields accurate eigenvalues[l4].
Now that the eigenvalue is known, the eigenvector can be improved. We
already have x,, = 1 and x,, = 0 for i # j, and the xsaterms can be calculated from
Equation 6.20. We found that the results are uniformly improved after one
iteration [ I 41. Using the previous values of the x, terms, we now calculate all x,
terms from a variation of Equation 6.21.
In Equation 6.24 we have omitted the off-diagonal terms, except the K$ which is
likely to have a relatively significant contribution. Next, we calculate the
improved x,. Equation 6.19 gives
RESPONSE OF SECONDARY SYSTEMS/97
When the accurate eigenvalues evaluated above are used, it was found that the
one iteration solution also gives very accurate eigenvectors[l4]. These eigenvec-
tors are in transformed coordinates. The eigenvectors in the original coordinates
can be obtained from Equation 6.2.
The procedure for evaluating the coupled eigenvalues and eigenvectors
corresponding to the uncoupled secondary system modes is similar. For ath
secondary mode, we take x, = 1 and assume xsB= 0 for a # P. Equations 6.16
and 6.1 7 give
The eigenvectors calculated above are also transformed into the original
coordinate using Equation 6.2.
When the secondary system constraint terms ~ ~ w , ,are , small or zero and the
square roots of energy mass ratio terms r y 2are of the order of the damping ratios,
Equations 6.23 and 6.30 can be solved to give closed-form values obtained by
Igusa and Der Kiureghian[18]. In detuned modes the eigenvalues of the coupled
system are approximately equal to the corresponding uncoupled eigenvalues,
+
h, = - C,plwp, i wpl J(l C-,);, and ha = -emom+ i wsaJ(l - c:a). In trans-
formed coordinates, the eigenvectors are given by Equation 6.20 and Equation
6.28, respectively. A primary mode i and the secondary mode a are considered
tuned when
in which e represents the allowable relative error in the response of the tuned
modes. The term PI, is called the tuning parameter. The coupled eigenvalues in
this case are
Corresponding eigenvectors can be calculated from Equations 6.20 and 6.25, and
from Equations ,6.28 and 6.31, respectively, using the values of A, and A,,
calculated from Equation 6.34. Equation 6.34 along with Equations 6.20, 6.25,
6.28, and 6.31 can also be used to calculate the eigenvalues and eigenvectors in
the untuned modes.
coordinates by vj.
According to Equation 6.2, the corresponding eigenvector in
the original coordinates, ijii, can be determined from
We shall repeat here some of the steps of Chapter 5 for the calculation of response
of nonclassically damped systems. Each complex mode shape gives two response
vectors.
U; = Y ; S & , UP = W , ~ P S & , (6.36)
where s:,and SL, are the spectral displacements for the ith modal frequency from
the displacement and the velocity spectra of the input motion for the primary
system, and
The term Xiin the above equation represents the conjugate of A,, and F, is given
by
where Ub is the static displacement vector of the coupled system when the
primary system support displaces by unity in the direction of the earthquake, and
+
a, = 2A, v i T M y i y i T C y i . (6.39)
Equation 6.38 can be written as
The motive behind the above rearrangement of the expressions is to write them in
forms which can be readily obtained. For example, the damping matrix C is not
explicitly defined, the transformed matrix is, see Equation 6.4.
Let us re-examine Equation 6.36. The vectors v! and w , y : represent the
displacement responses in mode i, when the spectral displacements from
displacement and velocity spectra both are unity. These vectors may be
considered to be normalized real vectors representing the complex modal vector
yr,. Unlike the complex vector vi, the real vectors y! and oivP are unique.
Therefore, we shall first use these vectors to evaluate the accuracy of the
approximate scheme.
The same nine coupled primary-secondary systems, which were presented in
Chapter 5, were analyzed in reference[14]. The coupled modal frequencies,
damping ratios, and the vectors \y:, o,yrP were evaluated for nine modes for each
of the nine systems using the exact method[2 11, the Gupta-Jaw method[14], and
the Igusa-Der Kiureghian (IDK) method [18]. The results for Case 2 are typical of
all the cases and are presented in Tables 6.1 and 6.2. The exact frequencies, and
damping ratios for the eight coupled modes along with the percent error
introduced by the Gupta-Jaw 'method and the IDK method[l8] are given in
Table 6.1. The exact modal vectors v!, and o , yp are shown in Chapter 5. The dif-
ferences between the modal vectors from the present algorithm and the exact vec-
tors were so small that they could not be shown graphically. In Table 6.2, only the
mean and standard deviation of the percent errors of the modal vectors are given.
The vectors.(y and o,\y; have the same unit, and as indicated earlier, they have a
90" phase difference. Often w,\y; vector is smaller in magnitude. T o avoid over-
emphasizing error in a small quantity, the percent error for the DOF was
evaluated against the vector sum of the elements of the two vectors.
Finally, the mean and the standard deviation of percent error in frequency,
damping ratio and the modal displacement in all the modes for all the nine cases
are summarized in Table 6.3. At the bottom of Table 6.3 are the cumulative sta-
tistics of all the cases. It is clear that the mean errors are generally small, showing
a lack of a consistent bias. The standard deviations of the percent error in
frequency, damping ratio and the modal displacement, on the other hand, show a
more significant spread between the two approximate methods. In all cases the
Gupta-Jaw method yields better results than the IDK method. It is so, as stated
I 'I
earlier, because the problems presented here d o not satisfy the assumption of the I ,
lightness of the secondary system, and the implied assumption of the lack of con-
straint offered by the secondary system made in Reference[lS]. It was found that
the difference between results from the Gupta-Jaw algorithm and those based on
IDK narrowed as the mass of the secondary system diminished[l4].
Case Gupta-Jaw IDK Gupta-Jaw IDK Gupta-Jaw IDK Gupta-Jaw IDK Gupta-Jaw IDK Gupta-Jaw IDK
All cases 0.01 -0.23 0.06 2.84 -0.26 0.06 0.51 12.77 0.17 0.68 1.26 10.72
Node Number n
@ Element Number e
Fig. 6.1 Example of a secondary systemjlS].
Further, in the present problem, the two nodes supporting the secondary system
have a high degree of correlation. Therefore. the two support displacements were
applied in-phase for calculating the 'static' component of the secondary system
displacements by the conventional method.
A comparison of nodal displacements and spring forces from the present
method and those from the conventional IRS method is made in Tables 6.4 and
6.5. On an average, the present method under-estimates the nodal displacement
by 3%, the conventional method over-estimates them by 58%. The standard
deviations of the percent error in displacement from the present and the
conventional methods are 6.9 and 44.1, respectively. The present and the
conventional methods have an average error in spring forces of - 11.2 and
1 16.1%, respectively. The standard deviations of the percent error in spring
forces from the two methods is 9.9 (present) and 46.9 (conventional). We note
that the error values in the present method are in the order of errors commonly I
introduced in the response spectrum method and are well within the acceptable
range. The average errors of 58% in displacement and of 116.1% in spring forces
from the conventional method are rather high. It shows a consistent bias of
significant over-estimation of the response values. A relatively large value of
standard deviations of the percent errors also indicates a considerable dispersion i
in the response values from the conventional method. In comparison, the
standard deviations from the present method are much smaller.
In modal displacement calculations from the conventional method, the effect
of the static displacement component is significant in only the first two cases. In
Table 6.4 Comparison of nodal displac&nents from the present method and the conventional floor response spectrum method[l5]
Equation 6.42 is identical to Equation 6.1, except for the right-hand side. On the
right-hand side Ub is a displacement vector obtained by statically displacing the
support by unity in the direction of the input motion, and u, is the ground dis-
placement. We will continue to use the notation of Section 6.2, in which subscript
p denotes a primary system property, subscript s a secondary system property,
and subscript c is a subset of p which is used to denote the primary system DOF
connected to the secondary system. From Equation 6.42, the equation of motion
for the secondary system can be written as
where us,is defined in Equation 6.5. Substituting Equation 6.44 in Equation 6.43
we get
It is customary to ignore the damping terms of the type on the right-hand side of
Equation 6.45; further we can write UbS= Us, Ubc.Equation 6.45 becomes
The vector { +
U,,, ii,) represents the total acceleration at the connecting DOF.
We can write in terms of the normalized mode shapes of the uncoupled
secondary system
The secondary system normal coordinate defined here, x', is different from that
defined in Equation 6.2, x,,. The latter is used to represent the displacement
vector of the secondary system relative to the primary system support displace-
ment, Us= $,xsa. Equations 6.46 and 6.47, and the use of orthogonality
a
conditions give
where A, and B,, are known constant coefficients. When the secondary system
does not offer any static constraint to the primary system the displacement vector
of the connecting DOF, Uc, will merely introduce rigid body motion in the
secondary system, and it will not cause any stresses. Therefore, in those cases, the
coefficient A, will be zero for stress related response values. On the other hand,
when the secondary system does offer static constraint, the stresses in the
secondary system generated by the vector U, can be significant, as is shown in
Section 6.1 1.
Equations 6.5 1 and 6.52 are the alternate equations for the coupled response
of the secondary system. We note here the definition of the instructure spectral
displacement.
Sea = I .vca 1max 9
-
r,, 0. Since we are going to deal with one secondary system mode (oscillator) at
a time, let us drop the subscript a, and denote the three oscillator parameters by
us, c , and r,.
In developing an equivalent SDOF system for the present analysis, no
equivalence for the ~ o i ,term
, exists; see Equations 6.13 and 6.17. To evaluate
~ o i , ,we
, need to know the properties of both the primary and secondary systems
at the time the IRS are being developed. The main advantage of an IRS method is
that the spectra for the secondary system input can be generated without any
knowledge of the secondary system. To retain this advantage, therefore, let us set
do;, = 0. Clearly, if the static constraint offered by the secondary system
significantly affects the coupled response of the primary system, the IRS method
(present or conventional) cannot be used. On the other hand, we have discussed
earlier the effect of static constraint on the coupled response of the secondary
system, which can be quite important and the present method is capable of
accounting for it. It is shown later that the static constraint can introduce
significant stresses in the secondary system even when it does not significantly
affect the coupled primary system response.
Let us now proceed to define the secondary system oscillator, whose
frequency and damping ratios are o , and c,, respectively. Let us assume that the
oscillator has a mass m,. For the normalized 'mode shape,' $, = 11Jm,,
u, = 1 , y, = Jm,. Equation 6.9 gives
Equation 6.54 presents a problem. We intend to develop the IRS for specified
mass ratios, just as the conventional IRS are developed for specified damping
ratios. For a given oscillator mass, Equation 6.54 would give a different mass
ratio with respect to each primary system mode. The solution of this dilemma is
as follows. The purpose of defining the r,or m,is to take into account interaction
between the primary and secondary systems. Interaction will be most significant
with the primary system mode. I, whose frequency o,, is closest to the oscillator
frequency o,.Therefore, let us assume that the specified mass ratio is between
oscillator and the primary system mode I. This assumption in conjunction with
Equation 6.54 give the oscillator mass
Values of r, for other primary system modes can be calculated from Equation
6.54. In the conventional IRS method, r, = 0, for each IRS curve csis held
constant and w, is varied. In the present case, r, # 0. For each IRS curve, we shall
hold r, and 5, constant, and then vary w,. For each new set of r, andc,, a new IRS
will be obtained. It is obvious that in the course of obtaining a set of several IRS
curves, the computer program will have to solve a large number of these coupled
problems. There is, therefore, a need to develop an efficient computational
algorithm.
Let us denote the coupled eigenvalue by 1, where 1 = -c +
w ioD,
w , = w J(l - c2);
in which w and ( are the appropriate coupled frequency and
damping ratio, respectively. Equations for the eigenvalue and eigenvector here
have been adopted from Section 6.3. Equations derived in Section 6.3 show that
there will be some change in all the eigenvalues of the primary system due to cou-
pling. However, the changes are usually small, except in the eigenvalue of the
mode 'I' whose frequency is very close to the secondary system frequency.
Therefore, we can write
We ignore the interaction between the secondary system and all the primary
system modes other than the Ith mode when evaluating the eigenvalues hi and A,.
Modifying the appropriate Equations 6.23 and 6.30 we get
and
c1
r,"' (of+ 2 0 , hi)
xs - wf + 2 0 , c s h i + h;
There are two more eigenvectors corresponding to the Ith primary system mode
and the SDOF oscillator, one for each pair of A, obtained from Equation 6.57.
x, =
rj"' (of+ 2wS XI) cl
o;, + r, o: 4- 2(w,<, + rjos<,)XI + Ah:
for all values of j.
L I
I
j :I/I
SDOF oscillator us. In the coupled mode corresponding to the ith uncoupled i
primary mode, the complex modal values of u, and us(w, and yrs,respectively) can 1
:I
I
The summation in Equation 6.61 is on all the primary system modes. In the
response spectrum method for nonclassically damped systems, Chapter 5, each
complex mode shape and its conjugate give two response vectors
and
where s:, and S:, are the spectral displacement and velocity for the ith coupled
modal frequency from the displacement and the velocity spectra of the input
motion for the primary system; S& and S;, are the corresponding values for the
x,
coupled modal frequency of the oscillator; and Xsare the complex conjugate of
1 , and I , , respectively. The terms Fl and F, are defined below.
II ' s:, S:,, s f s ,S:$.T o simplify the notations, we shall use the subscript i to include
all the coupled modes, henceforth. Thus, all the instructure spectral displace-
ments can be denoted by Sf, and S:,.
It can be seen from Equations 6.58-6.63 that when m, = 0, expressions of the
form 0/0 are obtained. This problem can be easily remedied by either carrying on
many of the algebraic substitutions, and in the process, eliminating all the
problematic m, (or, r,, rJ) terms; or by normalizing the x, and xj values in
Equations 6.58 and 6.59 by dividing by rfI2and r;l2, respectively, and by making
corresponding changes in Equations 6.60-6.63.
The modal combination equation for a nonclassically damped system is given
by
The term S;,, (a = d or v) represents the maximum value of F,, for an oscillator
with frequency o,,and damping ratio (,, in the ith coupled mode. Previously it
was denoted by S:',and S:, without the subscript a, because at that time only one
oscillator was under consideration. Equations 6.66 and 6.67 give
RESPONSE O F SECONDARY SYSTEMS/113
Let us define
(UC)~ = 11 1
r j a b
&$ ':J9
t w2 - i
111& I
XI. s:Ja.
j a b
and
Equation 6.70 constitutes the modal combination equation for the IRS
method. The desired spectral values along with various correlation coefficients
are defined in Equation 6.69.
In an IRS approach the properties of the secondary system (o,, C,, r,) are not
known in advance. In the conventional method, r, is assumed to be zero, and IRS
curves are drawn for several values of C,, by varying o,. In the proposed method r,
# 0, the spectral curves are drawn for several sets of (r,, C,,) values by varying o,.
Another variable is the connecting DOF. Separate IRS are evaluated for each
connecting DOF. Basically, for each point of an IRS curve, a coupled analysis of
the type described earlier is to be performed. The data from the coupled analysis
are saved to go back and evaluate the correlation coefficients, see Equation 6.69.
Each of the multitudes of the coupled analyses described above would give a
set of u t values. Which set should we use in Equation 6.69, and elsewhere? We
feel that the u: values from an uncoupled primary system analysis are a
reasonable approximation. Note that the primary system, itself, is classically
damped; therefore, the uncoupled analysis would give u:, = 0.
Fig. 8.2 Instructure response specirum (IRS).(a) fint floor, @) third floor, (c) rdp floor, El
Centro earthquake (SOOE,1940); primary system damping 7%; secondary system damping
2%. mass ratio 0[17].
Damping Values
Frequency, Hz
Fig. 6.3 Instructure response spectra (IRS). top floor, El Centro earthquake (SOOE, 1940);
mass ratio 0[17].
The effect of damping ratios on IRS is illustrated in Figure 6.3. Both IRS in
Figure 6.3 are at the top floor of the building for the El Centro earthquake. In one
case the primary and secondary system damping ratios are 7% and 2%,
respectively; and in the other case 1% and 0.5%, respectively. As would be
expected the spectral peaks are much higher for the case with lower damping
values than those for the case with higher values.
The effect of mass ratio on IRS is shown in Figure 6.4. Three IRS with mass
ratios of 0,O.O 1 and O.'1 are compared. All of the three spectra are at the top of the
building and for the building damping ratio of 1% and the oscillator damping
ratio of 0.5%. The higher mass ratios result in lower spectral peaks, as one would
expect.
Mass Ratio
-0
Frequency. Hz
Fig. 6.4 Instructure response spectra (IRS), top floor, El Centro earthquake (SOOE, 1940);
primary system damping 1%; secondary system damping 0.5% [17].
Several sets of correlation coefficients given by Equations 6.73 and 6.74 were
compared with those obtained numerically from Equation 69 in Reference [17]
and a reasonably good agreement was shown. A few of these are shown here in
Figure 6.5.
6.1 1 Response examples
The same primary-secondary systems used in Chapter 5, and in Sections 6.4 and
6.5, were used in Reference [17] to illustrate the response values from the present
IRS method. See Figure 6.1 for the node and the element numbers. Results from
the analysis of all nine systems for the El Centro (SOOE,1940) ground motion
using the time-history method, conventional floor response spectrum method
and the coupled response spectrum method are given in Section 6.5, Tables 6.4
and 6.5. We will compare those results here with the results from the present IRS
method. Tables 6.6 and 6.7 give a comparison of nodal displacements and
element spring forces, respectively, from the present IRS method and the
conventional floor response spectrum method with those from the time-history
method. Percent errors in the results from the two methods are also shown in the
same tables, using the time-history results as the reference or standard values. It is
clear that the results from the present IRS method are much closer to the time-
history results than those from the conventional floor response spectrum method.
We note here that the proposed IRS method is in fact an approximation of the
coupled response spectrum method. The response spectrum method is a well-
established method of seismic analysis. It can be looked upon as a tool to evaluate
an average seismic response for design purposes. Normally, a good correspon-
I
R E S P O N S E O F S E C O N D A R Y SYSTEMS/119
II
Proposed o Proposed o
Numerical - Nwnerkal -
C1-3.C2-6. We-1 Hz C1-J.C2-6,y - 1 0 Hz
1 10 100
Frequency. HZ Frequency. Hz
--g
-t
111
Proposed o
-
,
0'
0 Numerical
-1 I
1 10
I I
100
-1 1 I
1 10
I I
100
Frequency. Hz Freauency. Hz
Fig. 6.5 Comparison of correlation coefficients from the numerical procedure and the
approximate equation, El Centro earthquake (SOOE, 1940); primary system damping 7%;
secondary system damping 2%; mass ratio 0[17].
dence between the time-history results and the response spectrum results exists. A
total agreement between the two results is neither intended nor required. We
believe the orders of mean percent errors and the standard deviations of the
percent errors for the proposed IRS method in Tables 6.6 and 6.7 are well within
the acceptable margins.
A better measure of the accuracy of the present IRS method is given in Table
6.8 in which the results from the present IRS method are compared with those
from the coupled response spectrum method. The errors in response values,
especially in the spring forces obtained from the proposed method are very low.
There may be an impression[l9] that when the correlations between
responses from various support motions are accounted for, we need not consider
Table 6.6 Comparison of nodal displacements from the present method and the
conventional floor response spectrum method[l7]
Mean
Standard deviation
References
1. J. Penzien and A.K. Chopra, Earthquake Response of Appendages on a Multistory
Building, Proceedings. 3rd World Conkrence on Earthquake Engineering, Vol. XI, New
Zealand, 1965.
2. J.M. Biggs and J.M. Roesset, Seismic Analysis of Equipment Mounted on a Massive
Structure, in Seismic Design for Nuclear Power Plants, Ed. by R.J. Hansen, MIT Press,
Cambridge, Massachusetts, 1970.
RESPONSE OF SECONDARY S Y S T E M S I ~ ~ ~
Table 6.7 Comparison of spring forces from the present IRS method and the conventional
floor response spectrum method[l7]
Mean
Standard deviation
Table 6.9 Response components from the present IRS method[l7]
7.1 Introduction
We have pointed out in Chapter 6 that for various practical reasons it is
customary to perform the analyses of the primary system (building) and the
secondary system (HVAC, piping, equipment, etc.) separately, or to assume that
the two systems are uncoupled. It is shown there that the uncoupled analysis of
the secondary system introduces considerable error in the seismic response on the
conservative side. A method of performing the analysis of the coupled system is
presented in Chapter 6, which is based upon using an approximate technique to
evaluate the coupled mode shapes and frequencies. In fact the same method can
be employed to give the coupled response of both the primary and secondary
systems. There is another practical problem, however. in chronological order, the
design of primary system precedes the design of the secondary system. At the
time the primary system is being designed, only tentative information, if any,
about the secondary system is available. Therefore, the uncoupled analysis of
primary system is a fact which cannot easily be altered. There is a need to have
approximate criteria which can be used to evaluate the effect of decoupling on the
primary system response.
A decoupled analysis of the primary system can be rationally justified if the
decoupling results in a relatively insignificant error in the response calculation. A
necessary condition for small error in response would be a small change in the
frequencies of the primary system. In practice, this has also been taken to be the
sufficient condition.
Lin and Liu[l], United States Nuclear Regulatory Commission[2] and RDT
Standard[3] came out with decoupling criteria about the same time (1974-75).
They considered single-degree-of-freedom (SDOF) primary and secondary sys-
tems. Curves were presented between the secondary-primary system frequency
ratio, rf, and mass ratio, r,, to designate regions in which decoupling could or
could not be permitted. The three sets of curves are similar in shape. But in
certain ranges of rf, r, values, they may lead to widely different results. These
curves are characterized by abrupt discontinuities, lack sufficient justification,
and to some they portray the dynamics as very arbitrary. In developing their rela-
tionships between r, and r, Hadjian and Ellison[4] actually formulated the
frequency of the coupled SDOF-SDOF systems, thus leading to a much more
rational approach. When the primary system, or both the primary and secondary
systems have multi degrees of freedom (MDOF), approximate heuristic methods
have been used to evaluate the equivalent mass ratios, which can then be used in
conjunction with the criterion developed for SDOF-SDOF systems. Gupta and
Tembulkar[5,6] showed that it is not sufficient to limit the changes in frequency.
They developed approximate relationships to predict the change in response for
the SDOF or MDOF primary system connected to a SDOF or MDOF secondary
system. Approximate procedures similar to that used in Chapter 6 were
employed. The following treatment is primarily based on Gupta and Tembul-
kar[5,6].
Equation 7.2 will give two positive values of R F , and hence R, the frequency of
the coupled system. If decoupling has to be achieved, one of those frequencies
+
should be close to o,, the corresponding R Fclose to 1, say R: = 1 E, where E is
small. The other mode is not likely to contribute significantly to the response.
Substituting ( I+ E) for R: in Equation 7.2, we get
Mass Ratio r,
Fig. 7.2 Frequency and mass ratio (r, versus r,) curves for 5%, 10%and 15%changes in
the natural frequency [ 5 ] .
0.9, R: = 1.21, 0.81, E =0.21, -0.19. Note, a positive E means that the
frequency of the coupled system is greater than that of the uncoupled system.
The rf versus4r,,,curves for 5%, 10% and 15% changes in the value of o,are
shown in Figure 7.2. Roughly, E is positive for values of rf < I, and negative for
rf > 1. The region on the left side of any curve will assume an error less than that
used for the particular curve, As &*d expe&, t o limit the error, r, should be
small and rf away from unity. The
obtained by Hadjian and Ellison[
The uncoupled SDOF p r i m a j system
equal to the spectral displacement
coupled system is ,I. 1
1
}:{ {" - t 2 m s }
c I
= = rnSwi 1 !v
1 :
'!
Using the standard MDOF formulation, the relative displacement of the primary
mass in the coupled system can be shown to be
If we assume that the spectral acceleration SD does not change significantly
between the frequency of the uncoupled primary system and the coupled system,
the response ratio RR becomes
The 2DOF coupled system considered here will have two modes, and hence
two values of RFfor any set of r,, r, values. It can be shown that for both these
modes, the response ratio, R R , is always less than unity, i-e., the response of the
primary mass in the coupled analysis in any one of the two modes is always less
than that in the uncoupled analysis. Figure 7.3 shows r, versus r, curves for lo%,
20% and 30% reduction in the primary system response (RR= 0.9,0.8,0.7) in the
mode for which RR is closest to unity. For each reduction value there are two
curves. The 'acceptable' domain (which would limit the reduction) is below the
bottom curve and above the top curve for any given reduction.
Combined R,
in which R, denotes the frequency of the coupled system corresponding to the ith
uncoupled mode, o,, is the frequency of the ith uncoupled primary system mode,
w,, is the frequency of ath uncoupled secondary system mode, and r,,, is the
energy-mass ratio between the ath secondary system mode and the ith primary
system mode defined in Chapter 6. Equation 6.23 also has a ~ o i ,term , which
accounts for the static constraint offered by the secondary system. It is assumed
here that such constraint is negligible. Denoting RF; = R,/wp,, r/,, = oSa/op,,
Equation 7.7 can be rewritten as[6]:
Equation 7.8 can be solved iteratively for the desired RFivalue. The right-hand
side of Equation 7.8 is likely to be dominated by a few terms for which (r;, - R:~)
values are relatively small. To avoid iterations, one may use the equation with
only one term for which (ria - R:,) is the smallest, leading to what is called the
DECOUPLED PRIMARY SYSTEM ANALYSIS/131
I I I *
1 .o 2.0 30
Frequency Ratio, rc
Fig. 7.4 Variation in response ratios and their combinations with the frequency ratio[5].
single mode approximation or the SMA. The summation sign in Equation 7.8 is
dropped. The modified or the SMA equation is
R:, - ( 1 + r;. + rElar/,J R:, + r;,. = 0. (7.9)
The advantage of using Equation 7.9 is that it allows the use of r~ verses r,
curves, see Figure 7.2, for the SDOF-SDOF systems.
If the ith normalized modal vector of the uncoupled primary system is &,the
participation factor is y, and the spectral displacement is S,,, the response
displacement vector is given by y, $,SD,. It is assumed that in the coupled
system, the primary system mode shape remains approximately the same as that
in the uncoupled system, except that it is no longer normalized. Say $ j represents
the primary part of the normalized coupled system mode shape. The $ j = A $,.
where A is a scalar. Let the corresponding participation factor be y:, and assume
that S,, remains the same because the change in frequency is small. The primary
system displacement response in the coupled system becomes y; $hSDi.The
response ratio is given by
where U,,and Ub6are the displacement vectors of the primary and secondary sys-
tem, respectively, when the primary system support displaces by unity in the
direction of the earthquake; $,i and @iiare the secondary system displacement
vectors corresponding to the primary system vectors $., and +;;. To evaluate the
scalar A we note,
in which $,, is the ath uncoupled secondary system mode, $,i is part of the ith
uncoupled primary system mode containing the connecting degrees of freedom
only, and y, is a row of participation factors for each connecting degree of
freedom. Equations 7.10-7.13 give
D E C O U P L E D P R I M A R Y S Y S T E M ANALYSIS1133
The numerator and the denomintor of Equation 7.15 represent inertia forces for
the secondary and primary system, respectively. Further, for a SDOF-SDOF
system, Equation 7.15 becomes the square root of the familiar mass ratio,
J(rn,/rn,); hence the terminology.
Equation 7.14 gives the desired response ratio. As for the frequency ratio, we
can use the single mode approximation (SMA) for the response ratio also. Unlike
the frequency ratio, however, the SMA does not allow the use of the SDOF-
SDOF rfversus r,,, curves for the response change, Figure 7.3, due to the presence
of the inertia mass ratio, r,i,, term in Equation 7.14.
-
higher modes, the changes in frequencies and responses were very small, RFand
RR 1. Therefore, in Tables 7.4 and 7.5 the RFand RR values for the first three
modes only are reported. The present algorithms are able to predict the changes
accurately to within 1%.
Next, the effect of coupling on the fundamental mode of case 3 for all the three
coupled models in Figure 7.4 is investigated. The RF and the RR values are
Secondary mode , RF
Present Exact
Case r, rr Iterative SMA
D E C O U P L E D P R I M A R Y S Y S T E M ANALYSIS/135
Table 7.4 RF values for the coupled Model 1, Case 3 [6]
Secondary mode RF
I 2 Present Exact
Table 7.5 Response ratios for the coupled Model I , Case 3[6]
Response ratio R,
Present Exact
Primary Secondary
mode mode rf r~ ri Iterative SMA Story l Story 2 Story 3 Story 4 Story 5 Story 6
Oriln
222
o r i m
9 9 a!
0 0 0
o r i m
S S a!
0 0 0
o'a-
d Q I W
d w m
0 0 -
"88
86,:
0 d
8cr;
-
-99
do0
-(UM
reported in Tables 7.6 and 7.7. Two effects are clear. First, there is a clear
difference in the R,and R,values for the three models. Second, the present algor-
ithm can predict this difference. The agreement between the RFvalues from the
proposed algorithm and the exact values is excellent. For response ratio, the
agreement is reasonable, quite good for the top story, not so good for the bottom
story.
References
I. C.W.Lin and T.H. Liu, A Discussion of Coupling and Resonant Effects for Integrated Sys-
tems Consisting of Subsystems, Proceedings, Extreme Load Conditions and Limit Analysis
Procedures for Srrucfural Reactor Safiguards and Containmenf Structures, Paper U3-3,
Berlin, September 1975.
2. United States Nuclear Regulatory Commission, Standard Review Plan, Section 3.7.2,
Seismic Sysfern Analysis, June 1975.
3. RDT Standard F9-2T. Seismic Requirements for Design of Nuclear Power Plants and Test
Facilities, January 1974.
4. A.H. Hadjian and B. Ellison, Decoupling of Secondary Systems for Seismic Analysis,
ASME-PVP Conference. Reprint No. 84-PVP-59, San Antonio, Texas, June 1984.
5. A.K. Gupta and J.M. Tembulkar, Dynamic Decoupling of Secondary Systems, Nuclear
Engineering and Design, Vol. 81. September 1984. pp. 359-373.
6. A.K. Gupta and J.M. Tembulkar. Dynamic Decoupling of Multiply Connected MDOF
Secondary Systems, Nuclear Engineermg and Design, Vol. 81, September 1984,
pp. 375-383.
Chapter $/Seismic response of buildings
8.1 Introduction
The majority of buildings are analyzed and designed in accordance with the
building codes. An introduction to the historical development of the US and
international building codes is given by Berg[l]. Typically, the earthquake
loading is defined in terms of an equivalent lateral force, and a static analysis of
the building is performed. In recent years, building codes have adopted more and
more features of the formal dynamic structural analysis, while retaining their
original formats. Perhaps the most popular and relatively rigorous building code
presently in use in the profession is the Uniform Building Code (UBC)[2]. The
seismic design and analysis requirements of the UBC are based on the procedures
developed by the Structural Engineers Association of California (SEAOC)[3].
The developments of the UBC and the SEAOC requirements have been gradual
and evolutionary in nature. In a somewhat different approach, the Applied
Technology Council, a research and development organ of the SEAOC, under-
took to produce a model seismic code development in 1974 under the
sponsorship of the National Science Foundation. The effort was coordinated by
the National Bureau of Standards. Nearly a hundred scientists and engineers
contributed to the project report, 'Tentative Provisions for the Development of
Seismic Regulations for Buildings', popularly known as ATC3-06, or simply as
ATC3[4]. The report is being reviewed by engineers and building officials to
evaluate its utility, cost and effectiveness in leading to earthquake resistant
structures.
We will show in this chapter that the modern building codes, UBC, ATC3,
and others, use seismic response calculation procedures which are closely related
to the response spectrum method.
8.2 Analysis
Consider the ith mode of vibration, with circular frequency a,and modal vector
$,. Let us assume that the vector is normalized, $ i T $ ~; = 1, where M is the
mass matrix. Maximum displa&ment of the building in this mode is given in
Chapter 1.
where y, = (pTM U,,, the modal participation factor, and S,, is the spectral
acceleration corresponding to the frequency o,.As defined before, U,is the
displacement vector of the building, when the base of the building is displaced
-
statically by unity in the direction of the earthquake. For buildings with only a
horizontal degree-of-freedom in the direction of the earthquake, U,, 1, where 1
is a vector with every element equal to unity. The pseudo-static force which will
cause the displacement given by Equation 8.1 is given by
Equation 8.3 gives the pseudo-static force, which, when applied to the
building, will yield the displacement vector given by Equation 8.1. The elements
of mass matrix M have mass unit, $i yiSA,terms have acceleration units. As such,
we can also view the Fi force vector as a pseudo-inertia force vector. It is common
in the modeling of simple buildings to lump the masses associated with a story at
the corresponding floor level. The mass maxtrix M is defined as a diagonal
matrix; the rth diagonal being rn: the mass of the rth story. The pseudo-inertia
force of the rth story becomes
F : = mr@:~,SA,, w4)
in which $: is the rth element of the vector $;. The corresponding base shear is
If V, is known, based on Equations 8.4 and 8.5, the story force can be'etatyated
, . =
from
To define the base shear, Ifi, from Equation 8.5, we need to evaluate the
participation factor y,, which can be written as
SEISMIC RESPONSE OF BUILDINGS/ 141
When the vector +iis not normalized, +: M gi# 1, we can easily show that
Equation 8.8 becomes
Equation 8.6 does not change, whether the vector is normalized or not. Equations
8.6 and 8.9 can also be written in terms of story weights w r = mrg,
which is a modified form of Equation 8.1 1. It can be shown that the sum of all the
modal weights, W,, is equal to the total weight of the building W.
1
Equations 8.10 and 8.12 constitute the basis of evaluating the modal base
shear and the modal story forces in ATC3 in its modal analysis procedure. The
responses from several modes are combined by the square root of the sum of the
square (SRSS) method (see Chapter 3). The expression SAl/gis termed the modal
seismic design coefficient. We shall discuss this coefficient in greater detail later.
We shall now present the 'equivalent lateral force procedure'. In the design of
conventional buildings, much of the seismic force comes from the fundamental
mode. For buildings with relatively small periods (TG0.5 sec,/ 3 2 Hz), the fun-
Concentrated Forc
'.. '-.\ ,\
*
--
\ C
Parabolic
Distribution
)$.. v
\
';\.,
L
\-
Linear
Distribution 1 ',
Note in Equation 8.13, U7,is assumed to be equal to the total weight .'U
In the UBC, the effect of a higher period of the fundamental mode shape and
the effect of higher modes is accounted for by a concentrated force F, applied at
the top story.
w rhr SA
Fr = (V-F,), V=W-.
9
The overall effect of the ATC3 and UBC equations should be comparable. 1
The story shear is calculated by summing all the story forces above a t~
particular story. The structural elements within a particular story receive shear ,I
due to another source, the torsional moment which is generated due to
F
eccentricity between the centers of rigidity and mass. ATC3 recommends using
an accidental eccentricity of 5% of the building plan dimension perpendicular to
the direction, which is in addition to the known eccentricity, which may exist due
to designed architectural and structural features. ATC3 requires analysis for two
orthogonal horizontal components of earthquake separately. The effects of the
two components are combined before design. In the UBC procedure, on the other
hand, the design for the two horizontal components is done independently-the
effects of the two components are never combined. In the UBC procedure, the
accidental eccentricity is assumed equal to 5% of the higher of the plan
dimensions, irrespective of the direction of the earthquake under consideration.
Further, the known designed and the accidental eccentricities are not combined
in the UBC. The torsional moment due to the greater of the two is considered.
In the present UBC, the overturning moment is evaluated statically from the
calculated story forces. In the earlier versions, a reduction for out-of-phase
multiple modes was permitted. This reduction was eliminated from UBC after
SEAOC dropped it in 1970, at least in part because of bad results observed in the
Caracas, Venezuela earthquake of 1967. In that earthquake, several tall buildings
designed in compliance with seismic building codes similar to the UBC had
exterior column damage attributed to overturning moment. The exterior column
damage can be attributed to another effect not included in the UBC, as we
mentioned earlier, viz., the combined effect of the two horizontal components of
earthquake. This particular effect is accounted for by ATC3. Perhaps because of
this mitigating effect, ATC3 does allow some reduction in overturning moments
calculated statically from the story forces, to account for out-of-phase multiple
modes. ATC3 does not allow any reduction for the top ten stories. Between the
top ten and twenty stories a reduction factor which varies linearly between 1.0
and 0.8 is used. At the building foundations (inverted pendulum structures
excluded), an overturning moment reduction factor of 0.75 is recommended by
ATC3, in recognition of the observation that actual foundation overturnings or
the lack of them in past earthquakes indicate that the statically calculated
moments represent a significant over-estimation. T o avoid overturning, ATC3
requires that the resultant of the seismic forces and vertical loads at the I
foundation-soil interface remains within the middle one-half of the base of the
building. I
8.3 Building frequency
In case the building configuration and the member sizes have already been
worked out, the mode shapes and frequencies (periods) can be evaluated easily
using the available analysis programs. In fact, the complete building response can
be calculated using the response spectrum method, or using the equivalent ATC3
method. At the preliminary design stage, or in case of simple buildings when the
designer chooses to perform manual calculations, an accurate estimate of the
building frequency can be obtained using Rayleigh's method.
Let $ represent the assumed fundamental mode shape. The maximum strain
energy associated with the mode shape is $TK$, and the maximum kinetic
energy is 02$T M $, when o is the circular frequency of the fundamental mode
under consideration. Equating the maximum potential and kinetic energies,
we get
Let F denote the force vector which would give the $ displacement vector,
F = K $, Equation 8.16 becomes
The EPA and the EPV parameters are not directly related to the corresponding
maximum ground acceleration (a) and velocity (u) values. The two sets of values
are of the same order of magnitude. In many cases, a rough equivalence between
EPA and a, and between EPV and v can be seen, see Chapter 2. ATC3 further
defines two nondimensionalized coefficients A, and A, based on EPA and EPV.
The acceleration coefficient A, is numerically equal to the value of EPA in g
units:
The reasons for designing long-period buildings more conservatively include the
following: (1) tall buildings have a higher number of the degrees of freedom,
which increases the likelihood that the ductility requirements are concentrated in
a few stories of the building; ( 2 ) the number of potential modes of failure
increases with the number of degrees of freedom; (3) instability of a building is
more of a problem for a taller building.
In the buildings with very long period ( T > 4 sec), we know that the spectral
displacement becomes constant, see region DE in Figure 8.2. For this region, the
spectral acceleration is inversely proportional to T 2 .In view of the conservatism
intended for tall buildings, T 2 is replaced by T ~ ' ) .
3Avg
s* = -T4/3 ,for T 2 4 sec.
At T = 4 sec, Equations 8.22 and 8.23 give the same values of S,.
ATC3 has defined three types of soil profiles. The following description is
based on Berg's[l] abbreviated version:
SEISMIC RESPONSE O F BUILDINGS1147
particularly when the ground shaking is strong. ATC3 has accommodated this
observation by introducing a factor S, and other changes in Equations 8.23 and
8.24.
In the BC region of Figure 8.2, Equation 8.25 uses a reduction factor of 0.8 for
S3soil when A, is large (30.3). The soil coefficient S modifies the spectrum in the
longer period region. The value of S for the three soil conditions is given as
S = 1.0 for S,, S = 1.2 for St, and S = 1.5 for Sj.
Equation 8.25 gives an elastic design spectrum. As we discussed in Chapter 2,
a majority of conventional buildings are designed to undergo a significant
inelastic deformation when subjected to a strong motion earthquake. In effect,
the building absorbs the earthquake energy inelastically by a relatively larger
deformation and smaller strength than it would if it had to absorb the energy
elastically. The consequence of this design philosophy is a reduced strength
requirement, but a much greater emphasis on details which would make the
inelastic deformation possible without a building collapse. The ratio of the
maximum inelastic deformation in the structure to the yield deformation was
defined as the ductility ratio, p, in Chapter 2. A method of obtaining an inelastic
response spectrum from the corresponding elastic response spectrum was also
presented. In Figure 8.2, the spectral ordinates in the CDE region are divided by
the ductility ratio, p, and in the BC region the ordinates are divided by J(2p - 1).
ATC3 replaces both p and J(2p - 1) factors by a single response modification
factor, R. For calculating the inelastic deformations, another deflection amplifi-
cation factor C,,is used, which is similar to R, but not the same. Table 8.1 gives a
summary of the R-factors based on Berg[l]. Equation 8.25 becomes
1.2A,Sg 2.5Aag
For T a 4 sec,SA = 6
R T =I3 R '
2.0 A, g
4 ,for S, soil when A, 2 0.3;
R
Table 8.1 Response modification factor, R. (Based on Berg[l])
Bearing wall system (type 1) From 1- 114 for unreinforced masonry walls to
6-112 for light framed walls with shear panels
Building frame system (type 2) From 1- 112 for unreinforced masonry shear
walls to 7 for light framed walls with shear
panels
Moment-resisting frame system (type 3) From 2 for ordinary moment frames of
reinforced concrete to 8 for special moment
frames of steel
Dual system (type 4) From 6 for braced frames to 8 for reinforced
concrete shear walls
Inverted pendulum structure From 1-114 for ordinary moment frames of
steel to 2-1/2 for special moment frames of
either steel or reinforced concrete
3A,Sg
For T 2 4 sec, S, = ---
R ~ 4 / 3'
ATC3 gives almost identical spectral equations for the equival'ent lateral load
procedure, which is a single mode approximation, and for the modal analysis pro-
cedure.There are two exceptions. For the equivalent lateral load procedure, the
extra equation for the longer period buildings ( T a 4 sec) is not allowed to assure
sufficient conservatism in the approximate method. A reduced spectral value for
tall buildings may also be viewed as an incentive to perform a modal analysis.
The second exception is a reduction in spectral value in the short period range,
AB in Figure 8.2, in the modal analysis procedure in modes other than the
fundamental mode. The ATC3 equation for this range may be written as
A0 9
SA= -(0.8
R
+.4.0 0,for S, soil when T < 0.3 sec.
When T is very small, Equation 8.27 gives SA-- 0.8 A, glR. The numerator of 0.8
A, g is roughly the maximum ground acceleration. The spectral value is further
reduced by the response modification factor, R. As we pointed out in Chapter 2,
the ductility does not reduce the required resistance in the high frequency (short
period) range. Equation 8.27 is inconsistent with that observation. On the other
hand, in most buildings, the response in modes other than the fundamental mode
may not contribute significantly. Therefore, Equation 8.27 is unlikely to affect
the design response values much.
*
111
?j:i
jj j
According to the UBC, the spectral acceleration can be written as I,:
The UBC has five zones. the highest being seismic Zone 4 and the lowest Zone 0.
The latter has no seismic design requirement. For Zones 4- 1, the values of Z are
1, 3/4, 3/8 and 3/16, respectively. There are three categories based on the
importance of the building. For essential facilities the importance factor, I, is 1.5,
for large occupancy buildings it is 1.25, and for all other buildings, 1 .O. The build-
ing coefficient, K, depends on the type of structure, ranging in value from 213 to
4 / 3 for buildings and on up to 2.5 for elevated tanks. The K-factor, in effect,
lowers the margin of reserve strength required for structural systems that have
performed well in past earthquakes and raises the margin for systems that have
performed poorly. The site-structure resonance factor is a function of the ratio of
the building period to the site period and varies between I and 1.5. If the site per-
iod is not available, a factor of 1.5 must be used. However, the value of the pro-
duct CSneed not exceed 0.14. The UBC is intended to be used with the working
stress design method. The code allows a 113 increase in the allowable stresses.
Conversely, the seismic loads can be multiplied by a factor of 314.
As we pointed out earlier, the Uniform Building Code is an evolutionary code,
whereas, ATC3 was created from scratch based on current knowledge. Of course,
the current knowledge which went into A T C 3 includes the experience of the
profession with the UBC and other codes. It will be interesting to compare the
design values from the two codes. Let us consider the highest seismic zone
(A, = A, = 0.4, Z = I ) , site type S, without soil-structure resonance ( S = I for
ATC3 and UBC), and a moment resisting steel frame ( R = 8, K = 2/3). The
ATC3 equations give
The ATC3 response value is intended for ultimate load design. For a steel frame,
the capacity reduction factor is 0.9. T o obtain the nominal ultimate value we
should divide the above I.'/Wvalue by 0.9. T o compare the ATC3 value with the
UBC value we need to convert the ATC3 value into an equivalent working stress
value, which can be done by dividing the nominal ultimate value by a factor of
safety of 1.7. Completing both the divisions, we get
'
I
The UBC equations along with the 3/4 factor give
The V/ W values from ATC3 and UBC are compared in Figure 8.3, We observe
that the ATC3 and UBC curves in Figure 8.3 are comparable.
Let us consider another example similar to the first one, except that we now
have the site type S, in ATC3, and have the maximum soil-structure resonance
for the UBC. Both give S = 1.5. Following steps similar to those above, we get
The two V/W values are plotted in Figure 8.4. Again, the two codes have
comparable values.
Period, eec
8 - UBC
0 ATC3
$ 8 -
* 5 -
-E
Q
Q 4 -
F
@
g 3 -
U.
Period. sec
-
Fig. 8.4 Comparison of ATC3 and UBC seismic forces, A,
S,, S 1.5.
= A, = 0.4, Z = I , site type
From the maximum seismic intensity zones to the lowest, the ATC3 A, and A,
values go down from 0.4-0.05, by a factor of 8; the UBC Z values go down from
1-3/ 16, by a factor of 5.3 only. At a given location, the actual design values will
depend upon the respective maps. It is believed, however, that for low seismic
areas, the UBC requirement is relatively conservative. One reason may be that
the framers of the UBC seismic requirements, or those of its parent-the
SEAOC, paid much greater attention to the high seismic western United States.
The ATC3 committees, on the other hand, appear to have spent a great deal of ef-
fort in sorting out the seismicity of the eastern United States and have accounted
for such factors as the relative infrequency of earthquakes in many regions. We
may also add that ATC3 has considered the lower attenuation of seismic waves in
the eastern United States. Therefore, certain regions, which may have been
excluded from the seismic requirements in the UBC,are now included in the
ATC3 maps. A detailed discussion on this topic is beyond the scope of this book.
Interested readers should examine the respective seismic maps and other
supporting documents.
The range of the R factor, which accounts for ductility in the ATC3 procedure,
is 1.25-8. This means that an inverted pendulum having the same period as a
moment resisting steel frame will be designed for a seismic force which is 6.4
times that for the latter according to ATC3. The corresponding term in the UBC
is the K-factor, which varies from 2.5-0.67, giving the inverted pendulum force
3.7 times the steel frame force. Assuming that the forces for the steel frame are
comparable between UBC and ATC3, the ATC3 calculated forces for the low
ductility structures are likely to be relatively conservative compared to those
calculated from the UBC.
References
1 . G.V. Berg, Seiswtic Design Codes and Procedures, Earthquake Engineering Research
Institute, Berkeley, California, 1983.
2. International Conference of Building Oficials, Un$orm Building Code, Whittier, Califor-
nia, 1982.
3. Structural Engineers Association of California, Recommended Lateral Forces and Corn-
mentary, San Francisco, California, 1980.
4. Applied Technology Council, Tentative Provisions for the Development of Seismic
Regulations for Buildings, ATC3-06, National Bureau of Standards. Washington, DC,
1978.
Appendix/Numerical evaluation of
response spectrum'
in which all the variables are defined in Chapter 1, except the ground acceleration
history, which is now denoted by a ( t ) .A number of efficient methods are available for
integrating Equation A.I. An exact technique was developed by W.D. Iwan in an
unpublished study at the California Institute of Technology and was later reported by
Nigam and Jennings[2]. In this method, Equation A.l is solved analytically within
each successive time step assuming the ground acceleration varies linearly between
designated points.
1 Aa,
oD + 1;oC2)sin oD( t - ti)] - --.
- (C,
o' At,
In these expressions oDis the damped circular natural frequency, o, = o J( I - 1; ) , '
and C, and C2 are constants. These constants are evaluated by defining
u ( f = t i ) = U , , u(t = ti)= u,.
Thus, C, and C2 are
a 21; Aa,
c, = U i + -2- -- u , + ~ o u , + -1;a , + - -1 -2C2 Aa, . (AS)
o2 o3At, ' 0 o2 At,
[u,+l}
ui+I
= A([, o ,At,) + B(5, 0 ,At,) (01
a;+I
},
where
o
a2]= - e-w'l sin O D~ t, ,
J(1 - c2)
a22= e
-gwAr, cos oDAll - c
J(1 - C 2 )
and
6I I = e-cwA'~ [ (e5 )
W'A~,
+
W
o~
OD
+ (o325At, + $)cos oD --25
o3Ati
,
0
12 -
02At,
' sin o D A t i
WD
+-0321;A ti
If the record is digitized at equal time intervals, the coefficients of A and B are con-
-
stant for a given frequency. Hence, given the initial conditions for the single-degree-
of-freedom system, usually u(0) = u(0) 0,response computations proceed rapidly
by applying the recursion relationships defined by Equation A.6. Monitoring the
response quantities as computation proceeds enables the determination of the
maximum relative displacement, i.e. the spectral displacement. The calculations are
repeated for a family of frequencies for each selected damping value. Thereby an
entire set of elastic response spectra is developed for the given earthquake record.
The procedure described above can, of course, be applied to accelerograms
digitized at unequal time intervals. However, the evaluation of matrices A and B at
N U M E R I C A L E V A L U A T I O N O F R E S P O N S E S P E C T R U M / 155
each step of integration, i.e. for each Ati, increases the computation time considera-
bly. Experience has shown that this increase in computation time may be 100% or
more. To maintain computational efficiency for records digitized at unequal time
intervals, Nigam and Jennings[Z] recommend an approximate method involving
time coordinate rounding. However, with the development of uniform processing
and correction procedures, records are routinely digitized at equal time steps of 0.0 1
or 0.02 sec. Hence, it is unnecessary, insofar as the discussion here is concerned, to
consider the treatment of records digitized at unequal time intervals.
The time step used in the response computations is selected as the smaller of the
digitized interval of the earthquake accelerogram or some fraction of the period of
free vibration, for example TjlO. For systems whose natural period governs the
selection of Ati, i.e. for high frequencies, At, must be chosen so that an integral
number of time steps comprises the digitized interval of the accelerogram. This
restriction on At, preserves uniform time intervals and guarantees that response
quantities will be computed at times corresponding to those of the given earthquake
record. For example, suppose that the response of a system with T = 0.12 sec is to be
determined. In addition, assume that the earthquake accelerogram is digitized at
intervals of 0.02 sec. If the time step is not to exceed, say, T/10 or the digitized
interval, At, must be selected as 0.01 sec, providing two time steps between successive
digitized values of acceleration.
Aside from the uncertainties associated with the recording and processing of the
accelerogram itself, errors in spectral calculations result from approximations
employed in the numerical integration technique used for response computation. In
this sense, the method described herein is exact. However, error is introduced by dis-
cretization. That is, the true maximum displacement or velocity, i.e. the spectral
quantities, will not, in general, occur at one of the discrete times at which
computations are made. The maximum error results when the true maximum falls
midway between two consecutive time points, as depicted in Figure A.1. If the
The true spectral quantities are greater than those computed at the discrete time
points. By appropriately selecting the time step, however, the maximum error in the
spectral ordinates may be controlled. For example, the expression above gives 4.9%
error for At, = TIIO, 1.2% for T/20, and 0.3% for T/40. Thus, a time step
corresponding to At, = TI20 is generally adequate.
consider first the linear elastic response which follows unloading. For this case, the
equation of motion for ti s t S ti+,is
Aa,
ii + 26o1i+ w2(u- s) = -ai -- (t - r,), (A.1 1 )
A ti
where all symbols are as previously defined. This equation may be more conveniently
expressed as
Aii;
ii + 2@u+ 0 2 u = -a',--([
At,
- t,), (A.12)
where
a',= a ; - 02s, = a,,, - w2s. (A. 13)
The notation Aa', in Equation A. 12 is used for convenience since AZ,% hai.The solu-
tion for Equation A. 12 is given by Equation A.6 with the substitution of a', and a',+, for
a, and ai+t ,
(A.14)
in which the coefficients of matrices A and B are defined by Equations A.8 and A.9.
The set s required in Equation A.13 is computed a t the instant of unloading.
Following an excursion of positive yielding, the set is given by s = ( 1 - a) (uunl- u,);
following an excursion of negative yielding, s = (1 - a) (u,,, u,). In these +
equations, u,,, is the relative displacement computed at the instant of unloading. At
the same time, the current yield levels are updated. For example, following a positive
yield excursion, uyp= uunland uyn= u,,, - 2uy.
Now consider excursions of loading beyond the current yield levels for the bilinear
system. With reference to Figure A.2, the equation of motion for relative displace-
ments greater than the current positive yield level uypis
Aa,
ii + 26wu + 02(u, - s) + am2( u - u,) = -a, --
At,
(t - t,). (A.1 5 )
This differential equation applies for u > u,, until unloading is detected, when the
product ui X u,, < 0.Simplifying Equation A. 15 gives
A&
ii + 2 ~ , o z l+i o:u = -a', - -(t
At;
- ti), (A.16)
in which
(A.17)
and
Z i i 3 a, + w Z u y ( l- a ) , = a,,, + o 2 u Y ( 1- a). (A. 18)
Note that C,, and o,, equivalent properties associated with the strain-hardening
branch of the force-deformation model, are defined only for a > 0. For an extursion
of negative yielding, for u < u,,, Equation A.16 applies with the modification,
a',= a , - w2u,(1 - a ) , Zi+, = a,,, - o 2 u Y ( l- a ) . (A. 19)
The character of the solution of Equation A.16 may be underdamped (c2 c l),
critically damped (C, = I), or overdamped (4, > 1). However, for the majority of
bilinear systems of practical interest, the response is underdamped. For example,
c,
when C, = 0.05 and a = 0.02, 0.05 and 0.10, the largest value of is 0.051J0.02 or
0.35. Thus, the solution as expressed by Equation A. 14 holds with the substitution of
c2 and o2for C, and o in the elements of A and B given in Equation A.8 and A.9.
where ii, and a',,, are computed with a = 0, in accordance with either Equation A. 18
for positive yielding or Equation A.19 for negative yielding. The solution for
Equation A.20 may also be expressed by Equation A. 14 in which the elements of
matrices A and B are:
For the special case of no viscous damping (6 = O), the coefficients of A and B are:
NUMERICAL EVALUATION O F RESPONSE SPECTRUM/l59
The coefficients in Equations A.23 and A.24 may be obtained from those in Equation
<
A.21 and A.22 by taking the limit as approaches zero.
Thus, when CoAt, is < 0.06, the expressions in Equations A.25 and A.26 are used to
evaluate the elastoplastic coefficient a,, and those coefficients for matrix B.
Including terms in each series up to the eighth order provides results accurate to about
twelve significant figures.
-
motions at the instant at which the instrument is triggered and recording begins.
These initial motions may be expressed as a(0) = ao,v(0) = vo, and d(0) 4,where
a(t), u(t) and d(r) are, respectively, the ground acceleration, velocity and displace-
ment. The time coordinate 2, of course, is measured from the instant at which
recording commences. A difficulty arises when response computations are made for
systems subjected to base excitation with nonzero initial conditions. Namely, the
initial conditions for the single-degree-of-freedom oscillator are not known. T o clarify
this point, consider the initial conditions for the relative displacement and velocity
given by
where x ( t ) and x(t) are the absolute displacement and velocity of the mass,
respectively. It is apparent that the absolute motions, x(0) and x(0), depend upon the
ground motions not recorded, i.e. those before the instrument is triggered. Hence, the
x(0) and x(0) are unknown, and u(0) and u(0) are unknown.
NUMERICAL EVALUATION O F RESPONSE SPECTRUM/161
In spite of the foregoing problem, at-rest initial conditions are commonly assumed.
However, an inconsistency arises when considering very flexible systems, i.e. for
w -- 0. With u(0) = u(0) = 0 , Equation A.27 gives x(0) = do and x(0) = vo. For the
infinitely flexible system, these initial conditions are obviously incorrect since the
mass of the system must remain motionless for all time. Hence, the proper initial con-
ditions for the very low frequency systems result from x ( t ) = x(t) = 0 , from which
u(0) = -do and u(0) = -vo. However, for very high frequency systems, i.e. for w -,
-
co there is no relative motion between the mass and the ground, and the initial condi-
tions are precisely u(0) = u(0) 0 . In view of these limiting cases, it is clear that one
set of initial conditions does not apply for all frequencies. Accordingly, one early ap-
proach for treating records with nonzero initial motions was to change initial
conditions for the oscillator at some intermediate frequency.
Pecknold and Riddell[4] have proposed a successful method of treating the
problems encountered in response computations from records with nonzero initial
motions. In this method, a short acceleration pulse is added at the beginning of the
earthquake record. For this prefixed pulse, let ri, 6 and d denote, respectively, the
pulse acceleration, velocity and displacement. Also, assume that the pulse acts from 0
-
a t s H, or H a t S 0.The prefixed acceleration pulse consists superposition
of three influence functions which were derived by minimizing
the constraints
bH d(t)dT = vo and
The prefixed pulse is piecewise linear so that conventional integration methods yield
the velocity u, and displacement do at the end of the pulse. The ordinates of the pre-
fixed acceleration pulse are given by
where ci, = a(iAt).The pulse is divided into N ( N 2 3) intervals such that At = HIN.
The influence functions g,(i),g2(i)and g3(i)are cubic polynomials in the discrete
variable i and are given by
in which j = 1,2,3 and i = 1,2, . . .,N. The coefficients bj, c, and a,' depend upon the
number of intervals N a n d are given by
-
in which C = ( N ~ 1) (N' - 4).
The undamped spectra computed from Melendy Ranch record (N20W, 1982) with
and without a 2 sec prefix pulse are shown in Figure A.3. It is clear from this figure
that the correct asymptotic behavior at low frequencies is achieved only for the
spectrum computed from the record with the prefixed pulse. That is, at low
frequencies, the spectral displacement approaches the peak ground displacement, d,,
in this case 1.28 in. In fact, it can be shown that the low frequency asymptote for spec-
tra computed from records with nonzero initial motions corresponds to a constant
pseudovelocity equal in magnitude to the initial ground velocity, o,. This behavior is
clearly evident in Figure A.3 for the spectrum computed from the Melendy Ranch
record with no prefixed pulse, for which ( u, / = 1.17 in sec-'. In addition, note in
Figure A.3 that the significant differences between the spectra extend up to a
frequency of about 0.5 Hz. Above about 2 Hz, the spectra are identical, consistent
with the previous discussion regarding the initial conditions for high frequency
systems. Pecknold and Riddell(41 estimate that the frequency below which spectral
Frequency, Hz
Fig. A.3 Undamped spectra for Melendy Ranch record (N20W, 1982) with and without
prefix pulse.
NUMERICAL EVALUATION O F RESPONSE SPECTRUM/163
ordinates may be in error is j; = o,/(2xdP), which for the Melendy Ranch record is
0.15 Hz. It is evident from Figure A.3 that the spectral errors may extend to a
frequency several times the value given by the expression above.
References
I. J.M. Nau and W.J. Hall, An Evaluation of Scaling Methods for Earthquake Response
Spectra, Structural Research Series, No. 499, Department of Civil Engineering. University
of Illinois, Urbana-Champaign, May 1982.
2. N.C. Nigam and P.C. Jennings, Calculation of Response Spectra from Strong Motion
Earthquake Records, Bulletin of the Seismological Society ofAmerica, Vol. 59, No. 2, April
1969, pp. 909-922.
3. J.M. Nau, Computations of Inelastic Response Spectra, Journal of Engineering Mechanics,
ASCE, Vol. 109, No. 1, February 1983, pp. 279-288.
4. D.A. Pecknold and R. Riddell, Effect of Initial Base Motion on Response Spectra, Journal
of Engineering Mechanics Division, ASCE, Vol. 104, No. EM2, April 1978, pp. 485-491.
Author index