Ubc 1998-0359

D Y N A M I C A N A L Y S I S OF CIVIL ENGINEERING STRUCTURES
USING JOINT T I M E - F R E Q U E N C Y METHODS
by
C A M E R O N JOHN B L A C K
B . A . S c , The University of British Columbia, 1996
A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF
T H E REQUIREMENTS FOR THE D E G R E E OF
M A S T E R OF APPLIED SCIENCE
in
T H E F A C U L T Y OF G R A D U A T E STUDIES
(Department of Civil Engineering)
We accept this thesis as conforming
to the required standard/
T H E UNIVERSITY OF BRITISH C O L U M B I A
August 1998
© Cameron John Black, 1998

In presenting this thesis in partial fulfilment of the requirements for an advanced
degree at the University of British Columbia, I agree that the Library shall make it
freely available for reference and study. I further agree that permission for extensive
copying of this thesis for scholarly purposes may be granted by the head of my
department or by his or her representatives. It is understood that copying or
publication of this thesis for financial gain shall not be allowed without my written
permission.
Department of £l\JXL- EM^J/OgggJ^c^
The University of British Columbia

Vancouver, Canada
DE-6 (2/88)
ABSTRACT
The study of signals whose frequency content changes in time is prevalent in many academic
fields. The objective of this thesis is to demonstrate that joint time-frequency analysis is suitable
for the analysis of civil engineering vibration data.
A discussion of different joint time-frequency analysis methods is presented with emphasis on the
Wavelet Transform, the Wigner Distribution, Cohen's Class functions and the Short Time Fourier
Transform. Most of Cohen's class functions are not directly applicable to the analysis of civil
engineering vibration signals as they are not manifestly positive in the time-frequency plane. The
Wavelet and the Short Time Fourier Transforms are manifestly positive and appear to be suitable
for the analysis of civil engineering vibration data.
This thesis explores the use of joint time-frequency analysis through 5 case studies. These
include the analysis of ambient vibration data obtained from two bridges, data obtained from
shake table testing and strong motion data collected from 2 instrumented buildings. The joint
time-frequency analysis presented in the case studies makes use of the Short Time Fourier Trans-
form.
The dynamic behavior of 2 bridges is analyzed using ambient vibration data. It is shown that joint
Abstract
time-frequency analysis can be used to verify the stability of the dominant frequencies during the
course of testing as well as explain anomalous results obtained from frequency domain analysis.
During shake table testing of an unbonded concrete gravity dam model, upstream motion was
observed at certain combinations of amplitude and frequency of base motion. Joint time-fre-
quency analysis is used to improve the understanding of this phenomenon.
The most promising application of joint time-frequency analysis is for the interpretation of strong
motion data. The response of 2 instrumented buildings during the Northridge and San Fernando
earthquakes is studied using frequency and joint time-frequency analysis techniques. A function
called the Time Frequency Response Function is defined and used, to study many aspects of the
dynamic behavior of structures not explained through typical frequency domain analysis of strong
motion data. This includes the presence of coupling between modes of vibration and the temporal
location of modal response.
The case studies presented in this thesis demonstrate that joint time-frequency analysis is useful
for the study of civil engineering vibration data and should be studied further.
iii
TABLE OF C O N T E N T
ABSTRACT II
TABLE OF CONTENTS IV
LIST O F F I G U R E S VII
LIST O F T A B L E S X
ACKNOWLEDGMENTS XI
DEDICATION XII
CHAPTER 1 INTRODUCTION 1
1.1 Objectives 2
1.2 Scope 2
1.3 Outline 3
CHAPTER 2 BACKGROUND 5
2.1 Structural Dynamics 5
2.2 Frequency Domain Analysis 13
2.3 Joint Time-Frequency Domain Analysis 20
2.4 Literature Review 31
CHAPTER 3 E X P E R I M E N T A L DYNAMICS 33
3.1 Ambient Vibration 34
3.2 Strong Motion Data 38
CHAPTER 4 INTRODUCTION T O C A S E STUDIES 43
4.1 Ambient Vibration Data 43
4.2 Shake Table Data 43
4.3 Strong Motion Data . . 44
4.4 Presentation of Joint Time-Frequency Analysis 46
CHAPTER 5 AMBIENT VIBRATION O F A 3 S P A N BRIDGE 48
5.1 Introduction 48
Table of Contents
5.2 Description of the Bridge 49

5.3 Description of the Ambient Vibration Testing .50
5.5 Joint Time-Frequency Analysis 60
5.6 Conclusions 67
CHAPTER 6 A M B I E N T VIBRATION O F A SINGLE S P A N BRIDGE 68
6.1 Introduction 68
6.2 Description of the Bridge 69
6.3 Description of the Ambient Vibration Testing 71
6.6 Conclusions 87
CHAPTER 7 SLIDING O F A D A M M O D E L 89
7.1 Introduction 89
7.2 Experimental Model Description 90
7.3 Shake Table Testing 94
7.4 Analysis 99
7.5 Joint Time Frequency Analysis 99
7.6 Conclusion 109
CHAPTER 8 A N A L Y S I S O F T H E 1901 A V E N U E O F T H E S T A R S BUILDING 110
8.1 Description 110
8.2 Instrumentation 112
8.5 Conclusion 133
CHAPTER 9 ANALYSIS OF THE S H E R A T O N UNIVERSAL HOTEL 134
9.1 Description 134
9.2 Instrumentation 136
9.4 Joint Time-frequency Analysis 146
9.5 Conclusions 156
C H A P T E R 10 CONCLUSIONS 157
10.1 Applications of Joint Time-Frequency Analysis 157
10.2 Limitations of Joint Time-Frequency Analysis 160
10.3 Suggested Future Research 161
Table of Contents
C H A P T E R 11 REFERENCES 163
APPENDIX A T H E TIME F R E Q U E N C Y R E S P O N S E FUNCTION 167
A.1 The Time Frequency Response Function 167
A. 2 S D O F Example of the T F R F 173
APPENDIX B DATA ANALYSIS S O F T W A R E 176
B. 1 Ambient Vibration Data Analysis Software 176
B.2 Shake Table Data Analysis 177
B.3 Strong Motion Data Analysis Software 177
VI
Figure 2.1: Mass-Spring-Damper Representation of a S D O F System 6
Figure 2.2: Dynamic Free Body Diagram 7
Figure 2.3: Three Degree of Freedom Model 9
Figure 2.4: Real and Imaginary Parts of the F R F 19
Figure 2.5: Hanning Window Function 21
Figure 2.6: Comparison of the Spectrogram and the Wigner Distribution 27
Figure 4.1: Location of the San Fernando and Northridge Earthquakes 45
Figure 4.2: Example of Time-Frequency Plot 47
Figure 5.1: Photo of the University Drive/Crowchild Trail Bridge 50
Figure 5.2: Schematic of Sensor Locations 52
Figure 5.3: Averaged Normalized Power Spectral Densities 55
Figure 5.4: Averaged Normalized Power Spectral Densities (Log-Normal Scale) 56
Figure 5.5: Mode Shapes of the University Drive/Crowchild Trail Bridge 58
Figure 5.6: Wide Band T-F Plot of the Calgary Bridge 62
Figure 5.7: Narrow Band T-F Plot of the Calgary Bridge 63
Figure 5.8: Narrow Band T-F Plot of the Calgary Bridge - Low Frequency Range 64
Figure 5.9: Narrow Band T-F Plot of the Calgary Bridge - Very Low Frequency Range . 65
Figure 6.1: Photo of the Lindquist Bridge 70
Figure 6.2: Photo of the Top of the Bridge 70
Figure 6.3: Photo Showing the Underside of the Bridge 71

List of Figures
Figure 6.4: Sensor Locations and Directions of the Measurements 72
Figure 6.5: Log Normal Plot of the A N P S D s Obtained from March 31 Test 77
Figure 6.6: Log Normal Plot of the A N P S D s Obtained from April 1 Test 77
Figure 6.7: Mode Shapes of the Lindquist Bridge 79
Figure 6.8: Wide Band T-F Plot for the Lindquist Bridge 82
Figure 6.9: Narrow Band T-F Plot for the Lindquist Bridge 83
Figure 6.10: Narrow Band T-F Plot for the Lindquist Bridge - Low Frequency Range . . . . 86
Figure 7.1: Schematic of Dam Model 91
Figure 7.2: Photo of Dam Monolith Model 92
Figure 7.3: Photo Showing Plate Attachment to Dam Model and Table 93
Figure 7.4: Hydrostatic Load Assembly 94
Figure 7.5: Typical Dynamic Test Setup 96
Figure 7.6: The 5 Hz Harmonic Input Record 97
Figure 7.7: 5 Hz Earthquake Record 98
Figure 7.8: T F R Plot of 20 Hz Harmonic Input - No Upstream Motion 101
Figure 7.9: T F R Plot of 20 Hz Harmonic Input - Upstream Motion 102
Figure 7.10: T F R Plot of 22.5 Hz Harmonic Input - Upstream Motion 103
Figure 7.11: T F R plot of 15 Hz Earthquake Input - No Upstream Motion 105
Figure 7.12: T F R plot of 25 Hz Earthquake Input - Upstream Motion 106
Figure 7.13: Schematic Showing Upstream Motion at One Half the Forcing Frequency . 108
Figure 8.1: Photo of the 1901 Avenue of the Stars Building Ill
Figure 8.2: Schematic of Sensor Location During San Fernando Earthquake 113
Figure 8.3: Schematic of Sensor Location During the Northridge Earthquake 114
Figure 8.4: Measured Accelerations During the San Fernando Earthquake 115
Figure 8.5: Measured Accelerations During the Northridge Earthquake 116
Figure 8.6: F R F s of the 1901 Avenue of the Stars Building 119

List of Figures
Figure 8.7: Mode Shapes of the 1901 Avenue of the Stars Building 121
Figure 8.8: 1901 T-F Plot - Northridge Earthquake - Transverse Direction 124
Figure 8.9: 1901 T-F Plot - San Fernando Earthquake - Transverse Direction 125
Figure 8.10: Figure Showing Contribution of Heavy Shaking to F R F 127
Figure 8.11: 1901 T-F Plot - Northridge Earthquake - Longitudial Direction 129
Figure 8.12: 1901 T-F Plot - San Fernando Earthquake - Longitudial Direction 130
Figure 8.13: 1901 T-F Plot - San Fernando Earthquake - Longitudial
Direction - 0 to 0.6 Hz 132
Figure 9.1: Photo of Sheraton Universal Hotel 135
Figure 9.2: Schematic of Sensor Location During the San Fernando Earthquake 137
Figure 9.3: Schematic of Sensor Location During the Northridge Earthquake 138
Figure 9.4: Measured Accelerations During the San Fernando Earthquake 139
Figure 9.5: Measured Accelerations During the Northridge Earthquake 140
Figure 9.6: Frequency Response Functions for the Sheraton 144
Figure 9.7: Mode Shapes of the Sheraton Universal 146
Figure 9.8: Sheraton T-F Plot - Northridge Earthquake -Transverse Direction 148
Figure 9.9: Sheraton T-F Plot - San Fernando Earthquake - Transverse Direction . . . . 149
Figure 9.10: Sheraton T-F Plot - Northridge Earthquake - Longitudial Direction 152
Figure 9.11: Sheraton T-F Plot - San Fernando Earthquake - Longitudial Direction .... 153
Figure 9.12: Sheraton T-F Plot - San Fernando Earthquake - Longitudial
Direction - 0 to 0.8 Hz 155
Figure 11.1: S D O F Example Using the T F R F 175
IX
LIST OF TABLE
Table 5.1: Ambient Vibration Test Setup Locations and Directions 53

Table 5.2: Potential Natural Frequencies 57
Table 5.3: Summary of Mode Shapes 60
Table 6.1: Test Setup Locations and Directions for March 31, 1998 Test 73
Table 6.2: Test Setup Locations and Directions for April 1, 1998 Test 74
Table 6.3: Frequencies of Potential Modes 78
Table 6.4: Summary of Mode Shapes 81
Table 8.1: Frequencies of 1901 118

Table 9.1: Frequencies of Sheraton 143
X
ACKNOWLEDGEMENT
First I would like to thank my parents Trish and Mike Black, my brother Fraser and my fiance
Jennifer Royal for their unconditional support.
I gratefully acknowledge the support of my supervisor Dr. Carlos Ventura. His experience in the
field of dynamic testing and analysis has been invaluable to my research. His encouragement
toward, and interest in, my academic career is much appreciated.
I would like to thank Dr. Ricardo Foschi for being the second reader of this thesis and Dr. Helmut
Prion for his guidance during my graduate studies at the University of British Columbia.
Partial funding for this study was provided by ISIS Canada and the Natural Sciences and Engi-
neering Research Council of Canada. The financial support of B C Hydro is appreciated.
The technical assistance of Mr. Howard Nichol, U B C Earthquake Laboratory technician is appre-
ciated.
EDI Ltd. of Vancouver is acknowledged for the use of the software programs P2, V 2 and U2. I
would also like to thank Mr. Brian Schwartz and Dr. Mark Richardson of Vibrant Technology, Inc.
for their help with Me'Scope.
I would like to thank a number of current and former graduate students of the University of Brit-
ish Columbia for their assistance throughout my graduate studies with special thanks to Dr.
Andreas Felber, Mr. Vincent Lattendresse, Mr. Mahmoud Rezai, Mr. Tomas Horyna and Mr.
Jachym Rudolf.
I would like to thank Dr. Leon Cohen of the City University of New York for his valuable assis-
tance in discussions of joint time-frequency analysis and his interest in the particular application
of joint time frequency analysis to civil engineering.
xi
the memory of Nancy Lytle Black.
CHAPTER 1
Introduction
The perception of our surroundings can be attributed to a large degree on our ability to distinguish
frequencies which change in time. The detection of light of changing color and sound of varying
pitch is an ability which, to many of us, shapes the way we interact with the world. Time-fre-
quency analysis is thus fundamental to many academic fields including engineering, physics,
chemistry, astronomy, geophysics and mathematics.
Frequency domain methods are used widely in the field of civil engineering and in particular,
earthquake engineering. The dynamic properties of structures are characterized by the frequency
at which they vibrate. Earthquakes are described by their spectral, or frequency content. Building
codes estimate the demand on a structure during a seismic event by prescribing the level of accel-
eration to be overcome at different frequencies.
With the importance of frequency in earthquake engineering it is surprising that little research has
been done using joint time-frequency domain analysis. Many structures are assumed to behave
linearly. That is, that the natural frequencies of structures are independent of loading and thus do
not change in time. Although this assumption is false, the computational effort required to ana-
lyze the dynamic behavior of structures with properties that change in time is substantial.
l
Chapter 1 Introduction
1.1 Objectives
The objectives of this thesis are:
• To explore different methods for joint time-frequency analysis

• To demonstrate the applicability of joint time-frequency analysis to the analysis of civil
engineering vibration data
A large number of functions exist which can be used for joint time-frequency analysis. These
range from the widely used and physically intuitive Short Time Fourier Transform to the arcane
Wavelet Transform. A n infinite class of functions known as Cohen's Class functions can also be
used for joint time-frequency analysis. Each of the methods have positive and negative attributes
which make them particularly useful to a specific field. This thesis briefly discusses the methods
available for joint time-frequency analysis.
To demonstrate the application of joint time-frequency analysis to the analysis of civil engineer-
ing vibration data, a number of case studies making use of the STFT are presented. These include
the analysis of data obtained from ambient vibration testing of a three span overpass in Calgary,
Alberta and a single span logging bridge near Kamloops, British Columbia. A n interesting phe-
nomenon observed during shake table tests of a concrete gravity dam model is studied along with
the seismic response of two instrumented buildings during the San Fernando and Northridge
earthquakes.
1.2 Scope
It is not the objective of this thesis to replace standard analysis procedures but rather show that a
joint time-frequency analysis can be useful to further the understanding of dynamic behavior.
An attempt is made to interpret the results of the joint time-frequency analysis, but in many cases
a more comprehensive study would be required to adequately explain the observed behavior. A
comprehensive study would include the analyses of all of the measurements available for a certain
structure not just portions of records and selected measurements. As well, a non-linear finite ele-
ment model should be constructed to verify the observed results.
1.3 Outline
Chapter 2 presents background information including a review of the fundamental concepts in
structural dynamics and analysis in the frequency and joint time-frequency domains. Discussions
on the Fourier Transform, the Short Time Fourier Transform, the Wigner Distribution, Cohen's
Class functions and the Wavelet Transform are presented.
Chapter 3 outlines the fundamental concepts of frequency and joint time-frequency domain analy-
sis of ambient vibration and strong motion data.
An introduction to the applications of joint time-frequency analysis presented in this thesis is
given in Chapter 4.
Chapters 5 and 6 analyze the ambient vibration of two bridges. Chapter 5 analyzes a 3 span
bridge in Calgary, Alberta while Chapter 6 analyzes a single span bridge near Kamloops, British
Columbia. For each bridge, a standard frequency domain analysis is conducted followed by a
joint time-frequency analysis.
3
Chapter 7 presents a study conducted on data collected during shake table tests of a concrete grav-
ity dam monolith structure. A n attempt is made to better understand behavior observed during
testing.
Chapters 8 and 9 explore the application of joint time-frequency analysis for the analysis of strong
motion data. The records from two instrumented buildings during the California, San Fernando
and Northridge earthquakes are studied. Chapter 8 analyzes the 1901 Avenue of the Stars build-
ing while Chapter 9 analyzes the Sheraton Universal Hotel. A typical frequency domain analysis
is conducted to obtain the mode shapes and natural frequencies followed by a joint time-fre-
quency analysis to explain the presence of observed frequency shifts and the contributions of dif-
ferent modes of vibration during the earthquake.
Chapter 10 presents the conclusions of this thesis including suggestions for further research.
A brief study of the Time Frequency Response Function is presented in Appendix A . This
includes a derivation of the function, as well as an example of the application of this concept to
the analysis of single degree of freedom system.
Appendix B briefly describes the software programs used for the frequency and joint time-fre-
quency analysis presented in this thesis.
4
CHAPTER 2
Background
This chapter presents the background information needed to understand the analysis presented in
the following chapters. This includes an overview of the dynamic response of structural systems
followed by a general discussion of frequency and joint time-frequency domain analysis.
2.1 Structural Dynamics

To appreciate the theory and results presented later in this thesis, it is necessary to have an under-
standing of the basic concepts in structural dynamics. The following sections present a cursory
overview of the single and multiple degree of freedom systems. For a detailed investigation of
structural dynamics refer to Chopra (1995).
2.1.1 Single Degree of Freedom System
With proper assumptions, many structures can be idealized as Single Degree of Freedom (SDOF)
systems. Simple structures with nearly concentrated mass m, supported by a nearly massless
structure with stiffness k, in the lateral direction, can be analyzed as a SDOF system. An example
of such a structure would be a walkway canopy, with a heavy concrete roof supported on light
steel sections. To represent this structure as a SDOF system, the necessary assumptions would be
that the light columns have negligible mass and the roof is rigid and constrained to move in one
Chapter 2 Background
direction. Figure 2.1 shows a spring, mass, dashpot SDOF model which could represent the
above mentioned structure.
u(t)
c
n Pit)
M
—AAAAA-
— •
K
•0 0
Figure 2.1: Mass-Spring-Damper Representation of a SDOF System
The easiest way to develop the equilibrium equation of motion for this system is to draw a free
body diagram including the dynamic damping and inertia forces. D'Alembert's principle states
that a structure is in equilibrium at every instant of time with the addition of the fictitous inertia
force. This force acts in the opposite direction of the acceleration with a magnitude equal to the
product of the mass and the acceleration. The dynamic free body diagram of the mass, spring,
damper system is presented in Figure 2.2.
6
u(t) u(t)
Jh.
m
1^.
4 mu(t)
^ — —i
ku(t)
0 0
Figure 2.2: Dynamic Free Body Diagram
The equilibrium equation is:
mu(t) + cu(t) + ku(t) = p(t). (2.1)
Equation 2.1 assumes that the mass remains constant and the damping and stiffness are linear.
The damping and stiffness forces acting on the structure are given by the relative motion of the
mass and the ground where as, the inertia force is proportional to the absolute acceleration. For
earthquake excitation, the acceleration term u(t), in equation 2.1, would be replaced by
u (t) + u(t).
x Where u (t) is the ground motion relative to a reference point and u(t)
g is the
acceleration of the mass relative to the ground. The resulting equation of equilibrium is:
mu{t) + cu(t) + ku(t) = -mu (t).g (2.2)
7
Under free vibration, that is without any external force applied to the structure after initiation of
movement, the mass will oscillate harmonically at the damped circular natural frequency of the
structure co^. This frequency can be expressed as co^ m - (c/2m) . For low levels of
damping, the natural frequency of oscillation can be approximated by the relation co = Jk/m,
where co is the undamped, circular natural frequency, in rad/sec.
Once the equilibrium equation of motion has been formulated it is no longer a structural dynamics
problem but rather a mathematical one. Methods for the solution of the equations of motion are
described in detail by Chopra (1995).
2.1.2 Multi-Degree of Freedom Systems
It is only in very special cases that a structure can be modeled as a single degree of freedom sys-
tem. Generally the vibration of structures will require more degrees of freedom to properly char-
acterize its motion. It is common practice to lump the mass of a structure at its nodes, or points
were at least two members meet, or at the floor level. These nodes can have up to 6 degrees of
freedom, depending on the assumptions made which may constrain the node in some way. A n
example of a constrained node is one where motion in the vertical and rotational directions are
assumed to be negligible compared to the lateral directions. Figure 2.3 shows a three storey
building model with these attributes. The mass of the structure is lumped at the floor levels,
which are interconnected with columns of lateral stiffness k and dashpots with viscous damping
constant, c. This planar, 2D structure can be modeled as a three degree of freedom shear beam
type building; one degree of freedom to describe the horizontal motion of each floor.
8
-*-u(t) 3
-*U (t)
2
-•u,(t)
Figure 2.3: Three Degree of Freedom Model
The equation of motion of the M D O F system in Figure 2.3 would have the same form as the
SDOF equation with the constants m, k and c replaced with 3 by 3 matrices [M], [K] and /C], cor-
responding to the three degrees of freedom and the time dependent variables u(t) and p(t) replaced
by time dependent vectors { U(t)} and {P(t)} . The resulting equation for the M D O F system in
Figure 2.3 is given by Equation 2.3.
m 0 0 u\(t) u\(t) «j(f)

2c -c 0 4k -2k 0
0 m 0 (2.3)
u (t)
2 -c 2c -c u (t)
2
+ -2k 4k -2k u (t)
2 = P (t)
2
0 0 ? 0 -c c_ _0 -2k 2k_
"3(f) "3(f) " (0
3
P (t)
3
1
or,
[M]{U(t)} + [C]{U(t)} + [K]{U{t)} = {/>(*)}. (2.4)
Equation 2.4 is a system of coupled, 2nd order, differential equations in time. To solve this equa-
tion one can make use of the orthogonality property of the mode shapes. Orthogonality means
that one mode shape can not be represented by a combination of other mode shapes. The
undamped, free vibration response of a classically damped structure will have the form:
{U(t)} = £ a „ { ( ) ) „ } s i n ( 0 V + <p„) (2.5)

n
where {<])„} is mode shape vector for mode n, co„ is the circular natural frequency of mode n and
a and (p„ depend on the initial conditions of motion of the system.

n
For classically damped structures, the mode shape vectors of the undamped system form the basis
for the solution of the equation of motion just as the x, y and z planes form a basis for describing
a position in three dimensional space. That is, the structure will vibrate in a shape which is made
up of a linear combination of the normal mode shape vectors. If equation 2.5 is substituted into
equation 2.4, the result is the classical eigenvalue problem. Solution of this problem leads to the
natural frequencies and natural mode shapes of the system. For a discussion on solving the eigen-
value problem and the effects of non-classical damping see Humar (1990).
10
The mode shape vectors for a classically damped structure have the following orthogonality prop-
erties:
WifmWi} = m. and {^} [M]{^}

T
= 0 for i*j
•WifmWi} = k. and W.fmty} = 0 ' for i*j (2.6)
{<M [C]{(|>.} = c. and WifiCMj}

T
=0 fori^y
where { } represents the transpose of a vector and the parameters m,, k

T
t and c, are called
the modal mass, stiffness and damping coefficients associated with the ith mode of the structure.
For a non-classically damped structure the third set of properties, given in Equation 2.6, do not
hold. A classically damped structure can therefore be defined as one that satisfies this property.
This idealized and desirable situation results from assuming that the damping present in the struc-
ture is mass or stiffness proportional (see Humar, 1990).
The displacements of the force-excited system can be determined from the equation:
(2.7)
where y,(r) are the normalized coordinates which are determined from the solution of the uncou-
pled, viscously damped, single degree of freedom system represented by equation 2.8:
11
^ • ( 0 +^,.(0 + ^ . ( 0 = (t) Pi
(2.8)
where p (t) is the modal excitation given by:

t
Pi(t) = Wifpit). (2.9)
Solution of equation 2.8 involves the solution of an uncoupled, 2nd order ordinary differential
equation similar to that given by equation 2.1. Therefore, a n-degree of freedom system can be
broken down into a series of n, single degree of freedom systems, each with modal mass ra,,
modal stiffness k t and modal damping c, . These single degree of freedom systems can be
solved by a number of methods and then related back to the original structure using its associated
mode shape {()),} and equation 2.7. Methods for the solution of the equation of motion are pre-
sented in Chopra (1995).
2.1.3 Comment on Natural Frequencies
A major focus of this thesis is the application of methods that would help determine if natural fre-
quencies of a structural system change in time. However, for truly linear elastic systems, the nat-
ural frequencies, by definition, do not change and thus a structure's natural frequencies are
independent of the loading conditions and time invariant. A natural frequency which changes in
time should be thought of as an instantaneous or temporary natural frequency and the observed
change is relative to the actual natural frequency of the linear system.
12
The term "natural" frequency is used throughout this thesis to describe the dominant frequencies
of all systems, whether they are time invariant or not.
2.2 Frequency Domain Analysis

When examining the dynamic response of structures it is convenient to talk in terms of frequency.
As part of a typical analysis of structural vibration data, one is concerned with the frequency of
the excitation and/or the natural frequency of the structure. As was seen in the proceeding sec-
tion, the vibration of a structure at any point in time can be decomposed into a combination of its
mode shapes oscillating at their associated natural frequencies. If a structure is excited by a forc-
ing function with a frequency equal to a natural frequency of the structure, resonance occurs. At
resonance, the energy present in the input force is added, in phase, to the energy of the structure
and thus the response can grow to levels much larger than normal operating levels.
The introduction of the Fourier Series and Fourier Transform have greatly facilitated frequency or
spectral analysis. The basic mathematical representation of periodic signals is the Fourier Series,
which is a weighted sum of harmonically related sinusoids or complex exponentials. Jean Baptist
Joseph Fourier (1768-1830), a French mathematician, originally used such a series to describe the
phenomenon of heat conduction and temperature distribution through bodies (Proakis and Mano-
laksi, 1996). For an aperiodic signal of finite energy, analysis in the frequency domain is made
possible by the Fourier Transform. The Fourier Transform is a special case of the Fourier series
as the signal is periodic with infinite period. The frequency analysis of aperiodic, continuous time
signals involves the following transform pair:
13
Direct Fourier Transform: X((o) = \ x(t)e' dt m

(2.10)
—oo
oo
Synthesis Equation: x(t) = -z~- [ X(m)e~' 'd(a m

(2.11)
IK J
—oo
Equation 2.10 is the Direct Fourier Transform and Equation 2.11 is the synthesis, or inverse, Fou-
rier Transform. Most time signals which are analyzed today are discrete, or digitized, time sig-
nals. This representation of the signal requires the use of a slightly different transform pair.
The discrete value of x(t), at time t = r A , is written {x } and the series {x } is called a discrete
r r
time series. If T is the length of the signal in seconds and N is the number of discrete points then
the time step A is given by A = T/N. The range of Fourier components X is limited to k = 0
k
to k = N - 1, which correspond to harmonics of frequency co* = 2nk/T = 2nk/NA. The Dis-
crete Fourier Transform (DFT) pair is given by (Newland, 1993):
N- 1
DFT: X = ^ £ x e
k r
i ( 2 n k r / N )
k = 0,1,2,..., (N-l) (2.12)
r = 0
N- 1
Inverse DFT: x = J
r X e~k
K2nkr/N)
r = 0,1,2,..., (N-l) (2.13)
k=0
14
Methods for analysis in the frequency domain have been around since the early 19th century but
the computational effort required for complex, aperiodic signals made the use of this method
practically impossible. In the 1960's an algorithm was developed which enabled the calculation
of the Discrete Fourier Transform and Inverse Discrete Fourier Transform quickly and was hence
named the Fast Fourier Transform or FFT (Cooley and Tukey, 1965). This algorithm makes use
of the harmonic properties of the transform to greatly reduce the computation required. The intro-
duction of the FFT lead to an explosion of research in, and application of, analysis in the fre-
quency domain. Refer to Ramirez (1985) for a derivation of the Fast Fourier Transform.
Analysis which makes use of the Fourier Transform is said to be a frequency domain analysis.
Analysis in the frequency domain encompasses many different techniques depending on the
nature of the experimental data. There are a few basic representations of a signal, or signals, in
the frequency domain which are common to many applications.
If X (co) is the Fourier Transform of a signal, then the auto power spectrum G (co) is given by:
a ou
G (o»
aa = X (co)<(co)
fl (2.14)
where * denotes the complex conjugate. The auto power spectrum is a real valued expression
which is proportional to the energy, or power, in the signal.
15
For 2 time signals x (t) and x (t), the cross power spectrum is given by:
a h
G
a b ^ =X (C0)X W. (2.15)
a b
The coherence function "/^(co), is a measure of the likeness of two signals. It can be used to esti-
mate the amount of noise present in the input-output signals and the presence of non-linearities.
The coherence function between two signals "a" and "b" is given by:
y W = (2.16)
ab
G
a a ^ G
b b ^
where G ((a) is the cross spectrum and G (co) and G ((£>) are the auto spectrums of signals
ah aa hh
"a" and "b" respectively.
The Frequency Response Function (FRF), is used extensively in frequency domain analysis of
structures. It is defined as the response of a structure to an harmonic forcing function of unit
amplitude. The dynamic equation of equilibrium with a unit amplitude harmonic forcing function
is given by:
mx + cx + kx = e . im
(2.17)
If the response x(t) is assumed to be given by x(t) = H(®)e"°', then Equation (2.17) can be
rewritten as:
16
(- co m + /cue + k)H(a)e
2 iat
= e. im
(2.18)
Now the Frequency Response Function, H(co) can be written as the ratio of output motion to input
force:
//(co) = - 1
, . (2.19)
k - a r m + zcoc
Using the frequency ratio P=co/co Equation (2.19) can be rewritten to yield:
n
H(Q) = (2.20)
The F R F H((3), is a complex valued expression and more commonly expressed in terms of its
magnitude IH(P)I and phase angle <j)(P):
|ff(P)| = I . 1
= (2.21)
^(1-P ) 2 2
+ (2P^) 2
«|>(P) = atan^^j (2.22)
It is obvious from Equations 2.21 and 2.22 that, for small damping, the magnitude of the F R F
tends to infinity at resonance (P=l) and the phase angle moves through 90 degrees.
There are 2 ways in which the experimental F R F can be calculated. The choice depends on the
anticipated location of significant noise. If the input signal is denoted 'a' and the output signal as
17
' b \ then the two methods are:
*,«.) - ^ (2.23)
and
G
a *(°>)
Most analysis programs have only one of the Frequency Response Functions; usually H (co) as it {
is easier to compute (Ewins, 1984). (co) is the better estimate when there is more noise in the
output signal and H (a>) is a better estimate when there is more noise in the input signal.
2
The Frequency Response Function is a complex expression and therefore contains real and imag-
inary components. At resonance, there is a peak in the imaginary part of the F R F and a corre-
sponding zero value for the real part. Multiplying the numerator and denominator of Equation
2.20 by the conjugate of the denominator of Equation 2.20 yields:
m ) _ Kd-P ) -^) 2 2
( P )
" fc(l-P ) + 4 p 2 ^ 2
2 2
V-*)
which can be split into its real and imaginary components, as shown in Equation 2.26:
H m -1 iLz£i f i m (2 26)
W
- ( 3 2 ) 2 + 4(32^2 k ( l _ p2)2 + 4 p2£2- ^ >
18
It is obvious that at resonance the real part of Equation 2.26 is zero. Figure 2.4 shows the real and
imaginary parts of Equation 2.26 for low levels of damping (5%). It shows that the imaginary
part of the FRF is a maximum at (3=1. It should be noted however, that as damping increases, the
peak moves closer to the origin.
15
10
-5
-10
•15
F R E Q U E N C Y RATIO
— Real -^Imaginary
Figure 2.4: Real and Imaginary Parts of the FRF
One drawback of frequency domain analysis is that it can only be used for the analysis of linear,
time invariant response. If the dynamic properties of the structure change in time, these changes
will not be identified and will usually introduce anomalous results which complicate the analysis
procedure.
19
2.3 Joint Time-Frequency Domain Analysis

Joint time-frequency analysis involves the analysis of a signal in both the time and frequency
domain simultaneously. There are two basic approaches to analysis in the joint time-frequency
domain. The first approach is to initially cut the signal into slices in time and examine their fre-
quency content. The second approach involves the filtering of discrete frequency bands which are
in turn, sliced into discrete time bands and analyzed for their energy content. The first describes
the Short Time Fourier Transform and Cohen's Class functions, while the latter of the two
approaches describes the Wavelet Transform method of analysis.
2.3.1 Short Time Fourier Transform
Analysis using the Short Time Fourier Transform, (STFT) involves emphasizing the signal at a
desired time t and suppressing it all other times. This is accomplished by multiplying the time
signal s{t) by a window function w(t), centered at time t. The window function is chosen to
leave the signal unaltered at time t and near zero at distant times. The choice depends on the
nature of the data and the desired results. A rectangular window leaves the signal unaltered at
time t but it has poor frequency domain properties. The discontinuity of the window, in the time
domain, introduces ripples in the frequency domain. These ripples are known as sidelobes. To
alleviate the presence of large sidelobe oscillations, a window function which gradually decays
toward zero should be used (Proakis and Manolakis, 1996). Therefore, the Hanning window was
chosen for the analysis in this thesis.

For a Hanning window of M points, the time domain sequence has the form:
w(0 = -I 1 - c o s — J (2.27)
where T is the length of the window in seconds. A plot of Equation 2.27 for a window of 6.4 sec-
onds is given in Figure 2.5.
3.2
Time (sec)
Figure 2.5: Hanning Window Function
The resulting windowed signal has the form:
s (l)
t = s(T)w(T-t) (2.28)
21
where x is the running time variable and w(x -1) is the window function centered at time t.
If the time domain window size is taken to be small, the frequency band of the resulting spectrum
will be large and hence is called a wide band spectrum. Conversely, if the window size is taken to
be large in the time domain, the resulting frequency spectrum is denoted a narrow band spectrum.
The STFT of the windowed signal s,(x) is given by Equation 2.29 and represents the distribution
of frequency around time t.
SX<$) = " i = f e m%
s(x)w(x-t)dx (2.29)
1
J2n J
The energy density spectrum, or spectrogram, is defined as (Cohen, 1995):
-ZCOT
P(t, CO) = S ((0)
t = == f e *™"s(x)w(x-t)dx (2.30)
where | | denotes the absolute value of the expression.
To obtain the density of one variable in a joint density described by two variables, the other vari-
able is integrated out. The resulting density is known as the marginal density or marginal. The
summation of the energy distribution for all frequencies, at a particular time in a signal should
give the instantaneous energy, and conversely, the summation over all times, at a particular fre-
22
quency, should give the energy density spectrum, Therefore, ideally, a joint time-frequency spec-
trogram should satisfy:
a) for the instantaneous energy
oo
J P(t, CG)dCO = \s(t)\ 2

(2.31)
—oo
and
b) for the energy density spectrum
oo
J P(t, (0)dt = |S(C0)| 2

(2.32)
—oo
which are referred to as the time and frequency marginal conditions (Cohen, 1995).
As well as satisfying the marginals, the total energy of the signal should be independent of the
method used to calculate it. This is the case for distributions that satisfy Parceval's theorem,
which is given by:
oo oo
E = j \s(t)\ dt =
2
J \S((0)\ dw,
2
(2.33)
—oo —oo
23
where E represents the energy of the signal.
If a joint time-frequency spectrogram satisfies the marginals then it automatically satisfies the
total energy requirement, although the inverse is not necessarily true (Cohen, 1995).
One potential problem with the STFT is that the energy of the signal is scrambled with that of the
window. This has the effect of introducing energy into the windowed signal which is not present
in the original signal and therefore the time and frequency marginals are not satisfied. This failure
to satisfy the marginals means that results obtained from the STFT may be slightly distorted with
respect to the actual frequency content of the original signal in that, it will not give the correct
answers for averages of functions of frequency or time (Cohen, 1995). Even though the STFT
does not satisfy the marginals it can satisfy the total energy requirement if the window is normal-
ized to have unit energy (Cohen, 1995).
Another problem with the STFT, and all joint time-frequency analysis methods, is that the simul-
taneous resolution in the time and frequency domains is limited by the uncertainty principle. That
is, the frequency and time resolutions are interrelated. The uncertainty principle will be dis-
cussed in a following section.
Another problem with the STFT is the occurrence of leakage. A windowed spectrum is not local-
ized to a single frequency and thus the power of the original signal sequence is spread by the win-
dow to the entire frequency range of the window (Proakis and Manolakis, 1996). In a sense, the
power of the signal is "leaked" into the entire frequency range of the window. To reduce leakage,
a window with smaller sidelobes, such as the Hanning window is used (Proakis and Manolakis,
24
1996). When the length of the windowed section decreases the effect of leakage is more pro-
nounced as the power of the window becomes significant compared to the signal.
When using Short Time Fourier analysis, the choice of the window size is important as it controls
which frequencies, or periods, can be identified by the analysis. Usually it is preferable to have a
window which is longer, in time, than the longest dominant period in the signal analyzed. If this
criteria is not met, the Fourier Transform of the windowed signal may not accurately identify the
spectral content. In the case of long period oscillation, this consideration may seriously limit the
time resolution of a joint time-frequency analysis. For example, the analysis of a structure which
has a fundamental period of 2 seconds requires a windowed section with a minimum width of 2
seconds to accurately identify the fundamental period. As the length of the period increases, the
window size used in the analysis must increase thus limiting the time resolution of the joint time-
frequency analysis.
2.3.2 Wigner Distribution
The Wigner Distribution, W(t, co), was originally introduced in the field of quantum mechanics
and later introduced into signal analysis by Ville (Cohen, 1995). It is the prototype of a set of dis-
tributions that are qualitatively different from the spectrogram. The Wigner Distribution is said to
be bilinear in the signal as the signal enters twice in its calculation as shown in Equation 2.34.
oo
-/cox
dx (2.34)
— O O
25
The Wigner Distribution does not use a window function to estimate the frequency content at a
specific time but rather it folds the signal a time t and emphasizes overlapping frequencies.
A positive attribute of the Wigner Distribution is that it satisfies the time and frequency marginals
and hence Parceval's theorem. A serious consequence of this, at least for the analysis considered
in this thesis, is that a bilinear distribution which satisfies the marginals cannot be positive
throughout the time-frequency plane (Wigner, 1932). A time-frequency distribution which is not
manifestly positive makes the identification of a structure's natural frequencies considerably more
difficult and in some cases impossible. For signals with more than one dominant frequency, or
multi-component signals, such as those found in civil engineering applications, a negative distri-
bution can confuse the identification process. The physical significance of the joint time-fre-
quency distribution is lost and many of the useful functions in the frequency domain, such as the
Frequency Response Function, can not be represented clearly in the joint time-frequency domain
with the presence of negative values.
Another problem with the Wigner Distribution, and to a certain extent, all bilinear distributions, is
the presence of cross terms in multi-component signals. This so called interference results from
folding the signal. Due to interference, the Wigner Distribution is not necessarily zero at times
when the signal is zero and, as well, it is not zero for frequencies that do not exist in the spectrum
(Cohen, 1995). Figure 2.6 shows a comparison of the spectrogram and the Wigner Distributions
for two multicomponent signals. The Wigner Distribution shows significant interference which
makes the identification of distinct frequency components more difficult.
26
(After: Cohen, 1995)
Figure 2.6: Comparison of the Spectrogram and the Wigner Distribution
2.3.3 Cohen's Class Functions
The Wigner Distribution was the first example of a joint time-frequency distribution which was
qualitatively different then the Spectrogram. There exists however, a large family of joint time-
frequency distributions which are similar to the Wigner Distribution. Cohen has shown that
almost all time-frequency representations, of which there are many, can be obtained from:
27
- iQt-ixm + (9M
C(7, co) = —\\\s*(u-^x]s(u-\x](^(Q,x)e dudxdQ (2.35)
where (j)(8, x) is a two dimensional function called the kernel. The Kernel determines the distri-
bution and its properties. The variables u, x and 0 are different for each distribution. It is shown
by Cohen (1995) that the Kernel for the Wigner Distribution is 1 while the Kernel for the spectro-
gram is:
(2.36)
The application of bilinear transforms, such as those belonging to the Cohen Class functions
makes it possible to overcome the time-frequency resolution limitations encountered with the
Short Time-Fourier Transform, since they are not based on signal segmentation (Bonato et al.,
1997). As mentioned for the Wigner Distribution however, the Cohen Class Functions posses
serious problems which make their use impractical for the analysis of civil engineering structures.
These problems include the generation of spurious interference terms and the presence of nega-
tive values in the time-frequency plane.
2.3.4 Wavelet Transform
The Wavelet Transform has come into prominence in the last decade. Its application is primarily
in the field of electrical engineering and is used for such things as image compression.
The Wavelet Transform represents a time domain function f(t) as a linear combination of a fam-
28
ily of basis functions. The basis functions are generated by scaling (dilation or expansion) and
shifting of a single function fc(0 referred to as the mother wavelet (Daubechies, 1998).
Mother wavelets must be oscillatory and have amplitudes that quickly decay to zero away from
the center of the wavelet. A family of wavelet bases 4 /

a/p ( f ) , is given by:
(2.37)
where a is the scaling parameter and b is the shifting variable. The continuous Wavelet Trans-
form, W(a, b), is defined as the inner products of f{t) and fc(r) and is given by equation 2.38:
oo
(2.38)
— O O
where *F* {t) is the complex conjugate of ^

h h(t). The parameters a and b are directly related to
frequency and time respectively and, therefore, the Wavelet Transform can be used to analyze a
signal in the joint time-frequency domain. The form of Equation 2.38 is similar to the Fourier
Transform with the exponential term replaced by the wavelet function. The difference is that the
Fourier transform is a weighted sum of sines and cosines while the Wavelet Transform is a
weighted sum of wavelet bases functions.
An advantage of the Wavelet Transform is that it is a multi-resolution transform. That is, it ana-
lyzes high frequencies with good time resolution but poor frequency resolution and low frequen-
29
cies with good frequency resolution but poor time resolution (Bonato, 1997). The poor time
resolution of the Wavelet Transform, in the low frequency range, is due to the dilation of the
mother wavelet at low frequencies in order to capture oscillations in that frequency range. In the
high frequency range, the wavelet function is compressed to capture the high frequency oscilla-
tions. This has the effect of producing a wide frequency spectrum which is localized in time.
For more information on the Wavelet Transform see Newland (1996) or Rezai and Ventura
(1995).
2.3.5 The Uncertainty Principle
The uncertainty principle is an inescapable property of nature. It was first derived by W. Heisen-
berg in 1927 in the field of quantum mechanics. The physicist C.G. Darwin (grandson of Charles
Darwin) made the connection between the uncertainty principle in quantum mechanics to Fourier
Transform pairs. A statement made by Skolnik (Skolnik, 1980) on the use of the name uncer-
tainty principle, as applied to signal analysis, provides a useful introduction to the principle:
"The use of the word 'uncertainty' is a misnomer, for there is nothing uncertain about the 'uncer-
tainty relation' .... It states the well-known mathematical fact that a narrow waveform yields a
wide spectrum and a wide waveform yields a narrow spectrum and both the time waveform and
frequency spectrum cannot be made arbitrarily small simultaneously."
A signal, or phenomenon that is narrow in one domain must be broad in the other domain, which
is mathematically formulated:
30
AtAf> P (2.39)
The interpretation of P in Equation 2.39 is rather arbitrary and loses importance for practical con-
siderations (Dossing, 1998). What is important is that time and frequency are interrelated and
thus cannot both be made arbitrarily small.
More specifically, if T, and B, are taken to be the duration and bandwidth of time and frequency
of a normalized windowed segment, it can be shown that the uncertainty principle for the Short
Time Fourier Transform is given by (Cohen, 1995):
(2.40)
It is known that for an infinitely short duration signal, the bandwidth of the signal becomes infi-
nite. With this fact in mind, it is intuitively obvious that as the duration of the signal decreases the
frequency bandwidth approaches infinity which is confirmed by the inverse relation between the
standard deviation of time and frequency in Equation 2.40.
For further information on the uncertainty principle refer to Dossing (1998). A proof of the
uncertainty principle and Equation 2.40 is given by Cohen (1995).
2.4 Literature Review

Most of the literature on the use of joint time-frequency analysis is in the field of electrical engi-
neering. There exists relatively few papers on the subject of the joint-time frequency analysis of
31
civil engineering vibration data.
Bonato et al. (1997) investigated the use of Cohen's Class functions for the identification of mode
shapes and natural frequencies of bridges from ambient vibration data. They use relatively
advanced kernel filtering methods in attempt to overcome the problems with the interference
terms resulting from the use of the Wigner Distribution.
Huang et al. (1994) used the Wavelet Transform for the identification of the mass, stiffness and
damping characteristics of simulated structural systems.
Staszewski and Giacomin (1997) use the Wavelet transform for the analysis of mechanical engi-
neering data. They define a wavelet based FRF similar to the TFRF defined in this thesis, to study
road-vehicle interaction data.
32
CHAPTER 3
Experimental Dynamics
The use of analytical methods to calculate the natural frequencies and mode shapes of a structure
requires knowledge of the mass, stiffness and damping characteristics of each of the elements
within the structure. This information may be difficult to acquire, and in some cases impossible
to predict given the potential uncertainty in the boundary conditions or the construction quality.
Therefore, it may be necessary to determine the dynamic properties by experimental means.
There are numerous ways in which a structure can be tested dynamically. The methods used
depend on the size and characteristics of the structure. Typical forced vibration tests include:
shaker, impact and pullback tests. Shake table tests are generally applicable to relatively small, or
scaled structures while ambient vibration tests are suitable for obtaining dynamic characteristics
from a wide range of structures. Another important source of data is the recorded response of
instrumented structures during significant seismic events. This type of data is generally referred
to as strong motion data.
It is important to note that for this study nonparametric evaluation techniques are used. A non-
parametric study is usually only considered qualitative because of the inherent errors in the proce-
dure (Fenves and Desroches, 1994). A parametric study would include the matching or curve-
33
Chapter 3 Experimental Dynamics
fitting of an analytically derived expression for the Frequency Response Function with data
obtained experimentally. As these modal parameter estimation techniques are out of the scope of
this thesis they will not be discussed. For a detailed description of various frequency domain
methods for parametric estimation techniques see Ewins, (1984).
The nonparametric analysis of the strong motion data presented in this thesis is comprised of
visual inspection of frequency and joint time-frequency domain information such as the fre-
quency response functions and spectrograms. The inherent errors in this method stem from the
subjectivity of the decision making process. The decision of which peak corresponds to natural
frequencies, although sometimes obvious, is left to the researcher.
3.1 Ambient Vibration

Ambient vibration testing refers to the measurement of the response of a structure to the surround-
ing, or ambient, excitation. This excitation is provided by wind, human activity, micro-tremors,
traffic and operating machinery. Unlike forced vibration testing, one does not measure the input
into the structure. Natural frequencies and mode shapes are estimated by taking simultaneous
measurements at predetermined locations of the structure.
When using information obtained from ambient vibration measurements, the designer should be
aware that the dynamic properties of a structure at low levels of vibration may be different, and in
some cases significantly different, than those of the same structure subjected to greater levels of
vibration. This should considered when the designer calibrates a computer finite element model
of the structure being studied.
34
The most common method of analysis for ambient vibration data involves using frequency
domain techniques. These techniques are well established and will be discussed in the following
sections. There are however, other methods of analysis such as Random Decrement which are
also capable of analyzing ambient vibration data. The Random Decrement method is a time
domain method which uses the properties of the response of structures to white noise input to cal-
culate the dynamic properties. One advantage of the Random Decrement technique is that it is
capable of estimating the structural damping more accurately than frequency domain techniques.
For more information on the Random Decrement method see Asmusen (1997).
Joint time-frequency analysis enhances the standard analysis procedures as one is able to verify
that the response of the structure does not change significantly during the test. It has been shown
that temperature can significantly effect the frequencies of a structure (Farrar et al., 1998) and
therefore, it is of interest to conduct a joint time frequency analysis on ambient vibration data col-
lected over an entire day.
3.1.1 Frequency Domain Methods for the Analysis of Ambient Vibra-

tion Data
To use the techniques described in this section the structure under study must satisfy a number of
conditions. In general, the structure must not violate the following assumptions:
• The structure behaves as a linear system meaning that superposition is applicable.

• The ambient excitation has white noise characteristics.
• A l l of the modes of interest are excited.
Practice shows that the above assumptions are usually satisfied for most structures.
35
Estimating Natural Frequencies
The natural frequencies are estimated from the peaks in the auto spectra or the Power Spectral
Density, (PSD), of ambient vibration measurements. If X (co) is the Fourier Transform of the
fl
ambient vibration recorded, then the auto spectra G (co) is given by:
Ufl
G
a a ^ = ^
where * denotes the complex conjugate which is obtained by multiplying the imaginary part of
the Fourier Transform by -1. To avoid missing modes of vibration by calculating the PSD at a
node location, the PSDs from all points of the structure are averaged. This so called Averaged
Normalized Power Spectral Density (ANPSD) function gives a better estimate of the natural fre-
quencies (Felber, 1994).
Estimating Mode Shapes
Mode shapes are estimated from the relative motion between a reference location and other loca-
tions chosen for measurement. Each measured location will allow that point to be defined and
therefore, it is important to measure a sufficient number of locations to insure the accurate repre-
sentation of the vibration shape.
At a frequency corresponding to a natural frequency, a structure vibrates harmonically with a
deflected shape equivalent to its mode shape. A l l the information needed to construct the
deflected shape can be obtained from the transfer function T ((Q) between a measured point and
sh
a reference point. For two signals "a" and "b", the transfer function can be expressed as:
36
(3.2)
T
ab™ =
X ((o)X* ((o)
b b
G
«« ( ( 0 )
The absolute value of the transfer function gives the amplitude of the point relative to the refer-
ence sensor and the phase, between the two signals, gives the orientation. That is, if the phase
between the two points is zero, the points are moving in the same direction. Similarly, if the phase
is near 180 degrees the points are moving in opposite directions and the absolute value of Equa-
tion 3.2 would be multiplied by -1. The resulting expression is denoted the Modal Ratio.
For additional information on the theoretical derivation refer to Felber (1993), Diehl (1991) or
Luz (1986).
3.1.2 Joint-Time Frequency Domain Analysis of Ambient Vibration

Data
A joint time-frequency analysis lends itself particularly well to ambient vibration data as the exci-
tation is presumed to have white noise characteristics and thus the peaks in the spectrogram indi-
cate potential natural frequencies.
The spectrogram does not contain any phase information and thus it would not be possible to con-
struct mode shapes using the techniques described in section 3.1.1.
As stated earlier, the motivation for the joint time-frequency analysis is to determine if the domi-
nant frequencies of vibration remain constant over the course of the test, or if factors such as tem-
perature or level of input cause them to shift. Joint time-frequency analysis can also be useful in
37
the explanation of anomalous frequency domain results.
3.2 Strong Motion Data

Strong motion data refers to the data collected from instrumented buildings during significant
earthquakes. It is common for structures to be instrumented in seismically active zones. For
example, during the 1994 Northridge earthquake in California, the California Strong Motion
Instrumentation Program collected data from 77 instrumented structures (Finn et al., 1995).
The structural response information collected during a seismic event is used to study the behavior
of structures under seismic loading and their dynamic properties in general. For example, the
maximum displacement, velocity and accelerations recorded during different events are used in
the design of new structures. The records may be used to calculate the mode shapes and frequen-
cies of the structure. If the building was damaged during the earthquake the strong motion
records could be used to identify the severity and possibly the location of the damage.
There exists a number of methods available for the determination of the dynamic properties of
structures from strong motion data. The most prevalent being frequency and time domain analy-
sis techniques.
Before introducing the frequency domain and joint time-frequency domain analysis techniques, it
is useful to discuss the general behavior of buildings as it provides a starting point from which to
begin the analysis. It is known that buildings exhibit certain trends in the frequency ratios
between the first mode and successive modes. Shear type buildings usually have 1/3/5 ratios
38
where as bending type buildings have 1/5/17 ratios (Chopra, 1995). After the first mode is iden-
tified, the second mode frequency is probably between 3 and 5 times the first mode frequency.
The National Building Code of Canada estimates the fundamental period of a building by n/10,
where n is the number of stories (Associate Committee on National Building Code, 1995). If the
building is 25 stories high the first mode is probably around 2.5 seconds or 0.4 Hz.
3.2.1 Frequency Domain Analysis of Strong Motion Data
The peaks in the Frequency Response Function (FRF) are usually not sufficient to clearly estab-
lish modal frequencies. The level of noise, material non-linearity, minor damage and vibration
induced by the motion of non-structural components yield inconclusive FRF measurements. For
this study, the natural frequencies and mode shapes are identified using several different tech-
niques.
Estimating Natural Frequencies
The strong motion records are used to calculate the FRF and its components, the coherence func-
tion and mode shapes. This information combined with a general knowledge of building behavior
allows the natural frequencies to be estimated.
The F R F is the response of a structure to a unit harmonic input. For a harmonic excitation at a
particular frequency, it is intuitively obvious that the ratio of output motion to input force, at the
resonant frequency, would be greater than at all other frequencies. Therefore, when looking at the
FRF of building response, the peaks may indicate a possible mode of the structure. The frequency
response functions are calculated by taking the record at the base as the input into the structure
39
and all other measurements as the outputs. Strictly speaking, when an acceleration record is used
as the input rather than a force time history, the result is not a Frequency Response Function, but
rather a dimensionless ratio of input to output motion. If the mass of the building was know, the
dimensionless F R F could be divided by the mass to yield the standard FRF as you would have
output motion over input base shear. In this thesis the dimensionless FRF and standard FRF are
used interchangeably.
As mentioned in the previous chapter, F R F is a complex function with both real and imaginary
components. At a frequency which corresponds to a natural frequency, the real part of the F R F
should be zero and the corresponding imaginary component will peak.
The coherence function, y„ (co), is a real valued function of frequency that indicates the fraction
t
of response that is linearly related to the input force. For a structure that remains elastic and has
no extraneous inputs, or noise the coherence function should be unity. Experimental modal anal-
ysis shows that near resonance, the coherence function usually drops from unity. The lack of
coherence can be attributed to bias errors, as the frequency resolution of the analyzer is not fine
enough to describe the very rapidly changing functions encountered near resonance (Ewins,
1995). For the same reason, a lack of coherence is also usually seen at low and high frequencies.
The properties described above are applicable to a single degree of freedom system. For a multi-
degree of freedom system which has well separated modes, the response can be decomposed into
a series of SDOF systems. This is due to the fact that at resonance, the response will be domi-
nated by the mode in resonance and the contribution of all the other modes will be negligible.
40
Therefore, the methods presented for the analysis of SDOF systems are applicable to M D O F sys-
tems with well separated modes.
Estimating Mode Shapes
The mode shapes of an instrumented building are determined by plotting the relative amplitude of
the transfer functions (FRFs) between the input signal, measured at the base, and the output sig-
nals measured throughout the building. This, however, can be misleading if there are not enough
measurement points to accurately display the vibration shape. For instance, the animation of a
building with 1 sensor at the basement and 1 at the roof level will only be capable of animating
the first mode of the structure. In general, the highest mode which can be animated is one less
then the number of floors with sensors.
3.2.2 Joint Time-Frequency Analysis
If the modal characteristics of a structure are thought to be time variant a joint time-frequency
analysis can be useful in improving the understanding of the structure's behavior. These tech-
niques are used to investigate the "softening" phenomena observed during heavy shaking and for
the detection of damage during a seismic event. Softening refers to the reduction in stiffness
sometimes observed during periods of heavy shaking.
A Time-Frequency Response Function (TFRF) is approximated by taking the ratio of the output
time-frequency spectrogram to the input time-frequency spectrogram. The signal used for the
input is the recorded acceleration time history at the base of a structure and the output signal is
taken to be the acceleration time history at the roof of the structure.
41
As described previously, the time-frequency spectrogram P((£>,t) is obtained from
P((0,t) = ISJOl 2
(3-3)
where ^ ( O l 2
is the Short Time Fourier Transform of a time signal.
If P(co, t) in represents the time-frequency spectrogram of the input and P(co, t)out the output
time-frequency spectrogram, the TFRF is given by:
TFRF((x), t) = ° Ut
(3.4)
The TFRF can be used in much the same way as the FRF to estimate natural frequencies with one
exception. To obtain a joint time-frequency distribution, the STFT of the signal must be squared,
as shown in Equation 3.3, and thus the phase information is lost. This has the effect of limiting
the analysis to estimating natural frequencies from peaks in the TFRFs.
The TFRF is described in more detail in Appendix A .
42
CHAPTER 4
Introduction to Case Studies
Five examples of the application of joint time-frequency analysis to the civil engineering vibra-
tion data are presented in this thesis. These examples include the ambient vibration measure-
ments of 2 bridges, the response of a dam model during shake table testing and the response of
two instrumented buildings during the Northridge and San Fernando earthquakes.
4.1 Ambient Vibration Data

The dynamic characteristics of two bridges obtained from ambient vibration measurements are
studied in this thesis. A 3 span bridge located in Calgary, Alberta is presented in Chapter 5, and a
single span bridge near Kamloops, B.C. is presented in Chapter 6. For each bridge, a frequency
domain analysis is conducted to determine the natural frequencies and mode shapes followed by a
joint time-frequency analysis.
4.2 Shake Table Data

During the shake table testing of an unbonded concrete gravity dam model, an interesting phe-
nomenon was observed. At certain combinations of base excitation amplitude and frequency the
model moved upstream. That is, the model moved in a direction opposite to that of the applied
hydrostatic load. Joint time-frequency analysis is performed to better understand this behavior.
43
Chapter 4 Introduction to Case Studies
4.3 Strong Motion Data

A typical frequency domain analysis procedure assumes that the structure behaves in a linear,
time invariant manner and this assumption does not hold in some instances. During an earthquake
the forces carried by the structure are high and in some cases may cause non-linear behavior or
damage. Joint time-frequency analysis of recorded strong motions can be used to investigate non-
linear behavior and determine the temporal changes in frequency content.
To investigate the shift in the natural frequencies of structures often observed during seismic
events, the response of two instrumented buildings is determined through analysis of strong
motion records obtained from two earthquakes. The buildings analyzed are the The Sheraton
Universal Hotel (Sheraton) and the 1901 Avenue of the Stars (1901). These buildings are located
in Southern California, in the Los Angeles area. A similar study, involving instrumented building
in California was conducted by Rezai and Ventura (1996) in which damage was detected.
Each building has complete records from both the San Fernando and Northridge Earthquakes.
The San Fernando earthquake occurred on February 9, 1971 at 7:00 am PST. The hypocenter was
about 27 km from downtown L A with a focal depth of 8.5 km (Oakeshott, 1975). The earthquake
lasted around 45 seconds and had a magnitude of M L = 6.5. The Northridge earthquake occurred
on January 17, 1994 at 4:31 am PST. The hypocenter was about 32 km northwest of Los Angeles
with a focal depth of 19 km (Hall, 1995). The Northridge earthquake lasted around 45 seconds
and had a magnitude of M L = 6.6. A map showing the location of the San Fernando and
Northridge earthquakes and the two buildings is given in Figure 4.1
44
(Adapted from: Finn et al., 1994)
Figure 4.1: Location of the San Fernando and Northridge Earthquakes
45
Having more than one event record for each buildings allows for the comparison of the natural
frequencies obtained from different earthquakes. A frequency domain analysis is conducted on
each building to determine its natural frequencies and corresponding mode shapes. With this
information, an attempt is made to identify large shifts in the natural frequencies of the structures.
To further investigate the observed shift, a joint time-frequency analysis is conducted. Frequency
domain analysis can indicate the presence of a shift, but it is not able to determine whether the
shift was restricted to the heavy shaking or if the structure was damaged in the event. It is impor-
tant to determine if the shift was localized to a specific time or if the natural frequency shifted per-
manently after some point in time. The former would indicate a softening in the structural
components, while the latter would indicate possible damage to the structure.
4.4 Presentation of Joint Time-Frequency Analysis

In this thesis, the joint time-frequency analysis is presented in a tri-plot format such as the one
shown in Figure 4.2. Depending on the application, either the time-frequency spectrogram or
TFRF is shown as a two-dimensional contour plot. The time signals used in the analysis are
shown adjacent to the time axis of the time-frequency plot. In addition, a frequency domain rep-
resentation of the signals is presented along the frequency axis of the time-frequency plot. The
advantage of this format is that the information is presented in the time, frequency and combined
time-frequency domains simultaneously.
46
TFRF Magnitude
6P-2
CO
0.003
• 0.03
[0-1
0.0
60
50
40
CD
30
CD
E
20
10
-300 0 300 -15 0 15 0 2 3 4

Base Accel. Roof Accel. Frequency (Hz)
(cm/s )
2
(cm/s ) 2
Figure 4.2: Example of Time-Frequency Plot
47
CHAPTER
Ambient Vibration of a 3 Span Bridge
5.1 Introduction
The purpose of this study was to determine key dynamic characteristics of the University Drive/
Crowchild Trail bridge immediately after construction as a bench mark for future comparison.
This study was carried out in cooperation with ISIS (Intelligent Sensing for Innovative Structures)
and the City of Calgary in order to investigate the dynamic characteristics of the bridge by way of
ambient vibration testing. The dynamic characteristics of interest are the natural frequencies and
corresponding mode shapes in the vertical and transverse directions, as well as the torsional
modes of the bridge. Sections 5.1 through 5.4 of this chapter are based on a report prepared for
ISIS Canada by Black et al. (1997a).
The natural frequencies and mode shapes of the bridge were determined from ambient vibration
testing. The ambient vibration tests were conducted on August 15, 1997 with the aid of the HBES
(Hybrid Bridge Evaluation System) developed at the University of British Columbia (Felber,
1993).
A joint time-frequency analysis is conducted to ensure that the natural frequencies remain con-
stant throughout testing and to explain some anomalous frequency domain results.
48
Chapter 5 Ambient Vibration of a 3 Span Bridge
5.2 Description of the Bridge

The original University Drive/Crowchild Trail bridge was not capable of carrying the load of
today's heavier traffic and thus it was rebuilt using ISIS Canada technology. It is the first contin-
uous span, steel-free bridge deck in the world (ISIS, 1997). Construction of the bridge, located in
north-west Calgary, Alberta, was completed in mid August, 1997. It is a three span, 90 m over-
pass which carries the South-bound traffic of Crowchild Trail over University Drive (see Figure
5.1). The deck is a 185 mm thick, polypropylene fibre reinforced concrete slab, without any inter-
nal steel reinforcement.
A steel free deck eliminates the problem of corrosion within the deck. The removal of steel rein-
forcement is made possible by the addition of tension straps and polypropylene fibre reinforce-
ment. The straps are regularly spaced across the tops of adjacent girders to provide lateral
restraint and the fibre is included to control crack growth (Newhook and Mufti, 1996). The slab
overhang and barriers on each side are reinforced with glass fibre rods. The 9030 mm wide deck
is supported by five 900 mm deep steel girders. Figure 5.1 shows the North-South elevation view
of the bridge.
The bridge has been extensively instrumented by the University of Alberta in order to monitor the
new steel-free bridge deck technology. The instruments monitor the performance with respect to
the width and depth of the cracks and the fatigue performance of welded connections.
49
Figure 5.1: Photo of the University Drive/Crowchild Trail Bridge
5.3 Description of the Ambient Vibration Testing

The test was conducted on Friday, August 15, 1997 from 9:00 a.m. to 11:00 p.m. The temperature
was around 8 degrees Celsius with moderate to heavy rain for the majority of the test.
The University Drive/Crowchild Trail bridge was not open to traffic at the time of testing and thus
the ambient vibration was produced by wind, human activity and the traffic below. The maximum
levels of recorded vibration in the vertical and transverse directions were 1.7 milli-g's. These val-
ues were considered sufficient to proceed with further analysis of the data.
50
Eight accelerometers were used for the ambient vibration measurements. Forty-six locations
were chosen for measurements, 44 on the bridge deck and 1 near the base of each pier. A total of
14 setups were needed to accurately capture the natural frequencies and mode shapes. Two accel-
erometers were used as reference sensors for the first 11 setups and 3 accelerometers were used in
the remaining 3 setups. Figure 5.2 shows the location and direction of the ambient vibration mea-
surements taken during the test.
Table 5.1 gives the location of each of the sensors during the test. The first column is the setup
number. The second column lists the first five characters of the file name created. The remaining
three characters identify the channel number. The eight columns (labelled Chan. 1 through Chan.
8) refer to the 8 channels of the data acquisition computer. Each number-letter combination refers
to sensor location (see Figure 5.2) and the direction of measurement: vertical (V), longitudial (L)
or transverse (T). Sensor location 18 was used as the reference location (R) with the transverse
and vertical directions measured during all tests and the longitudial direction measured for tests
12-14. The attenuation of the signals are given in the last column of the table. The filter cut off
was set to 50 Hz, which corresponds to the natural frequency of the accelerometers. A total of
65536 samples per channel, per test were recorded at a sampling rate of 100 samples per second
(sps).
51
29830 mm 32818 mm 30230 mm

H M m w
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43
Is
Ill 1
-•5000 m
ill!
t
it
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42
PLAN VIEW
44 45
ELEVATION VIEW
Sensor positive
direction convention: t = ut of plane of page, o = into plane of page, « - = in plane of page
0
Figure 5.2: Schematic of Sensor Locations
52
Setup File Chan. Chan. Chan. Chan. Chan. Chan. Chan. Chan. Atten.
1 2 3 4 5 6 7 8 (dB)
1 CTB01 R18V 19V 18V 19V 18V R18T 18T 18T 18
2 CTB02 R18V 23V 22V 25V 24V R18T 22T 24T 18
3 CTB03 R18V 21V 20V 27V 26V R18T 20T 26T 18
4 CTB04 R18V 17V 16V 29V 28V R18T 16T 28T 18
5 CTB05 R18V 15V 14V 31V 30V R18T 14T 30T 18
6 CTB06 R18V 13V 12V 33V 32V R18T 12T 32T 18
7 CTB07 R18V 11V 10V 35V 34V R18T 10T 34T 18
8 CTB08 R18V 9V 8V 37V 36V R18T 8T 36T 18
9 CTB09 R18V 7V 6V 39V 38V R18T 6T 38T 18
10 CTB10 R18V 5V 4V 41V 40V R18T 4T 40T 18
11 CTB11 R18V 3V 2V 43V 42V R18T 2T 42T 18
12 CTB12 R18V 1V OV R18L 42V R18T OT 42L 18

**
13 CTB13 R18V 44L 44V R18L 45V R18T 44T 45T
14 CTB14 R18V OL — R18L — R18T ... 45L 18
Table 5.1: Ambient Vibration Test Setup Locations and Directions
** attenuation of 24 for channel 7 and 18 for others

The natural frequencies and mode shapes in the vertical direction were estimated using 64 records
of vertical ambient vibration measurements. Forty records from measurements in the transverse
direction were used and 7 in the longitudial direction. The torsional response was estimated from
the difference in the vertical motions obtained from opposite sides of the deck. This is possible as
the deck is assumed to be rigid and therefore the difference in the motion, at each side of deck,
gives a reasonable estimate of the torsion.
53
5.4.1 Natural Frequencies
The power spectral densities of the above measurements were averaged to give the Averaged Nor-
malized Power Spectral Density (ANPSD). The A N P S D permits a convenient way to display, in
one single plot, the most significant frequencies present in a series of recorded motions in a cer-
tain direction of the structure (Felber, 1993). Figure 5.3 shows the ANPSDs for the vertical,
transverse and torsional modes for two different frequency ranges, computed using 64 averages.
The most significant peaks in the frequency range of 0-20 Hz have been identified in Part B of
this figure. Other peaks in the 20-50 Hz range were also identified in the analysis, but were not
considered for further study since the frequency range of interest was 0-20 Hz.
Not all of the peaks in Figure 5.3 may correspond to a natural frequency (Bendat and Piersol,
1993) and additional tools are needed to confirm whether or not a peak of the ANPSD is associ-
ated with a natural frequency. Furthermore, a natural frequency of the bridge is not necessarily at
a peak of the ANPSD, but it may be in the vicinity of the peak. This is due to the presence of
damping in the system, as well as the effect of averaging multiple records.
More detailed information can be obtained from log-normal plots of the ANPSDs as shown in
Figure 5.4. The presence of peaks with low amplitude can be easily determined from these plots.
For example, a peak around 18 Hz can be easily seen in Figure 5.4, but not in Figure 5.3. This
happens to be one of the natural frequencies of the bridge, as will be described in the following
sections.
0 2 4 6 8 10 12 14 16 18 20
Frequency (Hz)
b) 0 to 20 Hz Range
Figure 5.3: Averaged Normalized Power Spectral Densities
55
100.000
10.000
1.000
Q
CL
0.100
0.010
0.001
Frequency (Hz)
100.000 -§
10.000 -4
Q
1.000 -i
CO
0.100
0.010 -g
0.001
6 8 10 12 14 16 18 20
Frequency (Hz)
Figure 5.4: Averaged Normalized Power Spectral Densities (Log-Normal Scale)
56
From the peaks in the ANPSDs it is seen that there are 15 possible modes below 20 Hz. A sum-
mary of the frequencies identified from this analysis is given in Table 5.2.
Frequency
Peaks
(Hz)
1 0.29
2 1.95
3 2.83
4 3.22
5 3.81
6 4.10
7 4.79
8 5.18
9 7.13
10 9.67
11 10.94
12 12.99
13 15.63
14 17.87
15 19.24
Table 5.2: Potential Natural Frequencies
5.4.2 Mode Shapes
The ambient vibration records obtained were used to determine the transfer function, coherence
and phase between the reference sensors and all other sensors. This information was used to con-
firm if the peaks in the A N P S D corresponded to natural mode shapes or to a mode of vibration at
that frequency. The peaks at 0.29 Hz and 1.95 Hz did not correspond to a natural mode of vibra-
tion and were disregarded from further analysis.
Selected mode shapes determined to be the natural modes of the vibration of the bridge are shown
in Figure 5.5. Each mode is shown in three views: plan, elevation and 3-dimensional. The modes
are not presented as sequential modes in a single direction, as numerous coupled modes exist.
Rather, the modes are shown in order of frequency.
Fundamental Vertical Mode (2.78 Hz) Fundamental Torsional Mode (3.13 Hz)
Figure 5.5: Mode Shapes of the University Drive/Crowchild Trail Bridge
58
Lightly Coupled Vertical/Torsion (3.76 Hz) Torsional Mode (4.05 Hz)
Figure 5.5: Mode Shapes of the University Drive/Crowchild Trail Bridge (Continued)
59
A total of thirteen modes were identified. Table 5.3 is a summary of the findings. The sequential
number of the mode and the frequency and period of each mode is presented with a brief descrip-
tion. The bracketed term indicates the sequential number in a certain direction and the * indicates
the presence of coupling.
Mode Fre- Period Description

quency (sec)
(Hz)
1 2.78 0.36 Fundamental Vertical Mode (1 ver)
2 3.13 0.32 Fundamental Torsional Mode (1 tor)
3 3.76 0.27 Coupled Mode [Vertical (2 ver*) / Torsion (2 tor*)]
4 4.05 0.25 Torsional Mode (3 tor)
5 4.64 0.22 Vertical Mode (3 ver)
6 5.18 0.19 Coupled Mode [Torsion (4 tor*) / Transverse (1 tra*)
7 7.13 0.14 Coupled Mode [Torsion (5 tor*) / Transverse (2 tra*)]
8 9.13 0.11 Vertical Mode (4 ver*)
10 12.84 0.08 Fundamental Transverse Mode (6 tra)
11 15.77 0.06 Torsional Mode (6 tor*)
Table 5.3: Summary of Mode Shapes
5.5 Joint Time-Frequency Analysis

A Short Time Fourier Analysis is used to obtain time information of the response. Due to the
length of ambient testing and the limitations on the array size of the software used for the calcula-
60
tion of the spectrogram, only a relatively small number of samples from each test can be analyzed
simultaneously. A 10 second segment from one accelerometer, during each test, was extracted
from each test and joined together to form a single file. The spectrogram, P((0, t), is calculated
for each file with different window sizes to obtain the best time-frequency resolution for a given
signal. The reference sensor was the obvious choice as the sensor because it remains in the same
location for the entire duration and thus should give consistent readings.
Each of the 14 tests recorded 65536 points sampled at 100 sps. From each test a 10 second sam-
ple (1000 points) was extracted and joined together to create a 14000 point record. This record
was analyzed using the STFT with two different window sizes; 64 point window yielding a wide
band spectrogram and a 256 point window yielding a narrow band spectrogram. At a sampling
rate of 100 sps, a 64 point window corresponds to a 0.64 second non-zero signal and the 256 point
window corresponds to a 2.56 second non-zero signal. Both of the signals lengths are greater than
the fundamental period of the structure. The wide and the narrow band Time-Frequency spectro-
grams are shown as contour plots in Figures 5.6 and 5.7, respectively. Figure 5.8 shows the nar-
row band time-frequency plot for frequencies close to the fundamental vertical mode (2.78 Hz).
This plot is shown because the response, and hence the spectrogram of the response is dominated
by the mode(s) near 10 Hz. Figure 5.9 shows the narrow band time-frequency plot for 0 to 3 Hz
range. The time-frequency domain information is presented in time-frequency plots as described
in Chapter 4.
61
The consequence of the uncertainty principle, discussed in Chapter 2, is clearly evident in these
figures as the wide band time-frequency plots lack the frequency resolution of the narrow band
time-frequency plots. This is shown as "spreading" along the frequency axis.
On first inspection of the time-frequency plot presented in Figures 5.6 and 5.7 there would appear
to be a slight shift at or near 10 Hz as the test progresses. The center of the contours shift slightly
to a lower frequency. This is a result of a lack of excitation at the frequencies in the area above 10
Hz rather than a shift. There exists two possible reasons for this lack of excitation. First, the time
segments chosen to represent the later tests may not have contained this excitation. To be certain
that this was not the case, a number of other segments were chosen and analyzed in a similar fash-
ion, yielding similar results. The other reason for the apparent lack of excitation is that the testing
was completed over a 14 hour span, with the last 4 tests being conducted in the late evening. This
may explain the decrease in excitation which is visible in the time history and inferred from the
contour plot.
From Figure 5.8, it is seen that there is no discernible shift in the fundamental vertical and funda-
mental torsional modes, represented by the A N P S D peaks at 2.78 and 3.13 Hz, respectively.
The A N P S D contains a peak at 1.95 Hz which was later discarded after analyzing the vibrating
shape at, or near, that frequency. It is of interest therefore, to study the joint time-frequency
response of that peak in an attempt to explain its origins. The time-frequency plot shown in Fig-
ure 5.9, shows little or no response at 1.95 Hz except during the first three tests (0-30 seconds). It
can be inferred from the plot that this frequency was only present for a brief time and is therefore
66
is not likely a natural mode of vibration.
5.6 Conclusions
The frequency domain analysis of ambient vibration data obtained at the University Drive/Crow-
child Trail bridge identified thirteen modes below 20 Hz. The fundamental frequencies in the ver-
tical and transverse directions were identified as 2.78 Hz and 12.84 Hz respectively while the
fundamental torsional mode was identified as 3.13 Hz. The bridge has a number of coupled
modes including several highly coupled modes at 3.76 Hz, 5.18 Hz and 7.13 Hz.
The joint time-frequency analysis was useful in that it showed that there were no significant fre-
quency shifts during the testing. The peaks near 10 Hz and those corresponding to the fundamen-
tal vertical and torsional modes of the bridge remained relatively stable during the test. There are
a few possible reasons for the lack of observable shifts. Testing was completed on a very cold,
rainy day, when the temperature remained relatively constant, as opposed to a sunny day when the
temperature may change by 10 or more degrees Celsius from mid-day to late evening. Another
reason is that the resolution of the joint time-frequency domain analysis is not capable, especially
for the case around 10 Hz, of identifying slight shifts in frequency. This lack of resolution is not a
significant problem as a shift of that magnitude would not affect the outcome of testing.
A joint time-frequency analysis of the peak near 1.95 Hz showed that this frequency was only
present for a very brief time and thus not likely a mode of the bridge. Although this conclusion
could be drawn from the standard frequency domain analysis alone, it is reassuring to see that the
peak in frequency is very localized in time.
67
CHAPTER 6
Ambient Vibration of a Single Span Bridge
6.1 Introduction
The results of ambient vibration testing of the Lindquist Bridge located near Kamloops, British
Columbia are presented in this chapter. This study was carried out in order to investigate the
dynamic characteristics of the bridge by way of ambient vibration testing (Black and Ventura,
1998a). It is a complimentary study to one conducted for Reid Crowthers, in which the bridge
was instrumented with 6 strain gages, four on the girders, at the third and center positions, and 2
on the straps at the same location (Cook et al. 1998). The Lindquist bridge is of particular interest
as it carries relatively high loads. It serves as a logging bridge transporting logging trucks over
Darlington Creek. Sections 6.1 through 6.4 of this chapter are based on a report prepared by
Black and Ventura (1998a).
Similar to the University Drive/Crowchild Trail Bridge described in Chapter 5, the purpose of this
study was to determine key dynamic characteristics of the Lindquist as a bench mark for future
comparison. The dynamic characteristics of interest were the natural frequencies and correspond-
ing mode shapes in the vertical and transverse directions, as well as the torsional modes of the
bridge.
68
Chapter 6 Ambient Vibration of a Single Span Bridge
Two complete ambient vibration tests were conducted, one on March 31 and one on April 1, 1998
with the aid of the HBES (Hybrid Bridge Evaluation System) developed at the University of Brit-
ish Columbia (Felber, 1993).
A joint time-frequency analysis is conducted to determine if the dominant frequencies shifted
through the testing and to explain anomalous frequency domain results.
6.2 Description of the Bridge

The original wooden bridge spanning Darlington creek was recently replaced with a steel-free
bridge deck supported by two steel girders. It is a 24 m single span bridge which carries logging
traffic. The deck is a 185 mm thick, polypropylene fibre reinforced concrete slab without any
internal steel reinforcement.
A steel free deck eliminates the problem of corrosion within the deck. The removal of steel rein-
forcement is made possible by the addition of tension straps and polypropylene fibre reinforce-
ment. The straps are regularly spaced across the tops of adjacent girders to provide lateral
restraint and the fibre is included to control crack growth (Newhook and Mufti, 1996). Figure 6.1
shows the North-South elevation view of the bridge; Figure 6.2 is a photo looking over the top of
the bridge and Figure 6.3 is a view from below the bridge showing the girders and the straps.
69
Figure 6.3: Photo Showing the Underside of the Bridge
6.3 Description of the Ambient Vibration Testing

The vibration measurements were performed twice, once on March 31 and again on April 1, 1997.
The temperature was around 6 degrees Celsius with mainly sunny conditions.
Excitation of the bridge was provided by wind and human activity. Eight accelerometers were
used for the ambient vibration measurements. Thirty locations were chosen for measurements, 26
on the bridge deck and 4 on the abutments. A minimum of 8 setups on each day were needed to
accurately capture the natural frequencies and mode shapes. Two accelerometers were used as
reference sensors and six were used as roving sensors.
71
Figure 6.4 shows the location and direction of the ambient vibration measurements recorded dur-
ing the test.
2 4 6 c5 10 12 14 16 18 20 22 24 26 28
i t, 1 :t 1 , 1 ]t i ,t 1 1 ± = £ 1
.3 5 7 9 11 13 15 17 19 21 23 25 27. 29
t Accelerometer direction
• Accelerometer direction
out of page
o Accelerometer
Figure 6.4: Sensor Locations and Directions of the Measurements
72
Tables 6.1 and 6.2 list the location of each of the sensors during the tests on March 31 and April
1, respectively. The first column of each table corresponds to the setup number and the second
column lists the first five characters of the file name created. The eight columns (labelled Chan. 1
through Chan. 8) refer to the 8 channels of the data acquisition computer. Each number-letter
combination refers to sensor location in Figure 6 and the direction of measurement, vertical (V),
longitudial (L) or transverse (T). A l l measurements were sampled at 200 sps for 327.67 seconds
giving 65536 points per channel, per test.
Chan. Chan. Chan. Chan. Chan. Chan. Chan. Chan. Atten.

Setup File 1 2 3 4 5 6 7 8 (dB)
1 LB01 R15V 15V 17V 14V 16V R15T 15T 17T *
2 LB02 R15V 13V 19V 12V 18V R15T 13T 19T **
3 LB03 R15V 11V 21V 10V 20V R15T 1 IT 2 IT 0
4 LB04 R15V 9V 23V 8V 22V R15T 9T 23T 0
5 LB05 R15V 7V 25V 6V 24V R15T 7T 25T 0
6 LB06 R15V 5V 27V 4V 26V R15T 5T 27T 0
7 LB07 R15V 3V 29V 2V 28V R15T 3T 29T 0
8 LB08 R15V IV 19V ov 18V R15T IT 19T 0
Table 6.1: Test Setup Locations and Directions for March 31,1998 Test
*segments 1-4: V=12, H=6; segments 5-8: V=6, H=6

**segments 1-5: V=24, H=12; segments 6-8: V=6, H=6
73
Atte
Chan. Chan. Chan. Chan. Chan. Chan. Chan. Chan. n.
Setup File 1 2 3. 4 5 6 7 8 (dB)
1 LB01 RUT 15V 17V 14V 16V R11V 15T 17T 0
2 LB02 RUT 13V 19V 12V 18V R11V 13T 19T 0
3 LB03 RUT 11V 21V 10V 20V R11V 11T 2 IT 0
4 LB04 RUT 9V 23V 8V 22V R11V 9T 23T 0
5 LB05 RUT 7V 25V 6V 24V R11V 7T 25T 0
6 LB06 RUT 5V 27V *4V 26V R11V 5T 27T 0
RUT *3V 29V *2V 28V IR11V *!3T !29T ##

7 LB07
8 LB08 RUT IV *15V OV *14V IR11V !1T !15T **
9 RUT 3V 4V 2V *6V R11V 3T *4T 0
Table 6.2: Test Setup Locations and Directions for April 1,1998 Test
** channels 6,7,8 at attenuation of 6; channels 1-5 at 0 attenuation

* redundant records determined to be of low quality

The vertical and torsional natural frequencies and mode shapes were estimated using 28 vertical
ambient vibration measurements on each day. The torsional response was estimated from the dif-
ference in the vertical motions obtained from opposite sides of the deck. This is possible as the
deck is assumed to be rigid and therefore the difference in the motion, at each side of deck, gives
a reasonable estimate of the torsion. Twenty-Eight records from measurements in the transverse
direction were used to obtain the natural frequencies and mode shapes in the transverse direction.
The testing on March 31 was conducted with the reference sensor at the center location. As a
result, the symmetric modes are well defined but the asymmetric modes are not. The problem
74
with the asymmetric modes is due to the normalization of the mode shape to the small value at the
center. The testing on April 1 had the reference sensor at the third point of the bridge and thus it
provides nicely defined asymmetric modes. Ideally, one would have two reference sensors, one at
the center and one at another point, not likely to be a node location, during each test. This how-
ever may be difficult due to hardware and time limitations. If only one test is conducted the refer-
ence position would be chosen in a location where a node is not likely to be present in order to
maximize the number of modes that can be properly identified.
Other testing activity was present during the ambient vibration testing and thus the data may
include spurious frequencies. These may include the revolutions of the truck or car engines in the
vicinity or a generator which was used for a power supply.
The power spectral densities of the above measurements were averaged to give the Averaged Nor-
malized Power Spectral Density (ANPSD). The A N P S D permits a convenient way to display, in
one single plot, the most significant frequencies present in a series of recorded motions in a cer-
tain direction of the structure (Felber, 1993). Figures 6.5 and 6.6 show the ANPSDs for the verti-
cal, transverse and torsional modes computed using 64 averages. The magnitude of the transverse
A N P S D should not be compared with the vertical and torsional magnitudes as it is normalized to
itself.
It is rare to have 2 complete ambient vibration tests for the same bridge, under similar ambient
conditions. It is of interest therefore, to look at the differences in the ANPSDs. The peaks at 13.3
75
Hz, 17.3 Hz, 18.2 Hz and 22.9 Hz are more pronounced for the March 31 test than the April 1 test.
The ANPSD calculated from the March 31 test has many peaks near 30 Hz, as compared to the
single dominant peak in the April 1 test. Another interesting difference is the shift in dominant
frequency observed near 40 Hz. The peak at 41.4 Hz is present in both ANPSDs but the peak
seen at 40.5 Hz in the April 1 test is absent in the March 31 test.
It is evident from the observed variation that results obtained from ambient vibration testing are
only approximate. The uncertainty arises from the variability in ambient conditions which affect
the measured frequencies.
The modal frequencies were estimated from the measurements taken on March 31 as the peaks
are generally more pronounced. Not all of the peaks in Figure 6.5 correspond to a natural fre-
quency (Bendat and Piersol, 1993) and additional tools are needed to confirm whether or not a
peak of the ANPSD is associated with a natural frequency. For more information on the normal-
ization scheme see Felber (1993). From the peaks in the ANPSDs it is seen that there are 9 possi-
ble modes below 50 Hz. A summary of the frequencies identified from this analysis is given in
Table 6.3.
76
100.0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Frequency (Hz)
Figure 6.5: Log Normal Plot of the ANPSDs Obtained from March 31 Test
100.0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Frequency (Hz)
Figure 6.6: Log Normal Plot of the ANPSDs Obtained from April 1 Test
77
Frequency
Mode (Hz)
1 6.3
2 7.3
3 13.2
4 17.3
5 18.2
6 22.9
7 26.1
8 30.2
9 41.4
Table 6.3: Frequencies of Potential Modes
6.4.2 Mode Shapes
The ambient vibration records were used to determine the transfer function, coherence and phase
between the reference sensors and all other sensors. This information was used to confirm if the
peaks in the ANPSD corresponded to natural mode shapes or to a mode of vibration at that fre-
quency.
The vibration shapes which were determined to be natural modes of the bridge are shown in Fig-
ure 6.7. Each mode is shown in three views: plan view at the top of the figure, elevation view at
the middle and 3 dimensional view at the bottom. The modes are not presented as sequential
modes in a single direction, as coupled modes exist. Rather, the modes are shown in order of fre-
quency.
78
Fundamental Vertical Mode (6.3 Hz) Fundamental Torsional Mode (7.32 Hz)
Torsional/Abutment (18.2 Hz) Torsional Mode (22.85 Hz)
Vertical Mode (30.2 Hz) Torsional Mode (41.4 Hz)
Figure 6.7: Mode Shapes of the Lindquist Bridge (Continued)
80
A total of eight modes were identified. Table 6.4 gives a summary of the findings. The sequential
number of the mode, the frequency and period of each mode is presented with a brief description.
The bracketed term indicates the sequential number in a certain direction.
Frequency Period
Mode (Hz) (sec) Description
1 6.3 0.16 Fundamental Vertical Mode (1 ver)
2 7.3 0.14 Fundamental Torsional Mode (1 tor)
3 13.2 0.076 Fundamental Transverse (1 tra) w/ torsional coupling
5 18.2 0.055 Torsional-Abutment Mode (3 tor)
Table 6.4: Summary of Mode Shapes

Each of the 8 tests conducted on March 31 recorded 65536 points sampled at 200 sps. From each
test a 10 second sample (2000 points) was extracted and joined together to create a 14000 point
record. This record was analyzed using the STFT to yield the joint time-frequency spectrogram.
At a sampling rate of 200 sps, the 64 point window corresponding to the wide band spectrogram
has a length of 0.32 seconds which is longer than the fundamental period. The narrow band spec-
trogram has calculated with a window length of 256 points, or 1.28 seconds.
As described in Chapter 4, the time-frequency information is presented in a time-frequency plot.
Figures 6.8 and 6.9 present the wide and narrow band time-frequency plots respectively.
81
Amplitude Frequency (Hz)

Figure 6.8: Wide Band T-F Plot for the Lindquist Bridge
82
Both the wide and narrow band time-frequency plots shown in Figures 6.8 and 6.9, respectively
indicate a marked shift in the dominant frequency near 30 Hz. This shift is most pronounced dur-
ing the 20 to 30 second segment of the data, corresponding to the third of eight setups.
These plots reveal the importance of presenting the time-frequency information in the tri-plot for-
mat. It is clearly evident in the time-frequency plot that the segment with the abrupt change in
frequency, as seen in the contour plot, also saw an increase in the amplitude of vibration as shown
in the acceleration time history. The time-frequency plot also shows that in the frequency
domain, the signal has multiple peaks near 30 Hz. With the use of joint time-frequency analysis it
is clear that the multiple peaks near 30 Hz are in fact a single mode which has shifted. If the anal-
ysis was conducted solely in the frequency domain, the researcher would have a difficult time
explaining, or explaining away, the multiple peaks seen near 30 Hz.
Even though the A N P S D does not show multiple peaks near the fundamental vertical mode (6.3
Hz) it is of interest to determine if there was a corresponding shift, like that seen near 30 Hz).
Figure 6.10 shows the time-frequency plot for the frequency range 0 to 20 Hz. A shift in the fun-
damental mode is not evident from the time-frequency plot. During the time segment in which
the mode near 30 Hz shifted, there was little response in the fundamental vertical mode. It
appears that the significant shaking, seen in the time history, was dominated by the mode near 30
Hz.
Figure 6.10 shows the existence of frequency content near 13 Hz which does not have a corre-
sponding peak in the ANPSD. This is the fundamental transverse mode. The reason that it shows
84
up in the vertical direction is that the transverse mode has torsional coupling as discussed in sec-
tion 6.4.2. The vertical reference sensor used for the joint time-frequency analysis also serves as
the torsional reference sensor and thus it contains torsional response that is not present in the ver-
tical ANPSD.
85
Amplitude Frequency (Hz)

Figure 6.10: Narrow Band T-F Plot for the Lindquist Bridge - Low Frequency Range
86
6.6 Conclusions
The ambient vibration of the Lindquist bridge was sufficient to identify 8 modes below 50 Hz.
The fundamental frequency in the vertical direction was identified as 6.3 Hz and in the transverse
direction to be 13.2 Hz. The fundamental torsional mode was observed at 7.3 Hz.
The results show consistent coupling of the torsional and transverse modes. In Modes 2, 3, 5, 6
and 8 both transverse and torsional components exist. This can be explained by the fact that when
the bridge deck deforms in a torsional mode the shape can not be formed without an associated
displacement in the transverse direction. For example, as the center of the bridge deck moves in
the transverse direction one side naturally lifts and one dips.
Mode 5 warrants particular attention as it shows large movement in the abutment. The abutment
at the north end is supported by longer piles than the south end as a result of the local soil condi-
tions. It appears that the long piles, in conjunction with the laterally flexible soil, have a rocking
mode of vibration at or near 18.2 Hz. This rocking motion, in turn, drives the bridge in a torsional
mode similar to the torsional mode at 22.85 Hz.
Joint time-frequency analysis of the ambient vibration data identified a significant shift in the fre-
quency of the 7th mode. This joint time-frequency information is very valuable as the standard
frequency domain analysis yields two peaks of approximately the same magnitude near 30 Hz.
An attempt to determine the natural frequencies near 30 Hz from a standard frequency domain
analysis is nearly impossible as one would not be able to explain the significance of both peaks.
This inability to verify the behavior of the bridge near 30 Hz may lead to skepticism of the results.
87
The joint-time frequency analysis of the data showed that it is not two modes but in fact one mode
that shifted in the third test.
The reason for the observed shift is not clear. It is possible that the mode near 30 Hz was driven
by ambient vibration with a dominant frequency equal to the shifted frequency. Due to the ampli-
fication of a structures response near one of its natural modes the response could be dominated by
a forcing function in close proximity to a natural mode of vibration. As the input into the struc-
ture is unknown it is not possible to calculate the Frequency Response Function (FRF) of the
bridge which would, theoretically, remove the response of the structure to a forcing function. If
the bridge's response was as a result of ambient vibration dominated by the frequency equal to the
shifted frequency one would still, however, expect to see the actual unshifted frequency.
The shift could be due to non-linearities in the bridge. If the stiffness was not linear the greater
level of excitation observed in the time history could cause a shift in the frequency. These non-
linearities however appear to be localized. This is evident from the fact that similar shifts were
not observed in other frequencies.
88
CHAPTER 7
Sliding of a Dam Model
7.1 Introduction
In the spring of 1998 a study was conducted at The University of British Columbia in order to
investigate certain aspects of the dynamic response of concrete gravity dams. The research was
carried out in collaboration with British Columbia Hydro and Power Authority (BC Hydro),
Maintenance, Engineering and Projects Division. This research is Phase 2 of a multi-phased
research project which aims to develop suitable techniques for determining equivalent static lat-
eral load coefficients for seismic assessment of existing dams. The results of these studies can be
found in the reports presented to B C Hydro (Horyna et al., 1997 and Black et al., 1998b). Sec-
tions 7.1 through 7.4 of this chapter are based on a report prepared for B C Hydro by Black et al.
(1998b).
A small scale model of a dam, to which different friction surfaces could be applied, was subjected
to harmonic and earthquake excitations of different frequency content and amplitude. The rela-
tive displacement between the model and the shake table was measured. The results from these
tests allowed the estimation of the static and kinetic coefficients of friction as well as the compu-
tation of the ratio of the dynamic force to net static force required to cause a prescribed level of
motion.
89
Chapter 7 Sliding of a Dam Model
During the course of the test an interesting phenomenon was observed. At certain combinations
of base excitation amplitude and frequency the model moved upstream. That is, under base exci-
tation the model moved in a direction opposite to that of the applied hydrostatic load. In an
attempt to explain this behavior a joint time-frequency analysis was performed. A description of
the experiment and the analysis is presented in the following sections.
7.2 Experimental Model Description

The model consists of a dam model, sliding surfaces and a hydrostatic load assembly. Sliding
occurs between a sliding surface attached to the bottom of the dam and one attached to the shake
table. The hydrostatic load assembly is designed to provide a constant pulling force in the down-
stream direction.
7.2.1 Dam Monolith
The dam monolith is approximately 1500 mm high and 480 mm wide (see Figures 7.1 and 7.2). It
was cast in July 1996 as part of Phase 1 of the project. The material used for the monolith was a
mix consisting of Portland cement type 10, perlite, Styrofoam, silica fume and water. The weight
composition of the mix was 42.2% cement, 40% water, 12.5% perlite, 3.6% silica fume and 1.7%
Styrofoam (Horyna et al. 1997). This mix was chosen in an attempt to satisfy similitude require-
ments between the model and an actual dam monolith, although it is understood that the intent of
this study was not to predict the performance of an existing dam but rather to compare static and
dynamic properties of the model. For a discussion of the similitude considerations see Horyna et
al. (1997).
90
225
1250
1250
Figure 7.1: Schematic of Dam Model
91
Figure 7.2: Photo of Dam Monolith Model
7.2.2 Sliding Surfaces
The upper and lower sliding surfaces measure 1150 mm by 480 mm and 1500 mm by 660 mm
respectively. The upper plate is clamped to the bottom of the dam and the lower plate is attached
to the shake table by way of a steel frame (as shown in Figure 7.3). These plates are unbonded
and the friction measured is a result of the upper plate attached to the dam moving over the lower
plate attached to the shake table.
Two different friction surfaces were used: a smooth surface and a rough surface. The smooth sur-
face was created by sanding the "cement milk" from the friction surface of the plates to give a
uniform surface. The rough surfaces were created by using an ultra-high pressure water jet to
remove the cement matrix leaving an exposed aggregate surface.
Figure 7.3: Photo Showing Plate Attachment to Dam Model and Table
7.2.3 Hydrostatic Load Simulation
The hydrostatic load was simulated by applying a pulling force in the downstream direction on
the downstream side of the dam model. This force was provided by a cable attached to the model
at the height corresponding to the resultant of the simulated hydrostatic load. This cable was
93
attached to a hanging weight equivalent to the required force. The mass was attached to two ver-
tical rods which constrained it to move only in the vertical direction. Friction between the mass
and the sliding rods was minimized by bronze-oilite bearings which provided a smooth contact
surface. A photo of the hydrostatic load assembly is given in Figure 7.4.
Figure 7.4: Hydrostatic Load Assembly
7.3 Shake Table Testing

The concrete gravity dam model was instrumented to measure its response in five different tests
for four different surfaces. The tests included: static, harmonic, earthquake, cyclic and impact.
A l l of the tests were conducted with the model placed on the shake table at the Earthquake Engi-
94
neering Research Facility at U B C .
7.3.1 Shake Table
The shake table measures 3m by 3m. It is moved by up to 5 actuators in various configurations.
It has a payload capacity of 156 kN, with a maximum displacement of +/- 7.6 cm in the horizontal
direction. The actuators are controlled by a state of the art Multi Exciter Vibration Control Soft-
ware program which performs closed loop control. For the tests described in this thesis, the table
was setup to produce only unidirectional motion in the horizontal direction.
7.3.2 Tests with Harmonic Input
A series of tests were conducted on each surface at nine harmonic frequencies ranging from 5 Hz
to 25 Hz with increments of 2.5 Hz.
The input signal for the 5 Hz harmonic test is shown in Figure 7.6 for illustration. This signal has
6 segments with different amplitudes, ranging from 1.5g to 2.0g, in increments of O.lg. This
record was used as the input signal to the shake table where it was reduced to desired amplitude
levels. For example, if the record was run at 20% then the maximum acceleration on the table
was 0.4g. Each test at a particular frequency was run multiple times with increasing amplitudes
in order to capture the amplitude of motion which caused the model to slide a prescribed amount.
Typically, the record for a certain frequency was run 5 to 10 times providing 30 to 60 different 10
second records at different amplitudes.

Figure 7.5: Typical Dynamic Test Setup
96
0.00 18.75 37.50 56.25 75.00

Time (sec)
Figure 7.6: The 5 Hz Harmonic Input Record
7.3.3 Tests with Earthquake Input Records
The setup and data collection for the earthquake tests were the same as for the harmonic tests.
A series of simulated earthquake tests were conducted at 9 dominant frequencies. Each test had
three different earthquakes with approximately the same dominant frequency. The records were
created by taking the Power Spectral Density (PSD) of three earthquakes and shifting the spectral
shape to obtain the desired dominant frequency. This was accomplished using the program SIM-
Q K E (Gasparini, D.A. and Vanmarcke, E.H., 1976). For this study, an earthquake test which is
denoted by a single frequency, for example the 20 Hz earthquake test, actually has a frequency
band of approximately 6 Hz. The frequency band runs from the dominant frequency minus 2 Hz
to the dominant frequency plus 4 Hz. For example, the earthquake test denoted 12.5 Hz has an
97
approximate frequency content of 10.5 Hz to 16.5 Hz with a dominant frequency of 12.5 Hz.
The input earthquake record for 5 Hz is shown in Figure 7.7 for illustration. The first earthquake
is a simulated earthquake with the same power spectral density (PSD) shape as the 1992 Califor-
nia Landers earthquake measured at the Joshua Tree Fire Station in the E/W direction. The actual
PSD was shifted in frequency to give the desired dominant frequency, which in this case is 5 Hz.
The second is a simulated earthquake with the same PSD shape as the 1994 California Northridge
earthquake recorded at the Tarzana Nursery in the E/W direction. The third earthquake is a simu-
lated earthquake with the same PSD shape as the 1979 California Imperial Valley earthquake.
10 20 30 40
Tim e (sec)
Figure 7.7: 5 H z Earthquake Record
98
7.4 Analysis
More than 500 shake table tests were conducted on the dam model. The analysis of the data was
comprised mainly of time domain data processing as the frequency of the excitation was con-
trolled and hence known. The main objective of the experimental part of this study was to deter-
mine the magnitude of prescribed base excitations which caused a specified level of motion
between a concrete gravity dam monolith model and its unbonded base.
The experimentally determined magnitude of the base acceleration required to move the model
was converted to a pseudo-static force equal to the mass of the model multiplied by the accelera-
tion of the model. This pseudo-static force was compared with the measured static force to give
the ratio of the dynamic to static forces required to displace the model a prescribed amount. This
ratio was determined for both harmonic and earthquake excitation and a comparison between the
ratios was made for each excitation and each different surface.
For more information on this study refer to Black et al. (1998b).
7.5 Joint Time Frequency Analysis

During the course of the study, an interesting phenomenon was observed. At certain combina-
tions of frequency and amplitude of base excitation, the dam model moved upstream overcoming
the hydrostatic load. This was generally observed during tests where the amplitude of excitation
approached l g and the frequency of base motion was greater than 20 Hz. It is not obvious why
the dam model would behave in such a manner.
99
In an attempt to better understand the upstream behavior of the dam a joint time-frequency analy-
sis was conducted. The measured acceleration at the base of the dam model was analysed using
the STFT.
The objective of the analysis was to identify dominant frequencies other than the forcing fre-
quency. A Hanning window of 128 points was used for all of the time-Frequency plots. With a
sampling rate of 200 sps, a window of 128 points yields a non-zero segment 0.64 seconds. The
dominant periods in this study are much smaller than the segment length and hence the spectro-
gram will easily capture the period range of interest.
Figures 7.8 and 7.9 show the time-frequency plots for two tests with 20 Hz harmonic base excita-
tion on the rough surface. Upstream motion is indicated by negative values of base motion. The
time-frequency plot of the first test shows that when the model moves downstream, the only dom-
inant frequency is the forcing frequency. During the second round of tests, upstream motion
occurred during the second through fifth segments. The time-frequency plot shows two other fre-
quencies, in addition to the forcing frequency, during upstream motion. The first at one half the
forcing frequency and the other at one and a half times the forcing frequency.
To verify this result another harmonic base excitation test was analyzed. The time-frequency plot
presented in Figure 7.10 shows the response of the model on a smooth surface excited at 22.5 Hz.
This plot confirms that during upstream motion, the model is vibrating at half and one and a half
times the forcing function.
100
Contour Legend
5000
CD
:::::::> 5000
3000
100000 Q
CO
CL
1000
90
;l •r
r,
•i 'I
'A
lj
!i •1
iL
'
75 /. -—
i;
I-
1!
-/
-
!!i- {
•1
:i
-
ji
60 c~>
vf .7
\
1
- ;l *
\
1:
;i
i
45
ii >
1: 1
:i )1
J-
30
1.
;l t
- ;i 4
1
|i i1
15
-1.20.0 1.2 0 5 10 15 0 10 20 30 40 50
Base Accel. Model Displ. Frequency (Hz)
(g) (cm)
Figure 7.8: TFR Plot of 20 Hz Harmonic Input - No Upstream Motion
101
Contour Legend
10000
CO
"> 5000 6000
..J 100000 Q
Q_
2000
90
75
--F.;>
I!
60
o
CD
CD
^4- E
45
30
1'-
15
•1.2 0.0 1.2 -10 -5 0 5 0 10 20 30 40 50
I Base Accel. M o d e l Displ. Frequency (Hz)
(9) (cm)
Figure 7.9: TFR Plot of 20 Hz Harmonic Input - Upstream Motion
102
Contour Legend
8000
CD
CO
LV.V.VJ 5000 5000
100000 Q
co
Q- 2000
90
IS
A *
75
i f
is 5-
60
o
«!• CD
IT
CD
E
45 i-
30
15
•1.2 0.0 1.2 -15 -5 0 10 20 30 40 50

(g) (cm)
Figure 7.10: TFR Plot of 22.5 Hz Harmonic Input - Upstream Motion
103
The vibration observed during earthquake excitation was considerably more violent than the har-
monic tests and thus the time-frequency analysis shows a much larger bandwith of response.
Figures 7.11 and 7.12 show time-frequency plots for the response of the model to earthquake
excitation. The first plot is for a test at 15 Hz in which no upstream motion occurred. During the
test shown in Figure 7.12 the earthquake input had a dominant frequency of 25 Hz and the model
moved upstream. While the model moved downstream, no frequency is observed at half the forc-
ing frequency. At one and a half times the forcing frequency, a few relatively small peaks are
observed. During upstream motion, the dam model vibrates at many frequencies with the most
prominent being the forcing frequency. The frequency band of excitation extends to half the forc-
ing frequency at times corresponding to large upstream motion.
104
' 1 1 1
• Q I I I I I I 1 1 1 1 1
-1.20.0 1.2 0 10 20 o 10 20 30 40 50

(g) ( )
cm
Figure 7.11: TFR plot of 15 Hz Earthquake Input - No Upstream Motion
105
An explanation for this behavior is that the dam model is rocking and possibly leaving the table at
certain times producing a net upstream displacement. A possible rocking motion which would
produce a frequency half that of the forcing frequency, for upstream motion, is shown in Figure
7.13. The figure shows the motion of the dam model and the shake table for two cycles of base
motion and the corresponding single cycle of upstream motion. The upstream motion is shown by
the dashed arrow.
107
7.6 Conclusion
A joint time-frequency analysis of the upstream motion of a dam model showed that during
upstream motion, the dam vibrates at frequencies other than the forcing frequency. This informa-
tion led to increased understanding and further study into the behavior. A standard frequency
domain analysis would not indicate the presence of the other frequencies for times when the dam
model moved upstream.
A possible explanation for the frequency observed at one half times the forcing frequency was
presented. This motion involves rocking about the downstream edge while the table moves in the
downstream direction.
109
CHAPTER
Analysis of the 1901 Avenue of the Stars Building
8.1 Description
The Avenue of the Stars building is located in the Century City area of Los Angeles at 1901 Ave-
nue of the Stars. This moment resisting, steel frame, office building has 19 stories above the
ground level with four parking levels below ground. The floor plan of the 1901 building mea-
sures 73 m long by 34 m wide (Murphy, 1973). A Photo of the building is shown in Figure 8.1.
The soil conditions at the site are generally fine sand. The building's foundation is comprised of
driven steel I-beam piles under the main structural tower and spread footings elsewhere. The lat-
eral load resisting system consists of four ductile steel moment-resisting frames in the major axis
direction and 5 X-braced steel frames in the minor axis direction (Murphy, 1973).
The epicenter of the Northridge earthquake was located 20 km north-west of the 1901 Avenue of
the Stars building and the peak ground acceleration at the site was 0.32g. The epicenter of the San
Fernando earthquake was approximately 39 km north-west of the building, with a peak ground
acceleration at the site of 0.15 g's.
Damage during the San Fernando earthquake was not severe. During the Northridge earthquake
bracing elements at the penthouse buckled and signs of distress and motion along the brace lines
110
Chapter 8 Analysis of the 1901 Avenue of the Stars Building
in upper floors were observed. Minor non-structural and content damage was observed but oper-
ation resumed within one day (Naeim, 1997).
$ll§! B p
(Adapted from Naeim, 1997)
Figure 8.1: Photo of the 1901 Avenue of the Stars Building
111
8.2 Instrumentation
The 1901 Avenue of the Stars building was instrumented with 7 strong motion accelerometers
during the San Fernando earthquake. These were positioned 3 at each of the basement, 10th and
roof levels, with an additional vertical sensor at the basement. A schematic of the sensor layout is
given in Figure 8.2. During the Northridge Earthquake, the 1901 Avenue of the Stars building
was instrumented with 15 strong motion accelerometers with 3 at each of the basement, 1st, 2nd
and roof as well as two on the 8th floor and an additional vertical accelerometer at the basement
level. A schematic of the sensor layout during the Northridge earthquake is shown in Figure 8.3.
The sampling rate during the San Fernando and Northridge earthquakes was 100 sps.
During the Northridge earthquake, 2 transverse records were obtained on each instrumented floor.
The transverse signals used in this analysis were obtained by adding the 2 transverse records and
dividing by a factor of two. Similarly, the torsional signals used for the analysis were obtained by
subtracting the 2 transverse records and dividing by 2. A l l other signals used in this analysis are
the original recorded accelerations. Figures 8.4 and 8.5 show the San Fernando and Northridge
earthquake signals used for the analysis presented in this chapter.
112
C
CO
c a!
+ o
o
o
o
C P
- 4 -
J '
Z
o £ «
S i
< « -r
CJ
N
o 3o 5c
rr
O) o W O
T>
en
CD z "0
4)
CD LU o
O 0) z 1 . —s—*
K
O
oO
Ml
• • • •
o c a a s •
+->
<n •« cO
i IP
CO ix
CD E
• 2 oil ? 0
A «s o es
o
Q
10 a" • <5
a u «0 5
© • J
CD • a I
OJ O C N ' >
C
• •
< o
w
O
9-
.ow.t-.eieor
.etc
(Adapted from: Shakal et al.,1994)
Figure 8.2: Schematic of Sensor Location During San Fernando Earthquake
113
r,oc
an
an
c;
• a - CO
Plan
» »—«
' OH ' -J
<"T •
o • PH
0-
' >H
, CO
. m
u
1 • *H'
«
o n *H
«*-. • o 2
. o • xS-si o
•o si o
Flo
o• c .
: E :: |
OS g
o• v
• OS- t- S •
I-
• ; j
H->
: oo : , c. 58 n
n o 1
*—> «T3
CO
z
o
Y~
<
o
o
_l
o
to
DO
z
UJ
O
to
co
I
CM
JS °
co 2
cow
C a.
< §
o 2
(Adapted from: Shakal et al., 1994)
Figure 8.3: Schematic of Sensor Location During the Northridge Earthquake
114
Transverse - Basement
O) 0.2
Accel. (
—
o
to b
6
Transverse - 9 th Floor
D) 0.2
"CD o.o
o
o
< -0.2
Transverse - Roof
a> 0.2
"CD 0.0
o
<t -0.2
Longitudial - Basement
S 0.2
"CD 0.0 — —-———-—-————
< -0.2
Longitudial - 9th Floor
3 0.2
ill 1 ll
8 0 0 ——-—~—-
ir.no
Longitudial - Roof
3 0.2
"CD 0.0 A V V w Y v v v v w ^ -
o
0 10 20 30 40 50 60
Time (sec)
Figure 8.4: Measured Accelerations During the San Fernando Earthquake
115
O) 0.3
CD 0.0
O
O
< -0.3
Transverse - 2nd Floor
CD 0.3
CD 0.0
O
O
< -0.3
Transverse - 8th Floor
CD 0.4
Transverse - Roof
CD 0.5
CD 0.0
O
O
< -0.5
o5 0.3
CD 0.0
O
O
< -0.3
Longitudial - 2nd Floor
"3 0.4
30 40 50 60 70 80 90
T i m e (sec)
Figure 8.5: Measured Accelerations During the Northridge Earthquake
116
Logitudial - 8th Floor
CO 0.4
CD 0.0
o
o
< -0.4
Longitudial - Roof
CD 0.4
CD o.o
o
o
< -0.4
Torsional - Basement
CD 0.2
CD 0.0
o
o
< -0.2
Torsional - 2nd Floor
cn 0.2
CD 0.0
O
CJ
< -0.2
Torsional - 8th Floor
CD 0.2
CD 0.0
o
o
< -0.2
Torsional - Roof
CD 0.2
CD 0.0 JUL
O
o
< -0.2
0 10 20 30 40 50 60 70 80 90
T i m e (sec)
Figure 8.5: Measured Accelerations During the Northridge Earthquake (Continued)
117

The modal frequencies of the 1901 Avenue of the Stars building are presented in table 8.1. These
frequencies were obtained from the methods described in Chapter 3. The magnitude of the fre-
quency response functions for the roof to basement pairs are presented in Figure 8.6. As indicated
in the figure, the plot at the top shows the FRFs for the transverse direction (shown as Y) followed
by the longitudial direction (shown as X ) in the middle and the torsional FRFs at the bottom.
There are no torsional frequencies calculated during the San Fernando earthquake as there were
not enough measurements taken to calculate the rotation at the base.
San Fernando Northridge

Mode
(Hz) (Hz)
TRA1 0.29 0.32

TRA2 1.27 1.22
TRA3 2.51 2.68
LON1 0.29 0.22
LON2 0.78 0.71
LON3 1.25 1.25
ROT1 NA 0.24
ROT2 NA 0.98
ROT3 NA 1.90
Table 8.1: Frequencies of 1901
The fundamental natural frequency in the longitudial direction shifted considerably between the
San Fernando and Northridge analysis.
118
a) Y direction
Frequency (Hz)
Figure 8.6: FRFs of the 1901 Avenue of the Stars Building
119
8.3.2 Mode Shapes
The first two mode shapes in each direction corresponding to the above frequencies are given in
Figure 8.7.
The torsional mode shapes presented in Figure 8.7 were obtained by taking the rotational values
of displacement and plotting them as a linear displacement. For example, the second mode
would be turning one way at the bottom and the opposite way at the top corresponding to the left
and right side of the undeformed shape.
The San Fernando mode shapes are not well defined, particularly in the transverse direction. This
may be the result of only having 2 sensors above ground level. The Northridge mode shapes are
interesting because of the relative displacement between the basement and first floor. For the
mode shape analysis, the first floor was often assumed to be the base of the structure and the
vibration shape above ground was taken to be the mode shape.
120
Northridge E q S a n F e r n a n d o E q
Figure 8.7: Mode Shapes of the 1901 Avenue of the Stars Building

A joint time-frequency analysis was conducted using the methods described in Chapter 3. The
time-frequency information is displayed in joint time-frequency plots which contain the input and
121
output time signals, the standard FRF and the TFRF. The TFRF is shown as a contour plot. The
contours have been limited to 3 for simplicity. A legend of the contour levels is shown in the top
left corner of each plot.
At a sampling rate of 100 sps, the length of a windowed signal with a window size of 256 points is
2.56 seconds. The longest fundamental period is 3.4 seconds. The windowed signal does not
capture one full oscillation of the structure's fundamental period and thus the results may be
slightly distorted. Only the narrow band TFRFs are presented for the 1901 Avenue of the Stars
building as the wide band windowed signal has a non-zero duration of one-fifth the length of the
fundamental period.
The following points should be considered when viewing time-frequency plots. The TFRF of
strong motion is made up of a series of peaks as the frequency content of the acceleration mea-
sured at the basement level is not constant. If the input was continuous in frequency and time, the
peak would be in the form of a ridge. This ridge would be present, at a particular frequency, for
all time.
The TFRF is cut at the highest contour level which does not correspond to the peak magnitude of
the TFRF. The width of the contour however, conveys information about the peak. A contour
with small width means the magnitude is close to the peak, or that the peak itself is very wide.
The spacing of the contours indicates the sharpness of the peak.
The amplitude of the TFRF does not indicate the presence of heavy shaking. It indicates which
frequency components are present at a certain time and their relative strengths.
122
The information contained in the TFRF is fundamentally different than the FRF. The FRF aver-
ages the frequency content over the entire time of the record. More specifically, it is the ratio of
the total energy contained in the output, at a particular frequency, over that in the input. The
TFRF conveys information on the relative energy levels, at a particular frequency, during a small
window of time. For this reason, the magnitude of the TFRF may be misleading. When the mag-
nitude of shaking at the roof level, for a given time and frequency, is greater than that at the base,
a peak will be shown. Hence, a large peak may be present during free decay of the vibration when
the input is very small.
Each time-frequency plot contains a vast amount of information on the structure's response to
earthquake excitation. Due to the scope of this thesis, a comprehensive study can not be con-
ducted for each time-frequency plot. A discussion of the important features, with respect to the
building's response, is presented instead. The general characteristics of the Time-Frequency
Response Function are discussed, as required to facilitate understanding of the information con-
tained in the TFRF.
Figures 8.8 and 8.9 show the narrow band time-frequency response plots for the Northridge and
San Fernando earthquakes in the transverse direction. The figure caption states the earthquake
and direction for each plot
123
TFRF Contour Magnitude
80
70
60
•i
50
o
CD
40
CD
E
i-
30
20
10
0
-300 0 300-600 0 600 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

(cm/s )
2
(cm/s )2
Figure 8.8: 1901 T-F Plot - Northridge Earthquake - Transverse Direction
124
co
60 I 1
I ' I 1
I 1
I 1
I 1
I 1
I 1
I 1
I 1
-200 0 200 -300 0 300 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
(cm/s )
2
(cm/s )
2
Figure 8.9:1901 T-F Plot - San Fernando Earthquake - Transverse Direction
125
The TFRFs show stable frequencies for both earthquakes. This is expected as the frequency
domain analysis showed little variation in the calculated frequencies.
One interesting feature of the TFRF is its ability to provide insight into the behavior of the FRF.
For example, the frequency fluctuations observed in the TFRFs around the second natural fre-
quency may explain the spreading in the peak of the FRF. As well it is seen that the first mode
does not dominate the response during heavy shaking. The magnitude of the peaks in the TFRF
shown in Figure 8.8, are less during the heavy shaking portion of the record than during the free
vibration portion.
For the case of a steady-state forced input, a structure will vibrate at the same frequency as the
driving force. An earthquake however is transient and steady state motion would not be achieved.
For this reason, there will be some modal response of the building during heavy shaking. The low
level of modal response seen in the TFRF does not necessarily mean that the building does not
vibrate in its modes during heavy shaking, just that the magnitude of the response, at a particular
frequency, is eclipsed by the magnitude of the input.
To explain this behavior further a typical frequency domain FRF is calculated using one segment
of data from heavy shaking (0 - 20 seconds) and one during free vibration (40-60 seconds). The
data used in the analysis is the recorded accelerations at the basement and roof, measured during
the Northridge earthquake in the transverse direction. The results are shown in Figure 8.10.
126
80
70
a) F R F for 40 to 60 s e c . Segment
60
CO
CO
50
rr
o
CD
£• 40
CD
0.5 1.0 1.5 2.0
E
Frequency (Hz)
30
b) F R F for 0 to 20 s e c . Segment
20
co
CO
10
0
-500 0 500-500 0 500 0.5 1.0 1.5 2.0

(cm/s2) (cm/s2)
Figure 8.10: Figure Showing Contribution of Heavy Shaking to FRF
127
Plot a) of Figure 8.10 shows that during the free vibration of the structure the modes are well
defined as the magnitude of the F R F during this time is high. Plot b) of Figure 8.10 shows little
contribution of the first mode to the response. There is no peak at 0.32 Hz, which corresponds to
the first natural frequency and the second mode contributes only slightly.
The magnitude of the TFRF, at a particular time, may be several orders of magnitude greater than
the FRF. It is shown in Appendix A that the TFRF has the same form as the FRF but its magni-
tude is squared and thus is greater than the FRF. Another reason for the large magnitude is that
the FRF is averaged, for a particular frequency, over the entire record. The small modal response
of the fundamental mode, relative to the input, observed during the heavy shaking is averaged
with the relatively greater response during free vibration. The TFRF is localized in time and
hence during free vibration the output, at a natural frequency, far exceeds the input which is
nearly zero. This is confirmed by the relative magnitudes of the FRFs in Figure 8.10.
Figures 8.11 and 8.12 present the narrow band time-frequency plots for the 1901 Avenue of the
Stars building in the longitudial direction.
128
Ji 100
1 900
j 1700
80
70
if
60
<i
•i
50
O
CJ)
40 CD
E
30
*
20
i)
10
0
-400 0 400 -400 0 400
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
(cm/s )
2
(cm/s ) 2
Figure 8.11:1901 T-F Plot - Northridge Earthquake - Longitudial Direction
129
'.::::::> 75
I I 400
r_r_r_rj 750
60
50
40
• i n o
i> 1 ii
CD
30
CD
E
ii8
• " 1
20 1 >
-. 1
v.
10
'•v.
0
-200 0 2 0 0 - 3 0 0 0 300
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Base Accel. Roof A c c e l .
Frequency (Hz)
(cm/s ) 2
(cm/s ) 2
Figure 8.12:1901 T-F Plot - San Fernando Earthquake - Longitudial Direction
130
Section 8.3 showed a shift in the frequency in the fundamental longitudial direction between the
two earthquakes. The narrow band TFRF calculated from the Northridge earthquake, shown in
Figure 8.11, shows that the fundamental longitudial mode has a relatively constant peak at 0.22
Hz. The TFRF for the San Fernando earthquake however, does show a shift.
Figure 8.12 shows a gradual decrease from 0.29 Hz to 0.22 Hz. The FRF appears to be dominated
by the strong peak during the first 5 seconds of shaking. Figure 8.13 presents a close up of the
time-frequency plot for the San Fernando earthquake in the frequency range 0 to 0.6 Hz. The
TFRF appears to have two dominant frequencies, one being around 0.32 Hz and one 0.25 Hz.
The FRF however, only indicates one peak at 0.29 Hz. The FRF averages the peaks over time and
thus it is possible that the two dominant frequencies combined in such a way that only a single
peak is visible in the frequency domain. The fundamental mode in the transverse direction is 0.32
Hz. If coupling exists between the transverse and longitudial directions, the fundamental mode in
the transverse direction could be influencing the FRF in the longitudial direction. The combina-
tion of two frequencies which yield a single peak is sensitive to the duration of the two signals, the
sampling rate and the amplitude of the signals. With the right combination of the above factors, it
can be shown that a time signal with two distinct frequency components can average to form a
single peak in the frequency domain.
131
60 /
1 / \ 1
«'/•' M • \
1 / \\ 1 i
\
•"v
i 1;
li 1
i
I
M V il> . ' A i; f
50 /
1
• I N 11 i; j
/
1
1 1 \
1 1 > i i
1 1
1 1 i\ i
- -
\,I
\
1 \ iI i
/
/
•
• y
40
i \
i. i
i
i -
o
i
30 - -
ii CD
i E
#
i—
*
0
i f
20
_
\ \ *\ V
, /
/ _
*\
\V \ 1
1
/ 1 » \
t |
I 1 i
\*' I
1
_ 1
f 1 ,
M
I ' \ 1
* Al i /
i A
1 /
,
\1 i /
• V
A
10 / \\<
f
i /. V
'\
ll
: i
J ji
i li
J It
M>
i 1
,! •' -ft.
0
-200 0 200-300 0 300 0.0 0.1 0.2 0.3 0.4 0.5 0.6
(cm/s )
2
(cm/s )
2
Figure 8.13:1901 T-F Plot - San Fernando Earthquake - Longitudial Direction - 0 to 0.6 Hz
132
8.5 Conclusion
The frequency domain analysis identified the first three natural frequencies in the longitudial and
transverse directions for both the San Fernando and Northridge earthquakes as well as the tor-
sional natural frequencies during the Northridge earthquake. The mode shapes of the building are
kinked at the ground level due to a discontinuity in stiffness.
The frequency domain analysis identified a shift in the fundamental longitudial mode. The reason
for this shift is not decipherable from frequency domain analysis alone. A joint time-frequency
analysis shows that the shift is not due to heavy shaking but that the peak at 0.29 Hz is actually a
combination of two modes; the fundamental longitudial mode vibrating at 0.22 Hz and the funda-
mental transverse mode at 0.32 Hz.
133
CHAPTER 9
Analysis of the Sheraton Universal Hotel
9.1 Description
The Sheraton-Universal Hotel is located in the Universal City area of Los Angeles at 3838 Lank-
ershim Boulevard. The building is a 20-storey concrete structure that serves as a hotel and con-
vention center. The construction of the Sheraton-Universal was completed in 1968 at a cost of
$7.5 million. It is a podium style building with a central tower, which runs from the fourth floor
to the roof. The dimensions of the tower are approximately 56 m long by 18 m wide. The base-
ment through the 3rd floor measures approximately 6 1 m long by 30 m wide. The tower portion
is separated from the rest of the building by expansion joints, which serve to seismically isolate
the tower from the rest of the building (Murphy, 1973). A photo of the Sheraton is shown in Fig-
ure 9.1
The structure is supported by reinforced concrete spread footings. The underlying soil deposits
consist primarily of bedded sandstone with deposits of clay and shale (Murphy, 1973). The lateral
load resisting system, for each direction, is a reinforced concrete moment resisting frame.
The epicenter of the Northridge earthquake was 19 km north-west of the Sheraton and a peak
ground acceleration of 0.33g was felt at the site. The epicenter of the San Fernando earthquake
134
Chapter 9 Analysis of the Sheraton Universal Hotel
was approximately 30 km north of the Sheraton, with a peak ground acceleration of 0.17g at the
site.
While no structural damage was apparent following the Northridge earthquake, serious nonstruc-
tural damage occurred in all floors. Six or seven window glasses were broken. The cooling tower
shifted off the bracket and extensive damage of the wall coverings including the bathroom tiles
and dry wall were observed (Naeim, 1997).
(Adapted from: Naeim, 1997)

Figure 9.1: Photo of Sheraton Universal Hotel
135
9.2 Instrumentation
During the San Fernando earthquake the Sheraton was instrumented with seven strong motion
accelerometers. Two at each of the basement, 11th floor and the 21st levels, plus a vertical sensor
at the basement. A schematic diagram showing sensor location is given in Figure 9.2. During the
Northridge earthquake the Sheraton Universal was instrumented with 16 strong motion acceler-
ometers. Three at each of the: basement, 3rd, 9th, 16th and roof levels with an additional vertical
accelerometer at the basement. A schematic diagram of the sensor layout is shown in Figure 9.3.
The sampling rate during both earthquakes was 50 sps.
During the Northridge earthquake, 2 transverse records were obtained on each instrumented floor.
The transverse signals used in this analysis were obtained by adding the 2 transverse records and
dividing by a factor of two. Similarly, the torsional signals used for the analysis were obtained by
subtracting the 2 transverse records and dividing by 2. A l l other signals used in this analysis are
the original recorded accelerations. Figures 9.4 and 9.5 show the San Fernando and Northridge
earthquake signals used for the analysis presented in this chapter.
136
C
CD
c
cu
CN E • a
CO
CO
CO 2 ..
CQ « 2
41
cc
«o
H
w *-»
3 (0
<*•» +^
O C
» o
to
z
g c c
o
Pla
<
u
o b
21st FIc
_l o
DC u_
o
to
LU
to
a
o
x .m
Mi
> o
in
O
+-»
CO
6 .2
CM '•P
• CO
X)
o >
o LU
LU
Ii5
o
X
JZ
tl
o .t.Ui
z
(Adapted from: Shakal et al., 1994)
Figure 9.2: Schematic of Sensor Location During the San Fernando Earthquake
137
c
JO
.J 2
CO « n,l
D_
I
i_ 9 c
0 o o CU
o E
u_ cu 8 „
T>
w
CO £ 2
CO «
CD
£ ~
o c
to O
Z
IT) •« ' '
o c c
<
Pla
JO
o c
JO
o n o
DL
o
o
J J
Roof I
3th Flc
rr 2
o sz
oo
z CN
CD
00
CU
+->
o CM O CD
X> _c
c o
oi
i— 4*r
o
+ -*
CIO a t c
o
O (0
C M CD
>
.5 JO)
o 6 LU
O 2 LU
t
>s
o
(«-*
ro
/>
X 0.
S
"§ to
u
o
Z
(Adapted From: Shakal et al., 1994)
Figure 9.3: Schematic of Sensor Location During the Northridge Earthquake
138
3 0.2
CD o.o * ^ ^ | f l j | f l | w V ^ — - ~
< -0.2
Transverse -11th Floor
3 0.1
CD 0.0
O
O
< -0.1
Transverse - Roof
CO 0.2
CD 0.0
O
O
< -0.2
3 0.2
CD o.o —
o
< -0.2
3 0.1
CD o.o
o
o
< -0.1
Longitudial - Roof
3 0.2
CD o.o
o
o
< -0.2
0 10 20 30 40 50 60
T i m e (sec)
Figure 9.4: Measured Accelerations During the San Fernando Earthquake
139
CD 0.3
CD 0.0
o
o
< -0.3
Transverse - 3rd Floor
CD o.3 r
CD 0.0
o
< -0.3
Transverse - 9th Floor
CD 0.4
CD 0.0
o
o
< -0.4
Transverse -16th Floor
CD 0.4
CD 0.0
o
o
< -0.4
Transverse - Roof
CD 0.4
CD 0.3
0 o.o •—"Hffi||^^^ •
o
o
< -0.3
0 10 20 30 40 50
T i m e (sec)
Figure 9.5: Measured Accelerations During the Northridge Earthquake
140
Longitudial - 3rd Floor
CD 0.3
CD 0.0
O
O
< -0.3
CD 0.3
CD 0.0
O
o
< -0.3
Longitudial -16th Floor
D) 0.3
o5 o.o I—^iiif!^^
o
< -0.3
Longitudial - Roof
CD 0.4
CD 0.0
o
o
< -0.4
Torsional - Basement
3 0.1 -
"CD o.o
o
< -0.1
Torsional - 3rd Floor
3 0.2
"CD o.o
o
o
< -0.2
0 10 20 30 40 50
T i m e (sec)
I4l
Torsional - 9th Floor
3 0 2
"a> o.o — i w ^ i f ( f t y v f M | l / \ l | A ^ — — ~ — ~
o
< -0.2
Torsional -16th Floor
3 0 2
"CD 0.0 •~~~-~^t^jljl^M<tl^rfil(^t\*j4f<r^^ — —~ ~~
o
< -0.2
Torsional - Roof
3 0.3
(I
"CD0.0 — i ( J w w y w J \ / l / i j ^ — — • — •
o 1 < uyn
< -0.3

The frequency domain procedure outlined in Chapter 3 was used to identify the first three natural
frequencies in each direction. The modal frequencies for the Sheraton Universal Hotel are given
in Table 9.1. Figure 9.6 shows the magnitude of the FRFs between the basement and roof signals
during the San Fernando and Northridge earthquakes. As indicated in the graph, the plot at the
top shows the FRFs for the Transverse direction (shown as Y) followed by the Longitudial direc-
tion (shown as X ) in the middle and the torsional modes at the bottom. There are no torsional
142
modes calculated during the San Fernando earthquake since there were not enough measurements
made to calculate the rotation at the base.
San Fernando Northridge

Mode
(Hz) (Hz)
TRA1 0.51 0.34

TRA2 1.42 1.10
TRA3 2.20 2.10
LON1 0.46 0.39
LON2 1.46 1.17
LON3 2.05 1.85
ROT1 NA 0.42
ROT2 NA 1.42
ROT3 NA 2.88
Table 9.1: Frequencies of Sheraton
143
a) Y direction
b) X direction
Northridge Eq
San Fernando Eq
c) Torsional
80
50
20
Frequency (Hz)
Figure 9.6: Frequency Response Functions for the Sheraton
144
The natural frequencies of the Sheraton, calculated from the Northridge earthquake records, are
consistently lower than those obtained from the analysis of the San Fernando earthquake records.
The reason for this shift cannot be determined from frequency domain analysis alone. It is possi-
ble that the larger amplitude of shaking during the Northridge earthquake caused the building to
behave in non-linear manner. This non-linearity may be due to damage incurred during the earth-
quake or the result of temporary reduction of the structural stiffness during heaving shaking.
9.3.2 Mode Shapes
The first two mode shapes corresponding to the above frequencies are given in figure 9.7. The
resolution of the mode shapes obtained for the Sheraton during the Northridge earthquake are par-
ticularly good as five levels of sensors measured the building response.
Similar to the 1901 Avenue of the Stars building, the torsional mode shapes presented in Figure
9.7 were obtained by taking the rotational values of displacement and plotting them as a linear
displacement.
145
Northridge Eq San Fernando Eq
Figure 9.7: Mode Shapes of the Sheraton Universal
9.4 Joint Time-frequency Analysis

The Sheraton Universal Hotel was analyzed using the same methods as those used for the 1901
Avenue of the Stars building. When sampling at 50 sps, a 256 point window gives a non-zero sig-
nal of 5.12 seconds. This signal length is sufficient as it is nearly twice that of the fundamental
146
transverse mode. Figures 9.8 and 9.9 present the narrow band time-frequency plots for the
Northridge and San Fernando earthquakes in the transverse direction.
147
CD 8
CO
':::::::< 50
I I 350
DC
r_~J"_ J_
650
35
Hi
30
25
20 o
CD
O
15 E
i-
10
0
-300 0 300 -400 0 400 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

(cm/s ) 2
(cm/s )
2
Figure 9.8: Sheraton T-F Plot - Northridge Earthquake -Transverse Direction
148
35
30
••i
25
20 o
CD
CD
15 E
10
0
-200 0 200 -200 0 200 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

(cm/s )
2
(cm/s )
2
Figure 9.9: Sheraton T-F Plot - San Fernando Earthquake - Transverse Direction
149
As seen earlier in the 1901 Avenue of the Stars building, the modal response relative to the input
is not pronounced during heavy shaking.
In Section 9.3 it was shown that the frequency of the fundamental transverse mode, calculated
from the Northridge earthquake, is substantially lower than that calculated from the San Fernando
earthquake; 0.34 and 0.51 respectively. The TFRFs shown in Figures 9.8 and 9.9 show relatively
stable ridges throughout the duration of shaking.
If the building experienced significant structural damage during the earthquake, the natural fre-
quencies of the building would change. If there was a temporary reduction in the structural stiff-
ness, the TFRF would indicate this change over time. It can be inferred therefore, that the
Sheraton did not suffer significant damage or structural softening during either earthquake.
The fundamental natural frequencies in the transverse mode can be approximated by:
(9.1)
for the Northridge earthquake and
(9.2)
for the San Fernando earthquake, where k and k represent the modal stiffness and m and m
n s n s
150
represent the modal mass of the Northridge and San Fernando earthquakes respectively. The ratio
of Equation 9.1 and 9.2 yields:
T —*0.5.
5
(9.3)
k
sn m
If the structural stiffness is assumed to be the same during both earthquakes, Equation 9.3 sug-
gests that the modal mass of the Sheraton Universal Hotel was twice as great during the
Northridge earthquake as the San Fernando earthquake. Although hotels are more susceptible to
changes in mass, as compared to office buildings, it is not likely that the mass doubled in the years
between the two earthquakes. Although no major structural damage occurred during the
Northridge earthquake, there was severe non-structural damage (Naeim, 1997). This level of
damage would not cause the stiffness of the system to decrease by a factor of 2 and thus some
other factor must be responsible for the decrease in frequency. It is possible that soil-structure
interaction could have reduce the frequencies during the Northridge earthquake. A more compre-
hensive study is required to verify this.
Figures 9.10 and 9.11 give the narrow band time-frequency plots for the longitudial response of
the Sheraton during the both earthquakes.
151
-200 0 2 0 0 -300 0 300 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Base Accel. Roof A c c e l . Frequency (Hz)
(cm/s )
2
(cm/s ) 2
Figure 9.10: Sheraton T-F Plot - Northridge Earthquake - Longitudial Direction
152
-200 0 2 0 0 -300 0 300 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
(cm/s )
2
(cm/s ) 2
Figure 9.11: Sheraton T-F Plot - San Fernando Earthquake - Longitudial Direction
153
A joint time-frequency analysis of the longitudial direction did not show the presence of any
major shifts in frequency. The most interesting feature of the longitudial response is the presence
of simultaneous peaks near 0.5 Hz in the San Fernando earthquake shown in Figure 9.11. The
TFRF shows 2 distinct ridges between 8 and 20 seconds. Figure 9.12 shows the frequency range
0 to 0.8 Hz. The peak at 0.46 Hz corresponds to the fundamental longitudial frequency. The con-
current peak at 0.51 Hz however cannot be the fundamental longitudial mode as a single mode
cannot have two simultaneous frequencies. The fundamental mode in the transverse direction has
a frequency of 0.51 Hz and thus it can be inferred from the TFRF that coupling exists between the
transverse and longitudial directions.
154
8
co
(0
DC
35
30 1
1
1
i
1
1 :
if;
'7' ;
25 1
i '/ J
i ' / li
•
•/ / ii
/ 1
i
20 - • -
o
CD
CD
-
15 -
£
i ( i-
i i
i, /*
11
i i
i
10
i i•
i i >
it
it
0
-200 0 2 0 0 -300 0 300 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
(cm/s )
2
(cm/s ) 2
Figure 9.12: Sheraton T-F Plot - San Fernando Earthquake - Longitudial Direction - 0 to 0.8
Hz
155
9.5 Conclusions
The frequency domain analysis of the Sheraton Universal Hotel identified the first 3 natural fre-
quencies and corresponding mode shapes in each direction. A large shift was observed in the fun-
damental transverse mode between the two earthquakes. During the Northridge earthquake the
calculated frequency was 33% lower than in the San Fernando earthquake.
A joint time-frequency analysis of the Sheraton Universal shows no significant shift in the TFRFs
calculated for the transverse direction during the Northridge and San Fernando earthquakes. This
suggests that the Sheraton Universal Hotel did not suffer any structural damage during either
earthquake.
156
C H A P T E R 10
Conclusions
Joint time-frequency analysis is prevalent in many fields. This thesis has shown that it is also
applicable to the analysis of vibration data obtained from civil engineering structures.
Several methods for joint time-frequency domain analysis were presented. The Wavelet Trans-
form, in theory, appears to be a suitable analysis tool although no analysis was conducted using
the transform. Many of the Cohen's Class functions, including the Wigner distribution however,
do not appear to be easily applied to the analysis of civil engineering data. They are more accu-
rate than other methods in terms of energy, but any bilinear distribution which satisfies the mar-
ginals is not manifestly positive. Negative values in the time-frequency plane complicate the
analysis and limit the physical significance of the results. The Short Time Fourier Transform,
which can also be expressed by the Cohen Class function, is a manifestly positive distribution. It
also has the advantage of being based on a fundamental analysis tool, the Fourier Transform.
10.1 Applications of Joint Time-Frequency Analysis

The Short Time Fourier Transform was used to illustrate the applicability of joint time-frequency
analysis to the analysis of civil engineering vibration data. It serves mainly as a supplemental
analysis technique to improve standard system identification analysis. Time-frequency analysis,
157
Chapter 10 Conclusions
however, also provides insight into the dynamic behavior of structures not available through fre-
quency domain analysis alone.
The ambient vibration of a 3 span bridge in Calgary, Alberta was analyzed using the standard fre-
quency domain approach to estimate natural frequencies and mode shapes. Joint time-frequency
analysis was subsequently used to demonstrate that the properties of the bridge did not change
significantly during the test. The joint time-frequency analysis showed that an unexplained peak
in the averaged power spectral density of the records was only present during a small portion of
the testing and thus not a natural frequency.
A similar study was conducted on a single span bridge near Kamloops, British Columbia. A stan-
dard frequency domain approach yielded an averaged power spectral density which had a number
of peaks near a certain frequency. The frequency domain analysis identified one natural mode of
vibration near that peak, but failed to explain the occurrence of the other adjacent peaks. A joint
time-frequency analysis, however showed that the numerous peaks were in fact one natural mode
whose frequency shifted in time.
Joint time-frequency analysis was used to investigate upstream sliding observed during shake
table testing of a concrete gravity dam model unbonded at the base. The analysis showed that,
during upstream motion, the dam model vibrates not only at the base excitation frequency, but
also at one half and one and a half times the forcing frequency. This behavior may have been the
result of the model rocking as it moved upstream.
158
The most promising application of joint time-frequency analysis is for the interpretation of strong
motion data. The Time-Frequency Response Function, (TFRF), presented in this thesis allows for
the study of many aspects of the dynamic behavior of structures not easily studied with frequency
domain analysis. This includes the coupling between modes, the temporal location of modal
response and the identification of the temporal location of shifts seen between earthquakes. The
use of joint time-frequency analysis to explain anomalous results in the frequency domain greatly
enhances the standard frequency domain analysis.
A joint time-frequency analysis can explain anomalous shifts in frequency. One such shift was
seen in the longitudial direction of the 1901 Avenue of the Stars building. A joint time-frequency
analysis showed that the shift is not due to heavy shaking but that the peak at 0.29 Hz is actually a
combination of two modes; the fundamental longitudial mode vibrating at 0.22 Hz during free
vibration and the fundamental transverse mode vibrating at 0.32 Hz during heavy shaking.
The frequency of the fundamental transverse mode of the Sheraton Universal Hotel was consider-
ably lower during the Northridge earthquake. A joint time-frequency analysis showed that the
frequencies were stable during both earthquakes and thus the shift in frequency must have been
caused by some other factor in the years other than damage or structural softening.
The TFRF confirms that the fundamental mode of the structure controls its response during free
vibration.
159
10.2 Limitations of Joint Time-Frequency Analysis

The practical application of joint time-frequency analysis is limited to relatively few functions.
The bilinear Cohen's Class functions are more accurate in terms of energy content than the Short
Time Fourier Transform, but they yield negative values in the time-frequency plane making the
application of the functions difficult. The Short Time Fourier Transform is manifestly positive
but if suffers from slightly distorted energy values as the energy of the window is combined with
the energy of the original signal.
The resolution of the STFT is also limited by the uncertainty principle. The time and frequency
distributions of a signal are directly related as is the standard deviation of the time and frequency
distributions. The standard deviations of the time and frequency distributions can not both be
made arbitrarily small and therefore there is a trade off between time and frequency accuracy.
Another problem with the Short Time Fourier Transform is leakage. The window function leaks
energy into the surrounding frequencies of a windowed signal. Leakage is minimized by choos-
ing window functions, such as the Hanning window, which slowly decay to zero.
In the dynamic structural analysis of civil engineering structures, the lower modes of vibration are
of primary interest. The window function used for the Short Time Fourier Analysis must be cho-
sen sufficiently wide to capture these long period oscillations. For this reason the time-resolution
of joint time-frequency analysis may be limited.

10.3 Suggested Future Research

The Wavelet transform should be explored in more detail with similar case studies as those pre-
sented in this thesis for the STFT. An in depth study of the TFRF is required to better understand
its behavior. This study may include the use of different window types and sizes to explore their
effect on the energy contained in the spectrogram. A study on the effect of the initial conditions
of the windowed signal and the effect of leakage should be made.
The TFRF function could be used to study many interesting aspects of the response of a structure
to strong motion input including the prediction of damage during a seismic event and the response
of base isolated buildings. The TFRF should be calculated between the basement and other floors
of a building to emphasize other dynamic characteristics of the structure.
A study should be conducted on how a function, similar to the TFRF, could be formed by taking
the ratio of the Fourier Transform of a windowed output signal over the Fourier Transform of a
windowed input signal. This function would yield magnitudes similar to the Frequency Response
Function as well as retain the phase information required to construct the structure's mode shapes.
It would be useful, and relatively simple, to write a customized computer program which would
calculate the TFRF directly from 2 time signals. This program should include different windows
and window sizes. The option to normalize windows to unit energy should be available, as this
would produce spectrograms with the correct total energy content. The ability to add and subtract
signals prior to the calculation of the time-frequency spectrogram and the TFRF should be incor-
porated in the program to allow for the calculation of the torsional TFRF. The program should
I6l
also include a smoothing function to allow for easier presentation and initial investigation of the
TFRF.
162
C H A P T E R 11
References
Asmussen, J.C. (1997) Modal Analysis Based on The Random Decrement Technique -
Application to Civil Engineering Structures. Ph.D. Thesis. Department of Building
Technology and Structural Engineering, Aalborg University.
Associate Committee on National Building Code (1995) National Building Code of Canada 1995,
National Research Council, Ottawa.
Bendat, J.S. and Piersol, A . G . (1993) Engineering Applications of Correlation and Spectral
Analysis. John Wiley & Sons, New York, NY.
Black, C.J., Ventura, C E . and Tsai, P. (1997a) Ambient Vibration Measurements of the University
Drive/ Crowchild Trail Bridge in Calgary, Alberta. Technical report submitted to ISIS.
Project Number: E Q 97-005.
Black, C.J. (1997b) Seismic Response of Four Instrumented Buildings During the Northridge and
San Fernando Earthquakes. Civil 598 Report prepared for C E . Ventura.
Black, C.J., Ventura, C E . (1998a) Ambient Vibration Measurements of the Lindquist Bridge.
Technical report. Project Number: E Q 98-003.
Black, C.J., Horyna, T., Ventura, C E . and Foschi, R.O. (1998b) Shake Table Testing of a Concrete
Gravity Dam Model Unbonded at Base - Phase 2. Technical report submitted to B C Hydro.
Project Number: E Q 98-001.
Black, C J . and Ventura, C E . (1998c) Seismic Response of Four Instrumented Buildings During
two Earthquakes. Proceedings of the XVIth International Modal Analysis Conference,
Santa Barbara, California.
Bonato, P., Ceravolo, R., and De Stefano, A . (1997) Time-Frequency and Ambiguity Function
Approaches in Structural Identification, Journal of Engineering Mechanics. December.
Chopra, A . K . (1995) Dynamics of Structures: Theory and Applications to Earthquake
163
Chapter 11 References
Engineering, Prentice Hall, New Jersey.
Cohen, L. (1995) Time-Frequency Analysis. Prentice Hall PTR, Englewood Cliffs, New Jersey.
Cook, S., Ventura, C.E., Jackson, S. and Black, C.J. (1998) Static and Dynamic Strain
Measurements of the Lindquist Bridge. Technical report prepared for Reid Crowther and
Partners Ltd. Project Numer: E Q 98-002.
Cooley, J.W. and Tukey, J.W. (1965) A n algorithm for the Machine Calculation of Complex
Fourier Series, Mathematics of Computation, 19.
Daubechies, I. (1988) Orthonormal Bases of Compactly Supported Wavelets. Comm. on Pure

andAppl. Math, Vol. 4, Nove.
Diehl, J.G. (1991) Ambient Vibration Survey: Application Theory and Analytical Techniques
Application Note No. 3. Kinemetrics: Pasadena, C A .
Dossing, O. (1998) Uncertainty in Time/Frequency Domain Representations, Sound and

Vibration, January.
EDI, Experimental Dynamic Investigations Ltd. (1995) U2, V2 & P2 Manual.
Ewins, D.J. (1984) Modal Testing: Theory and Practice, Research Studies Press Ltd. John Wliley
& Sons Inc. New York.
Farrar, C.R., Doebling, S.W. and Cornwell, P.J. (1998) A Comparison Study of Modal Parameter
Confidence Intervals Computed Using the Monte Carlo and Bootstrap Techniques,
Proceedings of the XVIth International Modal Analysis Conference, Santa Barbara,
California.
Felber, A.J. (1993) "Development of a Hybrid Evaluation System", Ph.D. Thesis, University of
British Columbia, Department of Civil Engineering, Vancouver, 1993.
Fenves, G.L. and Desroches, R. (1994). Response of the Northwest Connector in the Landers and
Big Bear Earthquakes, Report No. UCB/EERC-94/02, Earthquake Engineering Research
Center, The University of California at Berkeley.
Finn, W.D.L., Ventura, C.E. and Schuster, N . D . (1995) Ground Motions During the 1994
Northridge Earthquake, Canadian Journal of Civil Engineering: Vol. 22, Number 2, April
164
Gasparini, D . A . and Venmarcke, E . H . (1976) SIMQKE - A Program for Artificial Motion

Generation, User's Manual. Massachusetts Institute of Technology, Department of Civil
Engineering, Cambridge, Massachusetts.
Horyna, T., Ventura, C E . and Foschi, R.O. (1997) Shake Table Testing of a Concrete Gravity Dam
Model Unbonded at the Base. Technical report submitted to B C Hydro. Project Number:
EQ 96-006.
Huang, S.Y., Qi, G.Z. and Yang, J.C.S. (1994) Wavelet For System Identification, Proceedings of
the XHth International Modal Analysis Conference, Tennessee.
Humar, J.L. (1990) Dynamics of Structures, Prentice Hall, New Jersey.
ISIS (1997) Innovator Newsletter of ISIS Canada, April.
Luz, E., Gurr-Beyer, C. and Stoecklin, W. (1984) Experimental Investigations of Natural

Frequencies and Modes of the H D R Nuclear Power Plant by Means of Microtremor
Excitation, Proceedings of the 8th World Conference on Earthquake Engineering, San
Francisco, California.
Murphy, L . M . (1973) San Fernando, California, Earthquake of February 9, 1971. US Department

of Commerce. National Oceanic and Atmosphere Administration. Washington, D . C
National Instruments (1995) Joint Time Frequency Analyzer 3.1a Toolkit, Labview Executable.
Newhook, J.P. and Mufti, A . A . (1996) A Reinforcing Steel-Free Concrete Deck Slab for the
Salmon River Bridge, Concrete International, June 1996.
Newland, D.E. (1996) An Introduction to Random Vibrations, Spectral and Wavelet Analysis, 3rd
Ed, Addison Wesley Longman Limited, Essex, United Kingdom.
Proakis, J.G. and Manolakis, D.G. (1996) Digital Signal Processing, Principles, Algorithms, and
Applications, 3rd Ed, Prentice Hall, New Jersey.
Ramirez, R.W. (1985) The FFT Fundamentals and Concepts. Englewood Cliffs, New Jersey:
Tektronix.
Rezai, M . K . and Ventura, C E . (1996) Structural Dynamic Characteristics of a 12 Storey Steel

Frame Building During the 1971 San Fernando and the 1994 Northridge Earthquake, CSCE
1996 Annual Conference Proceedings: Volume lib.
Rezai, M . K . and Ventura, C E . (1995) Wavelet Transform Analysis of Acceleration Data Recorded
165
at Treasure Island Site During Loma Prieta Earthquake, Proceedings of the Second
International Conference on Seismology and Earthquake Engineering, Tehran, Islamic
Republic of Iran.
Schuster, N.D. (1994) Dynamic Characteristics of a 30 Storey Building During Construction

Detected From Ambient Vibration Measurements. M.A.Sc. thesis, University of British
Columbia, Canada.
Shakal, A., Huang, M . , Darragh, R., Cao, T.Q., Sherburne, R.W., Malhotra, P., Cramer, C , Sydor,
R., Graizer, V., Maldonado, G., Peterson, C. and Wampole, J. (1994) CSMIP Strong Motion
Records from the Northridge, California Earthquake of January 17, 1994. California
Department of Conservation, Division of Mines and Geology, Office of Strong Motion
Studies, Report OSMS 94-07, February.
Skolnik, M.I. (1980) Introduction to Radar Systems, Mcgraw-Hill Book Co.
Staszewski, W.J. and Giacomin, J. (1997) Applications of the Wavelet Based FRFs to the Analysis
of Nonstationary Vehicle Data, Proceedings of the XVth International Modal Analysis
Conference, Orlando, Florida.
Vibrant (1996) Me'Scope Modal. Vibrant Technology, Inc. Jamestown, California.
Wigner, E.P. (1932) On the Quantum Correction for Thermodynamic Equilibrium, Physical
Review, vol. 40
166
APPENDIX A
The Time Frequency Response Function
If the input into a structure is known, the Time Frequency Response Function can be used to esti-
mate the natural frequencies of the structure. It is analogous to the frequency domain FRF, in that,
it is the ratio of the output to input motion. The TFRF however, provides information in both the
time and frequency domains simultaneously. A proof of the TFRF is given along with a single
degree of freedom example.
A.1 The Time Frequency Response Function

As shown in Chapter 2, the Frequency Response Function, (FRF), is defined as the response of a
structure to a unit amplitude, harmonic function and is given by:
H ( 0 ) J = fM (A.,)
Therefore, the F R F can be calculated by dividing the Fourier Transform of the output signal by
the Fourier Transform of the input signal.
To prove the TFRF, it is easiest to show that it has the same form as the FRF. That is, show that
the windowed signal will yield a correct value of frequency and then extend it to all window loca-
167
Appendix A The Time Frequency Response Function
tions introducing the time component of the TFRF.
As described in Chapter 2, the time-frequency spectrogram, P(co, t), is obtained from:
P(co, 0 = I ^ C O ] 2
(A.2)
where |Su,(f)| is the Short Time Fourier Transform of a time signal.
If P(co, 0;« represents the time-frequency spectrogram of the input and P((H, t) out represents the
output time-frequency spectrogram, then the TFRF is given by:
P(to, t)
v
' 'in
The proof of Equation 3 begins with the dynamic equation of equilibrium for a single degree of
freedom system with a forcing function p{t) given by:
my{t) + cy{t) + ky = p{t) (A.4)
Multiplying both sides of Equation 4 by a window function w(t) yields:
my(t)w(t) + cy(t)w(t) + ky(t)w(t) = p(t)w(t). (A.5)
168
If:
x(t) = y(t)w(t), (A.6)
then:
x(t) = y(t)w(t) + y(t)w(t) (A.7)
and
x(t) = y(t)w(t) + 2 (t)w(t) + y(t)w(t).

y (A.8)
If a rectangular window w(t), in the time interval (t , t ), is chosen for the analysis, then:
a h
lfor t <t<t h
w(t) = ; a b
0 elsewhere
The first and second derivatives of the window function are given by:
w(t) = $(t-t )-b(t-t )

a b (A.9)
and
w(t) = &(t-t )-8(t-t ),

a b (A.10)
where §(t-t )a and 8(t-t ) h represent a Dirac delta function centered at time t and t , respec-
a b
169
tively and 8' is the first time derivative of the Dirac delta function.
Substituting Equations 9 and 10 into Equations 7 and 8 yields the following expressions for x{t)
and x(t) :
x(t) = y(t)w(t) + Y d(t-t )-Y Ht-t )

a a b b and (A.ll)
x(t) = y ( 0 w ( 0 + 2 | y 5 ( r - r ) - ^ 5 ( f - ^ ) | - r 5 ' ( r - ? ) - F ^ 8 ' ( r - ^ ) ,

a a a a (A. 12)
where Y = y(t ),
a a Y = y{t ), Y = y(t ) and Y = y{t ).
h h a a h h
If equations 11 and 12 are rearranged for y(t)w(t) and y(t)w(t) and substituted into Equation 5
the result is:
mx(t) + cx(t) + kx(t)-mB(t)-cA(t) = f(t) (A. 13)
where
f(t) = p(t)w(t),
A(t) = Y b(t-t )-Y b(t-t )

a a b b (A.14)
170
and
B(t) = 2 | F 5 ( . - . ) - y ^ - ^ ) | - F 5 ' a - y - r ^ ( r - y .
f l a f l (A.15)
Notice, that A(t) and B(t) are zero for all locations other than t and t
a b
Taking the Fourier Transform of both sides of Equation 13 yields:
2
(&-co m + /coc)X(co) - mB(d)) - cA(co) = F(co) (A.16)
where:
X(cd) and F(co) are the Fourier Transforms of x(t) and f(t) respectively and.
i(Ht n i(i)th
A(co) = Y e a
a
-Y e b (A.17)
and
B((0) = (2Y + i(0Y )e

a a
a
-(2Yb + mY )e
b . (A.18)
By defining an effective forcing function F((fl) = F(co) - mB((a) + cA(co) the windowed H((£>)
function at time t is given by:
171
H((ti) = z± . (A.19)
F ((0) t
Equation 19 has the same form as Equation 1 for the Frequency Response Function with the Fou-
rier Transform of the forcing function replaced by the Fourier Transform of a modified forcing
function which includes the effect of the end conditions, that is, the displacement and velocity of
y(t) at t = t and t =
a t.
h
Equation 19 should, in theory, yield the same FRF for each time shift of the window.
Therefore, if the center of the window, w(t), at any given time is defined by t , then c
Ht ((d) =>'H((0, t ) as t is translated through the record. Equation 19 can therefore be rewritten
c c
as:
H((0,t )
c = zz . (A.20)
The TFRF given by Equation 3 is obtained by sqauring the magnitude of t ) and F(co, t ) in
c c
Equation 20. The resulting expression is the spectrogram of the output divided by the spectro-
gram of the input with the corrections for the end conditions and is given by:
172
TFRF((0, t ) c =
c
out |*eM >| c
2
(A.21)
2
Therefore, if the input is corrected for the end conditions of the window function, the TFRF can
be defined as the square of the amplitude of the response of a structure to a unit harmonic input
function.
In practical signal analysis, the numerical evaluation of the spectrogram is done with a Hanning
window function which minimizes undesirable leakage effects. If a Hanning window is used for
the computation of the spectrogram in the numerator and denominator of Equation 21, it can be
shown that the product of A(f) and B(t) with the Hanning window produces zero values in the
interval (t „ t ).
( h As a consequence, there is no need to account for the end conditions in (t , a t ),
h
and therefore the TFRF calculated using the Hanning window function is a good approximation to
the square of FRF. It is only an approximation, as the Hanning window slightly distorts the spec-
trogram.
A.2 SDOF Example of the TFRF

The TFRF was calculated for a single degree of freedom system subjected to the Northridge
earthquake. The SDOF system has a natural frequency of 2 Hz and a damping value of 5%. The
resulting time-frequency plot is presented in Figure 11.1.
The TFRF shows a ridge at 2 Hz for the entire length of the record confirming the accuracy of the
173
TFRF. It can be inferred from the contour levels in Figure 11.1 that the amplitude of the TFRF is
around 0.09 which is the square of the amplitude of the FRF as predicted by Equation 21.
174
TFRF Magnitude
d)0.2
CO
LV.V.V." 0.003
I I 0.03
;0.1
0.0
60
50
40
o
JO,
30 CU
E
20
10
J
0
-300 0 3 0 0 - 1 5 0 15 0 2 3 4
Base Accel. Roof A c c e l . F r e q u e n c y (Hz)
(cm/s ) 2
(cm/s ) 2
Figure 11.1: SDOF Example Using the TFRF
175
APPENDIX B
Data Analysis Software
This appendix briefly describes the frequency domain and joint time-frequency domain software
programs used for the analysis presented in this thesis.
B.1 Ambient Vibration Data Analysis Software

Frequency Domain Analysis
The custom data acquisition program A V D A (Schuster, 1994) was used to record the ambient
vibrations of both bridges studied in this thesis.
The computer programs P2, U2 and V2 (EDI, 1995) were used to identify the natural frequencies
and mode shapes of each bridge. Program P2 was used to compute the Averaged Normalized
Power Spectral Density (ANPSD) for a series of ambient vibration records. Program U2 was
used to view signals quickly in order to decide whether the data obtained was satisfactory. It was
also used to calculate individual power spectral densities and the potential modal ratios needed for
the frequency and mode shape estimations. V2 was developed in conjunction with the program
U2 to illustrate and animate mode shapes obtained from ambient vibration data.
Joint Time-Frequencv Analysis
The Lab view JTFA Toolkit 3.1a (National Instruments, 1995) was used to compute the time-fre-
176
Appendix B Data Analysis Software
quency spectrograms. Due to the input characteristics no further analysis was required.
B.2 Shake Table Data Analysis

Joint Time-Frequency Analysis
The time-frequency spectrograms were calculated using the JTFA Toolkit (National Instruments,
1995). No further analysis programs were required.
B.3 Strong Motion Data Analysis Software

Frequency Domain Analysis
The dynamic properties (mode shapes and frequencies) of the buildings studied in this thesis are
calculated using the frequency domain techniques described above. To facilitate this, the program
Me'Scope Modal (Vibrant, 1997) was used. The time domain acceleration data recorded during
the seismic events is read into Me'Scope. Me'Scope allows the user to animate a structure by
building a model and assigning records to certain joints. By assigning the base records as inputs
and the others as outputs, Me'Scope calculates the frequency response matrix for the given struc-
ture. The frequency response information can be viewed in a number of ways: F R F magnitude,
co-quad plot (real vs. frequency and imaginary vs. frequency), phase, coherence and individual
power spectral densities. This information along with animation in the frequency domain allows
the user to determine the structure's natural frequencies and mode shapes.
Joint Time-Frequency Analysis
The strong motion data measured at the basement and roof levels are used to calculate the Short
Time Fourier Transform spectrogram. The spectrogram was calculated using the JTFA Toolkit
177
Appendix B Data Analysis Software
(National Instruments, 1995). The TFRF is constructed using a spreadsheet program to divide the
roof spectrogram by the basement spectrogram. To avoid any undefined points in the TFRF all
points with nearly zero magnitude are given a magnitude of 0.0001 prior to the construction of the
TFRF.
178

Ubc 1998-0359

Uploaded by

Copyright:

Available Formats

You might also like

Ubc 1998-0359

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ubc 1998-0359

Uploaded by

Copyright:

Available Formats

D Y N A M I C A N A L Y S I S OF CIVIL ENGINEERING STRUCTURES

USING JOINT T I M E - F R E Q U E N C Y METHODS

B . A . S c , The University of British Columbia, 1996

A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF

T H E REQUIREMENTS FOR THE D E G R E E OF

(Department of Civil Engineering)

We accept this thesis as conforming

to the required standard/

© Cameron John Black, 1998

Department of £l\JXL- EM^J/OgggJ^c^

The University of British Columbia

for the analysis of civil engineering vibration data.

for the analysis of civil engineering vibration data.

quency analysis is used to improve the understanding of this phenomenon.

location of modal response.

5.2 Description of the Bridge 49

Figure 2.2: Dynamic Free Body Diagram 7

Figure 2.3: Three Degree of Freedom Model 9

Figure 2.4: Real and Imaginary Parts of the F R F 19

Figure 2.5: Hanning Window Function 21

Figure 2.6: Comparison of the Spectrogram and the Wigner Distribution 27

Figure 4.1: Location of the San Fernando and Northridge Earthquakes 45

Figure 4.2: Example of Time-Frequency Plot 47

Figure 5.1: Photo of the University Drive/Crowchild Trail Bridge 50

Figure 5.2: Schematic of Sensor Locations 52

Figure 5.3: Averaged Normalized Power Spectral Densities 55

Figure 5.4: Averaged Normalized Power Spectral Densities (Log-Normal Scale) 56

Figure 5.5: Mode Shapes of the University Drive/Crowchild Trail Bridge 58

Figure 5.6: Wide Band T-F Plot of the Calgary Bridge 62

Figure 5.7: Narrow Band T-F Plot of the Calgary Bridge 63

Figure 6.1: Photo of the Lindquist Bridge 70

Figure 6.2: Photo of the Top of the Bridge 70

Figure 6.3: Photo Showing the Underside of the Bridge 71

Figure 6.4: Sensor Locations and Directions of the Measurements 72

Figure 6.7: Mode Shapes of the Lindquist Bridge 79

Figure 7.1: Schematic of Dam Model 91

Figure 7.2: Photo of Dam Monolith Model 92

Figure 7.4: Hydrostatic Load Assembly 94

Figure 7.5: Typical Dynamic Test Setup 96

Figure 7.6: The 5 Hz Harmonic Input Record 97

Figure 7.7: 5 Hz Earthquake Record 98

Figure 7.8: T F R Plot of 20 Hz Harmonic Input - No Upstream Motion 101

Figure 7.9: T F R Plot of 20 Hz Harmonic Input - Upstream Motion 102

Figure 7.10: T F R Plot of 22.5 Hz Harmonic Input - Upstream Motion 103

Figure 7.11: T F R plot of 15 Hz Earthquake Input - No Upstream Motion 105

Figure 7.12: T F R plot of 25 Hz Earthquake Input - Upstream Motion 106

Figure 8.5: Measured Accelerations During the Northridge Earthquake 116

Figure 8.6: F R F s of the 1901 Avenue of the Stars Building 119

Figure 8.10: Figure Showing Contribution of Heavy Shaking to F R F 127

Figure 8.13: 1901 T-F Plot - San Fernando Earthquake - Longitudial

Direction - 0 to 0.6 Hz 132

Figure 9.1: Photo of Sheraton Universal Hotel 135

Figure 9.5: Measured Accelerations During the Northridge Earthquake 140

Figure 9.6: Frequency Response Functions for the Sheraton 144

Figure 9.7: Mode Shapes of the Sheraton Universal 146

Figure 9.12: Sheraton T-F Plot - San Fernando Earthquake - Longitudial

Direction - 0 to 0.8 Hz 155

Figure 11.1: S D O F Example Using the T F R F 175