Вы находитесь на странице: 1из 172

Lecture Notes for

SE 221 - Earthquake Engineering

by

Prof. P. Benson Shing

Department of Structural Engineering


University of California, San Diego

Spring 2017
Table of Contents

Chapter 1 Introduction

Chapter 2 Causes and Characteristics of Earthquakes

Chapter 3 Seismic Response of Linear SDOF Structures

Chapter 4 Seismic Response of Nonlinear SDOF Structures

Chapter 5 Seismic Response of MDOF Structures

Chapter 6 Seismic Design and Code Requirements

Chapter 7 Seismic Hazard Analysis


CHAPTER 1

INTRODUCTION

As shown by the numbers in Table 1.1, a major earthquake can lead to significant casualties not
to mention the direct and indirect economic impact on the affected communities. Property
damage caused by a major earthquake could easily amount to tens of billions of dollars. The
magnitude-7.1 Loma Prieta earthquake in 1989 in the San Francisco Bay Area resulted in 7
billion dollars of property damage. Including the loss of business opportunities, the total loss has
been estimated to be around 10 billion dollars. The total economic loss caused by the 1994
Northridge earthquake in the Los Angeles area has been estimated to be 43 billion dollars. The
2011 Tohoku earthquake in Japan resulted in a total loss of 200 to 300 billion US dollars, which
was about 5% of the GDP of Japan.

Earthquake engineering is a multi-disciplinary field focusing on the mitigation of earthquake


hazards. It involves concerted efforts of seismologists, social scientists, architects, and structural
and geotechnical engineers to develop better earthquake prediction methods and preparedness
plans and policies, and to improve the design of earthquake resistant structures.

Japan has a long history of earthquake engineering studies dating back to the nineteen century. In
the US, the earliest building code provisions in seismic resistance design can be traced back to
the 1927 edition of the Uniform Building Code. Major efforts on earthquake engineering studies
in the US started right after a magnitude-6.2 earthquake struck Long Beach, California, in 1933
killing several hundred people and causing over one billion dollars in property damage
(converted to present dollar value). Earthquake engineering studies were interrupted by World
War II and were resumed in the 1950's. In late seventies, with lessons learned from the 1971 San
Fernando earthquake in California, which had a magnitude of 6.5, seismic design provisions in
the US were significantly revised. Improvements in seismic design standards were evident from
the performance of newer buildings and bridges in the 1989 Loma Prieta earthquake and 1994
Northridge earthquake. However, lessons from these earthquakes, such as the damage of steel
structures designed according to the most current code provisions at that time in the Northridge
earthquake, also indicated that there was still much to learn and improve in code and structural
design standards. It must be noted that earthquake hazards are not limited to the West Coast
regions of the US. There were major historical earthquake events in the eastern part of the
country, such as the 1755 Cape Ann earthquake in Massachusetts, which had a magnitude above
6.0, and the New Madrid earthquakes between 1811 and 1812 with magnitudes between 7.0 and
8.1. Even though the probability of a strong earthquake occurring in the Midwestern and Eastern
US is small, the consequence could be extremely severe if it occurs because structures in these
regions have been designed with standards much less stringent than those in the West Coast.
Table 1.1 also shows that earthquake hazard is a world-wide problem.

Moderate earthquakes occurring in the West Coast regions of the US and severe earthquakes that
happened in different countries in recent years have shown that structures designed according to
current code standards performed well in general and more or less achieved the life-safety
objective. However, major challenges facing earthquake engineers today include the retrofit and

1
strengthening of older structures to bring them to modern standards, improving seismic design
standards for regions subject to low probability but high consequence earthquakes, like the
Midwestern and Eastern US, and the design of earthquake resilient structures to minimize
structural and non-structural damage and the economic loss. However, our seismic design
methods are still not completely rational or disaster proof, and heavily rely on experience-based
assumptions. They may not necessarily guarantee a consistent level of safety for different
structural types.

Table 1.1 Some Major Earthquakes since 1990

Year Location Richter Death Toll


Magnitude
1991 Northern India 7.1 768
1992 Turkey 6.8 800
1992 Landers, California 7.4 0
1992 Egypt 5.4 500
1993 Hokkaido, Japan 7.8 200
1993 Guam 8.1 0
1993 Maharashtra, India 6.4 12,000-30,000
1994 Northridge, California 6.7 72
1995 Kobe, Japan 7.2 5500
1999 Kocaeli, Turkey 7.4 17,400
1999 Chi-Chi, Taiwan 7.6 2,400
2001 Bhuj, India 7.9 25,000
2003 Bam, Iran 6.6 43,000
2004 Sumatra, Indonesia 9.0 280,000
2005 Kashmir, Pakistan-India 7.6 80,000
2008 Wenchuan Earthquake, China 8.0 70,000-90,000
2010 Chilean Earthquake 8.8 550
2011 Tohoku Earthquake, Japan 9.0 19,000
2011 Christchurch Earthquake, NZ 6.3 185

Types of Earthquake Damage

Earthquake-induced structural damage can be classified into the following types.

1. Foundation displacements associated with fault breaks and landslides.


2. Foundation failures due to liquefaction.
3. Damage due to tsunamis, fire, etc.
4. Structural damage induced by seismic ground shaking.

2
Seismic Resistance Design Concepts

For many years, seismic resistance design has been based on the following philosophy.

1. Prevent any damage (structural and non-structural) for frequent minor earthquakes.
2. Permit non-structural damage but prevent structural damage for occasional moderate
earthquakes.
3. Permit structural damage but prevent severe damage or collapse for rare and severe
earthquakes.

The aforementioned approach recognizes the fact that it is economically not practical to produce
structures that can withstand rare and severe earthquakes without damage. This approach has
been gradually refined to become more quantitative by quantifying the seismic hazard and
defining the performance levels that can be associated with each hazard level so that a structure
can be designed to satisfy one or more performance criteria. This is called performance-based
seismic design. The concept of performance-based seismic design as adopted by ASCE/SEI 41-
06 for the seismic rehabilitation of existing buildings is illustrated in Table 1.2. As shown, one or
more performance criteria can be selected by the building owner or required by the building
official on the seismic retrofit of an existing building. ASCE/SEI 41-06 recognizes the fact that
seismic rehabilitation requires greater flexibility than the design of a new building. For example,
meeting both criteria k and p in Table 1.2 is considered the Basic Safety Objective, which will
result in a building that has a safety level comparable to that of a new building but could suffer
more earthquake damage than a properly designed and constructed new building; or a building
can be retrofitted to meet a less stringent objective of satisfying criteria g and l. In recent years,
the performance-based approach is moving towards probabilistic-based performance objectives
to address the fact that earthquake loads are non-deterministic, and design provisions and
analytical predictive tools are not perfect.

Table 1.2 Range of Rehabilitation Objectives*

Target Building Performance Levels


Earthquake Hazard Operational Immediate Life Safety Collapse
Level Occupancy Prevention
50%/50 Years a b c d
(Return Period = 72)
20%/50 Years e f g h
(Return Period = 225)
BSE-1: 10%/50 Years i j k l
(Return Period = 475)
BSE-2: 2%/50 Years m n o p
(Return Period = 2,475)
*ASCE/SEI 41-06: Seismic Rehabilitation of Existing Buildings, 2006.

3
For performance-based design, the ability to accurately assess the elastic or inelastic dynamic
response of a structure under earthquake loads is of primary importance. This course focuses on
analytical and numerical methods that can be used for this purpose, and on gaining a
fundamental understanding of the response of structures under earthquake loads through the
principles of structural dynamics. It is also intended to provide a good understanding of the
rationales behind seismic design methods and code provisions.

4
Reference:

1. ASCE/SEI 41-06. Seismic Rehabilitation of Existing Buildings. American Society of Civil


Engineers. New York, NY, 2006.

2. ASCE/SEI 41-13. Seismic Rehabilitation of Existing Buildings. American Society of Civil


Engineers. New York, NY, 2013.

Website Resources:

https://www.eeri.org (Earthquake Engineering Research Institute)


http://peer.berkeley.edu (Pacific Earthquake Engineering Research Center)
http://nisee.berkeley.edu (National Information Service for Earthquake Engineering)
http://earthquake.usgs.gov (US Geological Survey)
http://www.conservation.ca.gov (California Department of Conservation)
http://www.iris.edu (Incorporated Research Institutions for Seismology)
http://www.scec.org (Southern California Earthquake Center)

5
CHAPTER 2

CAUSES AND CHARACTERISTICS OF EARTHQUAKES

2.1 Earthquakes

2.1.1 Types of Earthquake

Earthquakes can have difference causes. However, all destructive earthquakes were related to
tectonic forces.

1. Tectonic Earthquakes - caused by tectonic forces; they are the most common type.
2. Volcanic Earthquakes associated with volcanic eruptions; but they are most likely related
to tectonic forces.
3. Collapse Earthquakes - caused by the collapse of roofs of mines and caverns, or by
landslides.
4. Explosion Earthquakes - caused by explosions such as nuclear explosions.

2.1.2 Plate Tectonics

Tectonic is a word that originates from a Greek word meaning "builder". Tectonic forces are
forces that are involved with the building of the earth's crust, such as the forces that fold
mountains, raise and depress large sections of the continents, cause fault movements, etc.

The earth's surface consists of several large and stiff plates (lithosphere) sitting on a layer of
relatively soft rock (asthenosphere) as shown in Fig. 2.1.

Lithosphere Subduction Zone Spreading Zone


(Ocean Trench) (Mid-Ocean Ridge) ~80 km

Asthenosphere Shallow
Earthquakes
Deep
Earthquakes Hypocenters
(foci)
magma
Fig. 2.1 Plate Tectonics

Interplate earthquakes are generated by tectonic forces at plate boundaries; they are responsible
for most of the world's seismic energy release. Five percent of the world's seismic energy release

6
occurs at ocean ridges (spreading zones) and over ninety percent at ocean trenches (subduction
zones).

Intraplate earthquakes occur within the plates; the mechanism is also related to tectonic forces
but it is not well understood. Examples are the series of earthquakes occurring in the New
Madrid seismic zone in Missouri between 1811 and 1812, the 1886 Charleston earthquake in
South Carolina, and more recently, the 2001 Bhuj earthquake in India. Such earthquakes are very
rare.

2.1.3 Elastic Rebound Theory

The elastic rebound theory was developed by Reid and published in 1911, based on observations
of fault slip in the 1906 San Francisco earthquake that had a Richter magnitude of 8.3. The
earthquake was caused by the slipping of the San Andreas fault over a length of more than 300
km. The maximum horizontal slip that occurred along the fault was 6 m. This theory is explained
in Fig. 2.2.

Fault

Imaginary line
before straining
the surface
Slip

Imaginary line
after straining

Straining Induced by Tectonic Forces Strain Release After Earthquake


Before Earthquake
Fig. 2.2 Elastic Rebound Theory

The elastic rebound of rock masses initiates vibrations that propagate through the earths crust as
seismic waves.

Types of fault:

1. Dip slip faults - classified into normal faults and reverse faults (Figs. 2.3a and 2.3b).

2. Strike slip faults - classified into left lateral faults and right lateral faults (Figs. 2.3c and
2.3d).

7
(a) Normal (b) Reverse

(d) Right Lateral


(c) Left Lateral

Fig. 2.3 Types of Fault

2.1.4 Focus and Epicenter

Focus (or hypocenter) is a point from which seismic wave is first generated. Epicenter is a point
on the earth's surface vertically above the focus.

Epicentral Distance

Epicenter Reference Position

Focal
Depth
Focal Distance

Focus

Fig. 2.4 Focus and Epicenter

8
Shallow earthquakes have focal depths less than 70 km and account for 75% of world's seismic
energy release. Intermediate depth earthquakes have focal depths between 70 and 300 km, while
deep earthquakes have focal depths more than 300 km. Most of the destructive earthquakes in
California have focal depths of 15 to 25 km.

2.1.5 Seismic Waves

There are two types of seismic waves (see Fig. 2.5):

1. Body Waves

(a) P Waves - longitudinal or compressive waves created by the motion of


particles moving back and forth in the direction of wave propagation;
velocities of P waves are usually between 5 and 7 km/sec and can be
estimated with the following equation derived from the theory of elasticity.

4
v P (k ) /
3

where k is the bulk modulus, is the shear modulus, and is the mass
density of the elastic medium.

(b) S Waves - shear or transverse waves created by the motion of particles


oscillating in a horizontal or vertical plane in a direction normal to the
direction of wave propagation; velocities of S waves are usually between 3
and 4 km/sec and can be estimated with the following equation.

vS /

2. Surface Waves

(a) Love (L) Waves - particles oscillate in a plane parallel to the earth's surface
and perpendicular to the direction of wave propagation.

(b) Rayleigh (R) Waves - particles oscillate in a plane perpendicular to the


earth's surface; they move both vertically and horizontally in an elliptical
path like ocean waves.

9
Fig. 2.5 - Seismic Waves (Bolt 1988)

10
Because of wave reflection, body waves can be significantly amplified when they reach the
ground surface. In general, vP vS vL vR . Hence, due to the different velocities, each wave
arrives at a different time (see Fig. 2.6). The different arrival times can be used to determine the
location of the focus. Accurate determination of the longitude, latitude, and focal depth of a
focus, and the origin time requires arrival times obtained at four to five different seismograph
stations.

Fig. 2.6 Accelerogram (Wakabayashi 1986)

Strong-motion seismographs are normally accelerometers used to measure ground motions close
to epicenters and provide information useful to engineers.

2.2 Size of an Earthquake

The intensity of an earthquake is a measure of the effects of the earthquake at a particular


location, while the magnitude is a measure of the absolute size of the earthquake.

2.2.1 Intensity Scale

Earthquake intensity depends on the magnitude of the earthquake, focal depth, epicentral
distance, local soil conditions, and geological features of the region. The most commonly used
intensity scale is the Modified Mercalli (MM) Scale developed by Wood and Neumann in 1931.
It has a scale from I to XII. It is based on observations of the effects of an earthquake on the
population, buildings, and ground at a particular site. Hence, it is very subjective and can be
unduly influenced by the quality of the structures at the location. Isoseismal lines are lines drawn
on a map defining zones of equal intensities. The isoseismal lines for the 1811 New Madrid
earthquake in Missouri are shown in Fig. 2.7.

11
Fig. 2.7 - Isoseismal Lines for the 1811 New Madrid Earthquake (Bolt 1988)

2.2.2 Magnitude Scale

Richter Scale

The magnitude of an earthquake can be measured in terms of different scales, most of which are
based on the Richter Magnitude Scale developed by Charles Richter (Bull. of Seismological
Society of. America, 25, 1935). The original Richter magnitude is expressed as

M L log A

where the subscript L represents the local magnitude, which is based on seismic waves recorded
at a short epicentral distance, and A is the maximum amplitude of seismic waves (in microns)
recorded on a standard (Wood-Anderson) seismograph at a distance of 100 km from the
epicenter. For seismic waves measured at different epicentral distances, the amplitudes need to
be adjusted using attenuation laws (see Fig. 2.8). It is not suitable for measurements taken at an
epicentral distance more than 600 km.

12
Amplitude
Attenuation Relation

Epicentral
100 km Distance
Epicenter

Fig. 2.8 - Attenuation of Seismic Waves

Modified Richter Magnitudes

Several modified Richter magnitude scales have been developed to account for different seismic
conditions.

mb : body-wave amplitude (normally based on P waves); it has the advantage that it is


not affected by the focal depth and is often used for deep focus earthquakes which
do not generate significant surface waves.

M S : surface-wave amplitude (normally based on surface waves that have a period near
20 sec.); it is often used for shallow earthquakes recorded at a long epicentral
distance. It better reflects the size of a large earthquake.

The magnitude scales M L and mb saturate at about 7, and M S saturates at about 8. In general,
M S M L for large earthquakes, and M S M L for earthquakes with M L 6 ~ 6.5 . Hence, the
following limits are imposed on the different magnitude scales:

M L or mb for earthquake magnitudes less than 7


MS for earthquake magnitudes between 6 and 7.5

The moment magnitude is used for earthquakes greater than 7.5.

Moment Magnitude

The moment magnitude (propsoed by Hanks and Kanamori, 1979) is a mechanical measure
which is defined as

13
log M o
Mw 6.1
1.5

where Mo is the seismic moment in N-m. The seismic moment can be estimated as follows.

M o aD a3/2

in which is the shear modulus, a is the size of the area that slipped along the fault, D is the
amount of slip, and is the seismic stress release. Because it is directly related to the actual
seismic stress release, it has no saturation point. For this reason, it is commonly used to rate large
earthquakes with magnitudes greater than 7.5.

The magnitude scale has no upper limit. However, magnitudes of past earthquakes seldom
exceeded 8.0. The 2004 Sumatra earthquake in Indonesia had a magnitude of 9.0. So was the
Tohoku earthquake in 2011.

2.2.3 Seismic Energy Release

Seismic energy release E is measured from seismic waves. The following relation between E and
M is accepted by most seismologists:

log E 11.8 1.5M

where E is measured in ergs. Hence, a unit increase of M leads to an increase of E by a factor of


30.

2.3 Attenuation of Ground Motions

Peak accelerations, velocities, and displacements ( a m , v m , d m ) are important characteristics of an


earthquake ground motion that affect the soil response and structural damage, and are, therefore,
important to engineers. They depend on

1. The magnitude of the earthquake.


2. The fault-distance or focal distance.
3. The geological features of the region.
4. The local soil properties.

Peak ground motions at a site depend very much on the attenuation rate of the seismic waves
propagating from the epicenter and the local soil amplification effect. A general attenuation
relation can be presented as follows (Bolt 2004).

Y b1 f1 M f 2 R f3 F

where

14
Y = peak ground motion (such as the acceleration, velocity, or displacement)

M = Richter magnitude or modified Richter magnitude of the earthquake

R = fault-distance or focal distance

F = a flag for the style of faulting (reverse, normal, strike slip)

where f1, f2, and f3 are given as follows:

f1 M eb2 M b3M :
2
In general, for high-frequency waves, b3 = 0, which reduces f1 to
the basic definition of the Richter magnitude M that is a
logarithmic measure of the amplitude of seismic waves.

f 2 R R b5
b4
: This function represents geometric damping assuming a
mathematical expression that is consistent with the solution of
wave propagation in an elastic medium.

f3 F eb6 F : This term accounts for the influence of the faulting mechanism.

The coefficients bi s can be determined by regression analysis using site specific data. Very
often, it has been observed that at a short epicentral distance, the spacing of the attenuation
curves due to different earthquake magnitudes becomes very small. This phenomenon is called
near-fault ground motion saturation. One way to account for this is to replace the function f 2
above by

f 2 R, M R f 4 (M )
b4

where

f 4 (M ) b7 eb8M

If b3 = 0 and b4 b2 / b8 , then Y becomes magnitude independent at R = 0. In this case, the model


has a complete saturation.

Example

The following attenuation relations were proposed by Esteva (1974) for firm soil conditions with
a focal distance greater than 15 km.

am 5600e0.8 M R 40
2

15
v m 32e M R 25
1.7

am d m 400
2
1 0.6
vm R

where

am is in cm/sec2

v m is in cm/sec

dm is in cm

R is the focal distance in km

2.4 Characteristics of Strong-Motion Records

Some ground acceleration records obtained in past earthquakes are shown in Fig. 2.9.
Characteristics of earthquake ground motions that affect the performance of a structure include:

1. Peak ground motions am , m , d m .

2. Frequency content.

3. Duration of strong shaking.

4. Duration and number of strong pulses.

These characteristics are affected by:

1. Earthquake source mechanism.

2. Regional geological properties that influence seismic wave propagation.

3. Epicentral distance.

4. Local soil properties.

5. Influence of soil-structure interaction.

Near-fault ground motions often contain strong long-duration pulses in the wave form. The
duration of these pulses may vary from 0.5 sec. to 5 sec. or more, and they are most obvious in
the ground velocity time history. The duration of a pulse may increase with the magnitude of an

16
earthquake. These pulses are caused by two main mechanisms (Bolt 2004). One is the directivity
effect and the other is the permanent offset of the ground called the fling step. The first
mechanism is related to fault rupture propagating toward the site and the slip direction is aligned
with the rupture direction (Somerville et al. 1997). The pulses are results of constructive
interference of the S-waves generated by the progressive fault rupture which happens to
progapate at almost the same velovity as the shear waves. For a strike-slip fault, the ground
displacement related to a directivity pulse is normal to the slip direction while that related to a
fling-step pulse is parallel to the slip. Thus, the two motions are in orthogonal directions.
However, for a dip-slip fault, the horizontal ground displacement related to a fling-step pulse can
also be normal to the fault. Examples of near-fault ground motions are the Sylmar record shown
in Fig. 2.9 and the Rinaldi Station record in Fig. 2.10a, both obtained during the 1994 Northridge
earthquake. The pulses in these records were introduced by the directivity effect. Figure 2.10b
shows the velocity pulse generated by the fling-step mechanism in the 1999 Chi-Chi earthquake
in Taiwan. As will be further considered in Chapter 4, long-duration pulses have strong
destructive power on structures.

Eathquakes used for structural design can be selected in a number of ways. The most common
ways are:

1. The use of strong motion records obtained from previous earthquake events that had
the source mechanisms and geological features similar to those that would be
expected for the site in consideration.

2. The use of artificially generated earthquake ground motions consistent with the local
geological features of the site.

3. Design spectra provided by building codes (See Chapter 7).

17
El Centro Record, 1940 1978 Miyagi-Ken-Oki
Imperial Valley Earthquake

Santa Monica Record, 1994 Sylmar Record, 1994


Northridge Earthquake Northridge Earthquake

1985 Mexico City


Fig. 2.9 Sample Ground Acceleration Records

18
(a) Rinaldi Station, 1994 Northridge Earthquake (Bozorgnia and Campbell 2004)

(a) Total = (b) + (c)

(c) Isolated pulse

(b) Remaining waveform

(b) Station TCU068, 1999 Chi-Chi Earthquake (Bolt 2004)

Fig. 2.10 Near-Fault Ground Motions

19
References:

1. Bolt, B. (2004). Earthquakes, 5th Ed., W.H. Freeman and Company, NY.
2. Bolt, B. (2004). Engineering Seismology, in Earthquake Engineering from Engineering
Seismology to Performance-Based Engineering, Bozorgnia and Bertero (Eds.), CRC Press.
3. Bozorgnia, Y. and Campbell, K.W. (2004). Engineering Characteristics of Ground Motion,
in Earthquake Engineering from Engineering Seismology to Performance-Based
Engineering, Bozorgnia and Bertero (Eds.), CRC Press.
4. Kramer, S.L. (1996). Geotechnical Earthquake Engineering, Prentice Hall.
5. Somerville, P.G., and Smith, N.F., Graves, R.W., and Abrahamson, N.A. (1997).
Modifcation of Empirical Strong Motion Attentuation Relations to Inlcude the Amplitude
and Durations Effects of Rupture Directivitiy. Seismological Research Letters, 68(1), 199-
222.
6. Wakabayashi, M. (1986). Design of Earthquake-Resistant Buildings, McGraw-Hill.
7. USGS Website: http://earthquake.usgs.gov (US Geological Survey)

20
CHAPTER 3

SEISMIC RESPONSE OF LINEAR


SDOF STRUCTURES

In general, a time-domain approach or a frequency-domain approach can be used to evaluate the


dynamic response of a structure to earthquake ground motions. This chapter presents the
fundamental concepts of these two approaches considering arbitrary earthquake load histories.
Because of the random nature of earthquake ground motion histories, close-form solutions of the
equation of motion for a structure are not attainable. Hence, we will focus on numerical methods
for time-domain and frequency-domain analysis. In addition to analysis methods, we will also
explore the characteristics of earthquake ground motions that have an important influence on
structural response.

3.1 Time-Domain Analysis using Duhamels Integral

The response of a structure to an arbitrary earthquake load history can be expressed in terms of
Duhamels integral, which is a convenient way to convey the concept of spectral response that is
central to earthquake resistance design. For an arbitrary load history, Duhamels integral can be
integrated using numerical methods, such as the trapezoidal rule, which are covered in many
structural dynamics books, even though they may not be computationally most efficient. Here,
we will not consider such a numerical approach, but will use Duhamels integral as a means to
introduce the concept of spectral response for earthquake loads.

M M

C K = C K

Fig. 3.1 - Single-Story Shear Frame

To begin with, we consider a linearly elastic, single-story, shear-frame structure as shown in Fig.
3.1. The equation of motion for the structure can be expressed as

M t + C + K =
0 (3.1a)

21
t (t ) + g ( t ) is the total structural acceleration. The above equation can be rewritten
in which =
as

M + C + K = f eff (t ) = Mg (t ) (3.1b)

or in an alternate form as

f eff (t )
+ 2 + 2 = = g (t ) (3.1c)
M

where

= the relative structural displacement with respect to the base support (a


superposed dot represents differentiation with respect to time)

g = ground acceleration

M = mass, K = stiffness, and C = damping coefficient

C C
= = (ratio of critical damping)
C c 2 M

K 2
= = undamped natural angular frequency
M T
(T is the undamped natural period)

We know that the displacement response of a structure to a unit impulse (with an infinitesimal
duration) occurring at a time instant is

h (t ) = e (t ) sin D (t )
1
(unit impulse response function)
M D

where

D = 1 2 (natural angular frequency of a damped structure)

Note that D for small damping (i.e., when < 10% ).

Consider an arbitrary load history such as the one shown in Fig. 3.2, and we want to evaluate the
response of a structure to this load history. This solution can be obatined from the unit impulse
response function by applying the principle of superposition. To this end, we divide the load

22
history into a sequence of pulses as shown in Fig. 3.2. The relative displacement response of the
structure to a single pulse at time , represented by the shaded area, is

h(t ) f eff d
d=

An impulse of
occurring at time

Fig. 3.2 - Load Function

With the principle of superposition, the total relative displacement response to the entire load
history is

(t ) h ( t ) f ( ) d
t t
= =
0
d
0 eff

which is called Duhamels integral (or convolution integral). Since f eff = Mg for earthquake
ground motion, we have

1
g ( ) e (t ) sin D ( t ) d
t
(t ) =

D 0
(3.2)

In the above equation, the negative sign can be dropped for simplicity. This integral can be
evaluated numerically for arbitrary earthquake load histories. The above equation can also be
written in the following form.

1 (t )
(t ) = F ( t , ) d
D 0

where

F (t , ) = g ( )e (t ) sin D (t )

23
(t ) = t

The total structural displacement is the sum of the relative structural displacement and the
ground displacement and is thus

t (t ) = (t ) + g (t )

However, the total displacement is not an important physical quantity as it is not related to the
deformation demand on the structure. The same can be said for the total structural velocity as it
is not related to the damping force. The relative structural velocity is obtained by differentiating
the relative displacement with respect to time as

d ( t ) d 1 (t ) 1 (t ) dF ( t , ) d ( t )
( t )
= =
dt D 0
= F ( t , ) d d + F ( t , ( t ) )
dt D 0 dt dt

Since 2 (t ) = t , we have the following expression from the above equation.

1
( t ) ( ) e
( t )
cos D ( t ) + d
t
g
1 2 0

where


= tan 1 (a phase angle)
1 2

On the hand hand, the total structural acceleration is an important response quantity as it is
directly related to the inertia force developed by the structure. Knowing ( t ) and (
t ) , we can
calculate the relative acceleration from the equilibrium condition (Eq. (3.1a)) as

1
(C + K ) = 2 ( )e
( t )
sin[ D (t ) + ]d
t

t =
M 1 0 g

where

2 1 2
= tan 1 (a phase angle)
(1 2 2 )
For small damping (i.e., when < 10% ), it is easy to see that the above response quantities can
be accurately represented by the following approximations.

24
1
(t ) ( )e ( t )
sin ( t ) d
t


g
0

( t ) g ( )e (t ) cos ( t ) d
t

t ( t ) g ( )e (t ) sin ( t ) d
t

3.2 Response Spectra

In seismic design and analysis, we are often interested in the maximum response of a structure.
As Eq. (3.1c) shows, this depends on the natural frequency and damping ratio of the structure.
For a given damping ratio, the maximum responses that can be induced by an earthquake ground
motion on structures of different natural frequencies can be expressed in terms of a response
spectrum, which is a plot of the maximum response against the natural period or frequency.

Spectral Displacement

As shown in Fig. 3.3, spectral displacement is defined as

S d ( , ) = (t ) max

Fig. 3.3 - Relative Displacement Response Time History

It can be obtained by evaluating the entire displacement response history using Duhamels
intergal or other numerical methods, and identifying the absolute maximum displacement. By
repeating this analysis for structures of different natural frequencies, one obtains a displacement
response spectrum.

25
Pseudo Spectral Velocity

Pseudo spectral velocity is not the true spectral velocity but a quantity that is defined as

PS ( , ) = S d

Pseudo Spectral Acceleration

Pseudo spectral acceleration is not the true spectral acceleration but a quantity defined as

PS a ( , ) = 2 S d

The above two quantities are called pseudo because they do not represent the true maximum
relative velocity and maximum total acceleration. For small damping, one can see from Eq. (3.2)
that

1
(t ) ( )e ( t )
sin ( t ) d
t


g (with the negative sign ignored)
0

and, therefore,

( ) e sin ( t ) d
t ( t )
PS g
0 max

( ) e sin ( t ) d
t ( t )
PS a g
0 max

By comparing the above expressions to the true maximum responses that can be obtained from
expressions dervied in the previous section for the true relative velocity and true total
acceleration, we can see that

PS a S a for small damping

= S a for = 0
and

PS S when is large and is small

where S and S a are the true spectral velocity and spectral acceleration, respectively. A
comparison of the pseudo and true spectral quantities for some ground motions can be found in
Chopra (2011).

26
The maximum elastic restoring force developed by a structure can be calculated as

f s ,max= K ( t ) max= KS=


d M 2 S=
d M PS a

Hence, one can see that the pseudo spectral acceleration is directly related to the elastic restoring
force or base shear developed by the structure, while the true spectral acceleration is related to
the inertia force which is equal to the sum of the elastic restoring force and the damping force.
Hence, the pseudo spectral acceleration is more relevant to the damage potential of a ground
motion as it is directly related to structural deformation. The two will, of course, be the same if
damping is zero. Because of this and for the convenience of constructing design spectra, which
will be discussed in a later chapter, the pseudo spectral quantities are more useful than the true
spectral quantities.

For a given ground motion record, spectral responses can be calculated for a range of structural
frequencies and dampings to generate spectral curves, as illustrated in Fig. 3.4.

g max

T
T

Fig. 3.4 - Response Spectra

Pseudo spectral velocity plots for the NS component of the El Centro ground motion recorded in
the 1940 Imperial Valley earthquake and for a ground motion recorded in the 1985 Mexico City
earthquake are shown in Figs. 3.5 and 3.6.

27
Fig. 3.5 - Pseudo Spectral Velocity Plots for the El Centro Record from the 1940
Imperial Valley Earthquake
Fig. 3.5 - Pseudo Spectral Velocity Plot for 1940 El Centro Record

Fig. 3.6 - Pseudo Spectral Velocity Plot for a 1985 Mexico City Record
(2% damping)

28
Tripartite Plots

Very often, all three spectral quantities are plotted in a single graph called a tripartite plot. A
tripartite plot has four logarithmic axes. Such a plot is possible because of the following
relations.

PS = S d log PS = log + log S d

1
PS = PS a log PS = log + log PS a

The above equations indicate that the log of the pseudo spectral velocity is linearly related to the
log of the natural frequency provided that the spectral displacement or pseudo spectral
acceleration remains constant. The slope of the linear relations is either one or minus one
depending on whether the spectral displacement or pseudo spectral acceleration is assumed a
contant. Based on these relations, the axes of a tripartite graph can be obtained and the response
quantities can be plotted as shown in Fig. 3.7.

Constant Sd

1
1

Constant PSa

Fig. 3.7 - Tripartite Plot

A sample tripartite plot for the El Centro record is shown in Fig. 3.8.

29
Fig. 3.8 - Tripartite Plot for the El Centro from the 1940
Imperial Valley Earthquake (0, 2, 5, 10, and 20% damping)

3.3 Shapes of Response Spectra

The shape of a response spectrum is influenced by a number of factors, including

1. Source mechanism.
2. Geological properties of the region.
3. Richter magnitude.
4. Epicentral distance.
5. Focal depth.
6. Local soil conditions.

Source Mechanism

The source mechanism refers to the mechanism of earthquake generation, such as the interaction
between different plates in a subduction zone. As discussed in Chapter 2, it is recognized that the

30
fault rupture mechanism has an important influence on the characteristics of a near-fault ground
motion because of the directivity effect.

Geological Conditions

Geological conditions affect seismic wave propagation and the frequency contents in earthquake
ground motions. However, because of the inhomogenity and complexity of the earths crust, such
influence cannot be easily quantified.

Richter Magnitude

In general, the response spectrum resulting from a small earthquake that is generated at a short
distance from the site will have relatively higher energy in the high frequency range than a large
earthquake generated far away from the site. This is illustrated in Fig. 3.9.

Epicentral Distance and Focal Depth

High frequency seismic waves attenuate faster than low frequency waves. Therefore, the longer
the epicentral distance (or the focal depth) is, the smaller will be the contribution of high
frequency waves in the earthquake ground motion.

Local Soil Conditions

The dynamics of soil strata can amplify or de-amplify seismic waves passing through the soil,
and, therefore, have a major impact on the frequency content of an earthquake ground motion
and the shape of a response spectrum, as illustrated in Fig. 3.10.

Fig. 3.9 Pseudo-Velocity Spectra (Seed et al.)

31
Fig. 3.10 Influence of Soil Conditions on Response Spectra (Seed

3.4 Time-Domain Analysis using Piece-Wise Linear Approximation for Excitation

For calculating spectral curves, numerical integration of Duhamels integral is not very efficient
from the computational standpoint. A more efficient approach was proposed by Nigam and
Jennings (1968) for constructing response spectra for earthquake ground motions. This method is
based on a piece-wise linear approximation of the ground acceleration as presented below.

Assume that an earthquake ground motion history is piece-wise linear with the digitized points
connected by straight lines as shown in Fig. 3.11.

Fig. 3.11 Piece-Wise Linear Acceleration

32
Based on the above assumption, for t i t t i +1 , we have

g (i )
g (t ) = g (i ) + (t ti )
ti

where

g (i ) = g (i +1) g (i )

t i = t i +1 t i

With the above approximation, the equation of motion for t i t t i +1 can be expressed as

g ( i )
+ 2 + 2 = g (t ) = g (i ) + (t ti )
t i

General solution to the above differential equation is

(t ) = c (t ) + p (t )

where

c (t ) = e (t t ) [ A sin D (t ti ) + B cos D (t ti )]
i

1 2
p (t ) = 2
g (i ) + i i (t t i )

g (i )
i =
t i

in which A and B are constants that depend on the initial conditions.

If the displacement and velocity at time t i are known in every time step i and are denoted by i
and i , then

1 2 2 1
A= + +
i
D
g (i )
2
i i

1 2
B = i + i
2 g (i )

33
The expression for the general solution becomes

2
= (t ) e ( t ti )
{ A sin ( t t ) + B cos ( t t )} + 1
D i D i 2 g (i ) + i i ( t ti )

Hence,

1 2
i +1 ( t=
= i +1 ) e ti { A sin D ti + B cos D ti } + 2
g (i ) + i i ti

d (t )
i +1 =
dt t =ti +1

The above two equations can be rearranged into a matrix form as

i +1 i g (i )
= A ( , , t i ) + B( , , t i )
i +1 i g (i +1)

where A and B are square matrices having the following forms.

a a12 b b12
A = 11 B = 11
a 21 a 22 b21 b22

It will be a simple exercise to find A and B. With the above equation, the response at any time
step ti +1 can be evaluated based on the known response at ti . Hence, the entire response history
can be obtained by recursive substitutions. Matrices A and B need to be evaluated only once if
t i is a constant. Furthermore, from the equilibrium condition, we can evaluate the relative
acceleration and, thereby, the total acceleration of the structure as follows.

i +1 = g (i +1) 2i +1 2 i +1

t i +1 = i +1 + g (i +1) = ( 2i +1 + 2 i +1 )

3.5 Time-Domain Analysis with Step-by-Step Numerical Integration

While the method presented in Sec. 3.4 is computationally efficient. It can only be applied to
linearly elastic structures as it relies on an analytical closed-from solution. Furthermore, for
multiple-degree-of-freedom structures, it requires the use of mode-superposition analysis, which
is computationally inefficient. For these reasons, step-by-step numerical time integration
methods are most often used for evaluating the linear and nonlinear response of a structure to
general dynamic excitations. They are computationally efficient and can be directly applied to

34
multiple-degree-of-freedom structures without requiring mode-superposition analysis. Hence,
they are the methods chosen for dynamic time history analysis in finite-element analysis
programs.

3.5.1 Time Discretization of Equations of Motion

The basic idea of a step-by-step time integration method is to replace the second-order
differential equation describing the motion of the structure by a set of algebraic equations, which
can be solved easily without resorting to solving the differential equations. A number of
numerical integration schemes are available for this purpose. To formulate such a scheme, let us
consider the response of a structure to the ground motion history shown in Fig. 3.12.

We first discretize the continuous excitation and response time histories into values at discrete
instants of time with small uniform intervals of t and require that the equilibrium condition of
the structure be satisfied at the discrete time instants only, i.e.,

M(ti +1 ) + C (ti +1 ) + K (ti +1 ) = f (ti +1 )

where t i +1 = (i + 1)t for i = 0, 1, 2, 3,...... . Note that equilibrium need not be satisfied within
any time interval between ti and ti +1 .

Fig. 3.12 Ground Acceleration

Next, we approximate the time variation of structural response within each time interval with
simple mathematical expressions. For example, we can approximate the displacement, velocity,
and/or acceleration of a structure at any instant of time, t i +1 , with finite difference expressions or
truncated Taylor series expansions. Alternatively, we can approximate the variation of structural
acceleration within a time interval with a temporal interpolation function similar to the spatial
interpolation used in finite element analysis. We will see later in this section that one can arrive
at identical schemes using different derivation methods. For the purpose of the following
discussion, we use the notation i +1 to represent the approximate numerical solution at time t i +1 ,

35
while (t i +1 ) is used to represent the true solution. The approximate solution still has to satisfy
equilibrium at discrete time instants. Hence, the equilibrium condition in a numerical scheme is
expressed in the following form:

Mi +1 + Ci +1 + K i +1 = f i +1

A scheme is termed explicit if i +1 can be expressed as a function of the response quantities at


the previous time step(s) only without the need to invert a term containing the structural stiffness.
Otherwise, a scheme is termed implicit. The distinction between explicit and implicit schemes is
most obvious if the structural response is nonlinear as it will be seen in the Chapter 4. Some of
the mostly commonly used explicit and implicit integration schemes are presented below.

3.5.2 Explicit Time Integration Schemes

Central Difference Method

In the central difference method, the equation of motion for a linearly elastic SDOF structure is
expressed as

Mi + Ci + K i = f i

Furthermore, the structural velocity and acceleration are approximated by the first and second
central difference equations:

i +1 i 1
i =
2t

i +1 / 2 i 1 / 2 i +1 2 i + i 1
i = =
t t 2

Substituting the above approximations into the equation of motion, we can solve for i +1 as
1
Ct 2 Ct
i +1 = M + t ( f i K i ) + 2 M i M i 1 (3.3)
2 2

This method is explicit since knowing i 1 and i , we can evaluate i +1 right away. It is a two-
step method as the solution at any time step requires the knowledge of the solutions in the two
previous steps. Hence, to start the computation, we need to know the values of both o and 1 .
However, we are normally given the initial conditions o and o only but not 1 . This may
create a problem at the beginning of the computation. There are a number of ways to overcome
this problem. The most rational approach is to define a fictitious response quantity, 1 , for time

36
equal to t and compute its value with the following procedure. Consider the central
difference approximations for o and o as follows.

1 1
o =
2t

1 2 o + 1
o =
t 2

Eliminating 1 by substitution, we can solve for 1 with the above two equations and obtain

t 2
1 = o to + o
2

The value of o in the above equation can be obtained from the equilibrium condition at to :

o =
1
( f o Co K o )
M

Hence, knowing o and o , we can evaluate 1 with the above two equations. With o and 1
known, we can carry out the recursive computation using Eq. (3.3) to find the displacement
response time history. The velocity and acceleration responses can be subsequently computed
with the central difference approximations.

3.5.3 Implicit Time Integration Schemes

Linear Acceleration Method

Fig. 3.13 Linear Acceleration Method

37
In the linear acceleration method, one assumes that structural acceleration varies linearly within a
time interval t as illustrated in Fig. 3.13. Hence, for ti t t1+1 ,


( ) = i + (i +1 i )
t

where = t ti and t= ti +1 ti .

With the above assumption, we can obtain the following expressions through successive
integrations. Integrating the above equation once, we have

2
( ) = i + ( )
i + t i +1 i d = i + i + 2t (i +1 i )

0

which can be integrated again to obtain

2
(i +1 i )d = i + i + i + (i +1 i )
2 3
( ) = i + i + i +
0
2 t 2 6t

At t = ti +1 , we have = t in the above equations. Therefore, from the above equations, we can
obtain the velocity and displacement at t = ti +1 as

t
i +1 = ( ) = i + (i + i +1 )
2
and
t 2
i +1= ( )= i + ti + ( 2i + i +1 )
6

The last two equations constitute the linear acceleration method. Together with the equilibrium
condition, the linear acceleration method comprises the following three equations.

Mi +1 + Ci +1 + K i +1 = f i +1 (3.4a)
t 2
i +1 = i + ti + (2i + i +1 ) (3.4b)
6
t
i +1 = i + (i + i +1 ) (3.4c)
2

Equation (3.4b) shows that i+1 depends on the acceleration at ti +1 . Therefore, it is an implicit
scheme. To use the above equations to compute the response time histories, we need to convert
the formulations into an explicit form by inverting a term that contains the structural stiffness.
To do this, we first rearrange Eq. (3.4b) to obtain

38
6 1
i +1 = i ti t 2i
2 i +1
(3.5)
t 3

Substituting Eq. (3.5) into (3.4c), we have

3 2 1
i +1 = i +1 i ti t 2i (3.6)
t 3 6

Substituting the last two expressions for the acceleration and velocity into Eq. (3.4a), the
equilibrium equation, we can obtain the following algebraic equation, which assumes the same
form as a static load-displacement relation.
~ ~
K i +1 = f i +1 (3.7)

where

~ 6 3
K = 2 M + C+K
t t

~ 6 3 6 t
f i +1 = f i +1 + 2 M + C i + M + 2C i + 2 M + C i
t t t 2

in which K is called the effective stiffness and fi +1 the effective force. Hence, Eq. (3.7) is called
the equivalent static form. Once we have this, we just need to solve the simultaneous equations
just like solving a static problem. In each time step, we can evaluate i +1 by solving Eq. (3.7). It
~
can be seen that the scheme is considered implicit because it requires the inversion of K , which
contains the structural stiffness K , to arrive at an explicit expression for i +1 .

Hence, knowing the initial conditions, i.e., 0 and 0 , of the structure, structural response can be
computed with the following procedure based on the linear acceleration method.

1. Formulate K .
2. Calculate 0 from 0 and 0 using the equation of motion.
3. Set i = 0.
4. Evaluate fi +1 .
5. Calculate i +1 using Eq. (3.7).
6. Calculate i +1 using Eq. (3.5).
7. Calculate i +1 using Eq. (3.6).
8. Set i = i +1, and go to step 4.

39
Constant -Average-Acceleration Method

Fig. 3.14 Constant-Average-Acceleration Method

In the constant-average-acceleration method, one assumes that structural acceleration is constant


within a time interval t and this constant is equal to the average of the accelerations at ti and
ti +1 , respectively, as shown in Fig. 3.14. Hence, for ti t t1+1 , we have

( ) = (i + i +1 )
1
2

where = t ti . Through successive integrations of the above expression, we have

1
( ) = i + (i + i +1 ) d = i + (i + i +1 )

0 2
2
and


(i + i +1 )d = i + i + (i + i +1 )
2

( ) = i + i +
0
2 4

At t = ti +1 , we have = t in the above equations. Therefore, from the above equations, we can
obtain the velocity and displacement at t = ti +1 as

t
i +1 = ( ) = i + (i + i +1 )
2
and
t 2
i +1= ( )= i + ti + (i + i +1 )
4

40
As a result, the constant-average-acceleration method comprises the following three equations.

Mi +1 + Ci +1 + K i +1 = f i +1 (3.8a)
t 2
i +1 = i + ti + (i + i +1 ) (3.8b)
4
t
i +1 = i + (i + i +1 ) (3.8c)
2

To evaluate the response i +1 in any time step, one can rearrange the above three equations into
an equivalent static form using the same procedure as that shown for the linear acceleration
method.

Alternative Derivation

The constant-average-acceleration method can also be derived from the trapezoidal rule, and is,
therefore, also called the trapezoidal method. To show this, we will convert the second-order
differential equation, representing the equation of motion, into a first-order differential equation
as follows. First, let us we define a response vector which contains and as follows.


y=

With this, we can write

y = f (y, f )

where


f (y, f ) = 1
M ( f C K )

The first row of the above matrix formulation represents an identity condition and the second is
the equation of motion.

By integrating the first-order differential equation, we have

ti +1
y (t i +1 ) y(t i ) = f ( y, f )dt
ti

Applying the trapezoidal rule to the above integral, we have

t
y i +1 = y i + (f i + f i +1 )
2

41
where

i +1
= ( y i , fi ) i and
fi f= = fi +1 f ( y=
i +1 , f i +1 )
i i +1

Hence, from the above equation, we have

i +1 i t i + i +1
= +
i +1 i 2 i + i +1

Note that the second row of the above matrix expression is the velocity approximation that is
used in the constant-average-acceleration method. Substituting it into the first row, we obtain the
approximation for the displacement response, resulting in exactly the constant-average-
acceleration method.

Newmark Method

The Newmark method is a general method encompassing a family of different integration


schemes such as the linear acceleration and constant-average-acceleration methods. This method
can be derived by considering the Taylor series expansions of the displacement and velocity
terms at time t i +1 about time t i as follows.

d t 2 d 2 t 3 d 3
( ti +1 ) =
( ti ) + t + +
dt t =ti = 2! dt 2 t t = 3! dt 3 t ti +1
i

d t 2 d 2
( t=
i +1 ) ( ti ) + t +
dt t =ti 2! dt 2 t =t
i + 2

where 0 1,2 1 . The last term in each of the above equations represents the remainder term.
Using the above expansions, we can approximate the displacement and velocity, i +1 and i +1 , as
follows.

t 2
i +1 = i + ti + i + t 2 (i +1 i )
2

i +1 = i + ti + t (i +1 i )

where the terms containing and are essentially the remainder terms since (i +1 i ) / t is
approximately proportional to d 3 / dt 3 . Because the actual values of the remainder terms, or, in

42
other words, the values of 1,2 , are not known, and are parameters to be chosen by the user
to arrive at a good numerical approximation. The choice of their values will be discussed later.

After rearranging the above two equations and adding the equilibrium equation, we have

Mi +1 + Ci +1 + K i +1 = f i +1 (3.9a)

1
i +1 = i + ti + t 2 i + i +1 (3.9b)
2

i +1 = i + t [(1 )i + i +1 ] (3.9c)

which constitute the basic equations for the Newmark method. The specific integration scheme
arrived at depends on the values of and selected. These parameters govern the stability and
accuracy of numerical solutions as will be discussed later.

The method is explicit when is 0. Otherwise, the method is implicit. When = 0 and
= 1 / 2 , the explicit scheme has exactly the same numerical properties as the central difference
method. With = 1 / 2 and = 1 / 6 , it becomes the linear acceleration method. With = 1 / 2
and = 1 / 4 , it becomes the constant-average-acceleration method.

Following the same mathematical manipulations applied to the linear acceleration method, we
can express the Newmark method in an equivalent static form as follows.
~ ~
K i +1 = f i +1 (3.10)

where

~ 1
K= M+ C+K
t 2
t

~ 1 1 1 t
f i +1 = f i +1 + M + C
+ M +
1C +
2 1 M +
2 C i

2 i i
t t t 2

In using the Newmark method, must not be less than 1/2. Otherwise, it will be unstable. When
= 0 , the method becomes explicit and the above equivalent static form becomes invalid as it
contains terms with in the denominator. Therefore, a different but simpler computation
procedure can be used for the explicit form. This can be a simple exercise to derive.

43
-Method (Hilber, Hughes, and Taylor)

The -method of Hilber, Hughes, and Taylor (Hughes 1987) comprises the following three
equations. Note that the approximations for the displacement and velocity are identical to those
in the Newmark method. The only change is in the equation of motion.

Mi +1 + (1 + )Ci +1 Ci + (1 + )K i +1 K i = (1 + ) f i +1 f i (3.11a)

1
i +1 = i + ti + t 2 i + i +1 (3.11b)
2

i +1 = i + t [(1 )i + i +1 ] (3.11c)

In the above equation, is an additional parameter introduced to improve the numerical energy
dissipation properties of the algorithm as will be discussed later. It is recommended that the
parameters satisfy the following conditions to achieve accurate numerical solutions with
minimum errors.

1/ 3 0 (3.11d)

=
1
(1 2 ) (3.11e)
2

=
1
(1 )2 (3.11f)
4

3.5.4 Selection of Integration Time Interval

There are two major considerations in the selection of an integration time interval t . One is the
stability of a numerical solution and the other is accuracy. Both are important considerations to
obtain reliable numerical results. However, numerical errors are inevitable but controllable with
any time integration schemes.

Stability

A time integration algorithm is stable if the numerical solution for a free-vibration response,
under any initial conditions, does not grow without bound. Figure 3.15 shows an unstable
numerical solution.

44
Fig. 3.15 Free-Vibration Solutions with the Central Difference Method

An algorithm is conditionally stable if it is stable only when t t cr , where t cr is the critical


step size. An algorithm is unconditionally stable if it is stable for any value of t . All explicit
integration schemes are conditionally stable, while a number of implicit schemes are
unconditionally stable. The critical step sizes for some commonly used integration schemes are
shown in Table 3.1. The critical step size depends on the damping as well as the natural
frequency ( ) of the structure. The values shown in the table are for zero structural damping. In
general, the increase of structural damping increases the critical step sizes.

The Newark method is unconditionally stable when

1
and
2 2

and it is conditionally stable when

1
and <
2 2

1
When < , the method is always unstable. When the Newmark method is conditionally stable,
2
the critical step size is given by the following expression.

1/ 2
( 1/ 2 ) + / 2 + 2 ( 1/ 2 )
2

tcr =
( / 2 )

45
From the above expression, we can see that when = 1 / 2 , structural damping does not
influence the critical step size. However, when > 1 / 2 , the increase of damping will increase
the critical step size.

Table 3.1 - Critical Integration Step Sizes

Integration Method Type of Method Critical Step Size ( t cr )


Central Difference Explicit 2

Newmark Method Implicit 3.464
1
= , =
1
2 6
(Linear Acceleration)
Newmark Method Implicit Unconditionally
1 1 Stable
= , =
2 4
(Constant-Average-
Acceleration)
Newmark Method Explicit 2
1
= , = 0
2
(Central Difference)
Wilson- Implicit Unconditionally Stable
when 1.37

The -method is unconditionally stable and has a second-order accuracy when

1/ 3 0 , =
1
(1 2 ) and = 1 (1 )2
2 4

Accuracy

A stable numerical solution does not necessarily guarantee an accurate solution. In general, the
smaller the integration time interval, t , is, the more accurate is the solution. One essential
condition for any numerical scheme is that it must be convergent, i.e., the numerical solution
must approach the exact solution as t goes to zero. Otherwise, we cannot use it with
confidence. The rate at which the numerical solution converges to the exact solution is
determined by the order of convergence of the numerical scheme. A scheme has an nth-order
convergence when the magnitude of the numerical errors is proportional to (t T )n , where T is
the natural period of the structure.

46
For the case of SDOF structures, the accuracy requirement, rather than the stability requirement,
determines the integration time interval, t . The required time interval depends on the natural
frequency or period of the structure. As a rule of thumb, for any scheme, t / T should be less
than 1/10 and preferably around 1/20 to obtain an accurate numerical solution. One way to
determine whether a solution is accurate enough is to repeat the analysis with reduced time
intervals until no significant change is observed in the solution.

To use a time integration scheme properly and be able to judge the quality of the numerical
results, it is important to understand the nature of the numerical errors. Numerical errors are
exemplified in the form of: 1) frequency distortion; 2) numerical energy dissipation (artificial
damping); and 3) amplitude amplification in the case of forced-vibration response. The
frequency distortion introduced by the central difference method is shown in Fig. 3.15. It can be
observed that the apparent natural period in the numerical solution shrinks as t increases.
However, the central difference method has no artificial numerical damping. The constant-
average-acceleration method has no artificial damping, but it introduces period elongation rather
than period reduction. It should be noted that artificial damping has the same property as viscous
damping (See Appendix A). The Newmark method has no numerical energy dissipation when

1
=
2

which is the case for the constant-average-acceleration method, and has numerical energy
dissipation when

1
>
2

PSv

No Damping

Some Damping

T
True Period Period Obtained in Numerical Solution

Fig. 3.16 - Influence of Frequency Distortion

47
The impact of frequency distortion on earthquake response analysis is illustrated in Fig. 3.16. As
shown, for undamped structures, an earthquake response spectrum has sharp spikes and troughs
with the spectral value changing quickly with respect to the natural period. Hence, even a small
frequency distortion may have a large impact on the peak response obtained. However, as long
as some damping exists, even if it is only a few percent of the critical damping value, an
earthquake response spectrum will be significantly smoother. In this case, a larger frequency
distortion can be tolerated without having a significant effect on the peak response obtained.
Hence, the selection of an adequate integration time interval also depends on the damping
property of the structure being analyzed.

Further discussion of this topic and an analytical procedure to derive the stability criteria and
quantify the numerical errors in an integration scheme are presented in Appendix A. It is also
shown in the appendix that an amplitude amplification effect can be introduced into a numerical
solution for a forced-vibration response. But this error normally does not dominate as compared
to the other errors.

3.5.5 Conditioning of a Numerical Operation

Round-off errors are inevitable in numerical computations because of the finite resolution of
digital computing. When a numerical operation is ill-conditioned, these errors can be
significantly amplified in numerical solutions. This will occur when an extremely small t is
used in step-by-step integration. To illustrate this problem, let us use the central difference
method as an example. The central difference method is formulated as follows.

Mi + Ci + K i = f i

i +1 i 1
i =
2t

i +1 / 2 i 1 / 2 i +1 2 i + i 1
i = =
t t 2

To find the displacement response at any time ti +1 , the numerical computation can be done in one
step with the following equation, which is derived in Sec. 3.4.2.

1
Ct 2 Ct
i +1 = M + t ( f i K i ) + 2 M i M i 1
2 2

The above formulation is not well-conditioned because it involves operations of the form
[ ]
O (1) + O (t 2 ) . When t is very small, the second-order term may vanish or wiped out by the
first order term, which will result in a significant round-off error. However, this problem can be
resolved by rearranging the computational procedure as suggested in Dahlguist and Bjorck

48
(1974). In this improved procedure, we define a new term z i = ( i i 1 ) / t and divide the
computations into two consecutive steps as follows.

1
Ct Ct
= M + t ( f i K i ) + z i M
2
z i +1
2

i +1 = i + tz i +1

This results in a numerically better-conditioned procedure as O ( t 2 ) no longer exists.


Alternatively, one can use the explicit Newark method with = 1 / 2 , which is mathematically
identical to the central difference method, but is numerically better-conditioned. The comparison
of the numerical results obtained with the central difference method using the conventional one-
step computation and the explicit Newmark method is shown in Fig. 3.17. It can be seen that the
central different method produces an erroneous solution when a very small integration time
interval is used, while the explicit Newmark method produces a good solution.

Fig. 3.17 - Numerical Conditioning ( )

49
3.6 Frequency-Domain Analysis

Frequency-domain analyses are carried out with Fourier transform. It not only provides an
efficient computational method but also good insight to how a structure responds to earthquake
ground motion histories. Basic concepts of Fourier transform are reviewed here.

3.6.1 Fourier Transform

Fourier transform is an extension of the Fourier series concept. As we all know, a periodic
function can be represented by a Fourier series. A non-periodic ground motion time history can
be expressed in terms of a Fourier integral. The nature of Fourier integral can be explained in the
following way. First, assume that the time history is periodic with a period T p as shown in Fig.
3.18. This is achieved by extending the time history with fictitious repetitive functions. The
resulting periodic function can be represented by a Fourier series. Then, by having T p
approaches infinity, the Fourier series become an integral, which is called the Fourier integral,
exactly representing the original ground motion time history. The Fourier integral expression for
a ground motion history is as follows. This can, of course, be used to represent any time history
functions, including displacements and velocities.


g (t ) =
1
2

C ( )e i t d

where


C ( ) = g (t )e i t dt

The above two equations are called the Fourier transform pair. The first equation (Fourier
integral) represents the inverse transform (IT), while the second equation is the forward
transform (FT).

Fictitious

Fig. 3.18 General Ground Acceleration History

50
Physical Interpretation of Fourier Transform

The forward transform of g ( t ) is


C ( ) = g ( t ) e it dt

Since C ( ) is complex, it can assume the following form.

C ( ) C=
= ( ) ei C ( ) cos + i C ( ) sin

From the forward transform expression, we can deduce that


C ( ) cos = g (t )cos tdt


C ( ) sin = g (t )sin tdt

from which values of C ( ) and can be obtained. Hence, the inverse transform can be
expressed as


g (t ) =
1 1
2

C ( )e i t d =
2

C ( ) e i ( t + ) d

From the above equation, we can see that an arbitrary ground motion time history can be
perceived as a summation of an infinite number of harmonic waves with frequencies and
phase angles , where varies continuously from minus infinity to positive infinity. C ( )
represents the amplitude density (i.e., amplitude per unit frequency) of each harmonic wave and
is called the Fourier amplitude. It is evident from the above equation that a ground motion time
history can be completely characterized by Fourier amplitudes C ( ) and phase angles ,
which can be plotted as spectral curves as illustrated in Figs. 3.19 and 3.20.

A Fourier amplitude spectrum provides useful information about the frequency content of an
earthquake ground motion. Phase angle spectra are normally not paid as much attention.
However, a phase angle spectrum dictates the shape of the envelope curve for a ground motion
history as well as other ground motion characteristics such as the duration of pulses present in
the record.

51
Fig. 3.19 Fourier Amplitude Spectrum Fig. 3.20 Phase Angle Spectrum

3.6.2 Differentiation and Integration in Frequency Domain

Differentiation and integration operations can be performed in frequency domain by means of


Fourier transform as shown below.

Transforms of Displacement, Velocity, and Acceleration

As mentioned previously, Fourier transform can be performed on any time histories, including
the displacement, velocity, and acceleration response of a structure.

Fourier Transform Pair for Displacement Response:

1
(t ) V ( ) eit d IT(V )
2
= = (Inverse Transform)


V ( ) = (t )e i t dt = FT( ) (Forward Transform)

Fourier Transform Pair for Velocity Response:

1
= ( t ) = V ' ( ) eit d IT(V ')
2


V ' ( ) = (t )e i t dt = FT( )

52
Fourier Transform Pair for Acceleration Response:

1
( t ) V '' ( ) eit d IT(V '')
2
= =


V ' ' ( ) = (t )e i t dt = FT()

We can perform differentiations on the displacement transform to have


(t ) = V ( ) dt (e )d i V ( )e i t d
1 d 1
=
i t

2 2

( )V ( )e d
d 2 i t
(t ) = V ( )
1 1
( e ) d = 2 i t

2 dt 2
2

By comparing the above two equations to the transform expressions for the velocity and
acceleration, we know that the following relations between the transforms of displacement,
velocity, and acceleration hold.

V ' ( ) = i V ( )

V ' ' ( ) = 2V ( )

With the above relations, one can do differentiation and integration by Fourier transform.

Differentiation

As an example, given (t ) , one can calculate (t ) with the following procedure.

1. Evaluate the forward transform of (t ) : V = FT[ ].

2. Calculate the forward transform of (t ) from V: V ' ' = 2V .

3. Evaluate the inverse transform of V": ( t ) = IT [V ''] .

The same can be done to calculate velocity.

Integration

Similarly, given (t ) , one can find (t ) with the following procedure.

53
1. Evaluate the forward transform of (t ) : V ' ' = FT[(t )] .

2. Calculate the forward transform of (t ) from V '' : V =


1
V '' .
2

3. Evaluate the inverse transform of V: ( t ) = IT [V ] .

3.6.3 Response through Frequency Domain

We can solve the equation of motion to calculate the response of a structure by Fourier
transform. This is the frequency domain approach. Consider the following equation of motion:

+ 2 + 2 = g

Perform forward transform on both sides of the above equation:

[ ]
FT + 2 + 2 = FT[g ]

The result is

2V + i 2 V + 2V = C ( )

which can be expressed as

2 i 2 2
2 + + V = C ( )

2
2 2

Solving the above equation for V, we have

C ( )
V= = C ( ) H ( )
(1 2 ) + i 2
2

where

H ( ) =
1
[(1 2 ) + i 2 ]
2

54
H ( ) is called the complex frequency response function, which can be physically interpreted as
the amplitude of the steady-state response of the structure to a complex harmonic ground motion
g (t ) = eit . Knowing H ( ) , one can evaluate the structural response with the following
procedure.

1. Evaluate the forward transform of the ground acceleration history: C ( ) = FT[g ].

2. Calculate the forward transform of the displacement response: V = C ( )H ( ) .

3. Evaluate the inverse transform of V: = IT [V ] .

The above three steps constitute an analytical procedure via the frequency domain. The meaning
of the term frequency domain is evident from the above procedure. In this procedure, the
random earthquake ground motion time history is decomposed into a set of harmonic waves. The
response to each harmonic wave is then computed and the total response is calculated by
superimposing the responses to these harmonic waves. Since it uses the principle of
superposition, it is only applicable to linearly elastic structures.

Finally, it is interesting to note that the complex frequency response function H ( ) is the
forward Fourier transform of M h(t ) , where

h (t ) = e (t ) sin D (t )
1
(the unit impulse response function)
M D

In other words, the following transformation relation holds:


( ) M h(t )e i ( t ) d (t=
H= ) M h(t )e it dt (3.12)

An illustrative plot of H ( ) is shown in Fig. 3.21.

Fig. 3.21 - Frequency Response Function

55
3.6.4 Discrete Fourier Transforms

For arbitrary ground acceleration time histories, Fourier integrals cannot be evaluated in closed
form. Hence, frequency-domain analysis is often performed numerically using discrete
transforms. Analogous to the continuous transform pair, a discrete Fourier transform pair is
expressed as

N 1
=g ( tm ) =
2 n =0
C (n )ei 2 nm / N IDT(C ) (3.13a)

N 1
t g ( mt )e i 2 nm / N =
C (n ) = DT(g ) (3.13b)
m=0
where

2
=
Tp
n = n
Tp
t =
N
t m = mt

in which T p is the total duration considered in the analysis and N is the number of time intervals
the total duration is divided into. In the discretization, integrals are replaced by summations. One
should note that discrete Fourier transform, similar to the Fourier series concept, retains the
assumption that the ground acceleration is periodic with a finite period T p as shown in Fig. 3.22.

Fictitious
Delay Time Delay Time

Fig. 3.22 Ground Acceleration History

56
The interval between the end of the ground motion and the beginning of the next period is called
the delay time. The longer the delay time is, the longer is the period T p . As the delay time
approaches infinity, discrete Fourier transform becomes the continuous transform.

Furthermore, based on the periodic nature of a harmonic function, we should note that following
relation holds.

e i 2m ( j + N ) / N = e i 2mj / N

N 1
Hence, by defining C n = C ( n ) , with j = -1, -2, -3,......., , and assuming that N is odd,
2
we can deduce from Eq. (3.13b) that

C N 1 = C 1
C N 2 = C 2
C N 1 = C N 1
+1
2 2

Since C j is the complex conjugate of C j , the C n s with n ranging from [(N-1)/2+1] to (N-1)
are simply the complex conjugates of the C n s with n ranging from 1 to (N-1)/2. Therefore, one
cannot distinguish the components in the series in Eq. (3.13a) with frequencies [(N-1)/2+1]
and above from the lower frequencies, and the highest frequency that can be captured in a
discrete transform is

N 1
N 1 =
2
2

with is called the Nyquist frequency. The above phenomenon is called the aliasing effect. As a
result, in performing discrete transform, the sampling frequency should always be at least twice
as large as the highest frequency we want to retain in the sample. A Fourier amplitude spectrum
obtained with discrete transform is illustrated in Fig. 3.23. Because of the aliasing effect, there is
no need to plot the spectrum beyond the frequency value of N 1 . Beyond N 1 , the spectral
2 2
curve simply provides a mirror image of the first portion as shown in the figure.

Discrete transform is always performed with a Fast Fourier Transform (FFT) algorithm, which is
available in computational tools like Matlab and even Microsoft Excel. How an FFT algorithm
works and its computational efficiency are explained in Appendix B.

57
Fig. 3.23 - Fourier Amplitude Spectrum

In using a computational tool like Matlab, it is important to realize that inverse and forward
discrete Fourier transforms can be expressed in a generic form as

N 1
Bm = AnW Nnm
n =0

for m = 0, 1, 2,........., N-1, where

W N = e i 2 / N

A positive sign in the exponential term represents an inverse transform, while a negative sign
represents a forward transform.

3.6.5 Frequency Domain Analysis using Discrete Transform

In theory, frequency-domain analysis using discrete transform can be conducted in the same
manner as using continuous transform. Nevertheless, from Figs. 3.21 amd 3.23, one can see that
the direct multiplication of C ( n ) with the continuous function H ( n ) over the frequency range
from 0 to N 1 will create a problem. One has the mirror image introduced by the complex
conjugate pairs while the other does not. Hence, over the range of n from 0 to (1), an
important feature (the mirror image of one another) of discrete transform is lost for the product
V ( n ) = C ( n ) H ( n ) . As a result, we can no longer perform an inverse transform to calculate
(t ) . A number of methods are available to circumvent this problem. One simple method is to
construct H ( n ) in such a way that it has complex conjugates when the frequency is beyond
N 1 . This can be achieved by defining H ( n ) as follows.
2

58
H ( n ) =
1
[(1 ) + i 2 n n ]
2 2
n

where
n
n = for 0 n ( N 1) / 2

( N n)
n = for ( N 1) / 2 < n ( N 1)

with

2
=
Tp

Another method is to replace the function H (n ) by its discrete counterpart, which can be
computed by FFT as follows by virtue of the relation shown in Eq. (3.12).

N 1
H (n ) =t M h ( mt )e i 2 nm / N
m=0

Similarly, if one wants to compute the velocity and acceleration using discrete inverse transform,
one has to assure that the mirror image will exist in V (n ) and V ''(n ) by defining

V ' (n ) = inV (n )

V '' (n ) = n 2V (n )

where

n = n for 0 n ( N 1) / 2

n =( N n) for ( N 1) / 2 < n ( N 1)

3.6.6 Accuracy of Response Evaluation using Discrete Transforms

Since discrete Fourier transform is based on the assumption that the ground acceleration time
history, g (t ) , is periodic with a finite period T p , the response obtained with discrete transform
will not necessarily represent the true response of the structure but rather the sum of steady-state
responses to a finite set of harmonic wave components. This may result in a solution that is not
accurate and does not satisfy the initial conditions as illustrated in Fig. 3.24 due to the fact that
some of the transient response in the solution is not captured. The numerical solution will,

59
however, approach the exact response as the delay time t d , shown in Fig. 3.24, goes to infinity.
The increase of the delay time reduces the frequency gap in the harmonic wave components in
the series and, thereby, leads to a better account of the transient response. In theory, an infinite
delay time is required to recover the exact response. Under this condition, we essentially have
the continuous transform. However, when a structure has some damping, a finite delay time will
suffice to obtain a reasonably accurate solution. In general, the delay time required depends on:

1. the structural frequency in that the higher the structural frequency is, the shorter is the delay
time required; and

2. the structural damping in that the larger the damping is, the shorter is the delay time required.

To determine an appropriate delay time, we can use a trial-and-error approach by gradually


increasing the delay time until the solution converges to the exact solution.

Fictitious
Part

Free
Vibrations Exact ( )

Numerical
Solution

Fig. 3.24 - Errors Introduced by Discrete Transforms

60
Correction Method

When damping is zero or very close to zero, the delay time required may be undesirably long. In
this case, we can still use a short delay time or zero delay time, and correct the numerical
solution with a correction function e(t), which, as illustrated in Fig. 3.24, is defined as

e( t ) = e ( t ) ( t )

in which e (t ) is the exact response and (t ) is the response obtained with discrete transforms.
The error function or correction function e(t) represents the transient (i.e., the free-vibration)
component that is missing in the numerical solution and can be determined from the analytical
expression for the free-vibration response of the structure, i.e.,

e(t ) = e(0) f (t ) + e(0) g (t )

where


f (t ) = e t cos D t + sin D t
1 2
1
g (t ) = e t sin D t
D

The functions f(t) and g(t) represent the free-vibration responses induced by an initial
displacement and velocity, respectively. The values of e(0) and e(0) are the errors in the initial
displacement and velocity, which can be identified and obtained from the numerical solution.

3.7 Additional Information Conveyed by Response Spectra and Fourier Amplitude


Spectra

3.7.1 Elastic Response Spectra

Elastic response spectra provide information on the peak responses induced by the earthquake
ground motion on linearly elastic structures for a range of structural frequencies and damping
levels. Hence, it is not surprising to note that the pseudo spectral velocity is related to the square
root of the maximum energy absorbed by the structure divided by the structural mass during the
earthquake excitation. This can be shown as follows. The maximum energy absorbed by a
structure during an earthquake excitation is

1
E max = KS d2
2

Hence, the maximum energy absorbed per unit mass is

61
1 K 2 1 2 2 1
m
E max = S d = S d = PS2
2M 2 2

By rearranging the above equation, we have

PS = 2 E max
m

3.7.2 Fourier Amplitude Spectra

A Fourier amplitude spectrum reflects the amplitudes of different harmonic waves in an


earthquake ground motion. Its shape can be close to the pseudo-velocity response spectrum for
undamped structures, as illustrated in Fig. 3.25. Furthermore, it related to the square root of the
energy in the undamped structure divided by the structural mass at the end of the earthquake
ground motion. This can be shown as follows.

Let us consider an earthquake ground motion with a duration t . Then, the Fourier amplitude can
be expressed as

2 2
C ( ) = g (t ) cos tdt + g (t )sin tdt
t t

0 0

The total energy in an undamped structure at time equal to t is

1 1
E= M 2 (t ) + K 2 (t )
2 2

Hence, the total energy per unit mass is

1 1 K 2 1 1
E m = 2 (t ) + (t ) = 2 (t ) + 2 2 (t )
2 2M 2 2

Based on Duhamels integral, the response of an undamped structure at time t can be expressed
as

(t ) = ( )sin (t )d
1 t


g
0

()
t = g ( )cos (t )d
t

62
Fig. 3.25 Comparison of Fourier Amplitude with Pseudo-Velocity Spectra

Substituting the above expressions into the equation to calculate the total energy per unit mass,
we have

1 t + t ( )sin (t )d
2 2
( )
2 0 0 g
Em =


g cos ( t ) d

1 t 2
t
2
= 0 g ( ) cos d + 0 g ( )sin d
2

Hence, we can see that

2 E m = C ( )

For the free vibration after t = t , we have

63
1 2
E m = max
2

where max is the peak velocity that occurs during the free vibration. Substituting the above
relation into the second last equation, we have

max = C ( )

Since max and PS v can be close to each other when damping is zero, the value of C ( ) can
be close to PS v for zero damping. Furthermore, since PS v is related to the maximum energy
absorption by a structure during excitation, it is expected that PS v > C ( ) , as shown in Fig.
3.25.

3.8 Intensity Measures of Earthquake Ground Motions

Engineers are interested in identifying the damage potential of an earthquake ground motion.
However, there is no single measure of the damage power of an earthquake ground motion that is
applicable to structures of all types with different properties. In this section, we will consider
some of the commonly used methods for assessing the level of impact of an earthquake ground
motion on structural response and damage.

3.8.1 Peak Ground Accelerations/Long-Duration Pulses

Even though peak ground accelerations are related to the maximum seismic forces that can be
delivered to a structure, they are not good indicators of the damage power of earthquake ground
motions as there are other factors that affect the response of a structure. For example, high-
amplitude acceleration spikes contributed by high-frequency waves (e.g. see Santa Monica
record in Fig. 2.9) will not be as destructive as long-duration lower-amplitude acceleration pulses
(e.g., see Fig. 2.10) as they will not carry a lot of input energy. Strong, long-duration pulses that
are characteristic of near-fault ground motions can be very damaging, especially to structures
with short periods as well be further discussed in Chapter 4.

3.8.2 Effective Peak Accelerations (EPA)

Since the actual peak acceleration does not properly reflect the damage effect of an earthquake
ground motion, effective peak acceleration (EPA) is often defined and considered in design.
Effective peak acceleration can be obtained, for example, from a ground motion record by
filtering out the high-frequency, high-amplitude spikes. There is no unique definition or method
to calculate EPA. However, the ultimate goal is to have ground motion records with the same
EPA have more or less the same damage power.

64
Fig. 3.26 Root-mean-Square Acceleration Calculated from the El Centro Record from
the 1940 Imperial Valley Earthquake

65
3.8.3 Root-Mean-Square Accelerations

The root-mean-square acceleration is defined as

1/ 2
1 td
= (t )dt
2
a rms g
td 0

It is a good indicator of the damage power of an earthquake ground motion and can be
interpreted as the rate of the frequency ensemble work. The frequency ensemble work represents
the sum of the work done by a group of structures that have a unit mass and have frequencies
ranging from 0 to infinity. It, therefore, represents the effect of an earthquake ground motion
over a range of structural frequencies. The main problem of evaluating a rms is to have a
consistent definition of integration limits. Trifunac (1975) proposed that

1/ 2
1 t95
= (t )dt
2
a rms g
td t5

with

t d = t95 t5

in which t 95 and t 5 are the time instants when the integral of the squared acceleration reaches
95% and 5% of the maximum valve, respectively, as illustrated in Fig. 3.26.

3.8.4 Response Spectrum Intensity (Housner)

To assess the intensity of an earthquake over a range of structural periods, Housner defined the
response spectrum intensity as follows.

T = 2.5
SI ( ) = PS (T , )dT
T = 0.1

where T is the natural period of a structure and is the damping ratio. This is illustrated in Fig.
3.27. It reflects the overall effect of a ground motion over a wide range of structural frequencies.
It is a good measure of ground motion intensity but may not be a good indicator of the damage
potential as it only considers the elastic response of a structure.

66
SI

0.1 2.5 T (sec.)

Fig. 3.27 Response Spectrum Intensity

References:

1. Chopra, A.K. (2011). Dynamics of Structures: Theory and Applications to Earthquake


Engineering, 4th Ed., Prentice Hall.
2. Clough, R.W. and Penzien, J.P. (1993). Dynamics of Structures, 2nd Edition, McGraw-Hill.
3. Bathe, K. (1982). Finite Element Procedures in Engineering Analysis, Prentice-Hall.
4. Hughes, T.J.R. (1983). Analysis for Transient Algorithms with Particular Reference to
Stability Behavior, in Computational Methods for Transient Analysis, T. Belyschko and
T.J.R. Hughes (Ed.), North-Holland.
5. Hughes, T.J.R. (1987). The Finite Element Method, Prentice-Hall.
6. Zienkiewicz, O.C. and Taylor, R.L. (1991). The Finite Element Method, 4th Edition, Vol. 2,
McGraw-Hill.
7. Argyris, J. and Mlejnek, H. (1991). Dynamics of Structures, North-Holland.
8. Dahlquist, G. and Bjorck, A. (1974). Numerical Methods, Prentice-Hall.
9. Bozorgnia, Y. and Campbell, K.W. (2004). Engineering Characteristics of Ground Motion,
in Earthquake Engineering from Engineering Seismology to Performance-Based
Engineering, Bozorgnia and Bertero (Eds.), CRC Press.

67
CHAPTER 4

SEISMIC RESPONSE OF NONLINEAR


SDOF STRUCTURES

As discussed in Chapter 3, the seismic response of a linearly elastic structure is governed by its
damping and natural period only. However, when a structure deforms beyond its elastic limit
under earthquake loading, its response and performance depend strongly on its strength and
ductility in addition to its period and energy dissipation properties. As will be discussed in this
chapter, ductility is an important consideration for the seismic design of a structure.

4.1 Strength and Ductility

Ductility is defined as the ability of a structure to undergo inelastic deformation. As shown in


Fig. 4.1, a brittle structure is one that will fail soon after reaching its elastic limit, while a ductile
structure can sustain a significant amount of inelastic deformation before collapse. For seismic
resistance purpose, it is desirable to have high ductility to avoid sudden failure or collapse. Since
it is practically not feasible to have a structure that will remain elastic all the way to a maximum
perceivable ground motion, we often have to rely on its ductility for safety and collapse
prevention. The basic philosophy of seismic design is that a ductile structure should be allowed
to deform more than a less ductile structure, and thus it can have lower design strength for the
same level of earthquake force.

Restoring
Force

Ductile

Brittle

Structural Displacement
Fig. 4.1 - Structural Ductility

Concrete and masonry are brittle materials, whereas mild steel is ductile. The use of reinforcing
steel in concrete and masonry structures is to enhance the ductility of these structures. However,
either over-reinforcement or under-reinforcement could lead to brittle behavior.

Figure 4.2 shows the behavior of reinforced masonry walls subject to in-plane cyclic
displacement reversals. The plots of the lateral load applied at the top of a wall against the top
68
displacement are called the hysteresis curves. It can be seen that the wall that was dominated by
flexure had a more ductile response that the one dominated by diagonal shear cracks. Introducing
too much flexural reinforcement or not enough shear reinforcement can lead to brittle shear-
dominated behavior. The capacity design method is to protect a structural member from this kind
of brittle behavior.

Flexure Dominated Wall Shear Dominated Wall

Fig. 4.2 - Reinforced Masonry Walls

4.2 Energy Absorption and Dissipation

The ability of a structure to absorb and dissipate energy has a significant implication on its
seismic performance. These properties are illustrated in Fig. 4.3. The amount of energy absorbed
69
by a structure at a certain deformation level is the total area under the load-deformation curve
between zero deformation and the maximum deformation reached. Part of this energy can be
recovered upon unloading. This is the elastic strain energy recovery. The energy dissipated by a
structure is the area under the load-deformation curve for one complete load cycle. The
irrecoverable energy is consumed by the structure through inelastic deformation. This energy
dissipation is often referred to as hysteretic damping or structural damping. A ductile structure is
able to absorb more energy than a brittle structure but it will not necessarily dissipate more
energy. To reduce structural response under earthquake loads, it is desirable to have a good
hysteretic energy dissipation capability, which is quantified in terms of the size of the area under
the load-displacement curve.

Force, r

Total Energy
Absorbed at
Plastic
Deformation Elastic Strain
Energy

Displacement,

Total Hysteretic Energy


Dissipated in One Cycle
(Hysteretic Damping)
Plastic Deformation

Fig. 4.3 Force-Displacement Hysteresis Curve

4.3 Ductility Factors

Ductility factors are defined to quantify the ability of a structure to withstand inelastic
deformation as well as the deformation demand of an earthquake excitation. A structure is safe
as long as the demand is less than the capacity. However, there is no consensus as to the best
way to quantify ductility. In fact, a number of ductility definitions have been used in research
and practice.

70
4.3.1. Ductility based on Maximum Displacement

For a structure with an ideal bilinear force-displacement relation as those shown in Fig. 4.4, the
ductility capacity can be defined as the ratio of the maximum displacement that can be sustained
by the structure to the displacement at first yield, i.e.,

max

y

This ductility factor is a dimensionless index that reflects the maximum deformation that can be
sustained by a structure as compared to its displacement at first yield, which is called the yield
displacement. It is easy to calculate.

Failure
Failure

Fig. 4.4 Bilinear Force-Displacement Relations

For the case of ductility demand, the same expression shown above can be used but the quantity
max becomes the maximum structural displacement induced by the earthquake load. For design,
we have to be sure than the demand is less than the capacity.

However, structural damage is not necessarily determined by the level of peak displacement
alone. It can also be related to the accumulation or history of inelastic deformation, such as the
total plastic deformation accumulated in the structure or the number of large-amplitude
displacement cycles the structure has been subjected to. Hence, it is desirable to generalize the
ductility definition.

Under cyclic displacement reversals, the maximum displacement response max may not reflect
the actual maximum inelastic deformation experienced by the structure because of the residual
plastic deformation p that can develop, as shown in Fig. 4.5. For the displacement cycle i
shown in Fig. 4.5, it is more appropriate to consider ia,max as the maximum displacement
excursion experienced by the structure than to use i ,max since it is the actual amount of plastic
deformation that will be of concern. If we are interested in the absolute maximum only, then the
ductility demand can be defined as

71
max ia,max
i

where max ia,max represents the absolute maximum displacement excursion experienced by the
i

structure throughout its entire response history (i.e., over all cycles i ). Evidently, this measure of
ductility is possible only if we have the force-displacement relations for the structure throughout
the response history.

Fig. 4.5 Displacement Cycle i

4.3.2. Ductility based on Cumulative Plastic Deformation

For structures that exhibit strength degradation under cyclic displacement reversals, as shown in
Fig. 4.2, the level of damage induced does not depend on the maximum displacement alone but
also on the cumulative plastic deformation sustained. The cumulative plastic deformation
developed in the displacement cycle i shown in Fig. 4.5 is

ip ip ip

Considering the total cumulative plastic deformation developed throughout the entire response
history, we can define the ductility capacity as

1

y

i
i
p

where the summation is over all displacement cycles up to failure. This expression can also be
used to quantify the ductility demand imposed by earthquake loading if one knows the response
72
time history and the load-displacement relation for the structure. Unfortunately, such information
is often not available for actual structures subjected to earthquake loads. Therefore, this ductility
index is only suitable for assessing the demand and capacity in analytical studies using numerical
simulations.

4.3.3. Ductility based on Total Hysteretic Energy

Another way of taking the entire response history into account is to consider the total hysteretic
energy dissipation. The total hysteretic energy dissipated before failure can be expressed as

E T E i
i

where Ei is the hysteretic energy dissipated in cycle i. Based on this, a dimensionless ductility
factor can be defined as

ET

1
ry y
2

where the denominator represents the elastic energy absorption capability of the structure.

4.3.4. Equivalent Elastic-Perfectly Plastic Systems

Most structures do not exhibit an ideal bilinear load-displacement relation even if the material for
the structure is elastic-perfectly plastic. This can be due to the progressive plasticization across a
section of a beam or column during bending and the sequential development of plastic hinges at
different locations in the structure, as shown in Fig. 4.6. It goes without saying that most
construction materials, including steel, do not possess a bilinear stress-strain relation.

2 2 r

1 1

Fig. 4.6 Progressive Yielding

73
In this situation, the yield displacement y can be defined in a number of ways. The simplest
way is to use the displacement at which yielding first occurs as shown in Fig. 4.7. However, this
may result in a very small yield displacement and, thereby, a very large ductility measure. This
will happen to the ductility measure for reinforced concrete structures if we use the first yield of
the reinforcement to define the yield displacement.

r
Failure

First Yield

Fig. 4.7 Ductility based on First Yield

To have an appropriate measure of ductility, it is often desirable to idealize the real behavior of a
structure by an equivalent elastic-perfectly plastic curve. Different alternatives to construct an
equivalent elastic-perfectly curve are shown in Fig. 4.8.

The idealized elastic-perfectly plastic curve shown in Fig. 4.8(a) has its elastic stiffness equal to
the initial stiffness of the original structure, while the one shown in Fig. 4.8(b) has its stiffness
based on the equal energy absorption criterion. Figure 4.8(a) is less conservative in the sense that
it over-estimates the energy-dissipation capability of a structure. The elastic stiffness of the
idealized system shown in Fig. 4.8(c) is given by the secant line through point P, which can be
defined as the point at which the first yield occurs or at which the load reaches a percentage (say
75%) of rmax .

To quantify the ductility capacity, we need to determine the maximum displacement, max , that
can be sustained by a structure without severe damage or collapse. A conservative approach is to
have the displacement that corresponds to the peak strength as max , as shown in Figs. 4.8(a) and
(b). Except for brittle structures, this approach is a bit too conservative. Very often, a structure is
still safe and operable even if it has been deformed to pass its peak strength. Hence, we may
consider the displacement at a post-peak strength of rmax as max , as shown in Fig. 4.8(c). The
value of should depend on the type of structure and the level of conservatism desired. For
most structures, such as ductile steel, reinforced concrete and masonry structures, can be
taken to be 0.8, and the yield strength of the equivalent elastic-plastic structure can be selected to
be somewhere between rmax and rmax to have more or less the same amount of energy
dissipation as the original structure. Among the three alternatives shown in Fig. 4.8, the one in
Fig. 4.8(c) is most commonly used.

74
r Actual behavior
r Same Area (Energy
dissipation)

(a) (b)

(c)

Fig. 4.8 Equivalent Elastic-Perfectly Plastic Structures

4.4 Response to Pulse Excitations

In general, the inelastic response of a structure to dynamic loads can only be evaluated
numerically using step-by-step integration. However, for some simple cases, analytical solutions
are attainable. We will consider two such cases, in which the base motions are short-duration and
long-duration acceleration pulses, respectively, and the structure assumes a simple elastic-
perfectively plastic behavior. These cases illustrate the importance of structural ductility in
seismic performance and the damaging power of near-fault ground motions, which are often
characterized by large velocity and acceleration pulses. In these two examples, the structure is
undamped and has an elastic-perfectly plastic behavior, as shown in Fig. 4.9. Note that the period
of the structure is defined in terms of its elastic stiffness as shown in the figure.

75
r K
M 1

Fig. 4.9 An Elastic-Perfectly Plastic Structure

4.4.1. Short-Duration Acceleration Pulse

Let us first consider the case of a short-duration acceleration pulse, as shown in Fig. 4.10(a). A
pulse is considered to have a short duration when its duration is much less than the natural period
T of the structure. A sketch of the response time history of the undamped structure is shown in
Fig. 4.10(b), which can be derived analytically using the momentum and conservation of energy
principles.

Assume that

t
(a)

r
T t (c)
(b)
Fig. 4.10 Response to Short-Duration Pulse

76
The maximum displacement response max induced by the pulse can be calculated with the
principle of conservation of energy. Let us first consider the following equation of motion.

M r f eff

where f eff Mg (dropping the minus sign for convenience). Multiplying both sides of the
equation by and integrating each side with respect to time from t d to t m , we have

tm tm tm
td
M dt r dt f eff dt
td td

The above equation leads to

1 1
M 2 (t m ) M 2 (t d ) Wr (t m ) Wr (t d ) 0
2 2

where Wr (t ) represents the cumulative work done by the restoring force r at time t. The last
equation simply states that the energy of the system is conserved. The first term on the left-hand
side of the equation is zero since velocity must be zero at t m . For a short-duration pulse, Wr ( t d )
is so small that it can be assumed zero.

Hence, we have

1 1
M 2 (td ) Wr (tm ) ry (max y )
2 2

If we simply integrate the equation of motion above with respect to time t from 0 to td , we have
the following momentum principle.

f r dt (td )
I
td
M (td ) eff
0 M

With r being negligible within the short duration t d , the term on the right-hand side of the first
expression above can be defined as the impulse induced by the effective force and is denoted by
I. Combining the last two equations, we have

I2 1
ry (max y )
2M 2

Rearranging the above equation, we have

I2 y
max
2 Mry 2

77
Hence, the ductility demand on the structure is

max I2 1 I 2 2 1
D
y 2 Mry y 2 2ry2 2

The last expression is based on the fact that y ry / K . On the other hand, if the ductility capacity
C of the structure is known, we can replace D in the above equation by C and rearrange the
equation to calculate the yield strength required to resist the pulse loading:

I
ry
2 C 1

If the structure were so strong that it would remain linearly elastic, the conservation of energy
requires that
2
I2 1 1 rmax
rmax max
e

2M 2 2 K

in which rmax and max


e
would be the maximum force and displacement developed by the linearly
elastic structure. Hence, we have

rmax I

We can now compare the strength required for an elastic-perfectly plastic structure to the
maximum restoring force that would be developed if the structure were to remain linearly elastic
under the same acceleration pulse. To this end, we define the strength ratio ry / rmax , which
can be expressed as a function of the ductility capacity based on the above equations.

ry 1

rmax 2C 1

The above expression tells us that the more ductile the structure is, the lower will be the yield
resistance required. One can also find the ductility demand on the structure for a given strength
ratio by rearranging the above equation and replacing C by D .

1 1
D 2 1
2

Table 4.1 shows the values of the ductility demand for different values. The relation between
the strength ratio and displacement demand is sketched in Fig. 4.11.

78
Table 4.1 Values of Ductility Demand with Different Strength Ratios for a Short-Duration
Pulse

1/2 1/3 1/4 1/5 0

D 2.5 5 8.5 13

K
1

Fig. 4.11 Relation between Strength and Displacement Demand

4.4.2. Long-Duration Acceleration Pulse

A pulse is considered to have a long duration when the duration of the excitation is substantially
greater than the natural period T of the structure. Let us consider a special case that the level of
the acceleration pulse remains more or less constant over a long duration as shown in Fig.
4.12(a). The response time history is sketched in Fig. 4.12(b) and can be derived analytically in
the same way as discussed above.

Within the duration of the pulse, the equation of motion can be written as follows.

M r f eff Mgo

Rearranging the above equation, we have

1
( f eff r )
M

79
From the above expression, it can be seen that if f eff ry , the acceleration will never be
negative. As a result, max as t d . Therefore, to have a finite max , we must have
f eff ry . This is considered to be the case here.

Assume that

t
(a)

r
T t (c)
(b)

Fig. 4.12 Response to Long-Duration Pulse

To find max , we multiply both sides of the equation of motion by and integrate each term
with respect to time from time equal to 0 to t m to obtain:

tm tm tm

0
Mdt rdt f eff dt
0 0

Since (tm ) = 0, the above equation can be simplified to

Wr (tm ) f eff max

where Wr ( t m ) is the cumulative work done by the restoring force. Hence, we have

1
ry max y f eff max
2

The above equation can be rearranged to give the required yield resistance in terms of the given
ductility capacity C as follows.
80
f eff
ry
1
1
2 C

Or, it can be rearranged to give the ductility demand for a given yield resistance.

1
D
f
2 1 eff
ry

If the structure were to remain linearly elastic, the conservation of energy requires that

1
f eff max
e
rmax max
e

Hence, the minimum resistance that would be required (i.e., the maximum force that would be
developed) if the structure were to remain linearly elastic would be:

rmax 2 f eff

To compare with the resistance required for an elastic-perfectly plastic structure that has a
ductility capacity C , we can again express the strength ratio as a function of C :

ry 1

rmax 1
2
C

One can also find the ductility demand on the structure for a given strength ratio by rearranging
the above equation and replacing C by D .

1
D
1
2

Table 4.2 shows the values of the ductility demand for different values.

81
Table 4.2 - Values of Ductility Demand with Different Strength Ratios for a Long-
Duration Pulse

3/4 3/5 1/2

1.5 3.0

Comparing the values in the above table to those in Table 4.1, we can see that a long-duration
pulse is much more demanding than a short-duration pulse for a structure that has limited
strength. For a long-duration pulse, reducing the strength of the structure can lead to a significant
increase in ductility demand. This example illustrates that near-fault earthquake ground motions
that have large long-duration pulses can be very destructive. This characteristic of near-fault
ground motions is discussed in Section 2.4 and PowerPoint slides. Figure 4.13 shows additional
examples of such ground motions.

Pulses
Derived Pacoima Dam Record
Pacoima Dam Record
(at base rock)

Fig. 4.13 Pacoima Dam Records from 1971 San Fernando Earthquake

4.5 Step-by-Step Integration of Nonlinear Equations of Motion

For a nonlinear structure, the most efficient and accurate way to evaluate its dynamic response to
an earthquake ground motion is to use a step-by-step integration method, which has been
discussed in Chapter 3 for linearly elastic structures. For nonlinear structures, the computational
procedures with explicit and implicit integration methods are quite different. Implicit methods
require the inversion of the structural stiffness, which may change from one time step to the next
for a nonlinear structure. Hence, the use of an implicit method for a nonlinear structure requires a
Newton-Raphson iterative approach, which is not needed for an explicit method.

4.5.1. Explicit Newmark Method

For a nonlinear structure, we can express the equation of motion as follows.

82
Mi1 Ci1 ri 1 f i 1

in which ri 1 is a nonlinear function of the displacement i 1 and the damping coefficient C is


assumed a constant. The hysteretic damping of a structure due to inelastic deformation is
accounted for in the load-displacement ( r ) relation. For an inelastic structure, ri 1 will not
only depend on the current displacement but also on the displacement history of the structure.
We assume that such a relation is known and can be represented by an appropriate constitutive
model.

For the explicit Newmark method, we have

t 2
i 1 i ti i
2
i 1 i t 1 i i 1

Substituting the last equation above into the equation of motion, we can solve for i 1 :

i 1 M tC 1 f i 1 ri 1 Ci t 1 i

The computational procedure can thus be formulated as follows.

1. Initialize the values of 0 , 0 , and 0 ; and set i = 0.

t 2
2. Calculate: i 1 i ti i
2

3. Calculate ri 1 from i 1 based on the given load-displacement relation of the structure.

4. Calculate: i 1 M tC
1
f i 1 ri 1 Ci t 1 i .
5. Calculate: i 1 i t1 i i1 .

6. Set i = i + 1.

7. Go to step 2.

Note that the response can be computed directly without any iteration.

4.5.2. Implicit Time Integration Methods

For nonlinear problems, an implicit integration method has to be used with a Newton-Raphson
iterative scheme. Such a solution procedure is illustrated below with the -method.

83
-Method

The -method can be expressed as

Mi 1 (1 )Ci 1 Ci (1 )ri 1 ri (1 ) fi 1 fi (4.1a)

i 1 ~i 1 t 2 i 1 (4.1b)

i 1 ~i 1 ti 1 (4.1c)

where

1
~i 1 i ti t 2 i (4.1d)
2

t 1
~ (4.1e)
i 1 i i

The above equations indicate that to evaluate the response i 1 , we have to know the restoring
force ri 1 , which, however, depends on i 1 . Hence, the response i 1 can only be obtained with
an iterative scheme using successive corrections. For example, first, we can make a guess of the
value of i 1 . Then, we calculate ri 1 based on the given load-displacement relation for the
structure. With the ri 1 obtained, we can revise our initial trial value for i 1 . This procedure can
be repeated until the above three equations (4.1a-4.1c) as well as the load-displacement relation
for the structure are satisfied. To have a fast convergence, a more systematic procedure has to be
used to carry out the above calculations. Such a procedure is presented below.

Consider that we have obtained the converged response quantities at step i, i.e., i , i , and i
that satisfy the equation of motion. We are now to find the solution at step i+1, i.e., the values of
i 1 , i 1 , and i 1 that will exactly satisfy Eq. (4.1a), by iterative corrections. At an iteration step
k, we have trial solution values, ik1 , ik1 , and ik1 , that do not satisfy the equation of motion.
Hence, for step k, we can write

Mik1 (1 )Cik1 Ci (1 )ri k1 ri (1 ) fi 1 fi Rik1 (4.2)

in which ri k1 is evaluated from the trial displacement ik1 , and Rik1 is the residual force error. If
we subtract this equation from the exact equilibrium condition, i.e., Eq. (4.1a), we have

Rik1 Mik1 (1 )Cik1 (1 )rik1 (4.3)

where

ik1 i 1 ik1
84
ik1 i 1 ik1

ik1 i 1 ik1

rik1 ri 1 rik1

1
1

Fig. 4.14 Secant and Tangent Stiffnesses

If we know the exact secant stiffness K sk,i 1 of the structure, which is defined in Fig. 4.14, we can
solve the above equation to obtain the solutions i 1 , i 1 , and i 1 as follows. As shown in Fig.
4.14,

rik1 K sk,i 1 ik1 (4.4)

Furthermore, based on Eqs. (4.1b) and (4.1c), we have the following expressions for iteration
step k.

ik1 ~i 1 t 2 ik1

ik1 ~i 1 tik1

Subtracting the above two equations from Eqs. (4.1b) and (4.1c), respectively, which represent
the exact numerical solutions, we have

85
ik1
k
i 1 2
t
(4.5)

k
i 1 t k
i 1 k
i 1
t

Substituting Eqs. (4.4) and (4.5) into Eq. (4.3), we have

ik1
R k
i 1 M 2 (1 )C ik1 (1 ) K sk,i 1ik1
t t

Rearranging the above equation, we have the following equivalent static form.
~
Kik1ik1 Rik1 (4.6a)

where

(1 )
C 1 K sk,i 1
~ 1
K ik1 M (4.6b)
t
2
t

If K sk,i 1 is known, we can solve the above equation for ik1 to obtain the exact correction.
Unfortunately, since we do not know i 1 a priori, we cannot evaluate the secant stiffness K sk,i 1 .
Hence, a Newton-Raphson iterative method is required. Like for any static problems, solving Eq.
(4.6) with a Newton-Raphson method does not require the exact secant stiffness, and can use the
tangent stiffness or even the initial stiffness instead. In that case, the correction ik1 in one trial
will not be exact and we have to repeat this process until ik1 approaches zero. Of course, the
choice of the stiffness quantity for the iterative corrections will affect the rate of convergence of
the trial solutions. A good estimate of K sk,i 1 will be the tangent stiffness K tk,i 1 , which is also
shown in Fig. 4.14. The original Newton-Raphson method uses the tangent stiffness. When the
initial stiffness of the structure is used in the iterative corrections, the method is called the
modified Newton method. This will result in a slower convergence but can be computationally
more robust when the structure has a strain-softening behavior.

Computation Procedure Based on the Newton-Raphson Method

We can summarize the iterative solution procedure as follows. Even though the tangent stiffness
is used here, the solution procedure with the initial stiffness will be the same.

1. Set i 0 ; initialize 0 , 0 , r0 , and 0 .

1 t 1 .
2. Evaluate ~i 1 i ti t 2 i and ~i 1
2
i i

86
3. Set k 0 and select a trial solution by assuming that

ik1 i 1
ik1 i 1
ri k1 r (ik1 )
ik1 0

4. Evaluate the tangent stiffness K tk,i 1 based on the given load-displacement relation (or
constitutive law) for the structure (structural members) and then the effective stiffness
K i 1 .

(1 )
C 1 K tk,i 1
~ 1
K ik1 M
t
2
t

5. Evaluate the residual error (with Eq. (2)).

Rik1 (1 ) fi 1 fi Mik1 (1 )Cik1 Ci (1 )ri k1 ri

6. Solve the following equation for ik1 .

~
Kik1ik1 Rik1

7. Evaluate the new trial displacement.

ik11 ik1 ik1

8. Evaluate rik11 .

9. Evaluate the new trial acceleration and velocity.

ik11
1
t
2

ik11 ~i 1

ik11 ~i 1 tik11

10. If ik1 set tolerance , go to step 12. Note that represents an absolute value.

11. Otherwise, set k k 1 and go to step 4.

12. Update the displacement, velocity, and acceleration values.

87
i 1 ik11
i 1 ik11
i 1 ik11

13. Set i = i + 1 and go to step 2.

One can see that an explicit method is more efficient than an implicit method for nonlinear
problems since the former does not require any iteration. On the other hand, explicit methods are
only conditionally stable and may have a severe limitation on the size of the integration time
interval t when the structure has a large number of degrees of freedom.

4.5.3. Stability and Accuracy Analysis

The stability and accuracy properties of a time integration method derived for a linearly elastic
structure, as discussed in Chapter 3, can be applied to a non-linear structure if we treat a
nonlinear structure as piece-wise linear. With this consideration, we can see that the stability and
accuracy of a time integration method will be governed by the tangent stiffness if the structure
deforms into the nonlinear regime. For a hardening nonlinearity, such as the one shown in Fig.
4.15, the stability will be governed by the maximum stiffness that will be developed in the
analysis. Fortunately, most of the structures that are of interest to us in seismic design and
analysis have the softening type of nonlinearity, which relaxes the stability condition as the
structure develops nonlinear deformation.

Fig. 4.15 Hardening Nonlinearity

It should be noted that the stability and accuracy of a numerical solution for a nonlinear problem
also depend on how well the nonlinear load-displacement relation of the structure is traced with
the size of the integration time intervals. Very large integration time intervals may introduce
spurious energy into or take energy away from a structure. These spurious hysteretic energies are
caused by the discrete sampling of the nonlinear load-displacement relation of the structure in
numerical integration as illustrated in Fig. 4.16. The solution may become unstable if spurious
energy is continuously added into the system. However, such energy effects may not be
88
significant as compared to the actual hysteretic energy dissipated by an inelastic structure, but
can have a major influence for a nonlinear elastic structure, which has no hysteretic energy
dissipation.

Actual Load Path in


Numerical Solution

Fig. 4.16 Nonlinear Elastic Structure

Furthermore, implicit schemes require iterations and always have residual errors. Large residual
errors may affect the accuracy of the numerical solution and may even cause instability.

4.6 Inelastic Response Spectra

Earthquake response spectra can be developed for inelastic structures. However, for inelastic
structures, the spectral response quantities do not only depend on the elastic period and damping
of the structure but also on its post-yield force-displacement properties. There are a several ways
to construct inelastic response spectra. Here, we will consider the ductility spectra and inelastic
yield spectra.

4.6.1. Ductility Spectra

Consider an elastic-perfectly plastic structure with the properties shown in Fig. 4.17.

K
1

Fig. 4.17 Ideal Elastic-Perfectly Plastic Structure


89
The equation of motion for the structure can be expressed as

M C r Mg t (4.7)

Dividing the above equation by M, we have

g t
r
2 (4.8)
M

where

K C

M 2 M

Note that the natural frequency and damping ratio are defined in terms of the elastic stiffness of
the structure.

To express the equation of motion in terms of dimensionless quantities, let us define the
following normalized displacement and force quantities:



y
r

ry

It is evident that 1 for an elastic-perfectly plastic structure. Dividing Eq. (4.8) by y , we


have

g t
2 2 (4.9)
y

Furthermore, let us define the following dimensionless resistance factors for the structure:

ry
(seismic resistance factor) (4.10)
Mg max

ry
cy (resistance coefficient) (4.11)
Mg

in which g max is the absolute value of the peak ground acceleration for the given ground motion
record and g is the acceleration due to gravity. We can see that represents the resistance of the

90
structure normalized by the maximum effective force introduced by the ground motion, and c y
is the structural resistance normalized by its own weight (i.e., the resistance per unit weight).

We can also write

cy


g max / g

Finally, we can introduce the following condition into Eq. (4.9):

g t Mg max 2g t 2 g t
. .
y K y g max g max

Hence, we have

2 g (t )
2 2 . (4.12)
g max

It should be noted that max given by the above equation represents the ductility demand on the
structure. From the above equation, we can see that for a given ground motion record, the
ductility demand on the structure depends on the natural frequency , damping , and
seismic resistance factor , and can be expressed as

max
F , ,

where F denotes a function of but not in a strict mathematical sense. We can construct plots
of the ductility demand versus or T for various values of and . These are called ductility
spectra. They differ from elastic response spectra in that the response quantity is the ductility
demand and that there is an additional parameter, the seismic resistance factor, to consider. Just
like elastic response spectra, these spectral curves are useful design tools. Examples of ductility
spectra are shown in Fig. 4.18 for elastic-perfectly plastic structures subjected to different
earthquake ground motions. It should be noted that these spectral curves also depend on the
specific nonlinear force-displacement relations of the structures. Therefore, different spectral
curves need to be generated for structures with different nonlinear load-displacement relations.

91
Fig. 4.18 Ductility Spectra with 5% Damping (Bertero et al. 1978)

To construct ductility spectra, one needs to solve the normalized equation of motion using a step-
by-step integration method for various values of T , , and . Once constructed, ductility spectra
can be used to find the required displacement ductility for the given T , , and c y of the structure
in the following manner.

cy
1. Calculate .
g max / g

2. Determine from the spectral curves based on the T , , and given.

Alternatively, one can determine the required resistance coefficient c y for the given T and ,
and the ductility that can be provided by the structure as follows.

1. Determine from the spectral curves based on the T , , and given.

g max
2. Calculate the required resistance coefficient: c y .
g

92
4.6.2. Inelastic Yield Spectra

There is another way to construct response spectra for inelastic structures. From the previous
section, we see that for a given ground motion,

F ( , , )

where ry /( Mg max ) . We can also rewrite the above relation as

G , ,

where G represents a function. Based on the definition of and the fact that ry K y , the above
equation can be expressed as

y
2 G , ,
g max

which can be simplified to the following form:

y G , , ,g max

The above relation indicates that for a given earthquake ground motion with a given peak
acceleration, we can determine the yield displacement required for an elastic-perfectly plastic
structure that will result in a ductility demand of . Hence, for each value of , we can calculate
the values of yield displacements y required for structures of different natural frequencies or
periods T. Plotting the values of y against or T results in a spectral curve which is called the
inelastic yield spectrum. This spectral curve represents a constant ductility demand and the
spectral value is the yield displacement required for a structure of specific period to be subjected
to that ductility demand. A set of spectral curves can be generated with each representing a
specific ductility demand. In addition, we can define the following pseudo-velocity and pseudo-
acceleration quantities.

PS y

PSa 2 y

The quantities y , PS , and PSa can be plotted against the natural frequencies in a tripartite
graph as shown in Fig. 4.19. Compared to ductility spectra, one major disadvantage of inelastic
yield spectra is that they require more time-history analyses to generate.

It should be noted that ry K y 2 M y M PSa . Hence, PSa reflects the maximum base
shear or yield force developed by the elastic-perfectly plastic structure. For the case of 1 ,

93
the inelastic yield spectrum is identical to the elastic response spectrum. It is also interesting to
note that as the natural frequencies exceed 30 Hz, all the curves will approach the peak ground
acceleration.

Elastic Response
Spectra
2

Maximum Ground
Acceleration

~30 Hz

Fig. 4.19 Inelastic Yield Spectra for a Given Damping Ratio

Similar to ductility spectra considered before, inelastic yield spectra are useful design tools. For
example, if we are given , , and c y , we can find the value of the ductility demand as
follows.

ry Mgc y g
1. Evaluate y provided: y cy .
K M
2
2

2. Determine from the spectral curves based on the , and y given.

If we are given the values of , , and the ductility demand , we can find the value of c y
required as follows.

1. Determine the value of y required from the spectral curves based on the , , and
given.
94
2 PSa
2. Calculate the resistance coefficient required: c y y .
g g

Equation (4.9) shows that if the ground acceleration is scaled up or down by a factor, the values
of y will have to be scaled by the same factor to have the same value of ductility demand .
Hence, for a given ground motion, the values of y , PS , and PSa in a tripartite graph can be
normalized by the peak ground displacement, velocity, and acceleration, respectively, as shown
in Fig. 4.20.

Fig. 4.20 Inelastic Yield Spectra for Elastic-Perfectly Plastic Structures


subjected to 1940 El Centro Ground Motion with 2% Damping (Note:
, 13 in./sec., and ; D, V, and A
represent , , and , respectively; y represents ground motion
quantities)

Similar to ductility spectra, inelastic yield spectra can be generated for structures with different
hysteretic behaviors.

95
4.6.3. Total Displacement Response Spectra

Since max y , if we multiple the spectral values of an inelastic yield spectrum by the
respective ductility, we obtain a total displacement response spectrum. As illustrated in Fig. 4.21
(in a slightly exaggerated fashion), total displacement response spectra for different earthquake
records show that for natural frequencies less 2 Hz, the maximum displacement response of a
structure is more or less the same regardless of the value of . Since y ry / K , this means
that for a give stiffness K, a smaller value of ry will result in a high ductility demand with the
maximum displacement remaining the same. Furthermore, for natural frequencies between 2 and
8 Hz, the energy absorbed by a structure is not much influenced by its yield strength. These
empirical observations provide useful relations between the elastic and inelastic responses of a
structure as discussed below.

Constant
Ductility

2 8
not much influenced
by , Hz Energy absorption not much
influenced by

Fig. 4.21 Total Displacement Response Spectra

96
For Frequency f 2 H z

In this frequency range, the maximum inelastic response of a structure, max


i
, is very close to the
maximum response of the corresponding elastic structure, max
e
, regardless of the yield strength.
Hence, we can write

max
i
e
max
y y

Rearranging the above equation, we have

max
e
y (4.13)

which indicates that for a given ductility demand , we can evaluate the required yield
displacement or yield strength from the elastic response of the structure.

For Frequency Range 2 H z f 8H z

For this frequency range, empirical observations have shown that the energy absorption by a
structure is not much influenced by its yield strength as illustrated in Fig. 4.22.

Equal Area

Fig. 4.22 Equal Energy Absorption for a Given Earthquake

Hence, assuming that

1 1
max
i
ry ry y K max
e2
(4.14)
2 2

97
we have

1 max
2
max
i e
1
y 2 y

or

max
e
y
2 1

The last equation allows us to evaluate the value of required y from the elastic response and the
given ductility demand.

The above empirical relations are useful for constructing an inelastic response spectrum from an
elastic response spectrum without time consuming analysis. This is often sufficiently accurate for
design purposes. In fact, Eq. (4.13) has been used as a basis to determine the values of the
structural response modification factor, R, in ASCE 7 for structures with different ductilities.

4.7 Design Spectra

Here, we consider some empirical methods for the construction of elastic and inelastic response
spectra that can be used for design.

4.7.1. Elastic Design Spectra

As illustrated in Fig. 4.23, we can identify three distinct regions in a response spectrum plotted in
a four-way logarithmic graph. For natural frequencies between 2 and 8 Hz, the curve has more or
less a constant pseudo spectral acceleration. For frequencies between 0.3 and 2 Hz, the curve has
more or less a constant pseudo spectral velocity. Between 0.1 and 0.3 Hz, it has more or less a
constant spectral displacement. This general tendency has been observed for ground motions
recorded at sites with firm soils, like the El Centro record. If we plot the peak ground motion
quantities in the same graph, we can see that the peak response of a structure can be estimated by
multiplying the respective peak ground motion in each of the three regions by an amplification
factor. This is illustrated in Fig. 4.24. For frequencies below 0.05 Hz, the spectral displacement
is essentially the same as the peak ground displacement; and for frequencies beyond 30 Hz, the
pseudo spectral acceleration is essentially the same as the peak ground acceleration. The rest of
the response spectrum curve can be completed by interpolation.

The above observations provide a way to construct an elastic design spectrum. For this purpose,
we need to determine the maximum peak ground acceleration, velocity, and displacement that
are probable for the site. These can be determined with a probabilistic analysis, which will be
discussed in Chapter 7, based on a selected return period. In lieu of such analysis, Newmark and

98
Hall (1982) recommended that for competent soil conditions, the peak ground velocity be 48
in./sec. for every 1g of peak ground acceleration, and for rock, the peak velocity be 36 in./sec.
for every 1g. Furthermore, they recommended that the factor ( g max g max / g2 max ) be around 6,
where the acceleration, velocity, and displacement quantities are in inches and seconds.

Fig. 4.23 Response Spectra for the El Centro Record from the 1940 Imperial Valley
Earthquake

The amplification factors A, V, and D for the displacement, velocity, and acceleration can be
determined from the empirical equations proposed by Newmark and Hall, which are based on a
statistical study assuming that the response quantities have a lognormal distribution. These
equations are presented as follows.

For an 84.1% probability of not being exceeded,

A 4.38 1.04 ln

V 3.38 0.67 ln

D 2.73 0.45 ln

where is ratio of critical damping in %. For example, when = 3%, A = 3.24.

For a 50% probability of not being exceeded,

99
A 3.21 0.68 ln

V 2.31 0.41ln

D 1.82 0.27 ln

0.05 0.1 0.3 2 8 30


Frequency, Hz
Fig. 4.24 - Construction of Elastic Design Spectrum

4.7.2. Inelastic Design Spectra

Based the empirical relations presented in Sec. 4.6.3, we can construct inelastic yield spectra
from elastic design spectra. The procedure is shown in Fig. 4.25. A sample inelastic yield
spectrum constructed by Newmark and Hall is shown in Fig. 4.26.

Limitations of Inelastic Design Spectra

The method proposed by Newmark and Hall to construct an inelastic design spectrum, as shown
in Fig. 4.25, assumes that it is directly related to the elastic design spectrum. However, we know
that this is not exactly true. For example, while viscous damping has a significant influence on

100
the response of an elastic structure, it has less influence on the response of an inelastic structure
because the latter has hysteretic damping. Furthermore, the duration of strong shaking and the
duration of a strong pulse in a ground motion can have a significant influence on the response of
an inelastic structure, but their effects are not necessarily reflected in an elastic spectrum. For
these reason, this method is not commonly used nowadays.

0.05 0.1 0.3 2 8 30


Frequency, Hz
Fig. 4.25 - Construction of Inelastic Yield Spectrum

101
Fig. 4.26 - Elastic and Inelastic Design Spectra (Newmark and Hall)

102
References:

1. Bertero, V.V., Mahin S.A., & Herrera, R.A. (1978). Aseismic Design Implications of
Near-Fault San Fernando Earthquake Records, Earthquake Engineering and Structural
Dynamics, Vol. 6, 31-42.

2. Bozorgnia, Y. and Campbell, K.W. (2004). Engineering Characteristics of Ground Motion,


in Earthquake Engineering from Engineering Seismology to Performance-Based
Engineering, Bozorgnia and Bertero (Eds.), CRC Press.

3. Chopra, A.K. (2011). Dynamics of Structures: Theory and Applications to Earthquake


Engineering, 4th Ed., Prentice Hall.

4. Newmark, N.M. and Hall, W.J. (1982). Earthquake Spectra and Design, Monograph,
Earthquake Engineering Research Institute.

103
CHAPTER 5

SEISMIC RESPONSE OF
MDOF STRUCTURES

5.1 Equations of Motion

The equations of motion for a multiple-degree-of-freedom structure can be formulated with the
dynamic equilibrium condition as follows.

fI fD r 0

where

f I = vector of structural inertia forces

f D = vector of structural damping forces

r = vector of structural restoring forces

Fig. 5.1 - Multi-Story Shear Frame

104
Since

f I Mv t

f D Cv

we have

Mv t Cv r 0

where

v t v ag

v 1 2 .............N
T
(structural displacements with respect to the base)

a = a vector that represents the rigid-body acceleration of the structure due to a unit
base acceleration

r Kv for a linearly elastic structure

Hence, expressing everything in terms of relative displacements, we have

Mv Cv r Mag

For the linearly elastic three-story shear frame shown in Fig. 5.1, we have

M 1 0 0
M 0 M2 0

0 0 M 3

C1 C 2 C2 0
C C2 C 2 C3 C3

0 C3 C 3

K1 K 2 K2 0
K K2 K2 K3 K3

0 K3 K 3

a 1 1 1
T

105
5.2 Mode-Superposition Analysis for Linearly Elastic Structures

For a linearly elastic MDOF structure, we can use the mode-superposition method to evaluate its
earthquake response. Even though this procedure is not very efficient and therefore, not a desired
choice for time history analysis, it is the basis for the response spectrum analysis procedure for
MDOF structures, which will be considered in the next section. In computer programs, the
equations of motion are normally solved directly using a step-by-step integration method.

To carry out the mode-superposition analysis procedure, we start with the coupled equations of
motion.

Mv Cv Kv Mag

Assume that damping is classical (i.e., the eigenvectors are orthogonal with respect to the
damping matrix). The earthquake response is then evaluated as follows.

1. Formulate and solve the eigenvalue problem to find n and n .

Kn n2 Mn

2. Formulate the equations of motion in normal coordinates Yn .

Ln
Yn 2nnYm n2Yn g
M n

where

M n nT Mn K n nT Kn

Ln nT Ma

n K n M n

The factor Ln M n is called the modal participation factor. The value of Ln depends on the
mode shapes and vector a.

3. Compute the modal response as

Ln
Yn t Vn t
M nn

where

106
Vn t g ( )enn (t ) sin n t d
t

The above expression assumes small damping so that nD n .

4. Calculate the relative displacement response in the geometric coordinates using superposition.
N N
Ln
v t nYn Vn t n where N is the number of degrees of freedom
n 1

n 1 M n n

5. Evaluate the elastic restoring forces developed by the structure.


N N
Ln
r Kv KnYn nVn t Mn
n 1 n 1 M n

Fig. 5.2 - Base Shear

6. For the structure illustrated in Fig. 5.2, where a 1 , the base shear can be calculated as
T

N N
Ln2
V0 ri 1 r nVn t
i 1 n 1 M n

107
The term Ln2 M n in the above equation is called the effective modal mass and it can be shown
that the sum of the effective modal masses is equal to the total mass of the structure (see pp. 627
of Clough and Penzien 1993), i.e.,
N N

Ln2 M n M i
n 1 i 1

Note that the base shear defined here does not include the damping force.

5.3 Response Spectrum Analysis for Linearly Elastic Structures

Response spectrum analysis can be applied to linear MDOF structures. Such analysis is allowed
by seismic design codes. With the mode-superposition analysis method, we have
N N
Ln
v t nYn Vn t n
n 1

n 1 M n n

The maximum value of Vn is

Vn max
PS ( n , n )

in which the pseudo-spectral velocity PS ( n , n ) has exactly the same definition as that for
SDOF structures except that the modal frequency and damping ratio should be used to find its
value. Hence, the peak displacements, restoring forces, and base shear contributed by each mode
n can be expressed as

Ln
v n ,max PS ( n , n )n
M n n

Ln
rn ,max n PS ( n , n )Mn
M n

Ln2
V0 n ,max n PS ( n , n )
M n

Once we have the above modal contributions, we can estimate the peak response of the structure
using a statistical method. For this purpose, let us consider a general response quantity R:
N
R t Rn t
n 1

108
in which R can be the displacement, restoring force, or base shear of the structure. If we know
Rn max , we may estimate the peak response R max using one of the following methods.

Absolute Sum

The peak response can be estimated as the absolute sum of the peak response for each mode, i.e.,

R max R1 max R2 max


.......... RN max

However, as shown in Fig. 5.3, the peaks of modal responses normally do not occur at the same
time instant or in the same direction. Hence, this method is way too conservative.

Mode i

Mode j

Fig. 5.3 - Modal Responses

SRSS (Square-Root-Sum-of-Squares)
109
The SRSS method is the most commonly used method for combining the modal responses of a
structure. The estimation is given by the following expression.

| R |max R1 max R2 max ............ RN max


2 2 2

This method is based on the assumption that the occurrences of the peak modal responses are not
correlated to each other. In most situations, this method provides a good estimate of the peak
response. Nevertheless, for cases when the modal frequencies are very close to each other, the
method can be unconservative. For those modes that have their natural frequencies close to each
other, their peak responses can be highly correlated and they can overlap with each other. Such a
situation can occur when the translational response of a structure is strongly coupled with the
torsional response. In this situation, the CQC method described below should be used.

CQC (Complete Quadratic Combination)

The CQC method is recommended for structures that have modal frequencies very close to each
other. In this method, the modal combination is carried out as follows.

R max R
n m
n ,max Rm ,max nm

in which the signs of the modal contributions Rn,max should be retained and nm is called the
cross-modal coefficient. For an earthquake excitation that has a long duration with respect to the
natural period of the structure and has a smooth response spectrum, the cross-modal coefficient
can be evaluated with the following expression (Der Kiureghian 1981).

8 nm n m 3/2
1/2

nm
1 2 2
4 n m 1 2 4 n2 m2 2

m
in which and n is the ratio of critical damping.
n

The above equation shows that nm 1 for n =m. For the modal frequencies that are widely
spaced, nm 0 for n m . In that case, the CQC method is essentially the same as the SRSS
method. Therefore, the CQC method is more general and the SRSS method can be considered as
a special case for the CQC method. Both methods can be derived from theories of random
vibrations.

110
Example

0.5 kip-sec.2/ft.

5'

5'

5' 5'
5 kips/ft. per column

Fig. 5.4 - 3-DOF Shear Frame

M = total mass

b Rotational Inertia:

For the structure shown in Fig. 5.4, we have

Mv Cv Kv Mag (t )

where

111
15 0 25 0 .5 0 0
K 0 15 25 M 0 0. 5 0


25 25 750 0 0 8.33

5.028 0.9656 1.000 0.2599



5.474 rad./sec. 0.9656 1.000 0.2599
9.725 0.0900 0.000 0.3345

0

a 1
0

The peak displacement response at each degree of freedom to the 1940 El Centro ground motion
has been obtained from a step-by-step integration analysis:

0.153 ft .

v max 0.327 ft.
0.020 rad.

The above values represent the exact solution and they will be compared to the values estimated
with response spectrum analysis using different modal combination methods. For this purpose,
the pseudo-spectral velocity for each of the three modes is obtained from the response spectrum
for the El Centro ground motion with 5% damping.

PS (1 ,1 ) 2.0 ft./sec.

PS (2 , 2 ) 2.2 ft./sec.

PS (3 , 3 ) 2.3 ft./sec.

Furthermore, the following quantities are evaluated

L1 1T Ma 0.4828

L2 2T Ma 0.5000

L3 3T Ma 0.1300

M 1 1T M1 M 2 M 3 1.0

From the above results, we have

112
0.185
L
v1,max 1 PS (1 , 1 )1 0.185

M11
0.017

0.200
L2
v 2,max PS (2 , 2 )2 0.200

M 22
0.00

0.008
L3
v3,max PS (3 , 3 )3 0.008
M 33
0.010

1. Absolute Sum

0.393

v max v 1,max v 2,max v 3,max 0.393
0.027

Comparing the above result to the exact solution, we can see that the estimate for the
displacement at DOF-1 is way too conservative.

2. SRSS Method

0.273

v max v 2
1,max v 2
2,max v 2
3,max 0.273
0.020

In this case, we underestimate the response at DOF-2 and way overestimate the response at
DOF-1. Note that in the above equation, the square of a vector is a vector that is obtained by
taking the square of each element of the original vector.

3. CQC Method

11 22 33 1.0

12 21 0.580

13 31 0.021

23 32 0.027

113
0.177

v max 0.343
0.020

We can see that the above estimation is very close to the exact solution for all DOF's.

Final Remarks

We should paid special attention to estimating the maximum base shear with any of the above
combination methods. It should be calculated directly from the maximum base shear contributed
by each mode and not from the estimated maximum restoring force developed at each story. This
is because the maximum restoring forces developed at different stories usually do not occur at
the same time instant in the response time history. The use of such values to estimate the
maximum base shear may way overestimate the total base shear. The same is true for the story
shear and the overturning moment at any story.

5.4 Reduction of Number of Degrees of Freedom

Under earthquake excitation, higher-mode responses are often insignificant in a structure with
many degrees of freedom. Hence, we can often ignore these higher-mode contributions by
considering an equivalent system that has a reduced number of dynamic degrees of freedom.
This can be achieved in a number of ways as discussed below.

5.4.1 Truncated Modal Analysis

In mode-superposition analysis, if the contribution of the modes higher than M is insignificant, it


will be sufficient to consider only the first M modes.
M
v nYn t
n 1

in which M N , where N is the total number of degrees of freedom. With this, it will not be
necessary to evaluate all the mode shapes and natural frequencies. One can evaluate only the first
M mode shapes and natural frequencies by using subspace iteration (see pg. 304 of Clough and
Penzien 1993). However, we should be aware of the fact that the higher-mode contribution will
be more significant in the restoring forces developed by the structure than in the displacement
response. Hence, in calculating the restoring forces, we should include more natural modes.

5.4.2 Truncated Modal Analysis with Static Correction

The dynamic amplification effect is often negligible for modes that have very high frequencies as
compared to the frequency of the forcing function. This is illustrated with the response spectrum
for harmonic loading in Fig. 5.5. In the figure, o is the amplitude of the steady-state response
normalized by the static amplitude, s f o / K , where f o is the amplitude of the harmonic
loading. As shown, when the natural frequency of a structure is much higher than the excitation
114
frequency (i.e., when is close to zero), the dynamic amplification effect essentially disappears
and the response can be considered static. This can happen to the higher modes of a structure
subjected to earthquake ground motions. However, even though the dynamic amplification effect
is negligible, their static amplitudes may still be non-trivial. To account for these contributions,
the truncated modal analysis discussed above can be improved by a static correction as follows.

1
Fig. 5.5 Steady-State Response Amplitude under Harmonic Loading

Considering that the contribution of modes higher than M is essentially static, one can divide the
total response into two components as follows.

v vd vs

where
M
v d n Yn t (dynamic response of the lower modes)
n 1

N
fn
vs
n M 1
n
K n
(static response of the higher modes)

in which Y n is dynamic response for mode n, K n is the generalized stiffness, and fn nT Ma g .


The static component can also be expressed as follows.

N
n nT 1 M n nT
vs Ma g or v s K Ma g

n M 1 K n n 1 K n

115
The second expression above simply means that the static response of the higher modes is equal
to the total static response subtracted by the static response of the lower modes. With this
expression, we can avoid calculating the frequencies and mode shapes for the high modes. As a
result, we have

M M
T
v nYn K 1 n n Mag


n 1 n 1 K n

In the above equation, the first term on the right-hand side represents a truncated modal analysis
while the second term represents the static correction.

5.4.3 Rayleigh-Ritz Method

Another approach to improve the computational efficiency is to use assumed mode shapes to
reduce the number of degrees of freedom. This is called the Rayleigh-Ritz method, which is a
generalization of the Rayleighs method. In this method, we assume that
M
v n Z n t
n 1

in which M N , n is an assumed mode shape that is called a Ritz vector, and Z n is a


generalized coordinate that has the same meaning as Y n but is associated with a Ritz vector
rather than the actual mode shape. For conciseness, the above equation can be written as

v Z

where

[ 1 2 .. M ]

Z Z1 Z 2 ....... Z M
T

By performing a coordination transformation similar to that in modal analysis with eigenvectors,


we have the following generalized equations of motion.

K Z f t
CZ
MZ

where

M T M
C T C
K T K
f T f
116
Note that the number of equations is now reduced from N to M. However, unlike the case with
the actual eigenvectors, the above equations are still coupled, i.e., the M* , K * , and C * matrices
are not diagonal. Hence, this reduced problem has to be solved in the following manner using the
mode superposition approach.

1. Solve the reduced eigenvalue problem.

K zn zn2 M zn

2. Formulate the equations of motion in normal coordinates.

M znYzn C znYzn K znY zn fzn for n = 1, 2,....., M

where

M zn Tzn M zn
C T C
zn zn zn

K zn K zn
T
zn

fzn Tzn f *

Assume that TznCTzm 0 for n m .

3. Solve for Yzn t .

4. Evaluate the displacement response.

Note that
M
Z znY zn t z Yz
n 1

where

z [ z1 z 2 zM ]
Yz Y z1 Yz2 .. Y zM
T

Hence, we can compute the displacement response as

v z Yz

In the about equation, z can be viewed as a matrix containing vectors that are
approximations of the eigenvectors for the original structure that has the stiffness and mass
117
matrices K and M . If the Ritz vectors are appropriately chosen, this approach can yield
more accurate solutions than a truncated modal analysis. Different methods to construct Ritz
vectors are discussed below.

Ritz Vectors based on Estimated Mode Shapes

One may select any vectors that are close to the expected mode shapes of the structure as Ritz
vectors. This is based on judgment. Examples of these are given in Fig. 5.6.

Fig. 5.6 Estimated Mode Shapes.

Fig. 5.7 Six-DOF Structure


118
Ritz Vectors by Means of Static Condensation (Guyan Reduction)

Ritz vectors can be derived with a static condensation procedure, which is often used to eliminate
the degrees of freedom that are not subjected to any force. For a dynamic problem, these degrees
of freedom have zero or negligible inertia associated with them. To illustrate this procedure,
consider the free-vibration response of the structure shown in Fig. 5.7, in which the rotational
inertia is negligible.

The equations of motion for the above structure can be expressed as

K tt K tr v t M tt 0 v t
K
rt K rr v r 0 0 v r

in which the subscript t represents the translational degrees of freedom and r denotes the
rotational degrees of freedom. When the inertia forces in the above equations are represented by
the force vector f I , the equations of motion can be written as

K tt K tr v t f I
K
rt K rr v r 0

which is similar in form to a static problem. The second of the above two equations leads to

v r K rr1K rt v t

Hence, we can write

v I
v t 1 v t v t
v r K rr K rt

in which represents a matrix containing Ritz vectors derived from the static condensation
procedure. Using the above Ritz vectors, the equations of motion can be reduced to

M*v t K * v t 0

where

M* T M M tt

K * T K K tt K tr K rr1K rt

in which M tt is simply a mass matrix without the rotational degrees of freedom, and K* is the
generalized stiffness matrix, which can also be considered as a condensed stiffness matrix.

119
Ritz Vectors Derived with Static Loading

Ritz vectors can be systematically derived with static analysis using successively updated static
loading that represents the inertia force developed by the structure (Wilson et al. 1982). In this
method, we must express the load vector in the following form.

f t Ff t

in which F is a vector that represents the spatial distribution of the dynamic loading. For the case
that the dynamic loading is induced by an earthquake ground motion,

F Ma and f (t ) g (t )

With the spatial distribution vector, Ritz vectors can be derived sequentially as follows.

First Ritz Vector:

Solve K 1 F for 1 .

Obtain a normalized vector 1 :

1
1 1
T
1 M
* 1/ 2
1

so that 1T M 1 1 . This is the first Ritz vector.

Second Ritz Vector:

Solve K 2* M 1 for 2* .

Conduct a Gram-Schmidt orthogonalization (introduce an orthogonality property):

2** 2* c1 1

where

c1 1T M 2*

As a result, we have

1T M 2** 0

Obtain a normalized vector 2 :

120
1
2 2**
**T
2 M
** 1 / 2
2

which is the second Ritz vector.

Third Ritz Vector:

Solve K 3* M 2 for 3* .

Conduct a Gram-Schmidt Orthogonalization:

3** 3* c1 1 c2 2

where

c1 1T M 3*

c2 2T M 3*

As a result, we have

1T M 3** 0

2T M 3** 0

Obtain a normalized vector 3 :

1
3 3**
**T
3 M
** 1 / 2
3

which is the third Ritz vector.

In general, to generate the i-th Ritz vector for i >1, we can use the following procedure.

1. Solve K i* M i 1 for i* .

2. Conduct a Gram-Schmidt Orthogonalization:


i 1
i** i* c k k
k 1

121
where

ck kT M i*

3. Obtain a normalized vector i :

1
i i**
**T
i M
** 1 / 2
i

The Ritz vectors generated with the above procedure satisfy the following condition.

T M M* I

However,

T K K *

will not be diagonal.

The Ritz vectors derived here have a physical meaning which can be explained as follows.
Consider the equations of motion expressed in the following form.

Kv Ff (t ) Mv

With the above expression, the dynamic response of a structure can be perceived as the sum of
static deflections induced by the applied loads Ff (t ) and inertia forces Mv . As it has been seen,
the first Ritz vector simply represents the static deflected shape of the structure due to the applied
loads and the subsequent Ritz vectors represent successive corrections of the deflected shape to
account for the inertia effect introduced by structural accelerations. It is interesting to note that
the use of these Ritz vectors can lead to more accurate results than a truncated modal analysis
that is based on the true eigenvectors when the same number of vectors is used. This is because
the higher-mode effects that are ignored in a truncated modal analysis are indirectly incorporated
in the Ritz vectors generated in this manner.

5.4.4 Rayleighs Quotient

The Rayleigh-Ritz method presented above is a generalization of the concept behind Rayleighs
quotient. Rayleigh's quotient, which can be used to provide a quick and reliable estimate of the
fundamental frequency of a MDOF structure, is defined as

T K
R
T M

122
in which is the assumed mode shape. This expression is analogous to that defined for a
continuous beam. It can be seen that Rayleighs quotient is equal to r2 when is equal to a
true eigenvector r of the structure. Rayleighs quotient has two important properties. First, it is
stationary with respect to the change of when is in the neighborhood of a true eigenvector
r . As a result of this, we can expect that R will be very close to r2 when is reasonably
close to r . Second, R 12 . Hence, we can conclude that when is reasonably close to 1 ,
Rayleighs quotient will provide a very good estimate of 12 , and this estimate is always higher
than the actual fundamental frequency of the structure. The shape vector can be estimated
with any of the methods mentioned previously considering only the fundamental mode shape of
the structure. The proof of the above properties of Rayleighs quotient for a continuous beam has
been provided in separate class notes. A similar proof can be derived for discrete systems.

5.5 Direct Step-by-Step Integration

For a MDOF structure, step-by-step integration can be applied directly to the coupled equations
of motion without going through the mode-superposition analysis procedure. For this reason,
step-by-step integration is a very attractive means for calculating the response time histories of
MDOF structures. This method entirely avoids the need to solve an eigenvalue problem.
Furthermore, it provides the only general means to evaluate the nonlinear response of a structure.
For MDOF structures, the computational procedure is exactly the same as that for a SDOF
structure.

5.5.1 Direct Integration with Newmark Method

For a linear MDOF structure, the Newmark Method is expressed as

Mv i 1 Cv i 1 Kv i 1 fi 1

1
v i 1 v i tv i t 2 v i v i 1
2

v i 1 v i t1 v i v i 1

With the same mathematical manipulation as that carried out for a SDOF structure, as described
in Ch. 3, we can obtain the following equivalent static equation.
~ ~
Kv i 1 fi 1

where

~ 1
K M CK
t 2
t

123
~ 1 1 1 t
fi 1 f i 1 M C
v M
1
C
v
2 1
M
2 Cv i

2 i i
t t t 2

For nonlinear MDOF structures, we can use the same iterative solution procedure developed for
a SDOF structure which is discussed in Ch. 4.

5.5.2 Stability and Accuracy

The stability and accuracy properties of a step-by-step integration scheme for a SDOF structure
can also be extended to a MDOF structure. For example, let us consider the Newmark method
presented in Sec. 5.5.1. Assume that the damping matrix is classical, i.e., it can be diagonalized
by the modal vectors. Based on the expansion theorem, we have

v i 1 Yi 1

in which is a matrix containing all the eigenvectors and Yi 1 is a vector containing the
displacement response in normal coordinates. Substituting the above expression into the first
three equations presented in Sec. 5.5.1 and premultiplying the first equation by T and the
second and third equations by T M , we obtain

Y
M C K
Y Y f
i 1 i 1 i 1 i 1

t 2 1 Y
Yi 1 Yi tY
Y
i 2 i i 1

Y
Yi 1 i
t 1 Y
Y
i

i 1
in which M , C , and K are the generalized mass, damping, and stiffness matrices that are
diagonal. The above expressions show that the direct step-by-step integration procedure can also
be cast in the form of a mode-superposition analysis procedure and the integration can be
performed for each mode independently. It follows that the stability and accuracy properties of
an integration scheme derived for a SDOF structure also apply to each mode of a MDOF
structure. The stability and accuracy of the solution for each mode i depend on t / Ti , where Ti
is modal period. As a result, the selection of an appropriate integration time interval t requires
the following considerations.

1. For the stability of the solution, the permissible time interval t will be governed by the
highest natural frequency of the structure.

124
2. For the accuracy of the solution, the time interval t will be governed by the highest
mode that has a non-trivial contribution to the total response.

For an unconditionally stable scheme, the first consideration is unnecessary.

5.6 Multiple Support Excitation

For a structure that extends over a long distance, such as a long-span bridge, each support can be
subjected to a ground motion that is different from those at the other supports. These
asynchronous ground motions can induce additional deformations and stresses to the structural
members. Consider the structure shown in Fig. 5.8 as an example. There is no external force
applied to the structure and each support of the structure is subjected to a ground motion time
T
history, gi , that is different from those at the other supports. Let v g g1 g2 .... , and v t
is a vector that contains the total displacement response of the structure with respect to a fixed
reference frame. The equations of motion for the whole system can be expressed as

M M sg
v C Csg
v K K sg
v 0
t t t

M (1)
gs v g
M gg Cgs v g
Cgg K gs v g
K gg f g

in which M, C and K can be interpreted as the mass, damping, and stiffness matrices of the
structure when the supports are fixed, i.e., when v g 0 , and M gs MTsg , Cgs CTsg , and K gs KTsg ,
which are the mass, damping and stiffness terms that couple the structural degrees of freedom to
the support motions. These matrices can be constructed from the stiffness matrix and consistent
mass matrix of each element of the structure. The damping matrix can be constructed as a
combination of the stiffness and mass matrices like Rayleigh damping. In the above equation, f g
contains the support reactions.

.
Fig. 5.8 Multiple Support Excitation

125
Our attention is however on the structural response, which is given by the first equation in Eq.
(1). We can rearrange this equation by moving the terms associated with ground motions to the
right-hand side and obtain the following expression.

Mvt Cvt Kvt M sg v g Csg v g K sg v g (2)

The right-hand side of the above equation represents the effective forces f eff , which are
calculated from the ground acceleration, velocity, and displacement. Once the effective forces
are obtained, the equations of motion can be solved in the same way as for a fixed-based
structure. However, the response obtained is the total response.

An example of the formulation of such a problem is given below.

Example 1

The frame shown in Fig. 5.9 has two supports with independent motions. All members of the
frame have the same bending stiffness and mass. Find K , K sg , M , and M sg for the frame.

In this case,
T
v t 1t 2t 3t

T
v g g1 g 2

Using the element stiffness and mass matrices shown in the figure, we can construct the stiffness
and mass matrices for the frame using a direct assembly procedure.

12 3L 3L
2 EI
K 3 3L 4 L2 L2
L
3L L2 4 L2

2 2
K sg 3 L 0
6 EI
L
0 L

732 22 L 22 L
mL
M 22 L 8L2 3L2
420
22 L 3L2 8L2

126
54 54
mL
M sg 13L 0
420
0 13L

Bending Stiffness = EI
L
Mass per Unit Length =

Beam Element:
1 3

EI, m
2 4
L

12 6 L 12 6 L 156 22 L 54 13L
6 L 4 L2 6 L 2 L2 2
3L2
EI mL 22 L 4 L 13L
Ke 3 Me
L 12 6 L 12 6 L 420 54 13L 156 22 L
2
6 L 2 L 6 L 4 L 13L 3L 22 L 4 L2
2 2

Fig. 5.9 Element


Frame with Two and
Stiffness Independent Support Motions
Mass Matrices

127
The solution of Eq. (2) is not very convenient as the effective forces have to be calculated from
the ground acceleration, velocity, and displacement. We can simplify this by separating v t into
two parts:

vt v s v (3)

where

v s represents the quasi-static response, and

v represents the dynamic response

The quasi-static response of the structure is the static displacements of the structure induced by
the motions of the supports as illustrated in Fig. 5.10. It can be obtained from Eq. (2) by omitting
the dynamic effects, i.e., by letting

M C0
M sg Csg 0

As a result, vt vs and we have

Kv s K sg v g

From the above equation, the quasi-static response can be expressed as

v s K 1K sg v g Av g (4)

Fig. 5.10 Quasi-Static Response

128
Hence,

v t v Av g (5)

Substituting the above expression into Eq. (2), we have

Mv Cv Kv [MA Msg ]v g [CA Csg ]v g [KA K sg ]v g

In general, the damping term on the right-hand side of the equation is relatively small as
compared to the inertia term and can, therefore, be ignored. Furthermore, the stiffness term on
the right-hand side should be zero since

KA K sg K K 1K sg K sg 0

Consequently, we have

Mv Cv Kv [MA Msg ]v g (6)

The above equations of motion can be solved for the dynamic response v with the effective
forces related only to the ground acceleration, like the case with single support motion. Once v
has been calculated, the total response can be calculated:

v t v Av g

The elastic restoring forces developed by the structure can be calculated as follows.

v t v Av g
f s [K K sg ] [K K sg ] Kv
v g vg

The above equation indicates that the elastic restoring forces depend only on the dynamic
response of the structure and not on the total response. This is understandable in view of the fact
that the quasi-static displacements of the structure are induced by the support displacements and
not by nodal forces while the dynamic response is induced by the nodal inertia. However, the
internal forces and stresses induced in the structural members depend on the total nodal
displacements v t and v g as shown in the following example. In frame analysis or finite element
analysis in general, element forces are calculated from the total element nodal displacements.

129
Example 2

Let us consider the shear frame shown in Fig. 5.11. The frame has only one degree of freedom
with two independent support motions. The roof mass is Mr and the mass per unit length of the
columns is m. Formulate the equation of motion.

Mr = 2mL

Bending Stiffness of
column section is EI

Fig. 5.11 - Shear Frame with Two Independent Support Motions

Based on Eq. (6), the equation of motion for the structure can be expressed as

M K [MA Msg ]v g

Carrying out a direct assembly procedure using the element stiffness matrix and consistent mass
matrix for the columns (see Example 1), we have

156 96
M 2mL 2 mL mL
420 35

9 9
M sg mL mL
70 70

24EI
K
L3

12 EI 12 EI
K sg 3
L L3

130
L3 12 EI 12 EI 1 1
A K 1K sg L3
24EI L3 2 2

48 48
MA mL mL
35 35

Substituting the above expressions into the equation of motion, we have

96 24EI 3 3 g1
mL 3 mL mL
35 L 2 2 g 2

Once the above equation is solved, the total response can be calculated as

1 1 g1 1 1
t Av g g2
2 g 2
g 1
2 2 2

The elastic restoring force is

24 EI
f s K
L3

The shear force in the left column is

12 EI t 12 EI 1 1
Vcl 3
( g1 ) 3 ( g1 g 2 )
L L 2 2

The shear force in the right column is

12 EI t 12 EI 1 1
Vcr 3
( g 2 ) 3 ( g1 g 2 )
L L 2 2

Note that,

f s Vcl Vcr

which is consistent with the equilibrium condition.

131
References:

1. Chopra, A.K. (2011). Dynamics of Structures: Theory and Applications to Earthquake


Engineering, 4th Ed., Prentice Hall.
2. Clough, R.W. and Penzien, J.P. (1993). Dynamics of Structures, 2nd Edition, McGraw-Hill.
3. Der Kiureghian, A. (1981). A Response Spectrum Method for Random Vibration Analysis
of MDF Systems, Earthquake Engineering and Structural Dynamics, Vol. 9, 419-435.
4. Wilson, E.L., Der Kiureghian, A., and Bayo, E.P., A Replacement for the SRSS Method in
Seismic Analysis, Earthquake Engineering and Structural Dynamics, Vol. 9, 1981, 187-
194.
5. Wilson, E.L., Yuan, M.-W., and Dickens, J.M., Dynamic Analysis by Direct Superposition
of Ritz Vectors, Earthquake Engineering and Structural Dynamics, Vol. 10, 1982, 813-821.

132
CHAPTER 6

SEISMIC DESIGN AND CODE REQUIREMENTS


In building codes, the seismic load demand on a structure is most often calculated using
the equivalent lateral force analysis method, which is a static analysis method. However,
this is only permitted for structures that have regular configurations, and uniform mass,
stiffness and strength distributions. For irregular structures, dynamic analysis is required.
Response spectrum analysis is a more commonly used dynamic analysis method than
time history analysis.

6.1 Modal Analysis Considering the Fundamental Mode Only

The equivalent lateral force analysis method in building codes is based on the modal
analysis and response spectrum analysis concepts. As discussed in Chapter 5, the
maximum base shear contributed by mode n of a structure subjected to a horizontal
ground motion can be expressed as (see Fig. 6.1)

Ln2 Ln2
V0 n ,max n PS (n , n ) PS a (n , n ) (6.1)
M n M n

where

Ln nT M 1
M n nT Mn

and Ln2 / M n is called the effective modal mass which satisfies the following condition.

N
Ln2 N
Wi
M
n 1

i 1 g
(6.2)
n

In the above expression, Wi is the weight of story i. With the effective modal mass, we
can define the effective modal weight as

Ln2
Wn g (6.3)
M n

Substituting the above equation into Eq. (6.1), we have the maximum base shear for
mode n expressed as

PSa ( n , n )
V0n ,max Wn (6.4)
g

133
Fig. 6.1 Multi-Story Frame under
Horizontal Ground Motion

The maximum restoring forces due to mode n, rn,max , developed by the structure can be
obtained as

Ln
rn,max PSa (n , n )Mn (6.5)
M n

which can be rewritten in the following form:

Ln2 Mn Mn
rn ,max PS a (n , n ) V0 n ,max (6.6)
M n Ln Ln

Based on the above equation, the lateral force acting on each story i can be expressed in
terms of the story weight:

Wiin
rin ,max V0 n ,max N
(6.7)
W
j 1
j jn

in which in is the ith element of the modal vector. Hence, if the response of a structure is
dominated by its fundamental mode, we can estimate the maximum base shear and
restoring forces induced by an earthquake as

134
PSa (1 , 1 )
V0,max W1 (6.8)
g

Wii1
ri ,max V0,max N
(6.9)
W
j 1
j j1

6.2 Equivalent Lateral Load Procedure in ASCE 7

The equivalent lateral force analysis procedure in ASCE 7 is essentially based on the
modal analysis method described above. In this procedure, the design base shear is first to
be determined by the following equation based on the response spectrum analysis
concept.

V CsW (6.10)

where

S DS I S D1 I
Cs for T less than TL
R TR

S DS = the design (pseudo) spectral acceleration (in gs) for the short period range

S D1 = the design (pseudo) spectral acceleration (in gs) for the period of 1 sec.

T = the fundamental period of the structure

TL = long-period transition period

R = the response modification coefficient (Table 12.2-1 in ASCE 7)

I = the occupancy importance factor (Table 1.5-2 in ASCE 7)

W = total weight of the structure

Equation (6.10) has essentially the same meaning as Eq. (6.8). In fact, the condition

S D1
S DS
T

in Eq. (10) represents a pseudo-acceleration response spectrum for 5% damping with the
pseudo-acceleration normalized by g as shown in Fig. 6.2. The curve S D1 / T in the
spectrum represents the constant pseudo-velocity region of the spectral curve. For periods

135
longer than TL, the constant spectral displacement curve governs. However, instead of
having the effective modal weight for the fundamental mode (as considered in Eq. (6.8)),
the total weight of the structure is used in Eq. (6.10). This is to conservatively
compensate for the higher-mode effects ignored in the modal analysis. In general, W1 is
equal to 60% to 90% of the total weight W of a structure. Note that Fig. 6.2 is only for the
equivalent lateral force analysis method and not for response spectrum analysis. For
response spectrum analysis, ASCE 7 has a similar but different spectral curve as
discussed in class before.

Fig. 6.2 Implied Design Spectrum for


Equivalent Lateral Force Analysis

The values of S DS and S D1 are determined as follows.

2
S DS Fa S s (6.11)
3

2
S D1 Fv S1 (6.12)
3

in which Fa and Fv are site coefficients that account for the local soil properties and their
values are given in ASCE 7 (Tables 11.4.1-1 and 11.4-2); and S s and S1 are the pseudo-
spectral accelerations to be obtained from the seismic maps in ASCE 7 for short (0.2 sec.)
and long (1 sec.) period structures. Except for coastal California, the seismic maps in
ASCE 7-05 (2006) are based on a probability of exceedance of 2% in 50 years (2475 year
return period), which is considered as the Maximum Considered Earthquake (MCE). The
reduction factor of 2/3 on the right-hand side of the above equations is to scale down the

136
spectral intensity to a level that has a 10% probability of exceedance in 50 years (475
year return period), which is considered as the Design Earthquake (DE). For coastal
California, the use of a probability of exceedance of 2% in 50 years would result in
ground motions much larger than what would be expected based on the characteristic
magnitudes of earthquakes on known active faults. Therefore, the spectral intensity for
coastal California is based on known active faults. In ASCE 7-10 (2010), the seismic
maps are based on a risk-targeted approach, i.e., targeted to have a uniform probability of
collapse. To this end, the maps in the previous edition have been modified to achieve a
1% probability of collapse in 50 years for the entire US. These have been discussed in
more detail in the Powerpoint on Response Spectra.

The response modification coefficient, R, is to account for the ductility of the structure. A
more ductile structure can sustain more inelastic deformation and can thus have a higher
value of R (i.e., lower seismic design forces). The factor also takes into account the
expected overstrength in a structure. The values of R for different types of structural
systems are given in Table 12.2-1 in ASCE 7. Historically, they were determined largely
based on empirical evidence and experience. A systematic approach (FEMA P695
procedure) to determine the values of this and other performance coefficients for new
structural systems have been recently developed (ATC 2009). In this approach, the R
factor for a new structural system is so determined that the probably of collapse of the
structure under the MCE will not be greater than 10%. However, that procedure requires
extensive nonlinear analysis (Incremental Dynamic Analysis) to determine the
probability of collapse.

The determination of the fundamental period, T, of a structure will be discussed later.


Once the design base shear, V, has been determined, the distribution of the lateral forces
is given by the following equation.

Wi hik
Fi N
V (6.13)
W h
j 1
j
k
j

where

hi is the height of level i from the base

k is equal to 1 for T = 0.5 sec. or less; it is equal to 2 for T = 2.5 sec. or higher; it
is equal to 2 or determined by linear interpolation between 1 and 2 for T between
0.5 and 2.5 sec.

It is apparent that the expression in Eq. (6.13) is based on Eq. (6.9), and that hik
represents the mode shape. The exponent k is used to account for the higher-mode
effects, which are more significant for a structure with a longer fundamental period.
Higher-frequency modes tend to shift more forces to the upper stories. For structures with

137
a short fundamental period (0.5 sec. or less), k = 1, which assumes that the fundamental
modal vector has a linear shape and that higher-mode effects are not important.

Estimation of Structural Periods

The value of the structural period T is required for the calculation of V and Ft . This can
be obtained by analysis using a structural model or by using the following approximation.

Ta Ct (hn ) x (6.14)

in which hn is the structural height in feet. The values of C t and x are given in Table
12.8-2 in ASCE 7 and they depend on the structural type. If an analytical model is used to
determine the period, the period obtained cannot exceed Cu Ta , where the values of Cu
are given in Table 12.8-1 in ASCE 7. This is to avoid an over-estimation of the natural
period of a structure, which will lead to a lower earthquake design load (as evident in the
shape of the spectral curve in Fig. 6.2).

Accidental Eccentricity

Where floor diaphragms are not flexible, ASCE 7 requires that the force Fi at each level
be applied at a point offset from the center of mass of the floor in the direction
perpendicular to the line of action of Fi by a distance that is 5% of the building
dimension at that level, as shown in Fig. 6.3. This is to account for accidental
eccentricity, which may induce higher seismic load demands on the lateral load resisting
elements of the structure.

0.05d

Calculated Center of Mass

Fig. 6.3 Accidental Eccentricity

Maximum Story Drift

138
ASCE 7 limits the story drift in a structure. As shown in Fig. 6.4, story drift, , is
defined as the relative displacement between two adjacent levels of a structure.

Fig. 6.4 Story Drift

The elastic displacement at level x of the structure, xe , is to be determined with elastic


structural analysis using the equivalent lateral forces determined with the aforementioned
procedure. The inelastic story displacement x is then computed with the following
empirical equation with an amplification factor Cd .

Cd xe
x (6.15)
I

in which the values of C d for different types of structural systems are given in Table
12.2-1 in ASCE 7. Drift limits for different structural systems are given in Table 12.12-1
in ASCE 7.

6.3 General Design Considerations

In addition to the code requirements, the following considerations are important in


seismic design.

1. Prevention of weak stories

Concentration of inelastic deformation in a single story is highly undesirable as this may


induce excessive inelastic deformation on structural elements in that story. The weak-

139
story phenomenon was responsible for a large number of structural failures in past
earthquakes. To avoid weak stories, one should use strong column-weak beam design and
avoid weak column-strong beam design, as illustrated in Fig. 6.5. One should also avoid
situations like open lower stories with walls in the upper stories.

Plastic Hinges

Soft Story

Strong Column-Weak Beam Weak Column-Strong Beam

Fig. 6.5 Prevention of Weak Stories

2. Avoid asymmetric structural systems

Asymmetric structural systems, such as that shown in Fig. 6.6, are undesirable as
asymmetry will introduce torsional response to the structure and, thereby, additional load
demands on the structural elements.

Seismic
Force Shear Walls

Building Plan View

Fig. 6.6 - Torsion Introduced in Asymmetric Structure


140
Moment-Resisting Frame Concentrically Braced Frame

Shear Link

Eccentrically Braced Frame

Fig. 6.7 - Moment-Resisting and Braced Frames

141
3. Maintain a balance between stiffness and ductility

The ductility demand from an earthquake ground motion is generally higher for a stiff
structure than for a flexible structure even when both structures have comparable
strengths. However, when a structure is too flexible, the total drift can be excessive.
Seismic design requires a good balance between stiffness and ductility. This
consideration is best illustrated with the different steel frame examples shown in Fig. 6.7.
Concentrically braced frames are stiff but may not be very ductile because of the post-
peak softening introduced by the buckling of the braces. Moment-resisting frames usually
exhibit a good ductility, but they are very flexible and may develop large story drifts
under a severe earthquake excitation. To have a good balance between stiffness and
ductility, eccentrically braced frames are becoming more popular. In an eccentrically
braced frame, braces are designed not to buckle and inelastic deformations are to be
concentrated in the shear links.

4. Provide adequate detailing for critical components

Most of the time, structural failure occurs because of poor design details. It is important
to have adequate design details in all critical components of a structure, such as beam-to-
column connections in a steel frame, beam-to-column joints in a reinforced concrete
building, and connection details in a precast concrete structure. For reinforced concrete
columns, adequate confinement and shear reinforcement is also of utmost importance to
assure good ductility.

5. Provide sufficient redundancy

The load resistance system of a structure should have sufficient redundancy so that
alternative load paths can develop when a few of its members fail. ASCE 7 has a
redundancy factor to penalize structures having a low redundancy. Structures with a low
redundancy will be subject to a larger design load.

6.4 Performance-Based Design

Current design codes focus mainly on collapse prevention. Nowadays, seismic design is
moving towards a performance-based approach. However, this concept is not entirely
new. For example, it has been widely accepted that a structure should not sustain any
structural or non-structural damage in the event of a minor earthquake that has a high
probability of occurrence during the lifetime of the structure, but it may sustain limited
structural and non-structural damage under a moderate earthquake. Under a severe
earthquake, a structure can have significant structural damage but it should not collapse.

142
Nevertheless, there were no quantities guidelines to achieve these performance
objectives.

The performance-based design framework as originally proposed in Vision 2000


(SEAOC 1995) suggested four levels of design earthquakes defined in terms of their
recurrence frequencies. It defined the performance objectives required for different
building types as shown in Table 6.1. The main goal of this approach was to yield
structures of predictable seismic performance.

Table 6.1 - Performance Objectives Recommended in Vision 2000

Fully Operational Life Safe Near Collapse


Operational
Frequent
Earthquake
O
Level
(50%-30 years)
Occasional
Earthquake
E O
Level
(50%-50 years)
Rare
Earthquake
S E O
Level
(10%-50 years)
Very Rare
Earthquakes S E O
(10%-100 years)
Note: O = Ordinary Buildings; E = Essential/Hazardous Facilities; S= Safety Critical
Facilities

The aforementioned concepts have been adopted by ASCE/SEI 41-06 for the seismic
rehabilitation of existing buildings as discussed in Chapter 1. The development of general
guidelines for performance-based design is currently undertaken by the Applied
Technology Council under the project ATC-58 (Hamburger 2004). These new
developments are moving away from the deterministic performance goals, as suggested
in Vision 2000, towards a probabilistic approach by defining the performance objectives
in terms of the expected annualized losses and the acceptable annual probability of
exceedance of economic losses and casualties of certain degrees.

The performance-based design method requires an iterative design procedure as shown in


Fig. 6.8. In spite of the fundamental framework that has been developed, research is
needed to provide the necessary database and tools to carry out performance-based

143
design. Reliable analytical tools to assess the inelastic performance of different structural
components and systems under various earthquake excitation levels are needed.

Select
Performance
Objectives

Develop
Preliminary
Design

Assess
Performance
Capability

Revise No Yes
Does Design Meet Done
Design
Objectives?

Fig. 6.8 Performance-Based Design Process


(Hamburger 2004)

6.5 Displacement-Based Design

The design approach adopted in ASCE 7 is force based in the sense that the load
resistance of the structure is the primary consideration. Even though the ductility capacity
of the structure influences the design force, it is not explicitly considered (but hidden in
the response modification coefficient R). An alternative design procedure, called the
displacement-based design, has gained attention in recent years. In this approach, the
displacement capacity of a structure is the first and primary consideration. Detailed
coverage of this topic can be found in Priestley et al. (2007) and will be further discussed
in class.

144
References:

1. ASCE/SEI 7-10: Minimum Design Loads for Buildings and Other Structures,
American Society of Civil Engineers, Reston, VA, 2010.
2. ASCE/SEI 41-06. Seismic Rehabilitation of existing buildings. American Society of
Civil Engineers. New York, NY, 2006.
3. Hamburger, R.O. Development of Next-Generation Performance-Based Seismic
Design Guidelines, Proc. of International Workshop on Performance-Based Seismic
Design Concepts and Implementation, P. Fajfar and H. Krawinkler (eds.), Pacific
Earthquake Engineering Research Center (PEER 2004/05), University of California,
Berkeley, CA, pp. 89-100, 2004.
4. Priestley, M. J. N., Calvi, G. M., and Kowalsky, M. J. Direct Displacement-Based
Seismic Design of Structures, IUSS, University of Pavia, Pavia, Italy, 2007.
5. Quantification of Building Seismic Performance Factors (FEMA P695), Applied
Technology Council, Redwood City, CA, 2009.
6. SEAOC. Performance Based Seismic Engineering of Buildings (Vision 2000).
Report, Structural Engineers Association of California, 1995.

145
CHAPTER 7

SEISMIC HAZARD ANALYSIS

This chapter covers some fundamental concepts in seismic hazard analysis. Seismic hazard
analysis is to estimate the probability of exceedance of earthquake ground motions of different
intensities in a given location over a given period of time. It is used to develop seismic maps in
design codes. For such analysis, seismicity data for the location is needed and probability
theories are used. Different methods have used for seismic hazard analysis. One relatively simple
method is described in this chapter.

7.1 Poisson Model for Earthquake Occurrence

Though not exactly appropriate, the probability of earthquake occurrence is often estimated with
the Poisson distribution as follows:

Pn m M , t
t n exp t (7.1)
n!

where Pn m M , t is the probability of having n earthquakes with magnitudes greater than M


over a time period t in a given area, and is the expected number of occurrences per unit time in
that area.

The Poisson model is based on the following assumptions:

(1) Earthquakes are spatially and temporally independent. (Independence)

(2) The probability of an event within a short time interval t is proportional to t and does not
change with time. (Stationarity)

(3) The probability that two seismic events will occur at the same place and at the same time is
zero. (Non-multiplicity)

Past earthquake events have shown that these assumptions are not exactly true as discussed in
Section 7.4.

7.2 Log-Linear Recurrence Relation for Expected Number of Occurrences

A log-linear relation between the expected number of occurrences and the magnitude of an
earthquake for a given area was proposed by Guttenberg and Richter (1954):

ln M (7.2)

where

146
= expected number of earthquakes with magnitudes greater than M per unit time in
a given area

M = Richter magnitude of an earthquake

= a parameter that describes the level of seismicity in a given area

= a parameter that describes the relative frequency of large and small earthquakes

Parameters and in the above log-linear recurrence relation can be determined from the
seismicity data for the given area by using a least-squares fit as shown in Fig. 7.1.

data points

Fig. 7.1 - Log-Linear Relation for Earthquake Occurrence

7.3 Probability of Earthquake Occurrence and Return Period

From the log-linear equation above, Eq. (7.2), we have

exp( M )

Substituting the above equation into the Poisson model as expressed in Eq. (7.1), we have

Pn m M , t
exp M t exp exp M t
n

n!

From the above expression, we can find the probability of having no earthquake with a
magnitude greater than M over a time period t in a given area as

Pm M , t Po m M , t exp exp M t

147
Hence, the probability of having at least one earthquake with a magnitude greater than M in time
t is

Pm M , t = 1 Pm M , t 1 exp exp M t

Based on the Poisson distribution, the return period of an earthquake with a magnitude greater
than M is

exp M
1
TR (7.3)

7.4 Limitations of Poisson Model

The spatial and temporal independence assumed in the Poisson distribution is not consistent with
the geophysical explanation of the earthquake generation mechanism. Furthermore, past
earthquake events have shown that larger earthquakes were usually followed by longer quiet
periods. Sometimes, earthqaukes occurred in clusters, such as foreshocks and aftershocks. To
account for these, other probibility models, such as self-exciting models and self-correcting
models or the combination of the two, have been proposed.

Models based on the elastic rebound theory have been proposed to account for the temporal
dependence of earthquakes. They are called the time-predictable model and slip-predictable
model. However, the main problem with these models is the lack of data to calibrate them.

7.5 Time-Predictable Model

The time-predictable model is based on the premise that the stress level in the earth's crust at
which slip occurs along fault lines is a known constant and that the rate of stress increase is also
a known constant. Based on these assumptions, time to the next earthquake can be predicted
from the time of occurrence and the size of the last earthqauke. However, the model cannot
predict the amount of stress drop when the threshold stress level is reached, and therefore, it
cannot tell the magnitude of the next earthqauke. The concept is shown in Fig. 7.2.

7.6 Slip-Predictable Model

The slip-predictable model is similar to the time-predictable model except that there is no
threshold stress level at which stress release takes place. Instead, stress release is a random
phenomenon that can occur at any time at any stress level; when it occurs, the stress always
drops to zero. Hence, the slip-predictable model cannot tell when the next earthquake will occur
but it can predict the magnitude of the next earthquake if it occurs at a given time by knowing
the time of occurrence of the last earthquake. The concept is shown in Fig. 7.3.

148
Stress

threshold level

constant stress rate

time
Cumulative

Slip

average slip rate

amount of slip
is related to the
magnitude of the
earthquake

time

Fig. 7.2 - Time-Predictable Model

149
Stress

constant stress rate

time
Cumulative

Slip average slip rate

amount of slip
is related to the
magnitude of the
earthquake

time

Fig. 7.3 - Slip-Predictable Model

7.7 Probabilistic Assessment of Peak Ground Motions

Engineers are interested in the probability of occurrence of maximun ground accelerations,


velocities, and displacements ( am , vm , and dm ) at the building site. The probabilistic assessment
of peak ground motions can be carried out with the following four steps (Kiremidjian et al.
1977). A more general coverage of this subject can be found in McGuire (2004).

150
Step 1: Develop an Idealized Model of the Earthquake Source

We can assume that the source of an earthquake is a point, a line, or an area, and that an
earthquake can occur at any point within the source. Each source is assumed to have a uniform
seismicity everywhere and have a uniform focal depth that represents the average depth of all
foci within the source.

Step 2: Determine the Earthquake Recurrence Freguency within the Source

Determine the parameters and in the log-linear recurrence relation based on the seismicity
data obtained for the source.

ln M

Step 3: Determine an Appropriate Attenuation Law for the Location (discussed in Chapter 2)

R (Focal Distance)

Step 4: Assess the Probability of Exceedance of Peak Ground Motions

Information from Steps 1-3 can be used with the Poisson model to arrive at the probability of
exceedance of peak ground motions. The following figure illustrates a curve that gives the
probability of exceedance of different levels of peak acceleration, A, for a particular site over a
time period t.

The above procedure is illustrated in the following example.

151
Example: Probabilistic Assessment of Peak Acceleration Generated along a Line Source

Step 1: Source modeling consider a line source as an example

x2 x

Line Source L
O d
Site of the

Structure
x
R(x)
h
x1

hypocenters

Step 2: Obtain a Log-Linear Recurrence Relation

ln M

The values of and are obtained from the seismicity data for the line source. Let

= /L

where is the expected number of earthqaukes with magnitudes greater than M per unit time
occurring along a unit length of the line source.

As a result, we have

ln = -M or = exp[-M]

where = -ln L

152
Step 3: Develop an Attenuation Law

Let us consider the attentuation law proposed by Esteva (a more general law is discussed in
Chapter 2).

b1 exp b2 m
am
R b5
b4

where

am = peak ground acceleration in cm/sec2


R = focal distance in km
m = Richter magnitude

Let us assume that the area has firm soils. Esteva suggested the following values for firm soil
conditions.

b1 = 5600
b2 = 0.8
b4 = 2
b5 = 40

Step 4: Conduct a Probabilistic Assessment

We first consider a unit length of the line source at a distance x and an earthquake magnitude
M ( x) for that location. Based on the Poisson model (see Sec. 7.3), the probability of having no
earthquake with magnitude greater than M(x) at that location over a time period t is

P ' m M ( x), t exp exp M ( x)t

Hence, for the entire line source, the probability of having no earthquake with magnitude greater
than M(x) is


2
x

Pm M ( x ), t exp exp ' M x dx t

(7.4)

x1

One should note that in the above expression, we specify the limiting magnitude of an
earthquake as a function of x. This is to consider the fact that the magnitude of an earthquake that
can induce a ground motion with a specific peak acceleration at a particular site depends on the
distance of the small line segment from the site, which is a function of x, and that an earthquake
can occur anywhere along the line source. However, if we assume that M is constant along the
line, then the above equation becomes

Pm M , t exp exp M t

153
With the attentuation law adopted here, the probaility of having a maximum ground acceleration
not greater than A along the entire line source can be expressed as

b1 exp b2 m 1

A 2

A, t P m ln R b5 b2 , t

b b4
P am A, t P
R b5
b4
b1

The above expression can be rewritten as

Pam A, t Pm M x , t (7.5a)

where
1



M ( x) ln R b5 b2
b2 b4
A
(7.5b)
b1

R( x ) x 2 d 2 h 2 (7.5c)

With Eqs. (7.5a), (7.5b), and (7.4), we can evaluate P am A, t . Once this is known, the
probability of having a maximum ground acceleration greater than A can be calculated.

Pam A, t 1 Pam A, t

The above probability function can be plotted against A to obtain a probability-of-exceedance


curve.

Modeling of Other Types of Source

The same procedure can be used for a point source and an area source.

For multiple line sources, we have

N
Pm M ( x ), t exp exp i' i M ( x ) dx t
i 1 Li

where N is the number of line sources and i and i are parameters for source i.

Return Period

The return period for a ground motion with am A' is

154
1
TR
p

where p is the probability of having a m A' in each year.

The value of p can be obtained from the probability-of-exceedance curve obtained above. If p
is equal to the probabilty of having am A' in each year, then the probabilty of having am A'
in r years out of n years is

n
Pn r p r 1 p
nr

where the binomial coefficient can be evaluated as

n n!

r r!n r !

Hence, using the above probability expression, we have

P am A' , t N years PN 0 1 p
N

Furthermore, since

P am A' , t N years 1 P am A' , t N years

we have

1 p 1 P am A' , t N years
N

From the above equation, we have

p 1 1 P am A' , t N years N

The above equation can be used to compute p from a probability-of-exceedance curve. The
design earthquake in ASCE 7 corresponds to a 90% probability of not being exceeded in 50
years. With the above equation, we can calculate that the probability of exceedance for the
design earthquake in each year is 0.002105, which corresponds to a return period of 475 years.

155
References:

1. Earthquake Engineering: from Engineering Seismology to Performance-Based Engineering,


Y. Bozorgnia and V. Bertero (Editors), CRC Press, 2004.
2. Kiremidjian, A.S., Shah, H.C., and Lubetkin, L. (1977). Seismic Hazard Mapping for
Guatemala, Report No. 26, John A. Blume Earthquake Engineering Center, Department of
Civil Engineering, Stanford University.
3. McGuire, R.K. (2004). Seismic Hazard and Risk Analysis, Monogragph MNO-10,
Earthquake Engineering Research Institute.

156
APPENDIX A

STABILITY AND ACCURACY OF TIME INTEGRATION SCHEMES

A.1 Stability Analysis

The stability condition of a numerical scheme can be analytically determined. To illustrate the
analytical procedure, let us consider the Newmark method with 0 (explicit scheme) and a
SDOF undamped structure.

Explicit Newmark Method 0

For the case of an undamped structure, the explicit Newmark method is formulated as follows.

Mi 1 Ki 1 fi 1 (A.1)

t 2
i 1 i ti i (A.2)
2

i 1 i t 1 i i 1 (A.3)

Substituting Eq. (A.2) into Eq. (A.1), we have

t 2 2 1
i 1 2i t 2i i f i 1 (A.4)
2 M

Substituting the above equation into Eq. (A.3), we have

2 t 2 t
i 1 2 ti 1 2 t 2 i 1 )ti f i 1 (A.5)
2 M

Multiplying Eq. (A.4) by t 2 and Eq. (A.5) by t on both sides and considering Eq. (A.2), we
have

x i 1 Ax i rf i 1

where

i 1

x i 1 ti 1
t 2
i 1

A-1
1
1 1
2
2

A 2 1 2
1
2
2
2
2

2


0
2
t
r
M
t 2

M

in which t . The above matrix equation is a recursive formulation, which may not be the
most efficient form for numerical computations, but it is convenient for the following stability
analysis.

Let us perform the following recursive substitutions:

x1 Ax o rf1

x 2 Ax1 rf 2 A 2 x o Arf1 rf 2
.
.
.
n
x n A n x o A n i rf i
i 1

For free-vibration response, the last term above is dropped and we have

x n Anxo

Let us consider the eigenvectors and eigenvalues of A by formulating the following eigenvalue
problem.

where

1 2 3

A-2
1 0 0
0 2 0

0 0 3

Assuming that the eigenvalues are distinct and the eigenvectors are linearly independent, we
have

A 1

A2 1 1 21

An n1

Consequently, the recursive equation for response evaluation can be expressed as

n
x n n 1 x o n i 1rf i
i 1

For free-vibration response,

x n n 1x o

Therefore, the solution for the free-vibration response is bounded if and only if n is bounded as
n . Since stability is defined as the condition that the numerical solution for free-vibration
response does not grow without bound, we can conclude that the algorithm is stable if and only if
n is bounded. This means that the spectral radius of A , defined as A max i , must be
less than 1 if the algorithm is stable.

To obtain the eigenvalues of A , one can solve the following characteristic equation:

A I 3 2 A1 2 A2 A3 0

where

A1 = one-half of the trace of A


A2 sum of the principal minors of A
A3 det A

For the explicit Newmark method, it can be shown that

A-3
2 1
A1 1
2 2

1
A2 1 2
2

A3 0

As a result, the eigenvalues of A are

1,2 A1 A12 A2 A iB

3 0
where

2 1
A A1 1
2 2

1/ 2
2 1
2

B A2 A 1
2
1
4 2

i 1

Hence, the properties of the eigenvalues depend on the values of .

2
(1) When , B is imaginary and 1, 2 are real roots.
1

2

2
(2) When , B = 0 and 1, 2 are real double roots.
1

2

2
(3) When , B > 0 and 1, 2 are complex conjugate roots.
1

2

For the case of free-vibration response, we have

x n n 1x o

A-4
From the above equation, we can express the free-vibration displacement response as

n c11n c2 n2

where c1 and c 2 are constants that depend on the initial conditions as well as the eigenvectors of
A.

Hence, when B > 0, for which the eigenvalues are complex conjugate roots, the displacement
response will be sinusoidal; and when B is imaginary or when B is 0, for which the eigenvalues
are real numbers, the displacement response will not be sinusoidal. The latter is, of course,
unrealistic. As shown above, the point at which the response changes from sinusoidal to non-
sinusoidal is

2

1

2

It is called a bifurcation point. It is evident that to have a stable solution, one must have 1,2 1 .
Form the expressions of the eigenvalues derived above, we can see that this is true if and only if

1

2

and
2

Therefore, the above two equations constitute the stability condition for the explicit Newmark
method. Both of these conditions must be satisfied in order that the method is stable. The spectral
radius of A is plotted against in Fig. A.1 with the stability limits and bifurcation points
indicated. This procedure can be used to analyze other integration schemes as well. For more
complicated cases, the eigenvalue problem can be solved numerically.

A-5
Fig. A.1 - Spectral Radius of A for the Explicit Newmark Method

A.2 Accuracy Analysis

The frequency distortion, numerical energy dissipation, and amplitude amplification effect
introduced by an integration method can be quantified analytically using the same procedure
developed for the stability analysis. Again, let us consider the explicit Newmark method as an
example.

Explicit Newmark Method 0

Free-Vibration Response

We see previously that the free-vibration displacement response of an undamped structure


obtained with the explicit Newmark method can be expressed as

n c11n c2 n2

where

1, 2 A iB

2 1
A 1
2 2

A-6
1/ 2
2 1
2

B 1
4 2

in which t . Consider only the case that 2 /( 1 / 2) , for which the numerically
computed response is sinusoidal. Otherwise, the numerical solution will be very different from
the exact solution, which must be sinusoidal for an undamped or under-damped case. As a result,
we only need to consider the case that 1,2 are complex conjugate roots. Hence, we can write

1,2 A iB ea ib

and, therefore,

n c1e na cos nb i sin nb c2 e na cos nb i sin nb e na g1 cos nb g 2 sin nb

with

a
1
2

ln A 2 B 2
B
b tan 1
A

Furthermore, we know that the exact analytical solution for the free-vibration response of an
underdamped structure can be expressed as

t nt e t g1 cos D t g 2 sin D t e tn g1 cos D tn g 2 sin D tn

where D 1 2 . By comparing the numerical solution to the exact analytical solution, we


can see that it is physically meaningful to express a and b as follows.

a t

b D t

in which we define D 1 2 . With the above definitions, we have

n e tn g1 cos tn g 2 sin tn

in which has the same physical meaning as , and has the same meaning as . These
quantities are the apparent frequency and damping ratio exhibited by the numerical solution.

A-7
Since the numerical solution in this case is derived with the condition that 0 , a non-zero
value of represents entirely artificial damping.

Finally, based on the expressions for a and b above, we have

1/ 2
1 1 2 2 1 B
1/ 2
a 2 b2
2

ln ( A B ) tan
2

t
2
t 4 A

a ln A2 B 2

t 2 t

Note that the values of A and B are functions of as shown above. The above equations can be
used to evaluate the apparent frequency and artificial damping. We can see that

A 1
1
and 0
2

A 1
1
and 0
2

Therefore, there is no numerical energy dissipation in the explicit Newmark method when
1 / 2 . The numerical energy dissipation and frequency distortion introduced by the explicit
Newmark method are shown in Figs. A.2 and A.3, respectively.

Fig. A.2 - Numerical Energy Dissipation with the Explicit Newmark Method

A-8
Fig. A.3 - Period Distortion with the Explicit Newmark Method

Forced -Vibration Response

For forced-vibration response, we have

n
x n A n x o A n i rf i
i 1

For 0 0 0 , it can be shown from the above equation that

n F Dn

where

t n 1
Dn f e t n i
sin D t n i
M D
i
i 1

F D te D t c 2 d 2
1/ 2

1
c
2

A-9
2
1
2
2

d 2
2B

c
tan 1 D t
d

We can note that D n is equivalent to a discretized form of Duhamels integral and that F is an
amplification factor introduced by the numerical method. The factor F is plotted in Fig. A.4. It
1
can be observed that for t / T , F is close to 1.
20

Fig. A.4 - Amplitude Amplification with the Explicit Newmark Method

A-10
Implicit Newmark Method

For the Newmark method with 0 , it can be shown that

O 2
1
2

T T
O 2
T

where T 2 / . Hence, it can be seen that when 1 / 2 , the artificial damping is an error of
the first order, while the natural period (or frequency) of the structure has an error of the second
order. The artificial damping ratios introduced by different implicit schemes are compared in
Fig. A.5, where the actual damping of the structure is set to zero. From this figure, one can see
that the -method with equal to -0.1 is most desirable as compared to the other methods
shown in the figure.

Fig. A.5 - Comparison of Artificial Damping Values Introduced by Implicit


Schemes

A-11
APPENDIX B

FAST FOURIER TRANSFORM

Both inverse and forward discrete Fourier transforms can be expressed as

N 1
Bm AnWNnm
n 0

with m = 0, 1, 2,........., N-1, and

WN e i 2 / N

A positive sign in the exponential term represents an inverse transform, while a negative sign
represents a forward transform.

For a forward or inverse transform, a straightforward evaluation of Bm involves the following


operations.


Bm A0 WNm A1 WNm A2 ......... WNm AN 2 WNm AN 1 (B.1)

Hence, there are N operations to evaluate each Bm , with one operation consisting of one
multiplication and one addition. To evaluate N number of Bm 's, there will be N 2 operations.

However, the number of operations can be significantly reduced by taking advantage of the
periodic property of WN and by arranging the computations in a more efficient way. This is the
basic idea of Fast Fourier Transform (FFT) algorithms. Even though there are a number of
different ways to implement FFT, the general concept is best illustrated with the following
approach. Consider that N is so selected that it can be expressed as

N 2M

where M is an integer. This means that N can assume only the following values: 2, 4, 8,.....,1024,
and so on. If this is the case, then the values of m and n can be evaluated with the following
binary operations.

m m0 2m1 4m2 ...... 2 M 1 mM 1


n n0 2n1 4n2 ...... 2 M 1 n M 1

where mi and n i must be either 0 or 1. As a result, a discrete Fourier transform can be expressed
as

B-1
1 1 1
B(m0 2m1 ...) .........
n0 0 n1 0
A(n
n M 1 0
0 2n1 ...)WN( m0 2 m1 ...)(n0 2 n1 ...)

To illustrate how the above computations can be performed in an efficient way, let us consider a
special case that N is 8, and, therefore, M is 3. Consequently, we have

1 1 1
B(m0 2m1 4m2 ) A(n
n0 0 n1 0 n2 0
0 2n1 4n 2 )W88( m1n2 2 m2n2 m2n1 )W84 m0n2 W82 n1 ( 2 m1 m0 )W8n0 ( 4 m2 2 m1 m0 )

Since

W88(int eger ) e i 2 (8 / 8)(int eger ) 1

we have

1 1 1
B(m0 2m1 4m2 ) A(n
n0 0 n1 0 n2 0
0 2n1 4n 2 )W84 m0n2 W82 n1 ( 2 m1 m0 )W8n0 ( 4 m2 2 m1 m0 )

The above summations can be carried out in a three-step sequence:

Step 1: Evaluate the inner-most summation.

1
A1 (m0 , n0 , n1 ) A(n
n2 0
0 2n1 4n 2 )W84 m0n2

where m0 , n0 , and n1 can be either 0 or 1. Hence, we have to evaluate the following


expressions.

A1 (0,0,0) A(0) A(4)


A1 (0,0,1) A(2) A(6)
A1 (0,1,0) A(1) A(5)
A1 (0,1,1) A(3) A(7)
A1 (1,0,0) A(0) A(4)
A1 (1,0,1) A(2) A(6)
A1 (1,1,0) A(1) A(5)
A1 (1,1,1) A(3) A(7)

Note that there are 8 addition operations in the above calculations.

Step 2: Evaluate the subsequent summation.

B-2
1
A2 (m0 , m1 , n0 ) A (m , n , n )W
n1 0
1 0 0 1 8
2 n1 ( 2 m1 m0 )

where m0 , m1 , and n0 can be either 0 or 1. As in Step-1 calculations, it requires eight


operations.

Step 3: Evaluate the outer-most summation.

1
Bm A3 (m0 , m1 , m2 ) A (m , m , n
n0 0
2 0 1 0 )W8n0 ( 4 m2 2 m1 m0 )

where m0 , m1 , and m2 can be either 0 or 1. Again, the above expression requires 8 operations.

Hence, total number of operations required to evaluate all N number of Bms is 24 (i.e., 3x8) or
less depending on the specific implementation adopted, while the straightforward approach using
Eq. (B.1) directly requires 64 (i.e., 82) operations.

From the above example, we can deduce that the total number of operations required in a Fast
Fourier Transform is M x N as compared to N 2 operations otherwise. While the saving in the
computational effort is small if N is a small number such as 8, the difference between FFT and
the straightforward approach is very significant when N is large as compared in Table B.1.

Table B.1 Comparison of FFT with Straightforward Evaluation

Number of Operations
Straightforward FFT FFT/
N M Approach N 2
(M N ) Straightforward
1024 10 1 million 10,240 1%
2048 11 4 million 22,528 0.5%
4096 12 17 million 49,152 0.25%

B-3

Вам также может понравиться