Evaluation of Seismic Damage of Tall Buildings From Acceleration Response Time Histories by A Modal Decomposition Approach

Evaluation of Seismic Damage of Tall Buildings from Acceleration
Response Time Histories by a Modal Decomposition Approach
by
Chananwath Sinthumongkhonchai
A thesis submitted in partial fulfillment of the requirements for the

degree of Master of Engineering in
Structural Engineering
Examination Committee: Prof. Pennung Warnitchai (Chairperson)

Dr. Punchet Thammarak
Dr. Thanakorn Pheeraphan
Nationality: Thai
Previous Degree: Bachelor of Engineering in Civil Engineering
King Mongkut’s Institute of Technology
Ladkrabang, Thailand
Scholarship Donor: HM King’s Scholarships
Asian Institute of Technology

School of Engineering and Technology
Thailand
May 2020
ACKNOWLEDGEMENTS
I want to give thanks to my Lord Jesus for influencing me to be faithful and patient on
finishing this study. Instead of being anxious with many difficulties, because of the hope I
have in him, I worked peacefully knowing this study will be completed.
I am thankful to my thesis advisor Prof. Dr. Pennung Warnitchai that he spent a great amount
of his time supervising me in every step and worked his best to help me finish the study. I
also would like to acknowledge my Committee members: Dr. Punchet Thammarak and Dr.
Thanakorn Pheeraphan for kindly listening to my presentations and asking useful questions
so that the documentation of this study can be further improved.
Without the opportunity from HM King’s Scholarships, I would not have experienced this
wonderful time in my life, studying at AIT and completing this study.
I would like to give special thanks to AIT Solutions staff: Mr. Thaung Htut Aung, Dr. Pramin
Norachan and Mr. Sudan Panday, for their kindness and wonderful time helping and sharing
knowledge and many valuable resources. I am grateful to Dr. Fawad Najam, PhD candidate
Phichaya Suwansaya, Mr. Phruek Chansukho, Mr. Kritcha Karunkritkul and Mr. Chanon
Chotmanee for their useful answers to the questions regarding this study.
I am also thankful to food vendors and AIT cafeteria staffs that I was full every day and
energized to proceed on the study because of the daily food they provided. I would like to
give thanks to AIT librarians for their great services, supporting this work.
I would also like to give thanks to my brothers and sisters at the RAC church for their prayers
and cares regarding my study and work at AIT, to my nearest friends for being my joy at
AIT despite many challenging situations and to my lovely family for their relentless contacts
and warm messages daily. I appreciate the love that they all have towards me.
I am also thankful for many other people in AIT (guards, classmates, staffs at STE
laboratory, cleaners and so on) that I have not mentioned, yet indirectly help me to complete
this study.
Chananwath Sinthumongkhonchai
July, 2020
ii
ABSTRACT
Under seismic events, there are some buildings having acceleration records measured. Some
guidelines expected that these acceleration records shall be utilized for engineers to
understand the dynamic behavior of those buildings under such seismic events in order for
the improvement of the future design of related buildings, and the guidelines even expected
that these acceleration records shall be utilized for rapid seismic damage assessment of those
buildings. However, there was no explicit and well-known accepted way available that these
acceleration records can possibly be utilized for seismic damage evaluation.
There were studies from many researchers reporting on the effectiveness of how seismic
demands of tall building can be predicted based on the concept of mode of vibration of tall
buildings; contribution from each mode was firstly calculated and then summation of those
contributions become estimated seismic responses. It was shown in those studies that even
though the responses were actually beyond the elastic range of tall buildings where the
concept of mode of vibration is theoretically invalid, the combined responses of those
separately calculated contributions were shown to be reasonable estimates of complex
seismic demands calculated by Nonlinear Response History Analysis procedure (NLRHA).
Therefore, in this study, the seismic damage evaluation scheme of tall buildings from
acceleration response time histories by a modal decomposition approach is proposed.
Applying the very same concept used in those studies, the proposed scheme is developed
and its effectiveness on how it is used to estimate various types of seismic demands once the
only known information under seismic events are measured acceleration records is to be
identified in this study. Since the formulation of the scheme is firstly developed here in this
study, the finite element commercial software, Perform3D, which can perform NLRHA
under selected ground motions are employed to generate acceleration time-histories of a
nonlinear numerical model under those selected ground motions. Then the proposed scheme
shall be applied to estimate seismic responses of that numerical model. This was done to
minimize the error that will happen in practice due to field measurement and another required
additional process called model updating, so that the inherent error due to the proposed
scheme itself (if any) can be realized and emphasized in this study.
The results of the study were found that some response types of seismic demands such as
floor displacements and inter-story drifts of the numerical model under selected ground
motions utilizing this proposed scheme were, in general, reasonably estimated. However,
because of the inherent error of the proposed scheme found in this study, other response
types of seismic demands were found to be fairly overestimated.
iii
TABLE OF CONTENT
CHAPTER TITLE PAGE
Title Page i
Acknowledgements ii
Abstract iii
Table of Content iv
List of Figures ix
List of Tables xx
List of Abbreviations xxi
1 Introduction 1
1.1 Background 1
1.2 Statement of the Problems 2
1.3 Objectives of the Research 3
1.4 Scope of this Study 3
2 Literature Review 4
2.1 Structural Health Monitoring and Evaluation of Potential
Damage of Civil Engineering Structures due to Extreme
Events 4
2.1.1 Damage Detection Based on Frequency Shift 9
2.1.2 Damage Detection Based on Traditional Approaches 13
2.1.3 Damage Detection Based on Strain Monitoring 15
2.1.4 Damage Detection Based on Model Updating 19
2.1.5 Damage Detection Based on Modal Decomposition 23
2.2 Operational Modal Analysis (OMA) 25
2.2.1 Preliminary Concepts and Fundamental Principles in OMA
25
2.2.2 Mathematical Background in OMA Based on This
Research 27
2.2.2.1 Fourier Transform 27
2.2.2.2 Singular Value Decomposition (SVD) 30
2.2.3 Stationary Random Data and Process 30
2.2.3.1 Basic Concepts 30
2.2.3.2 Spectral Density Functions 33
2.2.3.3 Errors in Spectral Density Estimates and Remedies
for the Errors in OMA 35
iv
2.2.4 Observability 36
2.2.5 Sampling Frequency and Filtering 38
2.2.6 Output-only Modal Identification Technique 38
2.2.6.1 Peak Picking Method 39
2.2.6.2 Frequency Domain Decomposition Method 42
2.3 Earthquake Analysis and Response of Multi-story Buildings 45
2.3.1 Earthquake Analysis and Response of Linearly Elastic
Buildings 45
2.3.1.1 Static Condensation 46
2.3.1.2 Orthogonality of modes 47
2.3.1.3 Modal Response History Analysis 48
2.3.1.4 Modal Response Contribution 49
2.3.1.5 Multi-story Building Subjected to Earthquake
Excitation 49
2.3.2 Earthquake Analysis and Response of Inelastic Buildings 53
2.3.2.1 Nonlinear Response History Analysis (NLRHA) 53
2.3.2.3 Formulation of Modal Pushover Analysis (MPA) 54
2.3.2.2 Formulation of Uncoupled Modal Response History
Analysis (UMRHA) 55
2.3.2.4 Recent Studies on Inelastic Response Estimation
using UMRHA approach 57
2.3.4.5 Modified Flag Shape Model 60
3 Methodology 62
3.1 Overview of the Research Methodology 62
3.2 Introduction to the Proposed Seismic Damage Evaluation
Scheme 62
3.2.1 Proposed Seismic Damage Evaluation Scheme: Part A 63
3.2.2 Proposed Seismic Damage Evaluation Scheme: Part B 63
3.3 Modal Decomposition Based on Measured Accelerations
from All Floors 66
3.4 Formulation of the Modal Decomposition Formula Based on
Measured Limited Number of Acceleration Records 67
3.5 How to Convert Measured Accelerations to the Relative
Accelerations Associated with the Dynamic Degrees of
Freedom of the building (for Modal Decomposition
Approach) 69
3.6 How to Relate Estimated Modal Accelerations to the Global
Seismic Demands (for Modal Decomposition Approach) 71
3.6.1 Demand Estimation Under the Elastic Range 71
v
3.6.2 Demand Estimation Under the Inelastic Range 74
3.6.2.1 Story Force Prediction Based on Estimated Floor
Acceleration Time History Vector 75
3.6.2.2 Story Force Prediction Based on Inelastic SDF
Relationships 76
3.6.2.3 Story Force Prediction Based on the Updated
Numerical Model 77
3.7 Case Study Building 81
3.8 Selection of Ground motions 83
3.9 Reference Methods 83
3.9.1 Reference Analysis Methods 83
3.9.1.1 Single Degree of Freedom – Uncoupled Modal
Responses History Analysis 85
3.9.1.2 Multi Degree of Freedom – Uncoupled Modal
Responses History Analysis 85
3.9.2 Reference Modal Decomposition Approach 86
3.9.3 Verification of Reference Methods 86
3.9.3.1 Verification within linearly elastic range 86
3.9.3.2 Verification beyond linearly elastic range 86
3.9.4 Special Study on Determination of How Many Modes of
Vibration Required to Accurately Estimate Various Kinds
of Seismic Responses 87
3.10 Procedure for Verifying the Proposed Seismic Damage
Evaluation Scheme Under the Simplified Predetermined
Scenario 88
3.10.1 Overview of the Simplified Predetermined Scenario of the
Verification Procedure 88
3.10.2 Verification of the Seismic Demands Estimated from the
Proposed Scheme 90
3.11 Additional Step to Enhance the Accuracy of the Proposed
Scheme on Account of Higher Mode Effects 90
3.11.1 Background behind the proposed correction procedure 90
3.11.2 The proposed enhancement procedure 90
4 Results and Discussion 92
4.1 Overview 92
4.2 Modal Analysis Results 92
4.3 Monotonic and Cyclic Pushover Analysis Results 92
4.3.1 Monotonic Pushover Analyses Within Elastic range 92
4.3.2 Hysteretic Behavior of the Numerical Model 92
vi
4.4 Verification of Reference methods 98
4.4.1 Verification results within linearly elastic range 98
4.4.1.1 Results Under 0.1× El-Centro Ground Motion (NS)
98
4.4.1.2 Results Under 0.3 × San Fernando Ground Motion
(NS) 104
4.4.1.3 Results Under 0.1 × Northridge Ground Motion
(NS) 104
4.4.1.4 Conclusion of the Verification Results Within
Linearly Elastic Range 104
4.4.2 Verification results beyond linearly elastic range 115
4.4.2.1 Results Under 1.0 × El-Centro Ground Motion (NS)
115
4.4.2.2 Results Under 3.0 × San Fernando Ground Motion
(NS) 128
4.4.2.3 Results Under 1.0 × Northridge Ground Motion
(NS) 141
4.4.2.4 Conclusions of the Verification Results Beyond
Linearly Elastic Range 141
4.5 Results of the Special Study on Determination of How Many
Modes of Vibration Required to Accurately Estimate Various
Kinds of Seismic Responses 153
4.5.1 Results of Special Study Under 1.0 × El-Centro Ground
Motion (NS) 153
4.5.2 Results of Special Study Under 3.0 × San Fernando Ground
Motion (NS) 153
4.5.3 Results of the Special Study Under 1.0 × Northridge
Ground Motion (NS) 153
4.6 Performance of the Proposed Seismic Damage Evaluation
Scheme Under the Simplified Predetermined Scenario 166
4.6.1 Estimation of Seismic Demands Under 1.0 × El-Centro
4.6.2 Estimation of Seismic Demands Under 3.0 × San Fernando
4.6.3 Estimation of Seismic Demands Under 1.0 × Northridge
4.7 The Performance of the Proposed Additional Step to Enhance
the Accuracy of the Proposed Scheme on Account of Higher
Mode Effects 208
4.7.1 Verification of Corrected Estimated Story Forces Under 1.0
× El-Centro Ground Motion (NS) 208
vii
× San Fernando Ground Motion (NS) 208
× Northridge Ground Motion (NS) 209
5 Conclusions and Recommendations 216
5.1 Conclusions 216
5.2 Recommendations for the Future Studies 219
References 220
viii
LIST OF FIGURES
FIGURE TITLE PAGE
Figure 2.1 (a) SIT building, (b) type of isolators, (c) layout of isolation system 4
Figure 2.2 Sensor layout 5
Figure 2.3 Response spectra of ground motions recorded during largest five
earthquakes between 2010 and 2012 for 5% damping: (a) EQ1 and
EQ2, (b) EQ3-EQ5. Note that DS-L2 are design spectra for Level 2
earthquake for soil condition S2 and S3. 6
Figure 2.4 (a) The first three identified mode shapes of the building, (b) time
variation of natural frequencies, and (c) time variation of damping
ratio identified from the main shock by system identification 7
Figure 2.5 Transfer functions of acceleration responses at the top of both

buildings during EQ1 and EQ2 ground motions in their own weak
axis 8
Figure 2.6 Influence of the shaking intensity on isolator's relative displacement

during the main shock for: (a) 1st mode, (b) 2nd mode, (c) 3rd mode
8
Figure 2.7 Long-term time sequential variation of the first mode: (a) natural
frequency from September 2010 to December 2011, (b) natural
frequency from January 2012 to December 2012 9
Figure 2.8 Dimension of test bridge: (a) elevation plan, (b) cross-section at the
center of main girder 10
Figure 2.9 (a) Results of vibration test, (b) change of natural frequency of the
1st vibration mode under failure process 11
Figure 2.10 (a) Layout of accelerometers at the top of the tower, (b) typical
acceleration record at the top of the tower due to ambient vibration,
(c) seismic response recorded on 21/06/2013 11
Figure 2.11 Variation of identified natural frequencies (global modes) versus

temperature between 17/12/2012 and 20/06/2013 12
Figure 2.12 Variation of natural frequency between 01/06/2013 and 10/07/2013:

(a) mode B1, (b) mode B2 13
Figure 2.13 Change in frequency-temperature correlation induced by the seismic

event of 21/06/2013: (a) mode B1, (b) mode B2 13
ix
Figure 2.14 Damaged building at the 8th floor by reduction of Young's modulus
and 18 measuring nodes are positioned equidistantly in the middle
of each floor, (b) Tensile damage of the Ophite tower numerical
model during L' Aquila earthquake 14
Figure 2.15 Damage index due to L'Aquila earthquake: (a) Mode shape curvature
method, (b) flexibility method 15
Figure 2.16 Sensor layout of the monitoring system 16
Figure 2.17 Typical sensor layout: (a) core walls, (b) mega columns and
outrigger trusses 17
Figure 2.18 (a) Vibrating wire strain gauge, (b) concrete coating, (c) a strain
rosette in the inner tube, (d) protection of cables 18
Figure 2.19 The variation of vertical strain from a strain gauge at the height of
121.2m on April 19, 2018 18
Figure 2.20 (a) Instrumented 26-story building, (b) 1st and 2nd translational
mode shapes of the instrumented building 20
Figure 2.21 Damage states of non-updated and updated FE model under a ground
motion 21
Figure 2.22 Regions selected to have different Moduli of Elasticity at the height
lower than 23 meters 23
Figure 2.23 Sensor layout on Roppongi Tower 23
Figure 2.24 Acceleration and displacement at the top of the building 24
Figure 2.25 Prediction of structural damage from drift angle and warnings for
furniture turnover and ceiling drop down from floor accelerations 24
Figure 2.26 Schematic representation of LTI system 26
Figure 2.27 Discrete Fourier Transform of the rectangular window 35
Figure 2.28 Typical sensor layout for building-like structures 37
Figure 2.29 Aliasing in digitized wave form 37
Figure 2.30 (a, b) APSD of signals in orthogonal directions, (c, d) coherence

functions of the mentioned signals and their reference signals in each
direction 40
Figure 2.31 Schematic chart of HPBW 41
Figure 2.32 Typical Singular Value plots 42
x
Figure 2.33 The parameters in FDD affecting damping ratios: (a) NFFT, (b)
types of window 45
Figure 2.34 (a) DOFs of the frame, (b) DOFs in the dynamic analysis 46
Figure 2.35 Typical plan of multi-story building 51
Figure 2.36 Modal decomposition of roof displacement due to 𝑝𝑒𝑓𝑓, 𝑛𝑡, n = 1,

2, 3 and 4: (a) due to 𝑝𝑒𝑓𝑓, 1𝑡, (b) due to 𝑝𝑒𝑓𝑓, 2𝑡, (c) due to
𝑝𝑒𝑓𝑓, 3𝑡, (d) due to 𝑝𝑒𝑓𝑓, 4𝑡 56
Figure 2.37 Typical pushover curves of a reinforced concrete core wall building 57
Figure 2.38 Envelop results obtained via UMRHA (MFS used) and NLRHA for
3.5 x Manjil Iran 59
Figure 2.39 Roof displacement from combined response from UMRHA and
NLRHA due to 3.5 x Manjil Iran: (a) using Flag shape model , (b)
MFS model 59
Figure 2.40 Summary of parameters in MFS 60
Figure 2.41 Construction of idealized tri-linear backbone curve 60
Figure 3.1 The flow charts of the proposed seismic damage evaluation: (a) part
A, (b) part B 65
Figure 3.2 Degrees of Freedom of any tall building stated in this study: (a)
theoretical dynamic DOFs of all floors, (b) theoretical dynamic
DOFs of some floors, (c) typical DOFs of some certain floors
required for the measurement set up 70
Figure 3.3 Movement of a floor relative to its center of mass 70
Figure 3.4 The flow chart summarizing the proposed force prediction process
beyond elastic limits based on Modal Decomposition approach 80
Figure 3.5 Finite Element modeling: (a) 3D view, (b) plan view 82
Figure 3.6 Ground motions: (a) 2.5%-damped response spectra, (b) El-Centro
(NS) x 1 time-history, (c) San Fernando (NS) x 3 time-history, (d)
Northridge (NS) x 1 time-history 84
Figure 3.7 Flow chart of typical procedure for verifying the reference methods
in this study 87
Figure 4.1 Mode shapes of three dominant modes in each direction 94
Figure 4.2 Pushover curves of modes of vibration in y direction: 95
xi
Figure 4.3 Inelastic SDF force-deformation relationships of modes of vibration
in y direction: (a) 1st mode, (b) 2nd mode, (c) 3rd mode 96
Figure 4.4 Comparison between SDF force-deformation relationship - derived

from Cyclic Pushover analysis and that predicted by idealized SDF
nonlinear system: 97
Figure 4.5 Verification of estimated roof-top displacement time-history 99
Figure 4.6 Verification of estimated roof-top (relative) acceleration time-

history 100
Figure 4.7 Verification of estimated Base-Shear time-history 101
Figure 4.8 Verification of estimated Moment story time-history of the 20th floor
102
Figure 4.9 Verification of estimated response envelopes 103

history 106
108
Figure 4.15 Verification of roof-top displacement time-history 110
Figure 4.16 Verification of estimated roof-top acceleration time-history 111
113

history 118
Figure 4.23 Verification of estimated Story force time-histories 120
xii
Figure 4.24 Verification of estimated Story force envelopes 120
Figure 4.26 Verification of estimated Moment time-history of the 20th floor 122
Figure 4.27 Comparison of estimated Base-Shear against estimated roof-

displacement of each mode under 1.0 × El-Centro ground motion
(NS) / 123
125

displacement of each mode under 1.0 × El-Centro ground motion
(NS) / 126
Figure 4.33 Mode by mode story force envelopes 127

history 131

displacement of each mode under 3.0 × San Fernando ground
motion (NS) / 136
xiii
138

displacement of each mode under 3.0 × San Fernando ground
motion (NS) / 139

history 143

displacement of each mode under 1.0 × Northridge ground motion
(NS) / 148
150

displacement of each mode under 1.0 × Northridge (NS) / 151
Figure 4.65 Verification of estimated relative floor acceleration envelope from

MD-All 154
xiv
Figure 4.66 Verification of estimated roof-top (relative) floor acceleration time-
history 154
Figure 4.67 Verification of estimated Moment story envelope from MD-All 155
Figure 4.68 Verification of estimated Shear-force story envelope from MD-All 156
Figure 4.69 Verification of estimated force time-histories at the base from MD-
All 157
All 157

MD-All 158

history 158
All 161
All 161

MD-All 162

history 162
All 165
All 165
Figure 4.83 Identification of nodes from which absolute acceleration time-

histories 167
Figure 4.84 Schematic representation of relative floor accelerations associated

with the center of mass of the jth floor 168
xv
Figure 4.85 Comparison between relative floor accelerations 176
Figure 4.86 Verification of estimated modal acceleration time-histories obtained

as a result of employing Modal Decomposition based on measured
limited numbers 177
Figure 4.87 Verification of estimated modal accelerations in frequency domain

obtained as a result of employing Modal Decomposition based on
measured limited numbers 178
Figure 4.88 Verification of estimated modal coordinate time-histories obtained

limited numbers 179
Figure 4.89 Verification of estimated modal coordinates in frequency domain

obtained as a result of employing Modal Decomposition based on
measured limited numbers 180
Figure 4.90 Verification of estimated response envelopes based on MD-Lim 181
Figure 4.91 Estimated relative floor accelerations associated with the floors,
having accelerometers mounted to, predicted by MD-Lim 182
Lim 183
Figure 4.93 Verification of estimated story force envelopes from MD-Lim 183
Lim 184
Lim 185
Figure 4.99 Plots dedicated to explanation on how the cumulative error occurred
once employing the force prediction method in subsection 3.6.2.1 for
MD-Lim to predict Shear-force story envelope under 1.0 × El-
Centro ground motion (NS): 186
Figure 4.100 Verification of estimated force time-histories in modal coordinate 187
xvi
Figure 4.103 Cumulative error, caused by double integration process, 191
Figure 4.104 Magnitude response of the Highpass filter employed to deal with low
frequency signals caused by double integration process 191
Figure 4.105 Esimated modal accelerations and erroneous estimated modal

coordinates caused by double integraton process in frequency
domain 192
Figure 4.106 Delay effect as a result of Highpass filter 192

limited numbers 193

limited numbers 194
Lim 196
Lim 197
Lim 198

limited numbers 201

limited numbers 202
xvii
Lim 204
Lim 205
Lim 206
Figure 4.129 Verification of the estimated story forces predicted by MD-Lim

employing the 1st version (time-history based) force prediction
method in subsection 3.6.2.3 after being further corrected by higher
mode contributions calculated by SDF-MRHA analysis procedure
under 1.0 × El-Centro ground motion (NS) 210

employing the 2nd version (envelope based) force prediction method
in subsection 3.6.2.3 after being further corrected by higher mode
contributions calculated by SDF-MRHA analysis procedure 210
Figure 4.131 SRSS envelopes of estimated Moment story envelopes along with 211
Figure 4.132 SRSS envelopes of estimated Shear-force story envelopes along with
211

under 3.0 × San Fernando ground motion (NS) 212

contributions calculated by SDF-MRHA analysis procedure under
3.0 × San Fernando ground motion (NS) 212
xviii
213

under 1.0 × Northridge ground motion (NS) 214

contributions calculated by SDF-MRHA analysis procedure under
1.0 × Northridge ground motion (NS) 214
215
xix
LIST OF TABLES
TABLE TITLE PAGE
Table 2.1 Summary of five largest earthquakes between 2010 and 2012 5
Table 2.2 Identified, updated and non-updated FEM frequencies of vibration 20
Table 2.3 Updating parameters 22
Table 2.4 Mode shape of the top floor identified from ambient responses 40
Table 3.1 Properties and dimensions of case study building 81
Table 3.2 Ground motion details 83
Table 4.1 Mass participation Ratio 93
Table 4.2 Calculated dynamic properties 93
Table 4.3 Root mean square of force time-histories in modal coordinate of the
first mode 174
second mode 174
third mode 174
xx
LIST OF ABBREVIATIONS
APSD Auto Power Spectral Density function

BFD Basic Frequency Domain method
CP Cyclic Pushover Analyses
CPSD Cross Power Spectral Density function
DBE Design Basis Earthquake
DOF Degree of Freedom
EFFD Enhanced Frequency Domain Decomposition
FFD Frequency Domain Decomposition
FFT Fast Fourier Transform
FRF Frequency Response Function
LTI Linear Time Invariant system
LTHA Linear Time History Analysis
MCE Maximum Considered Earthquake
MD Modal Decomposition
MD-All Modal Decomposition based on measured accelerations from all floors
MD-Lim Modal Decomposition based on measured accelerations from limited number
of floors
MDOF Multi Degree of Freedom
MP Monotonic Pushover Analysis
MRHA Modal Response History Analysis
MSF Modified Flag Shape model
NLRHA Nonlinear Response History Analysis
OL Operational level
PSD Power Spectral Density
PP Peak Picking method
SDF Single Degree of Freedom
SDF-UMRHA Single Degree of Freedom – Uncoupled Modal Response History Analysis
SDF-MRHA Single Degree of Freedom – Modal Response History Analysis
MDF-UMRHA Multi Degree of Freedom – Uncoupled Modal Response History Analysis
MDF-MRHA Multi Degree of Freedom – Modal Response History Analysis
SHM Structural Health Monitoring
UMRHA Uncoupled Modal Response History Analysis
VBDI Vibration Based Damage Detection
VBSHM Vibration Based Structural Health Monitoring
xxi
CHAPTER 1
1 INTRODUCTION
1.1 Background
In the early 2015, the “Guidelines and Implementing Rules on Earthquake Recording
Instrumentation for Buildings” from the Department of Public Works and Highways
(DPWH) of the Republic of the Philippines has been approved in the city of Manila. This is
in order to help support the instruction in the National Structural Code of the Philippines
2010 Volume 1, 6th Edition. It states "Unless waived by the building official, every building
in Seismic Zone 4 over fifty (50) meters in height shall be provided with not less than three
(3) approved recording accelerographs. The accelerographs shall be interconnected for
common start and common timing." These codes are made to primarily save lives of people
and to understand the real dynamic behavior of the buildings. This will also help evaluate
the current design practice and shall be important for future design. Also, one of the points
made in the objective is to understand the potential damage happening with the monitored
buildings due to the dynamic loading of earthquakes. As it is clearly seen here, there is the
attempt to study the dynamic behavior of buildings under earthquakes through the structural
responses i.e., accelerations, of those buildings and even aiming at identification of damage
from those events (DPWH, 2015). The very same subjects can also be found in the guidelines
of Los Angeles Tall Buildings Structural Design Council (LATBSDC, 2017).
Traditionally, the ground motion recorded at the base of a building can be used as the seismic
loading to its numerical model counterpart for the seismic evaluation after the building
experienced that earthquake ground motion though the method tells the analyst(s) nothing
about the performance of the real building except about the model “representing” the
behavior of the buildings. Instead, based on the understanding of the dynamic behavior of
buildings, the responses recorded at the upper parts of buildings have been used widely over
the last decades in the field of structural health monitoring (SHM).
There are many wide ranges of SHM, but the one related to using global responses of the
structures i.e., acceleration responses, is called Vibration Based Structural Health
Monitoring (VBSHM) in which its basis is that the global vibrational responses of a structure
can be related to the health of the considered structure (Karbhari, Guan, & Sikorsky, 2009).
The process of identifying dynamic characteristics such as damping ratio, natural
frequencies and mode shapes from the global vibrational responses due to ambient excitation
such wind and traffic loading is called System Identification (SI) or Operational Modal
Analysis (OMA). And many times, this information extracted from the structural responses
is utilized to identify the potential damage of the structure. The process is called Vibration
Based Damage Detection (VBDI). Currently, it draws a lot of attention from many
researchers to work on this field as their studies, and some papers published are related to
the structural health monitoring and even up to damage identification of tall buildings which
are associated with many techniques implemented. This area of study is very vital since they
can be used for seismic damage evaluation, specifically rapid one, after a building had
experienced a ground motion to answer the public whether the considered buildings are safe
for residents or not.
Under severe earthquake ground motions, buildings are expected to behave nonlinearly. It is
well known that the concept of mode of vibration of buildings (greatly based on the method
1
of superposition) is no longer valid once the considered structure has exceeded the elastic
limit, so the structural damage due to earthquake excitation, then, cannot be explained that
its damage is from the contribution of which vibration mode. However, based on the study
done starting from the study of Chopra and Goel (2001), they tried to formulate the multi-
mode pushover analysis under the assumption that the dominant contribution to responses
due to the modal inertia force pattern based on elastic mode shape is from that vibration
mode and the total responses can be considered as the combination from all mode
contributions based on elastic modal combination rule. Despite the fact that their estimate
method contradicts the well-known understanding of the method of superposition, this basis
developed in that paper can be proven reliable as a good estimate method for the evaluation
of non-linear seismic demands of high-rise RC shear wall buildings based on the studies
done by many researchers (Najam, 2017; Panday, 2017; Tahir, Pennung, & Phichaya, 2018;
Vasanthapragash, 2018).
In this research, the researcher aims to develop a novel technique predicated on the idea
stated above for seismic evaluation of tall buildings using limited acceleration responses
measured during a seismic event. These responses will be decomposed into each mode
contribution which are later utilized for seismic damage evaluation. Nonlinear numerical
model will be constructed to test the efficiency of the proposed method; responses, i.e. floor
accelerations from the numerical model is assumed to be the measured responses of a real
building, and this is done to examine the inherent error caused by the proposed method in
this study before it being implemented into further practice, if possible.
1.2 Statement of the Problems
Right now, researchers are paying their attentions more on the field of structural health
monitoring (SHM) of buildings in order to determine how well the structural condition of
those buildings are, especially after the structures experienced an earthquake ground motion
which will generally leave some potential damage to the structures. The structural health
monitoring (SHM) for some buildings is even more necessary if those buildings are historic
(Antonella, Carmelo, & Marco, 2015). Actually, many papers in the past can only figure out
that there is something changing to their considered structures which is identified only as
level I damage detection based on (Rytter, 1993), yet some researchers can figure out their
Vibration Based Damage Detection (VBDI) methods to not only be able to localize damage
(level II) and identify damage severity (level III) but also use the technique to assess the
current condition and the future prognosis of structures (level IV). However, those papers
are either unreliable or just in the initial state of developing the method – only applied with
numerical models or simplified cases.
Another thing is that, as mentioned earlier, in some countries, available data are only from
limited locations of a building, for example, in Philippines, only data are from the base,
middle floor and under roof top story (DPWH, 2015). Hence any newly developed scheme
should also be able to deal with limited condition that what the reality has.
Lastly, it is such a crucial responsibility for engineers to be able to assess the potential
damage happening in high-rise buildings subjected to earthquake ground motions, and
ensure the safety condition of those buildings. Otherwise, this may risk many people lives.
2
1.3 Objectives of the Research
• Proposing the seismic damage evaluation scheme based on the developed modal
decomposition approach for tall buildings under seismic events
• Formulating a modal decomposition approach utilizing only limited floor acceleration
records generated
• Identifying the effectiveness of the proposed scheme in terms of seismic demands
estimation under a simplified and predetermined scenario
1.4 Scope of this Study
Since this is one of the very first studies on seismic damage evaluation based on modal
decomposition approach, uncertainties, in this research, are minimized by studying the
effectiveness of this scheme on a nonlinear numerical model representing a real high-rise
building instead of any real building itself. Thus, most of the error can be eliminated, caused
by the on-site measurement and the fact that a nonlinear numerical model of a building never
represents exactly the same behavior under any seismic event as that of its real building
despite model updating process completed. The thing is, signals extracted from the model
under an input ground motion can be made noisy to some extent as a representation of the
reality, yet it is not incorporated in this study. As stated, this is the attempt to understand the
inherent error possibly caused by the developed modal decomposition approach alone.
Furthermore, the results of this research are greatly depending on the fact that the numerical
model has to be carefully made such that they can represent the real structural behavior of
the corresponding structure, and estimated ground motions can be measured from the floor
accelerations at the base level.
In this research, the aim is to complete the study with a building having RC core-walls as
the lateral load resisting system and RC slab-column frames carrying gravity load having 34
stories. The original model actually has unsymmetrical plans, yet for simplicity in this study
it is partly changed to has symmetrical plans (highly expected to behave in a Pure-sway
manner), and only for one direction that ground motions will be applied to the numerical
model to generate seismic responses.
3
CHAPTER 2
2 LITERATURE REVIEW
2.1 Structural Health Monitoring and Evaluation of Potential Damage of Civil

Engineering Structures due to Extreme Events
In this very first section of this chapter, literatures related to Structural Health Monitoring
(SHM) of civil engineering structures, mainly buildings, will be discussed. SHMs of
structures are very broad and have no clear explanation or steps on what have to be done;
they can be done, to illustrate, in order to check design assumption, understand dynamic
behaviors of the structures and emphatically in this research evaluate potential damage of
structures subjected to extreme events such as seismic events (or else in other literatures).
Checking the performance of the structure, whether it is according to the design assumption
or not and to understand the dynamic behavior of the structures, can be done through SHM
as will be shown in this paragraph. Once the strong ground motion or extreme wind come to
excite a considered structure one may want to know its current structural condition or (and)
potential damage occurring. In the study of Fujino and Siringoringo (2014), they used long
term structural monitoring records of accelerations over 3-year period to study the behavior
of an asymmetric base-isolated building in order to evaluate the serviceability performance
of the building after experiencing a strong seismic excitation over a period of time. Using
the structure records before and after large earthquakes to determine abrupt changes is the
very key to traditional SHM, and these changes can be referred to possible damage to the
structure. The very popular indicators used to do so is the dynamic characteristics of the
structure itself such as frequencies, mode shapes and damping ratios. The building in this
study is called SIT building consisting of two parts (14-story main building (M) and 7-story
annex building (A)), having braced steel frames as structural system, connecting at the corner
by elevator shafts and forming an L-shaped asymmetric structure.
Figure 2.1 (a) SIT building, (b) type of isolators, (c) layout of isolation system
4
The picture of SIT building is shown in Figure 2.1 (a). Between the basement and the 1st
floor, there are 146 units of isolators (Figure 2.1 (b)). For the instrumentation of the building,
21 triaxial accelerometers and four displacement meters are attached. The displacement
meters can measure the relative displacement between the basement and the 1st floor, i.e. the
deformation of isolators. Horizontal accelerations are set to measure in the x-direction and
y-direction with respect to the building’s orientation.
Table 2.1 Summary of five largest earthquakes between 2010 and 2012
Figure 2.2 Sensor layout
5
For the nomenclature of sensors, the first index relates to sensor type – A: accelerometer
and D: displacement meter, the second index relates to the building – M: main and A: annex
building, the third index represents the story level, and the final index shows location of the
sensor on each floor – S: South, E: East, and W: West. All sensor positions are shown in
Figure 2.2. During the long-term monitoring period, there are 5 earthquakes events out of
more than 100 ones, considered to be large according to their seismic intensity larger than 4
(JMA standard). The summary of those events is shown in Table 2.1. It is obvious that the
ground motion recorded to the March 11, 2011 which is the main shock (EQ1) and first
aftershock (EQ2) have the strongest power in the period range of 0.7-1.2 s (Figure 2.3). Later
on, it will be shown that the ranges roughly coincide with fundamental natural frequencies
of the building, especially its torsional mode. Also, it should be noted that the ground
motions recorded are still quite below the level 2 earthquake design response spectra for soil
condition s2 and s3 (specified for building design in Tokyo area), shown in Figure 2.3. By
comparing the spectra of recorded ground motions to the design spectra, one can roughly
estimate the performance of the structure after experiencing those seismic events by visually
looking at how large the identified spectra are compared to the design one. Based on the
records, large accelerations occur at the top of both main and annex buildings with respect
to the base acceleration which are not normal for base-isolated buildings – the opposite
should be the case, and it happened for both main shock and after events in the weak axis of
each building (Y direction for annex building, and X direction for main building). To identify
the reasons on what happened, a Multiple Input Multiple Output (MIMO) time-invariant
technique system identification was utilized in this study. The method requires a set of input-
output data which are the responses from triaxial accelerometers located at the basement
such as AA0S, AAOW, and so on as inputs and accelerations on the upper stories such as
AA1S, AM1W, and so on as outputs. As expected by the name of the SI method, modal
parameters are assumed to be constant during a certain time window that these input output
data are analyzed. For the non-linear responses of the building subjected to a large ground
motion, the inherent assumption of constant parameters may not be satisfied throughout the
response time history, so a piece-wise linear analysis was conducted using a certain length
of moving time window – during the divided period, the modal parameters were assumed to
be constant.
Figure 2.3 Response spectra of ground motions recorded during largest five earthquakes
between 2010 and 2012 for 5% damping: (a) EQ1 and EQ2, (b) EQ3-EQ5. Note that
DS-L2 are design spectra for Level 2 earthquake for soil condition S2 and S3.
6
After the system identification is done using the signal obtained during the main shock, it
results in three dominant modes as shown in Figure 2.4 (a). The first translation mode was
identified at the range of 0.44 to 0.6 Hz, the next mode is torsional mode identified at 0.58
to 0.68 Hz, and the second torsional mode was identified at 0.85 to 0.95 Hz. The slight
differences in frequencies of each mode indicates non-linearity of the structure. Based on
the figures, the natural frequencies of each mode reduced to the lowest values during the
largest excitation of the main shock time history which is shown elsewhere (Dionysius &
Fujino, 2014) and tend to increase slightly near the end of the excitation, but it did not
recover back to the condition prior to the excitation. In contrast, the damping ratios increased
during the largest excitation. For the first mode, the figure went up to 20% implying the full
engagement of dampers during largest excitation. It can be seen that the first mode shape has
high modal displacement between the basement and the 1st floor (isolator displacement) and
small values on the upper floors representing rigid movement of the top structure which is
the common characteristic of base-isolated building. For the first torsional mode, the annex
building is rigid but the main building, rotating around its center of mass, turns out to have
high displacement at the very top corners, and the third mode has the opposite characteristic
– the one twisting is the annex building causing large modal displacement at the top corner
away from the main building as shown in Figure 2.4.
Figure 2.4 (a) The first three identified mode shapes of the building, (b) time variation of
natural frequencies, and (c) time variation of damping ratio identified from the main shock
by system identification
7
Once the dynamic characteristic of the building has been investigated, the earthquake ground
motions are also needed to be examined as well in order to determine the reason why high
response seismic response occurred at the top of the base isolated building with respect to
the of the base ones. Revisiting the response spectra of EQ1 and EQ2 in Figure 2.3, one can
figure out that the very high energy of these ground motions is contributed from the range
of 0.8-1.4 Hz, and frequency domain analysis of the signals measured at the top of both main
and annex buildings in their own weak axis at the top of the building, i.e. AM14EX and
AA7SY, are examined as shown in Figure 2.5. It can be observed that the high acceleration
at the top of the main building during the events is mostly contributed from the 1st torsional
mode causing high modal displacement at the top corners, and in the same way, the top of
annex building was mostly contributed by the 2nd torsional mode. After that, the author
attempted to study the relationship between isolator deformation and the shaking intensity,
it is found out that only for the translational mode that when the shaking intensity is high,
the modal deformation at the base will be high portion compared to the top acceleration in
the same direction, but the tendency did not go the same way for both torsional modes. The
relationship mentioned can be seen from Figure 2.6.
Figure 2.5 Transfer functions of acceleration responses at the top of both buildings during
EQ1 and EQ2 ground motions in their own weak axis
Figure 2.6 Influence of the shaking intensity on isolator's relative displacement during the
main shock for: (a) 1st mode, (b) 2nd mode, (c) 3rd mode
8
Figure 2.7 Long-term time sequential variation of the first mode: (a) natural frequency
from September 2010 to December 2011, (b) natural frequency from January 2012 to
December 2012
The indicator used in the graphs is Relative Isolator Modal Displacement (RIMD) – it is the
ratio of the base displacement to the top one, and Sa is spectral acceleration roughly
presenting the shaking intensity. Moreover, the authors paid their attention in the frequency
change identified during many records over the long-term period; during each ground
motion, the system identification was applied to each segment of moving window having
the length equal to 50 s for the identification of dynamic characteristic, i.e. natural frequency
shown in Figure 2.7. During the EQ1, it was the time that the frequency of the first mode
reduced to the lowest value, but due to the fact that the values rose up over the period of
time. The authors concluded that it was because of the behavior of the property of rubber
bearing in base-isolators which is a temporary and recoverable reduction in its modulus after
experiencing cyclic straining time history. In this case, it can be inferred from the results that
there were non-potential damage to both structural and non-structural elements but to the
rubber bearing during these earthquakes over the monitoring time period unless it would
leave a permanent reduction in its natural frequencies as can be shown in the next paper to
be presented right after this paragraph. After visiting the paper, one can see the further use
of SHM in understanding the dynamic behavior of the building. The author used data
recorded along with data analysis to come up with reasonable explanation how comes the
acceleration responses at the top of both buildings were large compared to the base which is
not common to the expected behavior of the base-isolated building. Furthermore, one can
see a glimpse of damage detection technique used in the paper based on frequency change
in the fundamental mode, in which in this case there were no potential damage to the building
found.
2.1.1 Damage Detection Based on Frequency Shift
Considering damage detection in SHM, one of the most traditional way is to relate the change
in natural frequencies of the considered structure to potential damage since it is believed that
once damage happens to the structure, the stiffness of that structure reduces proportional to
how severe the damage is. Though, currently, there are many reliable and fast methods to
determine the natural frequencies of the structure after obtaining measured responses either
from ambient vibration or force excitation, the method alone is able to be used in real seismic
evaluation. In the study of (Salawu, 1997), the author did the literature review on damage
9
detection based on changes in frequencies. The author mentioned about the drawbacks of
this method; for example, if the damaged structural parts are not the important parts
influencing a certain vibrational mode or playing a vital role in resisting loading, a shift in a
frequency of the mode may not be enough for integrity monitoring or it can be said that
frequency shift may not be sensitive to that local damage. The author found that if the
frequency shift is less than 5%, one cannot say with confidence that there is a structural
damage occurring. And even the figure is higher than the threshold of 5%, 2 damage
locations in one scenario, to illustrate, may cause the reduction in frequency equal to 1
damage location in another scenario. This shows that the method is only categorized as level
I based on (Rytter, 1993). Moving to another paper, in order to acknowledge the structural
safety of a bridge, the load test was done on the existing prestressed concrete bridge up to
the failure point (Kato & Shimada, 1986); during the static load testing, the vibration
measurement of the bridge was also conducted in order to examine the dynamic
characteristic of the bridge, once the bridge had been made deteriorated. The bridge tested
in the paper is a prestressed concrete rigid frame having diagonal elements at both sides and
columns connected through hinges at the upper and lower ends (Figure 2.8). After the bridge
had been completed in its construction in the past five year, the bridge was planned to be
removed since the new highway interchange would be in its position. For the static load
testing scheme, there were six load steps to go, and the maximum load was set to be far
exceeding the design maximum capacity of the bridge, but the measurement of vibration was
done only four times: before testing, after the 4th cycle, the 5th cycle and the 6th cycle. Then
the load would be increasing up to the failure point. It should be noted here that after the
load step had reached to the target loading, which was right before the corresponding
vibration measurement, the load had to be removed first per the paper. In this study, the
horizontal and vertical vibrations were measured by velocity type transducers at various
positions on the bridge. Then, the data was processed for the system identification to obtain
frequencies and damping ratios of each vibration mode using the Fast Fourier Transform
method (FFT). The dynamic characteristic of the vertical modes identified are shown in
Figure 2.9. As it can be seen, once the applied static load (𝑃𝑁 ) approached the ultimate load
(𝑃𝑈 ), the fundamental natural frequency decreased rapidly representing in the graph by the
ratio of 𝐹𝑁 /𝐹𝐼 or the identified natural frequency (after the specified applied static load had
reached the target) over the initial natural frequency, i.e. before testing. It is also shown
elsewhere (Kato & Shimada, 1986) that at the case 2 in Figure 2.9 (a) of vibration
measurement, the damage did happen to the test bridge already – cracking of the concrete
was observed. The reduction in the frequency is even less than 5% at that point; based on
(Salawu, 1997), this could not even be considered as damage with confidence.
Figure 2.8 Dimension of test bridge: (a) elevation plan, (b) cross-section at the center
of main girder
10
Figure 2.9 (a) Results of vibration test, (b) change of natural frequency of the 1st
vibration mode under failure process
This is described by the fact that even though the cracking of concrete occurred after the
high cycle load step, the steel wires were still under the elastic limit, so when the load was
removed from the bridge and the vibration measurement was conducted, the identified
dynamic characteristics were then obtained in the state that the crack of the concrete may
have been closed by the prestressing force of the steel wire.
In the study of (Antonella, Carmelo, & Marco, 2015), the tower named Gabbia Tower which
is the historic tower with 54 meter in its height was surveyed, and was continuously
monitored for 8 more months after the tower experienced the ground motion of May 2012
in Italy. The tower was built by solid brick masonry, having masonry walls as load bearing
walls. The instrumentation, for the sake of continuous monitoring, has 3 piezoelectric
accelerometers (sampled at 200 Hz) and a temperature sensor measuring outdoor
temperature and storing data in every hour, and the data would be stored in the on-site PC.
This was done in order to observe the dynamic characteristics of the tower subjected to far-
Figure 2.10 (a) Layout of accelerometers at the top of the tower, (b) typical acceleration
record at the top of the tower due to ambient vibration, (c) seismic response recorded on
21/06/2013
11
field earthquakes expected to shake the tower during this continuous monitoring, and to
study the relationship between temperature and natural frequencies of the tower itself, and
to come up with the future strengthening plan for the tower in case of detecting abnormal
change. Based on the records between 8-month period, the maximum acceleration response
from ambient vibration at the top of the tower where 3 accelerometers were mounted as
shown in Figure 2.10 (a) is about 0.4 gals, and the typical acceleration response due to
ambient excitation measured at the top of the tower is shown in Figure 2.10 (b). In the paper,
it was found that the variation of the frequency of the fundamental mode B1 of the tower is
almost up to 18% difference, when considering the whole range of temperature variation
from -5 to 45 ℃ as can be seen in Figure 2.11. B1, B2, B3 and T1 mean the 1st bending, the
2nd bending, the 3rd bending and the 1st torsional mode, respectively. It can be noticed that
the higher the mode, the lower the frequency variation with temperature in percent were
found. Nevertheless, they were the variation of almost the whole year temperature; Figure
2.12 shows how frequencies of the mode B1 and B2 vary in a short period. It can be seen
that the variation is up to 5% per day for mode B1 (fundamental and 1st bending mode). The
frequencies increase as the temperatures rise, and it was concluded to be because of
temporary closure of superficial crack in masonry due to expansion of material. On
21/06/2013, there was a far-field ground motion exciting the tower causing the response at
the top of the building to be up to 20 gals (Figure 2.10 (c)), and one can observe from the
graph in Figure 2.12 that there was a sharp shift in natural frequencies of mode B1 and B2.
But the question is can this be the indicator pointing to potential damage to the tower? Before
this, it has been shown that the shift in frequency may be because of the temperature effect,
yet the shift was in a gradually fluctuating trend being opposite to the sharp change in its
trend as manifested in Figure 2.12.
Figure 2.11 Variation of identified natural frequencies (global modes) versus temperature
between 17/12/2012 and 20/06/2013
12
Also, the authors of the paper had the answer to the question by constructing the plot of the
relationship between identified frequencies and temperatures about 3 weeks before and after
the seismic event to manifest that the abrupt change occurring due to the seismic event was
not in consequence of the environmental effect but the deterioration of the tower itself; In
Figure 2.13, one can see that the correlations between the parameters change in both modes
after the event in 21/06/2013. In conclusion, the method based on frequency shift may not
be able to practically and easily be utilized to identify the severity and the locations of
potential damage happening in a structure subjected to seismic event or any strong source of
excitation, but through the observation of dynamic characteristic and environmental
parameters, i.e. temperature, one can distinguish between the environmental effect and
potential damage caused by the event.
Figure 2.12 Variation of natural frequency between 01/06/2013 and 10/07/2013:

(a) mode B1, (b) mode B2
Figure 2.13 Change in frequency-temperature correlation induced by the seismic event of

21/06/2013: (a) mode B1, (b) mode B2
2.1.2 Damage Detection Based on Traditional Approaches
In the past many decades, the Vibrational Based Damage Identification (VBDI) field
interested many researchers to work on the field, but mostly their work contributed to the
structural health monitoring of small structures such beams and plate-like structures. The
methods they used were based on examining changes in vibration responses since they are
related to the physical properties of the structure. Doebling, Charles, & Prime (1996) gives
13
a good summary of many traditional techniques used in VBDI. However, these methods can
hardly be utilized in the SHM of tall buildings. In the study of Frigui, et al. (2018), they used
many traditional methods together in order to develop a novel algorithm for detecting
potential damage and localization of damage. They tested the algorithm by applying it to a
numerical model representing a middle-rise 18-story reinforced concrete building named
Ophite tower (Figure 2.14). Before the algorithm being applied, the model was updated using
its identified dynamic characteristics. The algorithm has 2 main levels: the damage
realization and damage localization. First, the change in frequencies of each mode between
reference and identified states will be examined; if the change is higher than 5%, the damage
is assumed to happen somewhere in the building, but if it does not reach the threshold value,
the Modal Assurance Criteria (MAC) will be applied in order to check the correlation
between reference and identified mode shapes, and if the value become less than 0.8 (0
means not correlated at all and 1 means perfectly correlated) in even a single mode of
vibration, the damage is considered detected. Secondly, three techniques using reference and
identified mode shapes as inputs will be utilized independently in order to locate damage
locations. Regarding the limitation of this approach, it requires the acceleration responses in
every floor in order to construct mode shapes and used as the inputs in localization level; in
the real building, the sensors location must be mounted in every single floor as shown in
Figure 2.14 (a). And as it is manifested in the aforementioned papers, the threshold of 5% in
frequency change can be greatly affected by the environmental effect. Also, it has been
shown earlier that the damage occurring may not cause frequency change up to the threshold.
In the paper, L’Aquila earthquake ground motion scaled to have the PGA up to 0.6 g was
used to excite the numerical model; the results of potential damage can be schematically
shown by the software in Figure 2.14 (b). According to the results from the software, the
damage was concentrated in the 1st and 8th floors of the numerical model.
Figure 2.14 Damaged building at the 8th floor by reduction of Young's modulus and 18
measuring nodes are positioned equidistantly in the middle of each floor, (b) Tensile
damage of the Ophite tower numerical model during L' Aquila earthquake
14
Figure 2.15 Damage index due to L'Aquila earthquake: (a) Mode shape curvature method,
(b) flexibility method
Utilizing damage localization methods, the results in common could detect the damage in
the first floor as shown in Figure 2.15 (a) which was from Mode shape curvature method
despite the fact that the results implied the damage occurred in the 2nd floor, and the
localization of damage concentrated in 8th floor was identified by looking at high number of
damage index in flexibility method around node 8 to 9 (Figure 2.15 (b)). For the formulation
of damage index in each method is documented elsewhere (Frigui, et al., 2018). This can be
concluded that the algorithm developed in the paper requires high potential damage to
happen in the structure to be able to easily realize and localize any possible damage, and it
requires many responses (acceleration records in every single floor) during that high
amplitude of excitation for the identification of damaged state dynamic characteristics which
is the limitation in the scope of this research. In spite of ignoring these inherent limitations,
the main drawback of the algorithm is that it can only be categorized as level II VBDI
(Rytter, 1993) meaning there would be no clue what actually happened and how severe the
damage was in the floors having damage concentrating.
2.1.3 Damage Detection Based on Strain Monitoring
One of the simplest and straight forward way to monitor the structural health of a building
or even attempting to search for potential damage is to monitor strains of critical parts of the
building. Any structure is composed of many structural elements having many cross-sections
along its length, and in a cross-section, materials are the vital part forming the cross-section.
Given that the structure will lose its stiffness and have changes in its dynamic properties
once the material level in each cross-section has reached the inelastic limit; thus, observing
strain of materials in critical parts of the structure is the very direct way to detect damage in
the structure. The Shanghai Tower was under construction during the time a paper
documented about its SHM (Jia-Zhan, et al., 2013) and is 632 m supertall building, having
core wall inner tube, an outer mega-frame, and six levels of outriggers linking the inner tube
and the frame (Figure 2.16). Its SHM consists of more than 40 sensors designed to be used
in both construction and service state. The arrangement of the sensors on the building is
displayed in Figure 2.16. Sensors are categorized into 3 main groups: measuring loadings
15
Figure 2.16 Sensor layout of the monitoring system
16
such as wind pressure, structural temperature, and earthquake; structural response such as
settlement, inclination, displacement, strain and acceleration; and environmental effect such
as ambient temperature. There are some critical floors that vibrating wire strain gauges were
placed (before the pouring of concrete) and used to monitor the strain of both concrete and
embedded steel. The stress of materials had been calculated from the net strain meaning that
strain induced by shrinkage and creep were taken out of consideration already. Strain gauges
were attached in outrigger trusses, mega columns and core walls. And to capture this vital
behavior of the entire floor, the placement of strain gauges were such that typical elements
in different structural components, i.e. outrigger mega columns and core walls, arranged in
x, y and diagonal configuration were under monitoring (Figure 2.17). Given that the result
during the construction is very important since if there is something wrong, it is going to
affect the performance of the building during the service state; the strain was then monitored
carefully as the time passed under construction and used to compare with the results from
the Finite Element (FE) model of each structural state as well. The result was that they were
quite different to each other because of some reasons come up by the author such as stress
induced by temperature not being considered in the FE model. The very similar strain
monitoring under construction can also be found in other papers (Yong, Yi-qing, Peng, Wei-
yang, & Jan-ming, 2011; James & Pan, 2008). In the study of Yong, et al., instead of
measuring vertical strain, the strain at a point was measured in three directions making a
rosette pattern in order that the principal stress of concrete could be calculated. The SHM
was done on The Guangzhou New TV Tower (610-meter tall structure, having the tube in
tube structural system including reinforced concrete core wall inside and external steel
frame). Regarding strain monitoring, the vibration wires were (Figure 2.18 (a)) coated in the
cylinder form by concrete poured not long before the day of concrete pouring in the core
walls (Figure 2.18(b)), and then they were attached to rebars to avoid movement during the
pouring time of concrete (Figure 2.18 (c)). And finally, the cable is safely protected in the
steel pipe (Figure 2.18 (d)). During the construction period, there were some extreme events
to be worried about such as the hit from Typhoon Neoguri on April 19, 2008 (the 2-minute
measured mean wind speed at the height of 375 meter is 23.3 m/s). During the time around
18.00 to 19.00, the mean wind speed was highest over the day, and that caused the highest
vertical strain variation shown from a sensor mounted at the height 121.2 meter during that
Figure 2.17 Typical sensor layout: (a) core walls, (b) mega columns and outrigger trusses
17
Figure 2.18 (a) Vibrating wire strain gauge, (b) concrete coating,
(c) a strain rosette in the inner tube, (d) protection of cables
Figure 2.19 The variation of vertical strain from a strain gauge at the
height of 121.2m on April 19, 2018
time displayed in Figure 2.19. However, it should be noted that vibrating wires cannot
measure the dynamic strains, and the sampling rate was just 1/60 Hz – one data stored in one
minute for the strain monitoring from each strain gauge. This means that if the very strong
seismic event came to excite the tower within, say, 30 s duration, the strain measurement
may not be able to record useful data at all. To conclude the SHM based on strain monitoring,
if the correct type of strain gauges is utilized and attached in the critical part of the monitored
structure before the pouring of concrete and all phenomena causing variations in strain
except loadings are taken into account, the damage detection of the structure can be
successfully done by just monitoring all strain values whether they are exceeding the elastic
limits or not, especially during the extreme events such as earthquakes. Despite knowing the
real material conditions of the structure which is not given in any other VBDI in general,
there must be a lot of strain gauges attached to the critical parts of the structure; not including
the fact that if one wants to know the structural conditions in all ends of beams, ends of
columns, very top fibers of shear walls, and so on in both concrete and reinforcement,
imagine how much more strain gauges are needed. And more importantly, the decision of
this kind of SHM must be planned and designed well before the construction phase, and
certainly not applicable to any existing building. Lastly, this type of VBDI is categorized as
level III (Rytter, 1993) since it can be used to detect, localized damage and determine how
severe the damage is, but nothing informing about the future assessment of the monitored
structure or the performance of it under any susceptible extreme events to come.
18
2.1.4 Damage Detection Based on Model Updating
In structural health monitoring, especially damage detection scheme, one of the very few
methods that can reach level IV (Rytter, 1993) is Model Updating method. Through any
system identification, the real dynamic characteristics of the structure can be used to update
the numerical model of that structure, then one can obtain the intact numerical model without
damage or can even use the model in seismic evaluation to come up with potential damage
for both location and severity of damage. The method is basically minimizing the difference
between the obtained identified frequencies and mode shapes (mostly used as parameters
utilized in the error minimizing process (Maria & Celebi, 2019)) and those of numerical
model by varying certain structural properties in the model. The way these parameters are
chosen play a vital role in Model Updating method. To illustrate, if one does not consider
soil spring to be in the numerical model and only update other structural properties, in order
to make the dynamic characteristics of the model to be as close as those of the real structure
– more flexible due to soil structure interaction, the process will artificially soften the
structure as a result. The traditional way of utilizing model updating as VBDI is that after
updating the model with the identified modal properties from ambient response of the intact
structure, that updated linear model will, then, be used as baseline model so that when the
seismic event comes, the identification of the response under seismic event will be used to
update the linear model again, and the difference between both models may be inferred as
potential damage. Another way to look at VBDI through Model Updating is to use the
baseline model having updated from ambient response of the intact structure to be in non-
linear analysis and let the model deal with the damage by itself (Maria & Celebi, 2019).
Regarding parameters to be changed, since the mass of the structure is normally assumed to
be known parameter, Modulus of Elasticity and soil stiffness values are widely set as ones
to be changed. The thing is, one needs to consider about the ranges of possible values of
these parameters by either engineering judgement or information provided in existing
literatures. For the damping ratios to be used, in the literature of Maria & Celebi, the
identified damping ratios from SI of ambient responses can be used in the non-linear
analysis, i.e. non-linear performance evaluation, and let the non-linear model contributes to
hysteresis damping by itself.
In the study of Kaynardag & Soyoz (2017), the 26-story building, having core-walls and
outer columns as its structural system, consisting of three zones: basement, shopping mall
and residential zone, was instrumented with 13 accelerometers located at around the center
of each specified level (measuring x and y direction acceleration responses in six levels,
except one more located far away for torsional mode identification at the top, including
lowest level of basement, ground floor, top of the shopping mall zone, and three more levels
in the residential zone including the very top floor (Figure 2.20 (a))) in order for the authors
to gain insight about the characteristics of the structural modelling of a building and the real
building that may greatly differ because of assumptions made in FEM for the sake of seismic
performance assessment. After the FE model of the building was constructed, the system
identification method called Enhanced Frequency Domain Decomposition (EFDD) was
utilized to extract frequencies and mode shapes of the building, and the objective function is
formed in order to determine the appropriate values of uncertain parameters – in this case,
stiffness of vertical and horizontal soil spring, and modulus of elasticity.
19
Figure 2.20 (a) Instrumented 26-story building, (b) 1st and 2nd translational mode
shapes of the instrumented building
Table 2.2 Identified, updated and non-updated FEM frequencies of vibration
The authors, after decoupling linear dynamic behavior of the building, found out the number
of important modes to be used in error minimization. And as other literatures documented,
difference between identified and numerical model both frequencies and mode shapes
(represented by Modal Assurance Criteria (MAC)) of each vibration mode weighted with a
factor were used in the objective function. The factors represent how much participation
each mode has to the response contribution of the structure – in this case, modal participation
was chosen as the factors. The case where the objective function, i.e. error minimizing
function, gives the least value was chosen to represent the reasonable updated model. Despite
the fact, the authors realized that the changes in Elastic Modulus of materials only
represented the ambiguity of the real complex behavior of the super structure part, and
similarly, the change for soil stiffnesses were for that of soil structure interaction. The Figure
2.20 (b) compares the identified, non-updated and updated 1st and 2nd the mode shapes in a
translational direction, and the comparison for frequencies can be found in Table 2.2. For
the identified damping ratio, it was about 2% (the authors did not mention which mode the
values belong to). Then, 10 earthquakes were scaled based on the design spectrum of the
area and used in non-linear performance assessment for both updated and non-updated FE
models.
20
Figure 2.21 Damage states of non-updated and updated FE model under a ground motion
The results for both numerical models under an earthquake ground motion from the ten are
shown in Figure 2.21 (C = collapse, CP = collapse prevention, LS = life safety and IO =
immediate occupancy); the non-updated model turned out to experience more severe damage
in general throughout the building. Also, after that, since damping ratios to be used in the
numerical model seemed to be a big issue, the further study on the effect of damping ratio
selection was done by running the Non-Linear Response History Analysis (NLRHA) with
the input damping ratios in linear range equal to the identified value (2%) and the upper limit
specified by a code (five percent). As expected, the more damage can be found in the case
of 2% damping ratio as the input. In the literature of Maria & Celebi (2019), the author
specified what the aforementioned paper has just shown as “FEM Updating with Linear
Models for Seismic Performance Assessment” which is, in summary, to update the model
first and then assess the seismic performance of the structure by picking the expected ground
motions to excite the updated model in order to examine the consequences. Another category
benefiting from the utilization of Model Updating technique is called “Linear Models for
Damage Detection.” Some studies fall into the category (Dora, Mariella, Nicola, & Michele,
2012; Gentile & Saisi, 2007), but the situation is different than what was in the paper before
– the structure has already been realized to be in damage state due to some reasons, and the
linear model is used and updated based on the ambient vibration of the damage state. Such
papers are mostly related to historical building experiencing degradation of their own
materials. The idea behind these papers is that after specifying possible damage locations by
visual inspections or (and) any Non-Destructive Test (NDT) in order to roughly evaluate
which locations should have different material properties, any Model Updating technique
will be used to iteratively update Elastic Modulus of each location to the point where the
error between identified and numerical model dynamic characteristics are minimized. The
problem regarding this way of utilizing model updating technique is that the reliability of
the results greatly depends on: first, the lower and upper limits of updating parameters;
21
second, the judgement on either which location should have the same properties or in how
many regions, different properties should be assigned. And the thing is, the idea is not based
on reality – only a “magic” number will be used to represent the damage state of the entire
area which is impossible. Nevertheless, the results from these papers turned out that the
updated models were in some extent having good agreement of what the damage conditions,
obtaining by either visual inspection or (and) any NDT in the real structures, were. A 74-
meter tall bell-tower in the study of Gentile & Saisi (2007) was observed to have major
cracks on eastern and western load-bearing walls. The identification methods used in the
study were Peak Picking (PP) and Frequency domain decomposition (FDD) methods. Since
the tower connected to another nearby building at one side of the tower, the horizontal spring
was modeled to account for the contribution from the nearby structure, and the spring
stiffness was assigned to be one of updating parameters. Other ones were basically Moduli
of Elasticity of masonry at different locations; since the crack was observed to occur up to
the height about 23 meters, the authors assumed that at the height higher than 23 meters, the
masonry had the same constant Modulus of Elasticity. The rest were the Elastic Moduli of
different parts of the lower tower which were divided in to five regions based on the
engineering judgement of the authors (Figure 2.22). After applying two different methods of
Model updating, the results of updated parameters are shown in Table 2.3 (DR and IE are
names of those methods not emphasized in this research). It can be seen that in the
parameters representing the stiffness of the western and eastern sides shows reasonably and
locally lower values as expected since in the real tower, the crack was in those sides. And
the only thing that this updated model could do in the future was to be used as a baseline
model; in case that SHM of this tower was in need to be done, the very same technique could
be done and the results could be compared with those of this model. However, since it was
just a linear model, the seismic performance assessment cannot be done for the tower, yet
by having a calibrated model updated by ambient responses of the intact structure, this
damage detection technique based on linear model updating can be used to determine
potential damage in the future for the same structure once the response of the damaged
structure is used to update the baseline model. By comparing the difference between two
models, one can evaluate potential damage in the structure. To conclude, Model Updating
method is a very good way to make any numerical model match the dynamic behavior of
that of its corresponding structure, yet the method itself should not be directly used as VBDI
technique due to the uncertainties in selecting upgrading parameters. In the proposed seismic
evaluation scheme of this research, Model Updating will play a very important role in
applying the seismic evaluation technique from this research to practical implementation.
Table 2.3 Updating parameters
22
Figure 2.22 Regions selected to have different Moduli of Elasticity at the height
lower than 23 meters
Figure 2.23 Sensor layout on Roppongi Tower
2.1.5 Damage Detection Based on Modal Decomposition
Based on the literature of Tsuchihashi & Yasuda (2017), there are 12 buildings, under the
developer group named Mori Building, having rapid diagnosis systems using accelerometers
attached on some floors of the buildings. In the paper, the illustration of the system is
presented by using Roppongi Hills Mori Tower having 54 floors above ground and 6 floors
underground. The lateral load resisting system is moment resisting frame. Semi-active oil
dampers and buckling restrained braces are used to increase energy dissipating mechanism
to this steel building. The sensor layout on the building is shown in Figure 2.23. In this study,
one of the interesting works is the modal synthesis of responses due to Great East Japan
earthquake in 2011. Since acceleration records on the tower were measured during the time,
23
Figure 2.24 Acceleration and displacement at the top of the building
Figure 2.25 Prediction of structural damage from drift angle and warnings for
furniture turnover and ceiling drop down from floor accelerations
identification of dynamic properties such as frequencies and damping ratios were done, and
the authors found that damping ratio in the first mode was up to 4.7% indicating the
effectiveness of dampers. Based on identified properties, modal synthesis was done and the
generated acceleration at the top of the building from the combination of many sufficient
modes showed good agreement to the measured one as presented in Figure 2.24. This shows
that the complex behavior of the building can be described by the summation of each mode
contribution. In the figure, displacement at the same location by modal approach is also
plotted against one from double integration method on measured acceleration. Another work
is about effective real-time rapid diagnosis system, based only on acceleration responses
shown in Figure 2.23, story drift in the unit of radian and acceleration in the unit of gal (
𝑐𝑚/𝑠^2) at each floor are able to be predicted based on the least square approach which will
be soon discussed. According to predetermined pushover analysis, predetermined story drifts
are used to roughly refer to overall structural damage shown in Figure 2.25, and based on
predetermined floor accelerations, the warnings about possible turning over of furniture and
dropping down of ceiling shown in Figure 2.25 can be consulted with. These figures were
24
actually captured from a mobile application for residents in order for them to immediately
obtain rapid diagnosis regarding their safety during or after an earthquake ground motion.
Based on the equation (2.82), the measured acceleration should be able to be decomposed
into each mode contribution. The authors somehow obtain 3 reduced mode shapes (𝝓1𝑅 , 𝝓2𝑅
and 𝝓3𝑅 ) before hands (possibly from either identified mode shapes or numerical model)
and use them in the least square error approach to find out modal coordinates (𝑞𝑛 ) of each
mode. This is under the assumption that contributions from 3 modes are sufficient to describe
the complex behavior. The residual 𝑅 in the literature is mathematically shown as follows:
3 2 3 2 3 2
𝑅 = (𝑥1 − ∑ 𝑞𝑛 ∅1𝑛 ) + (𝑥20 − ∑ 𝑞𝑛 ∅20𝑛 ) + (𝑥40 − ∑ 𝑞𝑛 ∅40𝑛 )

𝑛=1 𝑛=1 𝑛=1
3 2 (2.1)
+ (𝑥54 − ∑ 𝑞𝑛 ∅54𝑛 )
𝑛=1
where 𝑥i and ∅i𝑛 are the acceleration record and the n-th mode shape value at the i-th floor.
Then equations of setting the partial derivatives of residual with respect to each mode modal
coordinate equal to zero in order to determine the combination of modal coordinate making
the least error can be formulated, and then by simultaneously solving those equations
altogether the modal coordinates from three modes at each time instant can be calculated.
Based on equation (2.81) and double integration approach, acceleration and story drift at
each floor can be found. And by repeating the process, continuous rapid diagnosis of
building conditions can be achieved. This approach is considered as level III based on
(Rytter, 1993).
2.2 Operational Modal Analysis (OMA)
This section is dedicated for summarizing what should be known about Operational Modal
Analysis for this research based on the literature of Carlo & Giovanni (2014). Rising of the
FE modeling is such a great tool in structural analysis, but in order to validate the reliability
of the FE modeling: many assumptions made, discretization of structural elements and so
on, the experimental test is needed for engineers to gain the physical insight of the dynamic
characteristics of the existing structures. Experimental Modal Analysis (EMA) was the very
first technique requiring known force excitation to be the input in the method for the
identification of dynamic characteristics of the structures, but when it comes to the
application in the large civil engineering structures, it is not impossible to excite them with
known forces. Due to the difficulty mentioned, the focus should be on Operational Modal
Analysis (OMA). It can be also called ambient vibration modal identification or output-only
modal analysis. These names are from the fact that ambient excitation caused by traffic, wind
and etc. can naturally excite the structure and the method requires only dynamic response of
the structure (without the known input loading).
2.2.1 Preliminary Concepts and Fundamental Principles in OMA
Staring with some terminologies and basic relationships in OMA, a signal is basically any
physical quantity varying with time, depending on a single or more variables. A system
converts an input signal into an output signal. In reality, the problem will be related to noise
25
(any other signals mixed with the interested one) level expressed in term of signal to noise
ratio (SNR) as follows:
𝐴𝑠
𝑆𝑁𝑅 = 20𝑙𝑜𝑔 (2.2)
𝐴𝑛
where 𝐴𝑠 and 𝐴𝑛 denotes the signal and noise amplitude in the very same units, and SNR
will be in the unit of dB. And if the SNR is quite low, the signal may be hardly separated
from the noise; that is why a great strategy planning on acquisition system must be taken
into consideration carefully. When ambient vibration measurement is done, the system under
response measurement is considered to be linear time-invariant system (LTI);
Figure 2.26 Schematic representation of LTI system
to put it simply, the dynamic characteristic of the structure during the measurement period
are not changing with time, and the linear combination of inputs will give the output response
equal to the linear combination of output response due to each individual input acting to the
system. Nevertheless, it should be noted that, in the other case in which the input load is not
as small as in the case of ambient excitation, the system experiencing large amplitude
excitation can no longer be valid to be assumed as LTI. One can describe the LTI system by
its response to standard excitation – unit impulse input. The dynamic response due to the
input is called impulse response function (IRF), being in time domain, but if the excitation
is unit amplitude sinusoidal force applied at every frequency in a given range in frequency
domain, the response in the very same domain is called frequency response function (FRF).
For both IRF− ℎ(𝜏), and FRF− 𝐻(𝜔), they can be theoretically realized by their own
structural properties of the structure including mass, stiffness and damping. For a single
degree of freedom system (SDOF), the LTI system can be schematically shown in Figure
2.26. f(t) and y(t) denotes input force and output response in time domain while 𝐹(𝜔) and
𝑌(𝜔) denotes those in frequency domain. The relationship between output response to
arbitrary input can be found using the following convolution integral IRF with f(t):
∞
𝑦(𝑡) = ∫ ℎ(𝜏)𝑓(𝑡 − 𝜏)𝑑𝜏. (2.3)
0
But the response in frequency domain can be easily determined by equation (2.4) which is
the reduction of equation (2.3) into just the simple multiplication expressed in equation (2.4).
𝑌(𝜔) = 𝐻(𝜔)𝐹(𝜔) (2.4)

As it is suggested from the equation, for LTI, the FRF only modify the amplitude and phase
of the input force and does not cause any frequency translation. Also, for multi-degree of
freedom system (MDOF), the equations (2.3) and (2.4) can be easily extended but with the
very same concept in the form of appropriate matrix notation.
Operation Modal Analysis (OMA) is the process to come up with the dynamic properties of
the LTI system of the structure by using only output signals of the dynamic response of the
26
structure itself. As suggested by the equation (2.4), the dynamic properties of the structure
will be confused by the dynamic properties of the input force excitations under ambient test,
and the fact that the inputs cannot be measured leads to error in identifying the LTI
properties. That is why there is a need to ensure enough amount of independent information
in the data which can be shown later on that the output power spectral density (PSD) matrix
can be expressed by the product of FRF matrix of the considered structure and the input
force PSD matrix. This is obvious that the rank of the output matrix (implying the number
of independent columns in the matrix) will not be larger than those of individual matrices of
the product. To illustrate, if the input force has its PSD matrix, having only rank one (the
ambient force pattern may only be proportional to the first fundamental mode shape only),
the output PSD matrix will not include other important dynamic characteristic of the other
modes causing erroneous estimation of dynamic characteristic of the structure through its
responses. The general rule in overcoming the problem about rank deficiency stated in the
literature (Carlo & Giovanni, 2014) is that the rank of FRF matrix can be maximized by
allowing a lot of sensors in the measurement, and the uncorrelated inputs lead to maximizing
the rank of input PSD matrix. This practically leads to the standard of the do in ambient
vibration measurement which is to spread sensors around the structure in order to capture
independent information further emphasized in section 2.2.4 Observability and also to avoid
placing sensors at nodes of interested mode of vibrations (deflected shape values of the mode
at any node are zero – or no movement contributed by that mode at the location).
2.2.2 Mathematical Background in OMA Based on This Research
2.2.2.1 Fourier Transform

Based on the very principle laid in Fourier series that any periodic function can be expressed
in terms of the summation of many sinusoidal functions including cosine and sine functions.
Any periodic function f(x), having its period equal to 2𝜋 can be represented as follows:
∞
𝑓(𝑥) = 𝑎0 + ∑(𝑎𝑛 cos 𝑛𝑥 + 𝑏𝑛 sin 𝑛𝑥) (2.5)

𝑛=1
𝑎0 , 𝑎𝑛 , 𝑏𝑛 , where n = 1, 2, 3, …, are constant numbers causing equation (2.5) to be valid are

called coefficients of the series, defined elsewhere (Erwin, 2011), but when it comes to the
general form representing a periodic function having its period equal to 2L, the formula in
equation (2.5) can be replaced by the equation (2.6).
∞
𝑛𝜋 𝑛𝜋
𝑓(𝑥) = 𝑎0 + ∑(𝑎𝑛 cos 𝑥 + 𝑏𝑛 sin 𝑥) (2.6)
𝐿 𝐿
𝑛=1
∞
𝑛𝜋
𝑓(𝑥) = 𝑎0 + ∑(𝑎𝑛 cos 𝑤𝑛 𝑥 + 𝑏𝑛 sin 𝑤𝑛 𝑥), 𝑤𝑛 =
𝐿
𝑛=1
And the following equation (2.7) are dedicated to express the coefficients 𝑎0 , 𝑎𝑛 , 𝑏𝑛 :
1 𝐿
(a) 𝑎0 = ∫ 𝑓(𝑥)𝑑𝑥
2𝐿 −𝐿
27
1 𝐿 𝑛𝜋𝑥
(b) 𝑎𝑛 = ∫ 𝑓(𝑥) cos 𝑑𝑥 n = 1,2, … (2.7)
2𝐿 −𝐿 𝐿
1 𝐿 𝑛𝜋𝑥
(c) 𝑏𝑛 = ∫ 𝑓(𝑥) sin 𝑑𝑥 n = 1,2, …
2𝐿 −𝐿 𝐿
when L approaches ∞, the coefficient 𝑎0 becomes zero and the discrete values of 𝑎𝑛 and 𝑏𝑛
which did depend on the number n from 1, 2, … in case of Fourier series turn out to be
continuous functions varying with continuous values of 𝑤𝑛 characterized in the equation
(2.6) can be expressed as follows:
1 ∞ 1 ∞
𝐴(𝑤) = ∫ 𝑓(𝜈) cos 𝑤𝜈 𝑑𝑣, 𝐵(𝑤) = ∫ 𝑓(𝜈) sin 𝑤𝑣 𝑑𝑣 (2.8)
𝜋 −∞ 𝜋 −∞
where 𝜐 is the variable used in the integration, and the periodic function can be defined in
the form of Fourier Integral as shown in the equation (2.9).
∞
𝑓(𝑥) = ∫ [𝐴(𝑤) cos 𝑤𝑥 + 𝐵(𝑤) sin 𝑤𝑥] 𝑑𝑤 (2.9)
−∞
By substituting equation (2.8) into equation (2.9) and applying the formula of cosine plus
knowing that cosine is the even function, the equation and limit of integration can be
transformed as follows:
1 ∞ ∞
𝑓(𝑥) = ∫ [∫ 𝑓(𝑣) cos 𝑤𝑥 − 𝑤𝑣 ] 𝑑𝑤 (2.10)
2𝜋 −∞ −∞
Given that sine is the odd function, the expression in equation is equal to zero.
1 ∞ ∞
∫ [∫ 𝑓(𝑣) sin 𝑤𝑥 − 𝑤𝑣 ] 𝑑𝑤 = 0 (2.11)
2𝜋 −∞ −∞
Then, the representation in equation (2.12) in place of equation (2.10) turns out to make
sense, and the expression is called complex Fourier integral and based on the Euler’s formula
in the equation (2.13) .
1 ∞ ∞
𝑓(𝑥) = ∫ ∫ 𝑓(𝑣)𝑒 𝑖𝑤(𝑥−𝑣) 𝑑𝑣 𝑑𝑤 (2.12)
2𝜋 −∞ −∞
𝑒 𝑖𝑥 = cos 𝑥 + 𝑖 sin 𝑥 (2.13)
By writing equation (2.13) as the product of exponential functions, we have
∞ ∞
1 1
𝑓(𝑥) = ∫ [ ∫ 𝑓(𝑣)𝑒 −𝑖𝑤𝑣 𝑑𝑣 ] 𝑒 𝑖𝑤𝑥 𝑑𝑤. (2.14)
√2𝜋 −∞ √2𝜋 −∞
Now, the expression inside the bracket as a function of w, denoted as F(x) is called the
Fourier transform of the function – f in the function of x in the place of v is
∞
1
𝐹(𝑤) = ∫ 𝑓(𝑥)𝑒 −𝑖𝑤𝑥 𝑑𝑥. (2.15)
√2𝜋 −∞
28
Substituting the above equation into the equation (2.14), it becomes
∞
1
𝑓(𝑥) = ∫ [𝐹(𝑤)] 𝑒 𝑖𝑤𝑥 𝑑𝑤, (2.16)
√2𝜋 −∞
and is called the inverse Fourier transform of F(x), but in reality, the signal x(t) recorded are
digitized at a certain frequency called sampling frequency − 𝑓𝑠 or in the other way, the signal
is digitized in equidistant time interval away equal to sampling interval ∆𝑡. The relationship
between these two parameters is
1
∆𝑡 = . (2.17)
𝑓𝑠
According to Shannon’s theorem, the maximum frequency ( 𝑓𝑚𝑎𝑥 ) considered in the
frequency analysis must be lower than half of the sampling frequency due to the negative
frequencies in the formulation of Discrete Fourier Transform (DFT). If the number of
samples of the digitized signal is N, the total length of the signal T can be expressed as
follows:
𝑇=𝑁∆𝑡 (2.18)
The resolution of the signal in the frequency domain is subsequently shown in
1
∆𝑓 = . (2.19)
𝑁∆𝑡
By the expressions shown, one can see that the time step interval is mostly dominated by the
considered frequency range and somehow the limitation in acquisition system. Thus, in order
to reduce the frequency resolution (smaller ∆𝑓), the measurement length can be extended as
a measure of it. The signal is recorded at the equidistant 𝑥𝑛 , which is
𝑥𝑛 = 𝑥(𝑛∆𝑡) n = 1,2, …, N-1, (2.20)

and the discrete frequency values used to compute the Fourier coefficients at each discrete
frequency is
𝑘
𝑓𝑘 = k = 1,2, …, N-1 (2.21)
𝑁∆𝑡
Then, the Fourier coefficients (defined also DFT) are as follows:
𝑁−1
𝑖2𝜋𝑘𝑛
𝑋𝑘 = ∑ 𝑥𝑛 𝑒 − 𝑁 k = 1,2, …, N-1, (2.22)
𝑛=0
and the inverse DFT is given by the equation below:

𝑁−1
1 𝑖2𝜋𝑘𝑛
𝑥𝑛 = ∑ 𝑋𝑘 𝑒 𝑁 n = 1, 2, …, N-1. (2.23)
𝑁
𝑘=0
29
Since the DFT requires N times N operations causing a lot of effort to work on the DFT, the
Fast Fourier Transform (FFT) developed can reduce the operation of DFT to just N log 2 N.
The Magnitude of the sinusoid component at frequency 𝑓𝑘 is given by the equation (2.24),
|𝑋𝑘 | = √[𝑅𝑒(𝑋𝑘 )]2 + [𝐼𝑚(𝑋𝑘 )]2 (2.24)

and the following equation is the phase shift of the sinusoid of the frequency 𝑓𝑘 .
𝐼𝑚(𝑋𝑘 )
𝜃𝑘 = arctan ( ) (2.25)
𝑅𝑒(𝑋𝑘 )
2.2.2.2 Singular Value Decomposition (SVD)
The rank of matrix A, denoted as r([A]), is the number of independent row(s) or column(s).
If a row is independent, there can be no way that the linear combination of other rows will
be equal to the vector of that row. To understand how each individual influence the entire
matrix A, there are various way of decomposing the matrix in order to observe the
characteristic of the decomposed matrix. The eigenvalue decomposition (EVD) represent the
decomposition of the invertible square matrix A as a product of three matrices:
[𝐀] = [𝐗][𝚲][𝐗]−𝟏 (2.26)

Where the columns of X are the eigenvectors of matrix A, and [𝚲] is the matrix containing
eigen values of the matrix A, at the diagonal elements of itself. While the EVD is the
decomposition of an invertible square matrix A, the Singular Value Decomposition (SVD)
can be stated as the extended version of EVD to rectangular matrices. When the matrix A is
real-valued and has the dimensions of L time M, and L≥M and r([A])≤M, the matrix A can
be decomposed as follows:
[𝐀] = [𝐔][𝚺][𝐕]𝐓 . (2.27)

where the matrix Σ is the diagonal element containing singular values, arranged in the
descending order, the number of singular values are equal to the rank of the matrix A, having
the same dimension of the matrix A, and the rest of the diagonal elements store zero values.
The matrices [U] and [V], being orthogonal matrices, have the dimensions of L time L and
M time M. In the case that A is complex-valued, the expression in the equation (2.27) can
be placed with the one having Hermitian operator instead of Transpose operator on top of
the orthogonal [V].
2.2.3 Stationary Random Data and Process
2.2.3.1 Basic Concepts

The ambient vibration responses, unlike the case of free decay oscillation response with
initial conditions in which the response can be predicted by an explicit mathematical
expression, are not able to be predicted by any mathematical terms but the probabilistic
terms. A random (or stochastic) process is the collection of all possible physical realizations
of the random phenomenon (Carlo & Giovanni, 2014). One can observe time history of two
sample functions, representing the random phenomenon of a random process, to see that in
general they are not the same at all despite the fact that they are from the same source or
system. A collection of sample functions called ensemble have to be used in order to identify
30
the characteristics of a random process. The mean value at the time 𝑡̅ of the ensemble can be
found by averaging all the values from each sample function 𝑥𝑘 (𝑡) at the time instance 𝑡̅,
and for the mathematical expression, we have
𝐾
1
𝜇̅𝑥 (𝑡̅) = lim ∑ 𝑥𝑘 (𝑡̅) (2.28)
𝐾→∞𝐾
𝑘=1
where K is the total number of sample functions of the ensemble, and 𝜇̅ 𝑥 is the average value.
And in the very same way, the autocorrelation is defined as the average of the product of the
value at the 𝑡̅ and 𝑡̅ + 𝜏 (𝜏 is time lag between two samples in the same time history) from
overall ensemble average. The mathematical expression is
𝐾
1
𝑅𝑥𝑥 (𝑡̅, 𝑡̅ + 𝜏) = lim ∑ 𝑥𝑘 (𝑡̅)𝑥𝑘 (𝑡̅ + 𝜏) . (2.29)
𝐾→∞𝐾
𝑘=1
when the autocorrelation function value does not vary with 𝑡̅, at a certain time lag value, the
random process is called stationary and the properties for the random process in the equations
(2.28) and (2.29) can, then, be reduced in their forms as follows:
𝜇̅𝑥 (𝑡̅) = 𝜇̅ 𝑥 (2.30)

𝑅𝑥𝑥 (𝑡̅, 𝑡̅ + 𝜏) = 𝑅𝑥𝑥 (𝜏)
If two sample functions 𝑥𝑘 (𝑡) and 𝑦𝑘 (𝑡) are stationary, the mean values of both sample
functions independent of time t are expressed by the expectation of each signal:
∞ ∞
𝜇𝑥 = 𝐸[𝑥𝑘 (t)] = ∫ 𝑥𝑝(𝑥)𝑑𝑥, 𝜇𝑦 = 𝐸[𝑦𝑘 (t)] = ∫ 𝑦𝑝(𝑦)𝑑𝑦 (2.31)
−∞ −∞
where p(x) and p(y) are the probability density functions of 𝑥𝑘 (𝑡) and 𝑦𝑘 (𝑡). The covariance-
functions independent of t are be defined according to equation (2.31) as follows:
𝐶𝑥𝑥 (𝜏) = 𝐸[(𝑥𝑘 (𝑡) − 𝜇𝑥 )(𝑥𝑘 (𝑡 + 𝜏) − 𝜇𝑥 )

𝐶𝑦𝑦 (𝜏) = 𝐸[(𝑦𝑘 (𝑡) − 𝜇𝑦 )(𝑦𝑘 (𝑡 + 𝜏) − 𝜇𝑦 ) (2.32)
𝐶𝑥𝑦 (𝜏) = 𝐸[(𝑥𝑘 (𝑡) − 𝜇𝑥 )(𝑦𝑘 (𝑡 + 𝜏) − 𝜇𝑦 ).
If the mean values for both the k-th samples of x and y are equal to zeros, the covariance-
functions are just equal to the correlation functions since we have
𝑅𝑥𝑥 (𝜏) = 𝐸[(𝑥𝑘 (𝑡))(𝑥𝑘 (𝑡 + 𝜏))

𝑅𝑦𝑦 (𝜏) = 𝐸[(𝑦𝑘 (𝑡))(𝑦𝑘 (𝑡 + 𝜏)) (2.33)
𝑅𝑥𝑦 (𝜏) = 𝐸[(𝑥𝑘 (𝑡))(𝑦𝑘 (𝑡 + 𝜏)).
The variables 𝑅𝑥𝑥 (𝜏) and 𝑅𝑦𝑦 (𝜏) are called auto-correlation functions of 𝑥𝑘 (𝑡) and 𝑦𝑘 (𝑡)
while 𝑅𝑥𝑦 (𝜏) are called the cross correlation function between 𝑥𝑘 (𝑡) and 𝑦𝑘 (𝑡). Given that
the mean value, mean square value and variance of 𝑥𝑘 (𝑡) is shown in the equation down
below:
31
∞
𝜇𝑥 = 𝐸[𝑥𝑘 (t)] = ∫ 𝑥𝑝(𝑥)𝑑𝑥
−∞
∞
(2.34)
Ψ𝑥2 = 𝐸[𝑥𝑘2 (t)] = ∫ 𝑥 2 𝑝(𝑥)𝑑𝑥
−∞
∞
𝜎 2 = 𝐸[(𝑥𝑘 − 𝜇𝑥 )2 ] = ∫ (𝑥 − 𝜇𝑥 )2 𝑝(𝑥)𝑑𝑥 = Ψ𝑥2 − 𝜇𝑥2 ,
−∞
the auto-correlation function 𝑅𝑥𝑥 (𝜏) and covariance-function𝐶𝑥𝑥 (𝜏) are maximized at 𝜏 =
0 (mathematically shown in the equation (2.35)), and 𝑅𝑥𝑥 (0) are, then equal to mean square
of the signal in equation (2.34) while 𝐶𝑥𝑥 (0) are equal to variance of the data.
|𝐶𝑥𝑥 (𝜏)| ≤ 𝐶𝑥𝑥 (0) (2.35)

|𝑅𝑥𝑥 (𝜏)| ≤ 𝑅𝑥𝑥 (0)
When the mean values and covariance functions of the random processes can be identified
from only two sample records by means of time averages instead of calculating the ensemble
averages, the two stationary process characterizing the random processes of both records are
called ergodic. It simply means that to compute the statistical properties of stationary random
processes, there is no need to collect a lot of data, but only a pair of sample records despite
the fact that theoretically, the signals need to be infinitely long. The time average mean
values of records can be expressed in the following equation.
1 𝑇
𝜇𝑥 (𝑘) = lim ∫ 𝑥𝑘 (𝑡)𝑑𝑡 = 𝜇𝑥 (2.36)
𝑇→∞𝑇 0
1 𝑇 (2.37)
𝜇𝑦 (𝑘) = lim ∫ 𝑦𝑘 (𝑡)𝑑𝑡 = 𝜇𝑦
𝑇→∞𝑇 0
It should be noted that the statistical properties, i.e. mean value, of the ergodic process of the
entire K samples can only be identified from any k-th sample. And as ergodic process
suggests that it is stationary, the parameter does not vary with time. In the same way, the
auto-correlation functions by mean of time average can be as follows:
1 𝑇
𝑅𝑥𝑥 (𝜏) = lim ∫ 𝑥(𝑡)𝑥(𝑡 + 𝜏)𝑑𝑡
𝑇→∞𝑇 0
1 𝑇
𝑅𝑦𝑦 (𝜏) = lim ∫ 𝑦(𝑡)𝑦(𝑡 + 𝜏)𝑑𝑡 (2.38)
𝑇→∞𝑇 0
1 𝑇
𝑅𝑥𝑦 (𝜏) = lim ∫ 𝑥(𝑡)𝑦(𝑡 + 𝜏)𝑑𝑡
𝑇→∞𝑇 0
In the practical application, the random processes are practically assumed to be ergodic due
to its simplicity of collecting only a sample function instead of the entire collection of
ensemble sample functions. Here are some criteria stated in the literature (Carlo & Giovanni,
2014) to practically assume that any random process is ergodic. the time averages such as
mean values (2.36) and auto correlation function (2.38) over a short period of time should
be roughly the same for those in the other short intervals, and the numbers should also be
the same for all sample functions.
32
2.2.3.2 Spectral Density Functions
Starting with the two sample functions 𝑥𝑘 (𝑡) and 𝑦𝑘 (𝑡), of finite length duration T,
characterized as stationary random process, their Fourier transforms which may be a bit
different in their mathematical forms from (2.15) because of the finite duration T are:
𝑇
𝑋𝑘 (𝑓, 𝑇) = ∫ 𝑥𝑘 (𝑡)𝑒 −𝑖2𝜋𝑓𝑡 𝑑𝑡 (2.39)
0
𝑇
𝑌𝑘 (𝑓, 𝑇) = ∫ 𝑦𝑘 (𝑡)𝑒 −𝑖2𝜋𝑓𝑡 𝑑𝑡 .
0
The one-sided auto- and cross-spectral density functions, which are normally used in the
practical application instead of two-sided ones, where the range of frequency varies from 0
to ∞. Using the derived equations (2.39), they can be expressed in the following equations:
1
𝐺𝑥𝑥 (𝑓) = 2 lim 𝐸[|𝑋𝑘 (𝑓, 𝑇)|2 ] 0 < 𝑓 < +∞
𝑇→∞𝑇
1
𝐺𝑦𝑦 (𝑓) = 2 lim 𝐸[|𝑌𝑘 (𝑓, 𝑇)|2 ] 0 < 𝑓 < +∞ (2.40)
𝑇→∞𝑇
1
𝐺𝑥𝑦 (𝑓) = 2 lim 𝐸[𝑋𝑘∗ (𝑓, 𝑇)𝑌𝑘 (𝑓, 𝑇)] 0 < 𝑓 < +∞
𝑇→∞𝑇
where * denotes complex conjugate. Though through the Fast Fourier Transform of samples
functions, one can determine their two-sided spectral density functions (either auto- or cross-
), it should be noted they can also be derived from the auto- and cross- correlation functions
based on the Winer-Khinchin relations shown in the equations (2.41).
∞ ∞
−𝑖2𝜋𝑓𝜏
𝐺𝑥𝑥 (𝑓) = 2 ∫ 𝑅𝑥𝑥 (𝜏)𝑒 𝑑𝜏 = 4 ∫ 𝑅𝑥𝑥 (𝜏) cos 2𝜋𝑓𝜏 𝑑𝜏
−∞ 0
∞ ∞
−𝑖2𝜋𝑓𝜏
𝐺𝑦𝑦 (𝑓) = 2 ∫ 𝑅𝑦𝑦 (𝜏)𝑒 𝑑𝜏 = 4 ∫ 𝑅𝑦𝑦 (𝜏) cos 2𝜋𝑓𝜏 𝑑𝜏 (2.41)
−∞ 0
∞
𝐺𝑥𝑦 (𝑓) = 2 ∫ 𝑅𝑥𝑦 (𝜏)𝑒 −𝑖2𝜋𝑓𝜏 𝑑𝜏
−∞
These equations manifest that the spectral density functions and correlation functions are
Fourier Transform pairs of each other. Suggested by both methods (2.40) and (2.41) on how
spectral density functions can be calculated, one can see that the auto-correlations are real-
valued while the cross-correlation ones are complex-valued. The cross-spectral density can
also be presented in their polar form:
𝐺𝑥𝑦 (𝑓) = |𝐺𝑥𝑦 (𝑓)|𝑒 −𝜃𝑥𝑦(𝑓) 0<𝑓 <∞ (2.42)
where the magnitude and phase are shown in the equation (2.43) .
2 (𝑓) + 𝑄 2 (𝑓)
|𝐺𝑥𝑦 (𝑓)| = √𝐶𝑥𝑦 (2.43)
𝑥𝑦
𝑄𝑥𝑦 (𝑓)
𝜃𝑥𝑦 = arctan
𝐶𝑥𝑦 (𝑓)
33
In some modal identification techniques such as peak-picking, one can use coherence
function which will soon be defined as the indicator whether the peak at each frequency is
characterized as what kind of vibrational mode, i.e. translational, torsional, or it is just the
peak influenced by noise., the coherence function is given by:
2
2 (𝑓)
|𝐺𝑥𝑦 (𝑓)|
𝛾𝑥𝑦 = (2.44)
𝐺𝑥𝑥 (𝑓)𝐺𝑦𝑦 (𝑓)
where:
2
0 ≤ 𝛾𝑥𝑦 ≤ 1 ∀𝑓. (2.45)
The approach, in practical application, based on the equations (2.41) is called Blackman-
Turkey procedure despite being referred to, will not be utilized in this research, but the
procedure named “Welch procedure” will used to calculate the power spectral density (PSD)
functions based on the equations (2.40). To illustrate, the long record representing a sample
function will be divided into 𝑛𝑑 contiguous segments, and each segment, having its length
equal to T = N∆𝑡 (2.18), will be used in the calculation of auto-spectral density function, and
through the averaging process over the 𝑛𝑑 sets of data, the expression is as follows:
𝑛𝑑
2
𝐺̂𝑥𝑥 (𝑓) = ∑|𝑋𝑖 (𝑓)|2 . (2.46)
𝑛𝑑 𝑁∆𝑡
𝑖= 1
N as being mentioned earlier is called block size determines how much solution of the
frequency response will be; the number of divided segments (𝑛𝑑 ) will deal with the random
error to be presented later. Since it is obvious right now that the length of the record is finite,
a strategy is needed to be introduced in order to deal with the occurring error; which is to
look at the finite record as the unlimited record 𝑣(𝑡) multiplied with a type of window 𝑢(𝑡),
say, rectangular window as an example mathematically manifested in the equation (2.47).
= 1 𝑤ℎ𝑒𝑛 0 ≤ 𝑡 ≤ 𝑇
𝑥(𝑡) = 𝑢(𝑡)𝑣(𝑡) 𝑢(𝑡) (2.47)
= 0 elsewhere
Then, to obtain the Fourier Transform of the signal x(t), the convolution (2.3) has to be used
with Fourier Transform of u(t) and v(t). The Fourier Transform of the rectangular window
is basically sinc function (Figure 2.27), and the presence of the side lobes of the high energy
component frequency will sometimes blur the nearby energy contributed from other
frequencies, especially when that nearby energy is relatively low. The phenomenon
mentioned is known as leakage, yet the phenomenon can be dealt with by picking appropriate
type of window in which for its Fourier Transform of the window, the energy does taper
down quickly and just lump a certain frequency.
34
Figure 2.27 Discrete Fourier Transform of the rectangular window
That such window can be Hanning window, having the comparatively low energy of its
nearby side lobs compared to the main lobe. It is the full cosine tapering window, and which
can be expressed as follows:
𝜋𝑡
1 − 𝑐𝑜𝑠 2 ( ) 0 ≤𝑡 ≤𝑇
𝑈𝐻𝑎𝑛𝑛𝑖𝑛𝑔 (𝑡) = { 0
𝑇 elsewhere
(2.48)
By using this of window, the error due to leakage can be minimized, yet the determination
of damping ratio can be erroneous if the Half Power Band Width method is utilized. This
happens when the Hanning window will lead to wider bandwidth. One can deal with the
effect by increasing the length of each segment up to the point where the results will be
having the same bandwidth despite different segment length. If the data is limited,
lengthening the segment length may be done by partially overlapping each contiguous
segment together.
2.2.3.3 Errors in Spectral Density Estimates and Remedies for the Errors in OMA
Through the understanding of where errors come from, one can determine how much length
of each record should be. After the experiments have been done over the total K experiments,
the estimate value 𝑥̂ is said to not be erroneous when it is equal to interested quantity x.
There are two kinds of errors: bias error and random error. The bias error occurs in the same
magnitude and direction during many repeated measurements under the very same condition.
The bias error is given by:
𝑏[𝑥̂] = 𝐸[𝑥̂] − 𝑥 (2.49)

Unlike the systematic error, i.e. bias error, the random error turns out to not be the same in
its magnitude and direction, when repeatedly measured under the same condition. The
variance of the estimate or random error is given by:
𝑉𝑎𝑟[𝑥̂] = 𝐸[(𝑥̂ − 𝐸[𝑥̂])2 ] (2.50)

It can be illustrated by the picture of an archer having his (or her) arrows missing the bullseye
despite by shot under the same wind condition. All the arrows may be left on the very top
left corner with respect to the bullseye suggesting the mean of the error was towards the
35
same direction, and this may be because of the bow itself causing all the arrows to be all the
time systematically missing the point. While the mean and the bias error can be predicted,
the variance or random error may be due to the archer himself (or herself) causing some
random errors as a result. As suggested by (2.34), the error in terms of mean square error
can quantitatively measure the intense of both bias and random errors as a whole, presented
in the equation (2.51).
𝑚𝑠𝑒[𝑥̂] = 𝐸[(𝑥̂ − 𝑥)2 ] = 𝑉𝑎𝑟[𝑥̂] + 𝑏[𝑥̂] (2.51)

A way to deal with the random error according to the literature (Carlo & Giovanni, 2014)is
to increase the number of segments 𝑛𝑑 for the average (2.46). The idea is that the error due
to randomness can be cleared out once a lot of data are used in the averaging process. For
the estimate for the number of segments, we have:
1
𝑛𝑑 = . (2.52)
𝑉𝑎𝑟[𝑥̂]
In the literature, it stated a standard to limit the calculated random error 𝑉𝑎𝑟[𝑥̂] which should
be about 1%, meaning that the number of segments have to be up to 100 segments. Anyway,
if the data is limited, overlapping technique can be implemented as mentioned earlier.
Regarding bias error, lengthening the segment length can reduce the average error and bias
error, and limiting to bias error to 2% can be met by following the equation (2.53).
𝑛𝑑
𝑇= 𝑇 (2.53)
𝜋𝜉1 1
where T is the length of one segment, and 𝑇1 and 𝜉1 are the fundamental natural period and
damping ratio, respectively. And the relationship between the total record length 𝑇𝑟 and the
segment length can be related as follows:
𝑇𝑟 = 𝑛𝑑 𝑇 (2.54)
2.2.4 Observability
The identifiability of modal properties of the structure does depend on both the attachment
method and the sensor layout. How sensors are attached on the structure matters because it
may accidentally limit the useful range of interested frequencies. To illustrate, a reduction
in the stiffness of the connection may limit the upper limit of the frequency due to the fact
that the connection behaves as if it was a spring leading to a spring mass system itself,
filtering high frequency range out. Despite this importance, the way sensors should be
attached will not be emphasized in this research and is responsible for those are working on
the on-site measurement conditions. In this section, the observability or the ability to identify
different modes; or to separate closely spaced modes will be discussed. Given that location
of sensors affects the identification of vibrational modes of the structure, one may need to
move sensors to avoid structural nodes for important modes unless the data from those
sensors will lose the information regarding those modes. As being introduced in section 0,
once sensors are placed at the node of a mode, the rank of FRF matrix from the response
recorded will be limited. The choice of placement of sensor locations is depending on some
factors such as: availability of sensors, needed information about mode shapes and the
reasons behind modal identification. From these factors stated, there will be no fixed sensor
layout from one project to another project. Anyway, the appropriate sensor layout can
36
generally be defined as being able to observe different types of modes, i.e. translational and
torsional, and maximize the rank of the FRF matrix (by spreading the sensors away from
each other to avoid different types of information). Also, specifically, when the information
about the mode shapes of the real structure are known (or predicted), effective sensor
locations can be achieved by placing sensors on the well-presented locations of vibrational
modes of interest – these locations show high modal displacements of interested modes.
Under the assumption of rigid floor diaphragms, a sensor layout suggested in the literature
(Carlo & Giovanni, 2014) can be shown in Figure 2.28 – the sensors are placed away from
each other, and at each floor, sensors are installed in the opposite corner and expected to
measure the response in two directions. In the case where, the available sensors are not
Figure 2.28 Typical sensor layout for building-like structures
Figure 2.29 Aliasing in digitized wave form
37
enough for the identifiability of dynamic properties of the structure, some sensors will be
assigned as reference sensors (having fixed location in all experiments) and roving sensors
(moved to different locations in the different experiments). However, in this research the
availability of sensors will not be emphasized.
2.2.5 Sampling Frequency and Filtering
The frequency limit called Nyquist frequency is defined as half of the sampling frequency,
and it is the upper limit of observable frequency range. This is why sampling frequency 𝑓𝑠 is
very important in the dynamic test setting, but there is also another cost of choosing too low
sampling frequency. To illustrate, the waveform in Figure 2.29 shows that the true signal in
analog form is sampled in such a low sampling frequency causing the constructed digitized
signal waveform to be misunderstood; this effect is called Aliasing. The reason, why it
matters, is that if there are high energy components in the frequency range higher than the
Nyquist limit, having digitized, they will be added up with the genuine lower frequency
components of the true signal causing error in the frequency spectrum. It should be noted
that aliased components cannot be separated from the true components once the analog
signal has digitized, so the only way to deal with this is to remove the higher frequency
components (higher than the Nyquist limit) by applying analog low pass filter, especially the
one with sharp cut-off frequency. This is normally referred as anti-aliasing filter. And
something to be concerned about is the transition band existing inherently in the filter; one
has to set such that the end of transition band will be right before the Nyquist limit unless
the frequency components in the transition band will eventually cause aliased signal.
Moreover, since the transition band will distort the information in the band, the information
should be discarded. Regarding digital filter, if the digitized signal still includes the
unwanted frequency components, various kinds of digital filter such as band-pass, low-pass,
high-pass and band-stop filters can be applied according to the purpose of the signal
modification. Specifically, low pass filter cuts the frequency higher than the cut off
frequency; in opposite, the high pass filter allows the high frequency components to pass;
band stop one excludes those in the frequency range while the band pass one only let those
in the band to pass. It should be noted that the ideal filter will have the sharp cut off frequency
while the real ones need to have the attenuation zone.
2.2.6 Output-only Modal Identification Technique
Operational Modal Analysis is based on some inherent assumptions: Linearity – the

combination of inputs will cause the response equal to the combination of responses due to
individual input; Stationarity – the dynamic characteristics of the structure are independent
of time; Observability – the effectiveness of the identified modal properties are greatly based
on the set up of sensor layout. Ideally, the inputs (force excitation) is white noise excitation
in which its input spectrums are constant; this means the all vibrational modes are equally
excited, and the output signals contains all useful information in them. However, the case is
that the input excitation has its own power spectrum which is in general not constant and
may be either time variant and time invariant; thus, the resulted spectrum due to the
excitation will be weighted by that of the input. Luckily, under the assumption that the force
excitation is broadband exciting important modes, the discrimination of structural system
properties can be done since they have narrowband properties and time-invariant.
38
2.2.6.1 Peak Picking Method
Peak Picking (PP) method is many times referred as Basic Frequency Domain (BFD) method
due to its simplicity based on the computation of auto-power spectral density functions (will
be after this referred to as APSD) and cross-power spectral density functions (will be also
after this referred to as CPSD). PP method is the SDOF method and based on the assumption
that around a resonance frequency (a natural vibration frequency), only a mode will dominate
the response around there. The method is quite flexible and not limited to only one certain
way on what have to be done in this method, but at least everything will be mainly based on
picking peaks shown on either APSD or CPSD plots. In the study of Karunkritkul (2019),
the author utilized APSD of acceleration responses at the top floor of the abandon building
for the sake of SHM of the building (only the top floor acceleration responses measured) for
both corners in both x and y directions for the identification of modal frequencies and
damping ratios, and in order to come up with the floor mode shape at the top (mode shape
of the building could not be constructed due to limited floor acceleration records only at the
top), the author, then, determined the CPSD between all signals and reference signals (the
ones having the highest amplitude and containing useful information for each mode of
interest), and per each resonance frequency, the mode shape of the floor was constructed
based on the movement at the center of mass. The X dominant mode, to illustrate, is the
vibrational mode at the frequency where the movement of the floor was mostly in X direction
as shown in Table 2.4 despite the fact that the mode is considered as a couple mode since
the center of mass moves in both translational direction and rotate around its center of mass.
In detail, before obtaining the movement at the center of mass, the mode shape of both
corners (including both x and y directions) must be firstly determined. The normalized
Fourier magnitude the i-th signal to the reference signal 𝑀𝑖 is shown in the equation (2.55)
along with the corresponding phase difference 𝜃̂𝑥𝑦𝑖 (2.56) between the two signals and the
mode shape value of the i-th signal 𝜑𝑖 (2.57) (implying the normalized movement of the
point, say, if the signal is at the bottom left corner in x direction, the mode shape value will
be of that position and direction).
𝑠̂𝑥𝑦𝑖
𝑀𝑖 = 𝑖 = 1,2,3, … , 𝑁𝑠𝑖𝑔𝑛𝑎𝑙 (2.55)
𝑠̂𝑥𝑥𝑖
𝜃𝑥𝑦𝑖 = 𝜃𝑦𝑖 − 𝜃̂𝑥
̂ ̂ 𝑖 = 1,2,3, … , 𝑁𝑠𝑖𝑔𝑛𝑎𝑙 (2.56)
𝜑𝑖 = 𝑀𝑖 cos (𝜃̂𝑥𝑦𝑖 ) 𝑖 = 1,2,3, … , 𝑁𝑠𝑖𝑔𝑛𝑎𝑙 (2.57)
where 𝑠̂𝑥𝑦𝑖 is Fourier magnitude of averaged CPSD between the i-th signal and reference
signal, and 𝑠̂𝑥𝑦𝑖 is Fourier magnitude of averaged CPSD between the reference signal itself
(averaged APSD of the reference signal). As implying by (2.40) in spite of different
notations, the equation (2.55) physically represent how much the Fourier magnitude of the
i-th signal is compared to the reference signal. It should be noted that 𝜑𝑖 varies based on the
which resonance frequency is the input of 𝑠̂𝑥𝑦𝑖 , and 𝑠̂𝑥𝑦𝑖 , and if the input is not any of the
resonance frequencies, it will either represent the combination of mode shapes at nearby
resonance frequencies or not useful value, say, blurred in noise zone since no domination
from any vibrational mode. And once the values at the measurement locations were
determined, the author was able to determine ones at the center of mass as mentioned, and
the results of the floor mode shape identified from the ambient vibration measurement of the
study (Karunkritkul, 2019) can be revisited again in Table 2.4. In the other literatures
39
Table 2.4 Mode shape of the top floor identified from ambient responses
Figure 2.30 (a, b) APSD of signals in orthogonal directions, (c, d) coherence

functions of the mentioned signals and their reference signals in each direction
(Carlo & Giovanni, 2014; Celebi, Huang, Shakal, Hooper, & Klemencic, 2013), the authors
recommended and used coherence function (2.44) in order to either determine whether peaks
in the PSD plots are due to disturbances or real frequencies, and used to identify whether a
mode is torsional mode. To exemplify, if the coherence function between two signals at any
peak suspected to cause by noise, say in the same direction, yields the value near zero, one
may assume that the his (or her) inference is valid. Another useful application about the
identification of the type of natural vibration mode is demonstrated in Figure 2.30, despite
the fact the both signals are in the different direction, the mark points in their coherence
functions with the reference signals in their own directions show the value about 1, implying
40
that the signals are correlated around this frequency range despite different directions and
the resonant frequency is likely to be of the torsional mode.
Regarding the limitation of the method, the good result is based on the frequency resolution
of the PSD plot, and will be when the damping ratio of the mode is low and modes are well
separated. Rough frequency resolution will cause the lobe of the resonant bell of the resonant
frequency to be big, and this makes picking a certain frequency (that it has the peak value)
very hard. Not only frequency resolution, but also the damping ratio, based on the half power
bandwidth method which will be later discussed, causes the lobe to be bigger as well. And
most importantly, when if the modes are closely spaced there is for sure to have some
contribution from any nearby vibration mode; and instead of allowing one to obtain the
estimate of individual mode shape, the results may be just operational deflected shape
contributed from multiple modes. Also, as mentioned earlier about the disturbance of the
PSD plot, the spectrum may be blur by the effect of noise making applying the technique to
be even harder despite usefulness of coherence function.
Figure 2.31 Schematic chart of HPBW
For the determination of damping ratio based on PP technique, the method itself as
mentioned earlier can only be used to determine the resonant frequencies and the
corresponding structural mode shapes. Thus, another appropriate technique which is
commonly utilized with PP is called Half Power Bandwidth (HPBW) to be used in the
determination of damping ratios. In the literature (Kijewski & Kareem, 2002), the author
stated the limitation on how this technique should be used: the damping ratio of the
equivalent SDOF systems (vibrational modes) should be less than 0.1. The schematic chart
in Figure 2.31 shows how HPBW works. At the vicinity of the resonant peak frequency, the
estimated auto power spectral density function (APSD) (the notation used in the chart is
𝑆̂𝑥𝑥 (𝑓)) of the frequencies around the peak will normally show the bell shape as shown in
the figure. About the half of the peak at two nearby sides, the associated frequencies will be
chosen (in this case 𝑓1 and 𝑓2 ). 𝐵𝑟 denoted as the bandwidth of the bell shape at half the
41
height is calculated by the absolute difference between 𝑓1 and 𝑓2 . Then the damping ratio 𝜉𝑛
of the corresponding vibrational mode is as follows:
𝐵𝑟 = 2𝜉𝑛 𝑓𝑛 (2.58)
2.2.6.2 Frequency Domain Decomposition Method
The method called Frequency Domain Decomposition (FDD) Method was firstly developed
in the literature (Brincker, Lingmi, & Palle, 2001). The technique is fundamentally related
to the PP method mentioned earlier by instead of picking the peaks in the frequency response
spectra of either APSDs or CPSDs, picking peaks on singular value plots of the matrix later
defined and denoted as power spectral density matrix (or PSD matrix). By doing so, the
method can overcome the limitation in the recently mentioned method which is that it is able
to identify closely spaced modes due to the fact that Single Value Decomposition technique
is used to examine the rank of the PSD matrix at each frequency, especially the ones at the
interested peaks. To illustrate, in a system there may be only one mode greatly excited by
the input despite the fact that another close mode may be lightly excited causing the SDOF
Figure 2.32 Typical Singular Value plots
bell (of the FRFs) of that mode to be blurred by the SDOF bell of the greatly excited mode;
by implementing this approach, the singular value plots (especially the first and the second
one) may show something shown in the Figure 2.32. – at the frequency about 2.6 Hz, it
shows a resonant peak in the svd2 (red color) signifying that at the point there is a
comparatively small contribution to the frequency component probably from a lightly
excited mode. The method, based on the ability of the analyst, allows the analyst to be able
to identify resonant frequencies of vibrational modes and acknowledge peaks caused by
disturbance. Compared to PP method in which even though the mode shapes can be
extracted, the results would be biased in the case of close modes, FDD method, however,
has an inherent assumption of geometrically orthogonal close mode shapes due to the
principle of SVD technique which will cause the identified mode shape (especially the one
with lower singular value) to be biased since the mode shape is forced to be orthogonal to
the dominant one. Anyway, the author concluded that even if that may be the case, the
42
estimate of the results is still accurate enough. Based on the theoretical background of modal
expansion of the response shown in the equation (2.59), the formulation of the FDD method
is developed (Carlo & Giovanni, 2014).
{𝑦(𝑡)} = [Φ]{𝑝(𝑡)} (2.59)

where {𝑦(𝑡)} is the vector containing displacement responses from many degrees of
freedom, [Φ] is the modal matrix containing mode shapes of the structure in each column
and {𝑝(𝑡)} is the vector containing modal coordinates. Based on the basic knowledge on
auto-correlation functions, the auto-correlation matrix of the responses [𝑅𝑦𝑦 (𝜏)] can relate
to that of the modal coordinates [𝑅𝑝𝑝 (𝜏)] by the following equation:
[𝑅𝑦𝑦 (𝜏)] = [Φ][𝑅𝑦𝑦 (𝜏)][Φ]𝑇 . (2.60)

By transforming (2.60) into the frequency domain by Fourier Transform, it gives
[𝐺𝑦𝑦 (𝜔)] = [Φ][𝐺𝑝𝑝 (𝜏)][Φ]𝐻 (2.61)
where [𝐺𝑦𝑦 (𝜔)] and [𝐺𝑝𝑝 (𝜔)] are denoted as PSD matrix of the responses and of the modal
coordinates, respectively. For instance, the PSD matrix of the response having the dimension
of m × m can be constructed as shown in the equation as follows:
𝐺𝑦1,𝑦1 (𝑓) ⋯ 𝐺𝑦𝑚,𝑦1 (𝑓)

[𝐺𝑦𝑦 (𝜔)] = [ ⋮ ⋱ ⋮ ]. (2.62)
𝐺𝑦1,𝑦𝑚 ⋯ 𝐺𝑦𝑚,𝑦𝑚 (𝑓)
Before moving forwards, the PSD matrix of the outputs can also be represented in the form
of input-output relationships through the FRF matrix 𝐻(𝜔), we have:
[𝐺𝑦𝑦 (𝜔)] = 𝐻(𝜔)[𝐺𝑥𝑥 (𝜔)] 𝐻(𝜔)𝐻 (2.63)

Where [𝐺𝑥𝑥 (𝜔)] is the input PSD matrix which is assumed to be constant over the considered
range of frequency – this will make sure that output PSD matrix will contain all information
of the structural characteristics. Nevertheless, in practice, the lack of fulfilling this
assumption will cause the input PSD matrix to lose some informative components. When
applying SVD to the output PSD matrix, it gives:
[𝐺𝑦𝑦 (𝜔)] = [𝑈][𝛴][𝑈]𝐻 . (2.64)

Despite the (2.27) containing two unitary matrices (containing the left and the right singular
vectors), [𝐺𝑦𝑦 (𝜔)] is Hermitian and positive definite causing both the left and the right
vectors to be equal, and is complexed value – the transpose operator changed to Hermitian
operator. And [𝛴] is the matrix containing singular values in its diagonal elements, and it
should be noted that the diagonal non-zero value elements can relate the rank of the PSD
matrix in each frequency 𝜔 – this is the key concept to separate any close modes, and inform
us that at the considered frequency, how many modes contribute to the structural response
in the frequency domain. As (2.61) implies that the out PSD matrix can be decomposed to
contributions from each mode at each frequency, the equation (2.64) can be concluded that
useful singular vectors (counted from the rank of matrix) will represent mode shapes of
nearby dominant modes, and the singular values can also be related to the spectra of
43
equivalent SDOF systems. In the literature (Carlo & Giovanni, 2014), the example is given
that if at a certain frequency 𝜔𝑘 , the frequency response is dominated by only 1 dominant
mode, the PSD matrix can be expressed as follows:
[𝐺𝑦𝑦 (𝜔𝑘 )] = 𝜎1 {𝑢1 }{𝑢1 }𝐻 (2.65)

where 𝜎1 is the first singular value, and {𝑢1 } is the first singular vector. And in this case (say
that 𝜔𝑘 is the k-th resonant frequency), the k-th estimated mode shape is:
{𝜙̂𝑘 } = {𝑢1 (𝜔𝑘 )}, (2.66)

and the first singular value represent the APSD of the equivalent SDOF system of the
frequency 𝜔𝑘 . Anyway, the APSD of the equivalent SDOF system of the vibrational mode
is defined by the set of singular values around the local peak, characterized by the similar
singular vectors, and the bell shape of the singular value plot around the peak (Figure 2.32)
will later be used in damping ratio determination. And here are some literatures
implementing this technique in their studies (Gentile & Saisi, 2007; Cimellaro, Pianta, &
Stefano, 2012).
In the original paper of Brincker, Lingmi, & Palle (2001), they did not state how the damping
ratio can be determined from the original FDD method, but in the method called Enhanced
Frequency Domain Decomposition (EFDD), the damping identification is included and the
natural frequencies can be determined without the effect of frequency resolution. The
method continued from the point that the equivalent SDOF system (having bell shape) of the
first (or else) singular values around the peak belongs to the corresponding mode of
vibration; The EFDD method gives the detail on how far singular values should be included
as belonging to the mode. The concept is that the first (in the case that the first singular value
is of interest) singular vectors around the peak value must be similar to the one at the peak
point. The threshold value is defined by the value of MAC (Modal Assurance Criteria)
between two mode shapes being not less than 0.8 in general (Carlo & Giovanni, 2014), or
can be called MAC rejection level. The MAC index is:
𝐻 2
|{𝑢𝑗 } {𝜙̂𝑘 }|
𝑀𝐴𝐶 = 𝐻 𝐻
(2.67)
({𝑢𝑗 } {𝑢𝑗 }) ({𝜙̂𝑘 } {𝜙̂𝑘 })
where {𝑢𝑗 } is the considered singular vectors at the frequencies around the peak
corresponding to the k-th mode. MAC value tells how much the two vectors correlate: the
index will be zero when they are orthogonal to each other and equal to one when they are
basically the same but different only by scaling. And as long as the index is higher than 0.8
the singular values are still counted as a part of the equivalent SDOF PSD function. Then
the bell shape function associated with the k-th mode will be transformed to the time domain
auto-correlation function of the k-th mode by Inverse Fourier Transform. Then, the well-
known logarithmic decrement technique (2.68) will be used to identify the damping ratio on
the free decay behavior of the auto correlation.
𝑢𝑖
𝛿 = ln ≈ 2𝜋𝜉𝑘 (2.68)
𝑢𝑖+1
44
Figure 2.33 The parameters in FDD affecting damping ratios: (a) NFFT, (b)
types of window
where it is under the assumption that the damping ratio is quite small, 𝛿 is logarithmic
decrement, 𝑢𝑖 and 𝑢𝑖+1 are the two successive peak values. And, since it is the SDOF
response, the damped frequency 𝑓𝑘𝑑 can be found by counting the time for a cycle to pass
the axis in a second, then the natural frequency can relate as follows:
𝑓𝑘𝑑 = 𝑓𝑘 √1 − 𝜉𝑘2 (2.69)
where 𝑓𝑘 and 𝜉𝑘 are the natural frequency and damping ratio of the k-th mode, respectively.
Nevertheless, this method also faces the same problem of frequency resolution affecting the
accuracy of the results in PP method. Based on the literatures (Hyo & Byung, 2018; Yukio,
et al.) , the longer the signal used as the input of the method, the lower the changes in the
results; in the study of Hyo & Byung on the SHM of a supertall building (555 meters tall),
the relationship between the identified damping ratios by FDD method and number of
samples to be used in the DFT (NFFT) can be shown in Figure 2.33 (a). Also, the types of
window to be used in the method were also being varied to examine the effect, but the plot
in Figure 2.33 (b) manifest that it was not that influential to changes in damping ratio
estimation. These imply how important the effect of segment length is for FDD method (also
for PP).
2.3 Earthquake Analysis and Response of Multi-story Buildings
2.3.1 Earthquake Analysis and Response of Linearly Elastic Buildings
Buildings are categorized as one of the very complex types of structures, yet can be
successfully modeled in the Finite Element software comprising of many elements and
nodes. Despite complexity of the numerical model, the governing equation of motion in the
elastic range is pretty much the same in all buildings, including inertia force term, damping
force term, restoring force term and importantly the external force term. This very last
section of the chapter will be used by the author to demonstrate how much insight one can
gain about the dynamic characteristics or behavior of the response by only examining the
dynamic governing equation of motion of the building.
45
2.3.1.1 Static Condensation
In the dynamic analysis of buildings, especially when the excitation is a ground motion
(either one or both directions), some degree of freedoms (DOF) can be written in terms of
others for the sake of simplicity in the analysis. Vertical and rotational DOFs are generally
those to not be included in the dynamic analysis and called static DOFs; one of the reasons
is that the external force excitations are not in these two directions (no associated external
force terms) and these DOFs do not produce much of inertia force terms (in practical
application, they are neglected) causing no inertia force terms in the governing equation. In
the literature (Chopra A. K., 2012), the author gives an example of a frame of two story and
two bays consisting of eight DOFs and under the assumption of elements having no axil
deformation effect shown in Figure 2.34. And as shown in the figure, only two translational
degree of freedoms are included in the dynamic analysis and are referred to as dynamic
DOFs. The governing equation of motion (without damping term) can be as follows:
𝒎𝒕𝒕 𝟎 𝒖̈ 𝒕 𝒌 𝒌𝒕𝟎 𝒖𝒕 𝒑 (𝑡)

[ ] { } + [ 𝒕𝒕 ] {𝒖 } = { 𝒕 } (2.70)
𝟎 𝟎 𝒖̈ 𝟎 𝒌𝟎𝒕 𝒌𝟎𝟎 𝟎 𝟎
where 𝒎𝒕𝒕 is the mass matrix of dynamic DOFs; 𝒖𝒕 and 𝒖𝟎 are displacement vectors of
dynamic and static DOFs, respectively. 𝒖̈ 𝒕 and 𝒖̈ 𝟎 are the acceleration vectors of those
DOFs. 𝒌𝒕𝒕 and 𝒌𝟎𝟎 are the uncoupled corresponding stiffness matrix of both types of DOFs
while the coupling terms are related by 𝒌𝒕𝟎 = 𝒌𝟎𝒕 𝑇 . Finally, 𝒑𝒕 (𝑡) is the external force
vector of the dynamic DOFs. From (2.70), the two following equations can be written as:
𝒎𝒕𝒕 𝒖̈ 𝒕 + 𝒌𝒕𝒕 𝒖𝒕 + 𝒌𝒕𝟎 𝒖𝟎 = 𝒑𝒕 (𝑡) (2.71)

𝒌𝟎𝒕 𝒖𝒕 + 𝒌𝟎𝟎 𝒖𝟎 = 𝟎. (2.72)
Since there is no inertia force term for the static DOF, the equation (2.72) allows a static
relationship between two types of DOF as shown in (2.73).
𝒖𝟎 = −𝒌𝟎𝟎 −𝟏 𝒌𝟎𝒕 𝒖𝒕 (2.73)

By substituting (2.73) into (2.71), it gives
̂ 𝒕𝒕 𝒖𝒕 = 𝒑𝒕 (𝑡)
𝒎𝒕𝒕 𝒖̈ 𝒕 + 𝒌 (2.74)
Figure 2.34 (a) DOFs of the frame, (b) DOFs in the dynamic analysis
46
̂𝒕𝒕 is defined as condensed stiffness matrix defined by
where 𝒌
̂𝒕𝒕 = 𝒌𝒕𝒕 − 𝒌𝑻𝟎𝒕 𝒌−𝟏

𝒌 𝟎𝟎 𝒌𝟎𝒕 (2.75)
Thus, it can be shown (2.74) that the static DOFs can be eliminated from the equation of
motion, and only dynamic DOFs (that dynamic force excitation applied in the directions and
causing inertia forces) can also be present in the equation of motion.
2.3.1.2 Orthogonality of modes

Unlike single degree of freedom (SDOF) structures, multi degree of freedom (MDOF)
structures have many vibrational modes characterized by different mode shapes and
frequencies in each mode where these dynamic characteristics can be determined from the
matrix eigen value problem stated in the following equation:
𝒌∅𝒏 = 𝜔𝑛2 𝒎∅𝒏 (2.76)

where k is the condensed stiffness matrix (which will later, for simplicity, called stiffness
matrix), 𝒎 is the mass matrix of the structure and 𝜔𝑛 and ∅𝒏 are the n-th eigen value and
eigen vector solved from (2.76) and referred to as the n-th natural frequency and mode shape
of the structure, respectively. The matrix used to contain these mode shapes as its columns
is called modal matrix which can be expressed in the following equation:
𝝓𝟏𝟏 𝝓𝟏𝟐 … 𝝓𝟏𝑵

𝝓 𝝓𝟏𝟏 … 𝝓𝟐𝑵
𝚽 = [𝝓𝒋𝒏 ] = [ 𝟐𝟏 ] (2.77)
⋮ ⋮ ⋱ ⋮
𝝓𝑵𝟏 𝝓𝑵𝟐 … 𝝓𝑵𝑵
where N represents the total vibration modes of the structure. One of the properties of mode
shapes which is proven elsewhere (Chopra A. K., 2012) is orthogonality between each mode
and for the mathematical expression, we have
𝝓𝑻𝒏 𝒌𝝓𝒓 = 𝟎 𝝓𝑻𝒏 𝒎𝝓𝒓 = 𝟎 (2.78)

where n≠r, the mode shapes of the structure having k and m as their structural properties
and solved from (2.76) are said to be orthogonal to each other with respect to stiffness and
mass matrices. Followed by the definition of modal stiffness 𝑲 and modal mass stiffness M
matrices, we have
𝑲 = 𝚽 𝑻 𝒌𝚽 𝑴 = 𝚽 𝑻 𝐦𝚽. (2.79)
Following the orthogonality condition property, only diagonal elements exist in both
matrices, they are called the modal stiffness 𝐾𝑛 and modal mass 𝑀𝑛 of the n-th mode and
can be defined as follows:
𝐾𝑛 = 𝝓𝑻𝒏 𝒌𝝓𝒏 𝑀𝑛 = 𝝓𝑻𝒏 𝒎𝝓𝒏 . (2.80)

And both of them are related by the following equation:
𝐾𝑛 = 𝜔𝑛2 𝑀𝑛 . (2.81)
47
For the proof of (2.81), it can be found elsewhere (Chopra A. K., 2012). Using the mode
shapes, the relationship between displacement vector of dynamic DOFs 𝒖 can be related to
the displacement vectors in the modal coordinate 𝒒 which is:
𝑁
𝒖 = ∑ 𝝓𝒓 𝑞𝑟 = 𝚽𝒒 (2.82)
𝑟=1
where 𝑞𝑟 is a scalar value also called as modal coordinate, and 𝒒 = 〈𝑞1 𝑞2 … 𝑞𝑛 〉𝑻 .

Inversely, if the mass matrix and the n-th mode shape are known, and based on the
orthogonal property, the displacement in modal coordinate can be determined by pre-
multiplying (2.82) by 𝝓𝑻𝒏 𝒎 in both sides, it yields
𝑁
𝝓𝑻𝒏 𝒎𝒖 = ∑(𝝓𝑻𝒏 𝒎𝝓𝒓 )𝑞𝑟 . (2.83)

𝑟=1
Only when r = n, the 𝝓𝑻𝒏 𝒎𝝓𝒓 will not be equal to zero; thus
𝝓𝑻𝒏 𝒎𝒖 = (𝝓𝑻𝒏 𝒎𝝓𝑛 )𝑞𝑛 . (2.84)

And by some manipulation, it gives
𝝓𝑻𝒏 𝒎𝒖 𝝓𝑻𝒏 𝒎𝒖
𝑞𝑛 = 𝑻 = (2.85)
𝝓𝒏 𝒎𝝓𝑛 𝑀𝑛
2.3.1.3 Modal Response History Analysis
In generally, the governing equation of motion of MDOF structure including the damping
term can mathematically be shown as follows:
𝒎𝒖̈ + 𝒄𝒖̇ + 𝒌𝒖 = 𝒑(𝑡) (2.86)

where k is, as already mentioned, condensed stiffness matrix and c is the damping matrix. It
can be seen that governing equation is coupled because there are coupling terms, generally
in most of the matrices. But one can turn (2.86) into the modal coordinate where MDOF
governing equation will be turned into just N SDOF equations associated with each
vibrational mode. This can be done by substituting (2.82) into (2.86) and pre-multiplying
(2.86) by 𝝓𝑻𝒏 , it gives
𝑁 𝑁 𝑁
∑ 𝝓𝑻𝒏 𝒎𝝓𝒓 𝑞̈ 𝑟 (𝑡) + ∑ 𝝓𝑻𝒏 𝒄𝝓𝒓 𝑞̇ 𝑟 (𝑡) + ∑ 𝝓𝑻𝒏 𝒌𝝓𝒓 𝑞𝑟 (𝑡) = 𝝓𝑻𝒏 𝒑(𝒕) (2.87)
𝑟=1 𝑟=1 𝑟=1
which can also be rewritten as

𝑁
𝑀𝑛 𝑞̈ 𝑛 (𝑡) + ∑ 𝝓𝑻𝒏 𝒄𝝓𝒓 𝑞̇ 𝑟 (𝑡) + 𝐾𝑛 𝑞𝑛 (𝑡) = 𝑃𝑛 (𝑡) (2.88)

𝑟=1
where 𝑃𝑛 (𝑡) is defined as 𝝓𝑻𝒏 𝒑(𝒕), and 𝑞̈ 𝑛 (𝑡) and 𝑞̇ 𝑟 (𝑡) are acceleration and velocity
responses in modal coordinate, respectively. And if the system has classical damping
48
(careful explained elsewhere (Chopra A. K., 2012)) – simply meaning 𝝓𝑻𝒏 𝒄𝝓𝒓 term is equal
to zero, the equation (2.88) can be reduced to the following equation:
𝑀𝑛 𝑞̈ 𝑛 (𝑡) + 𝐶𝑛 𝑞̇ 𝑟 (𝑡) + 𝐾𝑛 𝑞𝑛 (𝑡) = 𝑃𝑛 (𝑡) (2.89)
where 𝐶𝑛 is defined as 𝝓𝑻𝒏 𝒄𝝓𝒏 . Here by considering (2.89), one can see that the n-th modal
coordinate equation is governed by some parameters (𝑀𝑛 , 𝐶𝑛 and , 𝐾𝑛 ) and one input force
as in the case of single degree of freedom structure.
2.3.1.4 Modal Response Contribution

To understand how each mode contributes to the response of the structure, the input force
vector 𝒑(𝒕) is then decomposed to have the time-varying part 𝑝(𝑡) and the time-independent
part s which is also called spatial distribution; the mathematical equation is
𝒑(𝑡) = 𝒔𝑝(t) (2.90)
The spatial distribution 𝒔 can be, in order to separate the contribution to each mode,
expressed as follows:
𝑁 𝑁
𝒔 = ∑ 𝒔𝒓 = ∑ 𝛤𝑟 𝒎𝝓𝒓 (2.91)
𝑟=1 𝑟=1
By pre-multiplying (2.91) by 𝝓𝑻𝒏 both sides, it gives
𝝓𝑻𝒏 𝒔
Γ𝑛 = , (2.92)
𝑀𝑛
and the contribution from the n-th mode to 𝒔 is called 𝒔𝒏 which is
𝒔𝒏 = 𝛤𝑛 𝒎𝝓𝒏 . (2.93)
The equation below is the inertia force (𝑓𝐼 ) contributed from the n-th mode.
(𝒇𝑰 )𝒏 = −𝒎𝒖̈ 𝒏 (𝑡) = −𝒎𝝓𝒏 𝑞̈ 𝑛 (𝑡). (2.94)

One may notice that 𝒔𝒏 is proportional to the time-independent part of the modal inertia
force. The expansion shown in (2.91) is useful because 𝒔𝒏 will only excite the n-th
vibrational mode only, and the response due to the excitation will be totally contributed from
the n-th mode only.
2.3.1.5 Multi-story Building Subjected to Earthquake Excitation

For the 2D governing equation of motion of the MDOF structure subjected to earthquake
ground motion 𝑢̈ 𝑔 , it is as follows
𝒎𝒖̈ + 𝒄𝒖̇ + 𝒌𝒖 = 𝒑𝒆𝒇𝒇 (𝑡) (2.95)

where 𝒑𝒆𝒇𝒇 (𝑡) is referred as the effective earthquake-force defined in the following equation:
49
𝒑𝒆𝒇𝒇 (𝑡) = −𝒎𝒊𝑢̈ 𝑔 (𝑡) (2.96)
where 𝒊 is denoted as influence vector, in this case, equal to the vector having all entries
equal to one with the length of N – or the number of (dynamic) DOFs (the thorough way of
defining influence vector can be found elsewhere (Chopra & Goel, 2001)). By comparing
(2.96) to (2.90), one can expand the time-independent part as follows:
𝑁 𝑁
𝒎𝒊 = ∑ 𝒔𝒏 = ∑ 𝛤𝑛 𝒎𝝓𝒏 (2.97)
n=1 n=1
This is done in order to separate the contribution of the earthquake induced force into each
vibrational mode. The new parameter is, then defined accordingly as:
𝐿𝑛 = 𝝓𝑻𝒏 𝒎𝒊, (2.98)

and when relate both (2.98) and (2.97) to (2.92), it gives the following equation when the
excitation is the earthquake induced force.
𝐿𝑛
𝛤𝑛 = (2.99)
𝑀𝑛
where 𝛤𝑛 is also called modal participation factor implying how much each mode contribute
to the dynamic response of the structure despite the fact that it is just indirect and does not
gives genuine insight of that. And by relating (2.89) and (2.99), one can rewrite the
governing equation in the form as follows:
𝑞̈ 𝑛 + 2𝜉𝑛 𝜔𝑛 𝑞̇ 𝑛 + 𝜔𝑛2 𝑞𝑛 = −𝛤𝑛 𝑢̈ 𝑔 (𝑡) (2.100)

where right now the n-th mode equivalent SDOF is governed by n-th mode damping ratio
𝜉𝑛 and frequency 𝜔𝑛 and the modal participation factor, but (2.100) can further be written
such that the governing equation is governed by only 𝜉𝑛 and 𝜔𝑛 which is
𝐷̈𝑛 + 2𝜉𝑛 𝜔𝑛 𝐷̇𝑛 + 𝜔𝑛2 𝐷𝑛 = −𝑢̈ 𝑔 (𝑡) (2.101)

where 𝐷𝑛 is another form of modal coordinate related to the original modal coordinate 𝑞𝑛 .
The relation between them is:
𝑞𝑛 (𝑡) = 𝛤𝑛 𝐷𝑛 (2.102)
By considering (2.101), one can see that it is pretty much a SDOF system subjected to a
ground motion.
To illustrate, how the governing equation can be formed for any multi-story building and
how the coupled mode shapes as a result of coupled behavior in its dynamic characteristics
can be dealt with in modal analysis and later on be used and mentioned again for multi-mode
(or so called modal pushover analysis), the author of the literature (Chopra A. K., 2012),
gives an example of a multi-story building in which its plans (for all stories) are symmetrical
about x-axis and unsymmetrical about y-axis, subjected to a ground motion 𝑢̈ 𝑔𝑦 (𝑡) which is
the ground motion in y direction. Saying that all the floor diaphragms have the same radius
50
of gyration 𝑟 (where the mass moment of inertia of the j-th floor, having the floor mass 𝑚𝑗
is shown in equation (2.103), 𝐼𝑂𝑗 = 𝑟 2 𝑚𝑗 ).
𝐼𝑂𝑗 = 𝑟 2 𝑚𝑗 (2.103)
However, to avoid confusion regarding coupling system, the governing equation without
damping term of the multi-story building (Figure 2.35) having symmetrical plan in two
directions subjected to the ground motion 𝑢̈ 𝑔 having x and y translational and rotational
about z axis components is given in the equation (2.104) for the initial study. In Figure 2.35,
the dynamic DOFs of the buildings are set to be the movement of the center of mass in each
floor in x, y and rotational 𝜃 about z axis directions, and to make it symmetrical in two
directions, the stiffness of the frame B and C are set to be equal, and the eccentricity of the
frame A with respect to the center of mass (e) is set to be zero.
𝒎 𝒖̈ 𝒙 𝒌𝒙𝒙 0 0 𝒖𝒙
[ 𝒎 ] {𝒖̈ 𝒚 } + [ 0 𝒌𝒚𝒚 0 ] {𝒖𝒚 } = −𝒑𝒆𝒇𝒇 (𝑡) (2.104)
𝑰𝑶 𝒖̈ 𝜽 0 0 𝒌𝜽𝜽 𝒖𝜽
where 𝒑𝒆𝒇𝒇 (𝑡) is manifested in the following equation
𝒎 𝟏 𝟎 𝟎
𝒑𝒆𝒇𝒇 (𝑡) = [ 𝒎 ] ({𝟎} 𝑢̈ 𝑔𝑥 + {𝟏} 𝑢̈ 𝑔𝑦 + {𝟎} 𝑢̈ 𝑔𝜃 ) (2.105)
𝑰𝑶 𝟎 𝟎 𝟏
(where 1 = 〈1 1 … 1〉𝑇 and 0 = 〈0 0 … 0〉𝑇 ); 𝒎 is the mass matrix containing mass
of each floor as an entry in its diagonal element; 𝑰𝑶 is the mass moment of inertia matrix
working in the same way as 𝒎 but containing 𝐼𝑂𝑗 ; 𝒌𝒙𝒙 , 𝒌𝒚𝒚 , and 𝒌𝜽𝜽 are the uncoupled
stiffness matrix of x, y and 𝜃 DOFs, respectively; 𝒖𝒙 , 𝒖𝒚 and 𝒖𝜽 are the vectors individually
containing the DOFs in their directions which are:
Figure 2.35 Typical plan of multi-story building
51
𝒖𝒙 = 〈𝑢1𝑥 𝑢2𝑥 … 𝑢𝑁𝑥 〉𝑇 𝒖𝒚 = 〈𝑢1𝑦 𝑢2𝑦 … 𝑢𝑁𝑦 〉𝑇 (2.106)
𝒖𝜽 = 〈𝑢1𝜃 𝑢2𝜃 … 𝑢𝑁𝜃 〉𝑇
It should be noted that the stiffness matrices can be formulated by direct stiffness method
which is emphasized here in this research. By observing the equation in (2.104), one can see
that the governing equation can actually be independently divided into 3 main equations
corresponding to their own directions. It means that if the building is symmetrical like the
example shown and is subjected to the x directional ground motion (which is proportional
to the x axis of the building), the building will be excited only in x direction and there will
be no movement in y and rotational about z axis directions. Anyway, in general cases, the
stiffness matrix 𝒌 in the equation of motion can be shown as follows:
𝒌𝒙𝒙 𝒌𝒙𝒚 𝒌𝒙𝜽

𝒌 = [𝒌𝒚𝒙 𝒌𝒚𝒚 𝒌𝒚𝜽 ] (2.107)
𝒌𝜽𝒙 𝒌𝜽𝒚 𝒌𝜽𝜽
where all coupling matrices are present in the stiffness matrix 𝒌. Getting back to the
originally stated example of the one-way unsymmetrical building in y axis, the governing
equation of motion without damping can be visited herein:
𝒎 𝒖̈ 𝒙 𝒌𝒙𝒙 0 0 𝒖𝒙 𝒎 𝟎
[ 𝒎 ] { 𝒚} + [ 0
𝒖̈ 𝒌𝒚𝒚 𝒌𝒚𝜽 ] {𝒖𝒚 } = − [ 𝒎 ] {𝟏} 𝑢̈ 𝑔𝑦 . (2.108)
𝑰𝑶 𝒖̈ 𝜽 0 𝒌𝜽𝒚 𝒌𝜽𝜽 𝒖𝜽 𝑰𝑶 𝟎
As one can observe that, there is no coupling terms of stiffness matrices between x and other
directions DOFs, and only y component ground motion excites the building; thus, the
equation in (2.108) can be separated into 2 sub-equations as follows:
𝒎𝒖̈ 𝒙 + 𝒌𝒙𝒙 𝒖𝒙 = 𝟎 (2.109)

𝒎 𝒖̈ 𝒚 𝒌𝒚𝒚 𝒌𝒚𝜽 𝒖𝒚 𝒎 𝟏
[ ]{ } + [ ] {𝒖 } = − [ ] { } 𝑢̈ 𝑔𝑦 . (2.110)
𝑰𝑶 𝒖̈ 𝜽 𝒌𝜽𝒚 𝒌𝜽𝜽 𝜽 𝑰𝑶 𝟎
Since there is no excitation (assumed to have no initial displacement and velocity) in (2.109),
there will be no movement in the x direction while (2.110) suggests that there will be coupled
movement in y and rotation about z axis direction due to coupling terms in the equation of
motion. For the 2 equations above, the matrix eigen value problems of the system can be
independent solved based on (2.76). The mode shapes in x direction are 𝝓𝑥𝑟 where r = 1,2,
…, N and the coupled mode shapes of y and rotation about z axis 𝜃 are 𝝓𝑛𝑦𝜃 where n = 1,2,
…, N and 𝝓𝑛𝑦𝜃 = 〈𝝓𝑛𝑦 𝝓𝑛𝜃 〉𝑇 . On the other hand, if the case was that the building is
unsymmetrical in both directions (assuming that all coupling sub matrices in the stiffness
matrix are there), the typical coupling mode shapes are:
𝝓𝑛 = 〈𝝓𝒏𝑥 𝝓𝑛𝑦 𝝓𝑛𝜃 〉𝑻 where n = 1, 2, …, N (2.111)

And to illustrate the effect of modal response contribution in this very example, the modal
expansion of effective earthquake force is to be carried out. The effective earthquake force
for the 2D frame in (2.96) turns out to be:
52
𝒎𝒊
𝒑𝒆𝒇𝒇 (𝑡) = − { } 𝑢̈ (𝑡) = −𝒔𝑢̈ 𝑔𝑦 (𝑡) (2.112)
𝟎 𝑔𝑦
The spatial distribution related to (2.112) in (2.97) turns out to be:
𝟐𝑵
𝒎𝒊 𝒎𝝓𝑛𝑦
𝒔 = { } = ∑ 𝛤𝑛 { 2 }. (2.113)
𝟎 𝑟 𝒎𝝓𝑛𝜃
𝒏=𝟏
And, 𝐿𝑛 in the equation (2.98) becomes:
𝒎𝒊
𝐿𝑛 = 〈𝝓𝑛𝑦 𝑇 𝝓𝑛𝜃 𝑇 〉 { } = 𝝓𝑛𝑦 𝑇 𝒎𝒊. (2.114)
𝟎
The modal mass formulated in (2.80) can be further extended to be as follows:
𝝓𝑛𝑦
𝑀𝑛 = 〈𝝓𝑛𝑦 𝑇 𝝓𝑛𝜃 𝑇 〉 [𝒎 2 ]{ } (2.115)
𝑟 𝒎 𝝓𝑛𝜃
Using (2.99), one can determine the spatial n-th mode modal inertia force pattern of the
building in the example, it gives
𝑠𝑦𝑛 𝒎𝝓𝑛𝑦
𝑠𝑛 = {𝑠 } = 𝛤𝑛 { 2 } (2.116)
𝜃𝑛 𝑟 𝒎𝝓𝑛𝜃
where the equation (2.116) implies that in order to only excite the n-th mode (of the coupling
behavior in y and 𝜃), lateral forces 𝛤𝑛 𝑚𝑗 𝜙𝑛𝑦𝑗 , j = 1,2, …N and torque 𝛤𝑛 𝑚𝑗 𝜙𝑛𝜃𝑗 , j = 1,2,
…N are needed to be applied simultaneously to the center of mass of the corresponding j-th
floor. This example laid the principle of modal pushover analysis to be discussed in the next
sub-section.
2.3.2 Earthquake Analysis and Response of Inelastic Buildings
2.3.2.1 Nonlinear Response History Analysis (NLRHA)

What makes things change from the analysis of the linearly elastic buildings is the fact that
right now the resisting force in (2.86) is not having constant relationship with displacement
vector 𝒖 anymore, but is said to be path dependent (Chopra A. K., 2012). It is manifested as:
𝒇𝒔 = 𝒇𝒔 (𝒖) (2.117)
; thus, the inelastic governing equation of the structure subjected to a ground motion becomes
𝒎𝒖̈ + 𝒄𝒖̇ + 𝒇𝒔 (𝒖) = −𝒎𝒊𝑢̈ 𝑔 (2.118)

It is well known that NLRHA requires a lot of computational time for the analysis due to the
fact that at each time step the resisting force must be calculated under the repeated process
of updating the stiffness matrix. All of the hysteretic and nonlinear behavior of all elements
must be taken care of, and the P-Δ analysis is of need to be included since the large amplitude
of deformations is expected to happen. Despite its rigorous formulation, Chopra & Goel
(2001) proposed the estimated analysis methods based on the concept of modal response
contribution: Uncoupled Modal Response History Analysis (UMRHA) and Modal Pushover
Analysis (MPA).
53
2.3.2.3 Formulation of Modal Pushover Analysis (MPA)
For the linearly elastic system, in order to obtain the peak response contributed from the n-
th mode, the numerical model (with dead load and p-delta effect taken into consideration)
will be pushed by the modal inertia force pattern 𝑠𝑛∗ , it is
𝒔𝒏∗ = 𝒎𝝓𝑛 (2.119)

The structure will be pushed up to the target roof displacement 𝑢𝑟𝑛 (r denotes roof) which is
𝑢𝑟𝑛 = 𝛤𝑛 𝜙𝑟𝑛 𝐷𝑛 (2.120)

where 𝜙𝑟𝑛 is the value of mode shape at the roof position and 𝐷𝑛 is the value from the
displacement response spectrum of the considered ground motion corresponding to the
damping ratio and period of the n-th mode of the structure. Then, the responses, as a
consequence, are the responses due to the 𝒑𝒆𝒇𝒇,𝒏 (𝑡) (of the ground motion). And, for the
responses due to the ground motion, a modal combination rule can be applied to calculate
the responses from significant modes. In MPA of linearly elastic system, there is nothing
different from Response Spectrum approach (RSA) to estimate any response from a
considered ground motion.
Regarding inelastic system, as already mentioned in UMRHA, applying 𝒔𝒏∗ in MPA does not
only cause the response contributed from the n-th mode but weakly contributed from others
as well. And the structure will be pushed to the n-th determined roof displacement for each
mode; any response at that point of each mode will be combined based on the modal
combination rule (The authors of the literature (Chopra & Goel, 2001) stated that it lacks a
rigorous theoretical background behind it, but under the assumption of weak coupling of
modes, the results should be reliable). The thing different from MPA in linear system is that
𝑢𝑟𝑛 cannot be easily found from response spectrum curve but needed to take into account
the effect of hysteretic behavior of buildings, post-yield stiffness of the idealized curve,
strength reduction factor and so on. The determination of 𝑢𝑟𝑛 or so-called target
displacement, which is greatly required if the MPA have to be utilized for response
estimation, for the variety of high-rise RC shear wall buildings thoroughly studied can be
found in the literature (Najam, 2017) which will not be emphasized in this research. But
anyway, the approach is still needed to construct the relation between base shear 𝑉𝑏𝑛 due to
𝒔∗𝒏 and roof displacement 𝑢𝑟𝑛 for the considered numerical model (the result is called
pushover curve) since they can be transformed into 𝐹𝑠𝑛 /𝐿𝑛 - 𝐷𝑛 relation needed to be utilized
in UMRHA approach. They are related to each other as follows:
𝐹𝑠𝑛 𝑉𝑏𝑛 𝑢𝑟𝑛

= , 𝐷𝑛 = . (2.121)
𝐿𝑛 𝐿𝑛 𝛤𝑛 𝛤𝑛 ∅𝑟𝑛
There are two types of pushover curve: monotonic pushover curve and cyclic pushover
curve. Pushover curve of the n-th mode is basically the relationship between base shear and
roof displacement due to 𝒔∗𝒏 . The first type can be constructed by monotonically push the
structure with increasing amplitude of static force, yet proportional to 𝒔∗𝒏 in one direction
(anyway, this process can be controlled by the increment of roof displacement). The curve
not only provides the analyst insight about damage progression of the structure due to the n-
th mode contribution when the roof drift increases (as can be typically shown in Figure 2.37
as an example from the study of Najam (2017), the monotonic pushover curves are plotted
54
in dashed and bold lines), but also be used in the process of idealizing the equivalent force-
displacement relationship of the n-th inelastic SDOF system in order to obtain either bi-
linear (for flag shape model) or tri-linear (modified flag shape model) backbone curve of the
idealized model. Second, the cyclic pushover curve is also another main key to constructing
the idealized force-displacement relationship which will later be manifested that some
parameters are needed to be extracted from the cyclic pushover curve. The curve can be
constructed by increasing the amplitude and also reversing the direction of the force vectors
allowing the analyst to obtain the hysteretic behavior of the n-th mode inelastic SDOF system
of the structure shown as normal thickness line in the aforementioned figure.
2.3.2.2 Formulation of Uncoupled Modal Response History Analysis (UMRHA)

For the linearly elastic system, this method is basically Modal Response History Analysis
(MRHA) except the fact the MRHA totally filters the contribution from not related modes
by orthogonality property when the governing equations are turned into modal coordinate
(by pre-multiplying both sides with 𝝓𝑻𝒏 . Based on the modal expansion of effective
earthquake force 𝒑𝒆𝒇𝒇 (𝑡) into the combination of 𝒑𝒆𝒇𝒇,𝒏 (𝑡) (or the contribution from the n-
th mode to 𝒑𝒆𝒇𝒇 (𝑡)), the governing equation in (2.95) is replaced with 𝒑𝒆𝒇𝒇,𝒏 (𝑡) and turned
into modal coordinate for simplicity. Then, calculating the response from all modes and just
combined the obtained results together by normal superposition approach.
When it comes to the inelastic system, the governing equation of motion due to the 𝒑𝒆𝒇𝒇,𝒏 (𝑡)
is as follows:
𝒎𝒖̈ + 𝒄𝒖̇ + 𝒇𝒔 (𝒖) = −𝒔𝒏 𝑢̈ 𝑔 (𝑡) (2.122)

In the inelastic system, the solution to (2.122) can no longer be described by:
𝑢𝑛 (𝑡) = ∅𝑛 𝑞𝑛 (𝑡) (2.123)

because modes other than n-th will also contribute to the responses. Actually, the concept of
modal analysis cannot be used in the inelastic system, but it is insightful to describe the
inelastic system with the corresponding linear dynamic properties. Anyway, as per
discussed, the solution can be described by the following equation:
𝑁
𝑢𝑛 (𝑡) = ∑ ∅𝑟 𝑞𝑟 (𝑡) ≈ ∅𝑛 𝑞𝑛 (𝑡) (2.124)

𝑟=1
Despite the fact, it is convinced to be assumed that the contribution from the n-th mode
should be dominant due to the excitation 𝒑𝒆𝒇𝒇,𝒏 (𝑡) since in the elastic system, the whole
contribution is just from the n-th mode. To investigate the mentioned assumption, the author
of the literature analyzed the numerical model of 9 story steel building using NLRHA on
(2.122) for all significant modes. After obtaining the solutions, the author decomposed them
based on the concept of modal decomposition (2.85) to see the effect of other modes once
the building started to be in the inelastic range. The results in Figure 2.36 shows that
contribution from each mode can be barely observed except roof displacement due to
𝒑𝒆𝒇𝒇,𝟐 (𝑡) contributed up to 25% from mode 1 (was zero % in the elastic range). Anyway,
the results in general imply that the response can be approximated by the second portion of
(2.124). By substituting the approximation into (2.122) and pre-multiplying by 𝝓𝑻𝒏 , it yields:
55
𝐹𝑠𝑛
𝑞̈ 𝑛 + 2𝜉𝑛 𝜔𝑛 𝑞̇ 𝑛 + = −Γ𝑛 𝑢̈ 𝑔 (𝑡) (2.125)
𝑀𝑛
where 𝐹𝑠𝑛 is defined as the non-linear hysteretic function of the n-th modal coordinate 𝑞𝑛
(under the assumption that all other modal coordinates contribute very little to the solution),
and can be expressed as follows:
𝐹𝑠𝑛 = 𝐹𝑠𝑛 (𝑞𝑛 ) = 𝝓𝑇𝑛 𝒇𝑠 (𝑞𝑛 ) (2.126)
Figure 2.36 Modal decomposition of roof displacement due to 𝒑𝒆𝒇𝒇,𝒏 (𝑡), n = 1, 2, 3 and
4: (a) due to 𝒑𝒆𝒇𝒇,𝟏 (𝑡), (b) due to 𝒑𝒆𝒇𝒇,𝟐 (𝑡), (c) due to 𝒑𝒆𝒇𝒇,𝟑 (𝑡), (d) due to 𝒑𝒆𝒇𝒇,𝟒 (𝑡)
Using (2.102), the equation in (2.125) can be transformed into the modal coordinate 𝐷𝑛 as:
𝐹𝑠𝑛
𝐷̈𝑛 + 2𝜉𝑛 𝜔𝑛 𝐷̇𝑛 + = −𝑢̈ 𝑔 (𝑡) (2.127)
𝐿𝑛
Right now, this n-th mode inelastic SDOF system having 𝐷𝑛 as its displacement response
can be defined by the small oscillation properties (or of linear system)–natural frequency 𝜔𝑛
and damping ratio 𝜉𝑛 – of the n-th mode MDOF system and the relation between 𝐹𝑠𝑛 /𝐿𝑛 and
𝐷𝑛 or force-deformation relation describing the non-linear behavior of this inelastic SDOF
system. The displacement of this system will be substituted into (2.123) in order to obtain
the floor movements and further used to calculate story drift. And then any other response
quantity 𝑟𝑛 (𝑡) due to 𝒑𝒆𝒇𝒇,𝒏 (𝑡) can be determined after that. To obtain the responses due to
56
𝒑𝒆𝒇𝒇 (𝑡), the superposition method is used to sum all responses up despite the fact that it is
strictly valid only for linear system. The sources of error which may happen (Chopra & Goel,
2001) may be from: (1) the fact that the coupling effects are ignored in (2.125) and the
responses due to 𝒑𝒆𝒇𝒇,𝒏 (𝑡) are forced to be proportional to the n-th mode shape (2.123)
which is later avoided in most parts of MPA; (2) the superposition which is strongly valid in
linear system is violated to be used; (3) the way, the relation between 𝐹𝑠𝑛 /𝐿𝑛 and 𝐷𝑛 is
constructed (later to be described) is based on reliability and accuracy of the idealized model.
2.3.2.4 Recent Studies on Inelastic Response Estimation using UMRHA approach

Originally, UMRHA was developed in the study of Chopra & Goel (2001) for the sake of
convincing that despite inelasticity, the MPA (based on the very same basic assumption) are
to some extent capable of predicting the responses of the building due to a seismic ground
motion. However, in their study, the steel moment resisting frame was used as the numerical
model, and the problem is that unrealistic behavior of bi-linear relationship of 𝐹𝑠𝑛 /𝐿𝑛 - 𝐷𝑛
(having the same stiffness for the case of loading, unloading and reloading even in the case
of the structure being inelastic). Also, in the paper, the authors did not state about cyclic
pushover yet in which after that was commonly used as the tool in forming 𝐹𝑠𝑛 /𝐿𝑛 - 𝐷𝑛
relation. Published in 2018, the study of Mehmood, Warnitchai, & Suwansaya was intended
to implement UMRHA as the new practical analysis procedure since it can approximately
decompose the complex response of the structure into contribution from each mode
characterized by their linear dynamic properties. Working on four different concrete core
wall tall buildings with their stories ranging from 20 up to 40 with the torsional coupling
behavior and two different characteristic ground motions, they found that the structures
experienced self-centering affect once the crack fully closed due to gravity load from the
cyclic pushover curves (that is why the idealized model of flag shape was chosen with the
potential of representing self-centering effect), only the first mode behaves nonlinearly and
Figure 2.37 Typical pushover curves of a reinforced concrete core wall building
57
if other bi-linear models were used, the results did not show up in a reasonably good
agreement with those from NLRHA compared to the case of flag shape model used. The
very same insightful results can also be found in the study of (Najam, 2017). However, based
on the study and the attempt to develop the appropriate and realistic idealized model of
Pandey (2017), he found that once the amplitude of the earthquake excitation is large and
being able to cause permanent displacements (or residual displacements) in the n-th mode
cyclic pushover curve which generally occurs once the roof drifts are up to 0.8-1.2% of the
total height (based on the buildings in the study of Najam (2017)), and at this much drift
ratio, it causes the yielding of steel reinforcement in shear walls being the main reason of
residual displacements. The flag shape model which exhibits no residual displacement but
self-centering effect gave a bad result of representing cyclic pushover curve. He then
developed the so-called Modified Flag Shape (MFS) model and came up with rules on how
MFS should behave for loading, unloading and reloading situations by observing the
behaviors of real cyclic pushover curves of the structures (which are the same as those in the
study of Najam (2017)) subjected to many loading protocols. Rules and how to predict the
residual displacement in MFS based on his work (Panday, 2017) will be later discussed in
the next sub heading. Nevertheless, the earthquake excitations in his study were not large
enough to cause any drift ratio to exceed the yield point denoted in MFS; thus, there was no
records of MFS’ potential for idealizing in large amplitude response case. Anyway, the
results show that by utilizing the novel idealized model in UMRHA, it gives a better results
compared to the traditional one since not only it can also be extended to the case where
residual displacements occur but MFS was edified such that it can represent degradation of
n-th mode equivalent SDOF system pushover curve in the idealized curve. Lastly, in the
study of Vasanthapragash, (2018), the author’s work was partly intended to fulfill the gap
not done in the previous study (Panday, 2017) which is to observe the results of MSF once
the amplitude of the earthquake excitations are high enough to cause residual displacements.
In his study, many scaling factors were used to scale the amplitude of ground motions, and
the results for low amplitude excitation agreed with the previous work that flag shape model
utilized in UMRHA is also reasonably good enough to estimate the responses. But when it
came to large amplitude earthquakes, it caused the phenomenon that the combined time
histories from UMRHA behaved as if it had time lag compared to the case of NLRHA as
can be seen from the comparison in Figure 2.39. In all mentioned studies, they not only
presented the results in terms of time histories of contribution from each mode (UMRHA),
combined response (UMRHA) and those from NLRHA, but also in terms of height-wise
envelopes (or maximum values) of roof displacement, story drift, shear force and moment.
For example, Figure 2.38 shows the associated envelope results due to 3.5 x Manjil Iran
ground motion. It should be noted that for NLRHA the peak values were directly picked
while for UMRHA, the peak values were not determined solely from each mode contribution
but combined response. In conclusion, it should be reminded that despite the fact that
UMRHA can decompose complex responses into those of each mode contribution, they are
only approximate methods due to many inherent assumptions stated in the section 2.3.2.2.
Not only that, in the study of Vasanthapragash (2018), the author also dedicated a section to
examine the fundamental errors causing by performing both analyses in different software.
By studying the linear behavior of a building through the analyses using Modal RHA (done
in Opensees) and RHA (done in Perform 3D), theoretically, there should be no discrepancies
for the results from analyzing linear structures because of the fact that in linear analysis there
is no assumption in Modal RHA unlike UMRHA, but there are, practically still some
parameters stated in his work to be initially considered to avoid the initial error causing by
58
performing both analyses in different software individually having some inherent
limitations.
Figure 2.38 Envelop results obtained via UMRHA (MFS used) and
NLRHA for 3.5 x Manjil Iran
Figure 2.39 Roof displacement from combined response from UMRHA and
NLRHA due to 3.5 x Manjil Iran: (a) using Flag shape model , (b) MFS model
59
2.3.4.5 Modified Flag Shape Model
Modified Flag Shape (MFS) model as earlier mentioned has been realized to be reasonable
to represent the hysteretic behavior of the n-th mode inelastic SDOF in the study of Panday
(2017). To construct the MFS model of 𝐹𝑠𝑛 /𝐿𝑛 - 𝐷𝑛 relationship, at first, the tri-linear
backbone curve must be established from the real monotonic pushover curve (in the form of
𝑉𝑏𝑛 - 𝑢𝑟𝑛 relationship) as can be shown in Figure 2.41. The commencement points of real
cracking and yielding of shear walls can be identified from monotonic pushover data, and
then by minimizing the area between the curve and a tri-linear curve defined by points A, B
and C (where A will be later called idealized cracking point, B will be later called idealized
yielding point and C is just the same point at the end of the curve appropriately predefined);
as a result, the appropriate tri-linear backbone curve is identified. And from the associated
cyclic pushover curve, the 𝛽 factor which will be later used to define flag depth can be
identified. Then, the idealized curve will be transformed into force-deformation relationship
using the equation (2.121).
Figure 2.41 Construction of idealized tri-linear backbone curve
Figure 2.40 Summary of parameters in MFS
60
The slope of OA in Figure 2.41 is denoted as 𝑘1 (initial stiffness), the slope of AB is denoted
as 𝑘2 (post crack stiffness) and the last slope is denoted as 𝑘3 (post yield stiffness). 𝛼1 and
𝛼2 are used to define the ratios between 𝑘2 to 𝑘1 and 𝑘3 to 𝑘1 , respectively. The idealized
cracking point A (which later will be loosely called cracking point and also the same way as
yielding point) is specified as (𝐹𝑐𝑟 , 𝐷𝑐𝑟 ) and the yielding point (B) is specified as (𝐹𝑦 , 𝐷𝑦 ).
The flag depth is defined as a portion of cracking force 𝐹𝑐𝑟 , which is 𝛽𝐹𝑐𝑟 . These parameters
are summarized in the Figure 2.40. Regarding rules of MFS, they can be in overview
described in 3 states: linear - the maximum deformation (𝐷𝑚𝑎𝑥 )𝑛 does not exceed (𝐷𝑐𝑟 )𝑛 ;
post crack - (𝐷𝑚𝑎𝑥 )𝑛 does not exceed (𝐷𝑦 )𝑛 ; and post yield - (𝐷𝑚𝑎𝑥 )𝑛 exceeds (𝐷𝑦 )𝑛 . The
behaviors of force-deformation relations are categorized as loading (where the force
increases), unloading (where the force decreases) and reloading (where the force after
unloading starts to load again). In the linear state, the curve is basically moving as a linear
relationship between force and deformation. For the rules of the ranges of post crack and
post yield, the author made them up to try to mock the real behavior of cyclic pushover
curves in his study. These rules and examples are well documented in his study (Panday,
2017) and not repeated herein. Lastly, since in post yield range residual deformation will
happen and the way to predict them in the idealized models is to generate a relationship
between the maximum displacement and residual deformation of the real building based on
data from cyclic pushover curves, and they will be subsequently used in the prediction of
residual displacements.
And this completes chapter – Literature Review.
61
CHAPTER 3
3 METHODOLOGY
3.1 Overview of the Research Methodology
According to the background stated in section 1.1 that accelerometers have been mounted
on some buildings in only some certain floors; for example, per the mentioned guidelines
the locations are at the top, middle and either lowest basement or ground floor level. The
code expected that this information can be utilized for seismic damage evaluation of those
buildings after experiencing a ground motion for the sake of life safety of residents.
However, in the field of SHM, there is no acceptable promising scheme to utilize solely
measured accelerations under a seismic event for seismic damage evaluation of that tall
building. Thus, the objectives of this study are set around the idea, and in this chapter, the
methodology to achieve them is to be explained, featuring the proposed seismic damage
evaluation scheme, the formulation of modal decomposition (MD) formula based on
measured accelerations from some floors, and procedure on how the proposed scheme along
with the proposed MD formula will be verified in this study. It should be noted that to
examine the inherent error originated in assumptions made in the formulation of this
approach, the error in practice including those from measurement and model updating
process are not taken into account in this research; thus, the scenario for effectiveness
verification is to be established and evaluated using a nonlinear numerical model modeled
in the finite element commercial software Perform3D in which the researcher can have
uncertainties in control.
Remark: readers should be noted that notations from chapter 3 and on will be consistent, yet
not the same as those from chapter 2 (those equations are just copied directly from the
reference sources without changes). The author of this thesis will set his own notation in the
sequential expression from one section to the next section, and the variables that have already
been explained in the aforementioned section (starting from section 3.1) may not be
explained again.
3.2 Introduction to the Proposed Seismic Damage Evaluation Scheme
This section is dedicated to introduction of the proposed method to evaluate seismic damage
of any considered tall building experiencing a seismic event. The only needed data under the
seismic event to perform the seismic damage evaluation is the measured acceleration signals
from 4 levels of the building as shown in Figure 3.2 (c). Those four levels correspond to the
lowest basement or ground floor (or wherever the floor acceleration is very less affected by
the motion of the building itself and represents the ground motion exciting the entire
building), the floors associated with the height around one-third, two-thirds and the highest
floor possible. As can be seen in the figure, 2 biaxial accelerometers are required to be
attached in each floor far away from the center of mass (except at the case of lowest floor
where only one accelerometer is required and able to be installed at any location) in order to
capture torsional response of the building (Karunkritkul, 2019) and the reason why they are
placed vertically as demonstrated in the figure is on the ground of observability (Carlo &
Giovanni, 2014). The proposed scheme is divided into two main parts: Part A and Part B.
Part A is to be done before any seismic event to prepare all necessary information to be
ready, and once an earthquake comes to hit the considered building, Part B, mainly Modal
62
Decomposition of measured accelerations will be conducted, and the results will be further
utilized in the detailed calculations (to be explained later in this chapter) to finally come up
with indicators for seismic damage evaluation.
3.2.1 Proposed Seismic Damage Evaluation Scheme: Part A
Part A – This part has to be completed, hopefully, before any potential ground motion comes
to shake the considered building. It can be looked at as preparation part. The main steps of
part A can be summarized as follows:
• A-1 Carefully creating the finite element model of the considered building (all main
parameters probably causing extreme discrepancies between dynamic behaviors of
real building and numerical model such as soil springs, boundary condition and so
on must be taken into consideration) and represent the possible nonlinear behavior
with any appropriate nonlinear idealization in the model.
• A-2 During the intact condition of the building, obtain identified dynamic
characteristics of the considered building (i.e., mode shapes and natural frequencies)
by full ambience measurement test (using reference – rover station technique briefly
mentioned in the section 2.2.4). Results from this step will be further used for both
model-updating process and Modal Decomposition of measured acceleration
responses.
• A-3 Updating the numerical model of the building by any technique available to
make it closely represent the considered existing building. Then the updated model
shall be used in the next step.
PS: Based on the past research done on model updating. It is convinced to the author
that updating selected properties of the structures by matching linear dynamic
properties of the numerical model to the identified ones is not sufficient. Since these
parameters only represent how the measured building may vibrate in the linear
range, that is why special consideration should be put on nonlinear behavior of the
updated numerical model as well.
• A-4 Realizing nonlinear hysteretic behaviors of the updated numerical model.
Starting by conducting Cyclic Pushover analyses on the updated numerical for each
mode of vibration in order to obtain every interested Cyclic Pushover curve (Base
Shear – roof displacement relationship). They can be mapped with any SDF
hysteresis behavior model such as Flag Shape or Modified Flag Shape models
(Panday, 2017). The input of each mode idealized SDF model is modal coordinate,
and the corresponding output is the modal inelastic force of that SDF. And as will be
explained later, these quantities can be used in the calculation of story force
prediction of the considered building under any seismic event.
3.2.2 Proposed Seismic Damage Evaluation Scheme: Part B
Part B – This part is to be done rapidly after any seismic event take places. Acceleration
signals measured by accelerometers attached on some certain floors will be proceeded
through the modal decomposition process to extract the quantities (called modal
accelerations) quantifying contributions from each individual mode, which will play a vital
role in the evaluation of seismic damage under the seismic event. The main steps of part B
can be summarized as follows:
63
• B-1 Proceeding with the measured acceleration signals
(For further details, please visit section 3.5)
Firstly, obtaining acceleration records under the seismic event measured through the
accelerometers mounted on some specific floors and either at the base or the lowest
basement (assumed to be the measured ground motion).
PS: In this thesis, the focus will be based on simultaneous acceleration records from
the stories about a third, two thirds and the highest story in which in each floor the
accelerometer lay out will be as shown in Figure 2.28 or Figure 3.2 (c). The reason
behind the set-up is totally based on observability mentioned in section 2.2.4.
Secondly, dealing with noisy signals using any filtering technique.
Thirdly, subtracting each filtered acceleration record of every floor by the filtered
base acceleration of its own direction.
PS: All records to be used in the developed Modal Decomposition method must be
relative accelerations, NOT absolute accelerations (as measured). And in this thesis,
the very big assumption made is that at either the base, the lowest basement or
wherever the base accelerations are taken, it is not contaminated by the motion of
the superstructure itself; thus, they represent the true ground motion shaking the
entire structure.
Finally, converting the filtered signals to acceleration time-histories associated with
the dynamic DOFs of the building.
• B-2 Applying the novel modal decomposition method (which requires identified
mode shapes from A-2) developed in this thesis in order to obtain the variables called
estimated modal acceleration, q̈̂ n . The modal decomposition based on measured floor
accelerations from limited floors is explained in the section 3.4.
• B-3 Relating estimated modal accelerations to estimate seismic demands of various
response types. By first, doubly integrating the estimated modal acceleration, q̈̂ n into
estimated modal coordinate, 𝑞̂𝑛 for every mode. These solved variables are then
utilized to estimate floor displacements, inter-story drifts and story forces of the
entire building. For further detail, please see section 3.6.
• B-4 Utilizing envelopes of calculated quantities as indicators in the seismic damage
evaluation, including maximum floor accelerations to check falling-over possibility
of objects with their available fragility curves, maximum inter-story drifts to check
the potential damage to non-structural element, i.e., infill-walls with their associated
available experimental tests and Maximum shear forces and moments in each
structural component to check their potential structural damage with their associated
available experimental tests.
PS: Firstly, in this study, those available resources as references to typical potential
damage are not compiled here in; secondly, local force prediction is not the goal of
this study. For further study on this, a literature (Mehmood, 2015) mentioned how
local force prediction can possibly be done.
The overview of the proposed seismic damage evaluation based on Modal Decomposition
approach is presented in Figure 3.1. It should be noted that in Figure 3.1(b), it implies that
step B-4 can be completed by either computer or humans (or both).
64
Figure 3.1 The flow charts of the proposed seismic damage evaluation: (a) part A, (b) part B
65
3.3 Modal Decomposition Based on Measured Accelerations from All Floors
In this subsection, Modal Decomposition approach based on measured acceleration records

from all floors (MD-All) will be explained. A building, being shaken under a ground motion,
vibrates (obviously) in lateral directions in a complex way, yet, in the context of elastic
analysis, it is actually known that the complex vibration of a building can actually be
described by a set of dynamic single degree of freedom governing equation as noted in the
equation (2.101). By working with these equations and known input, i.e., a ground motion,
this forward way of analysis method will lead to the prediction of combined complex
responses of the building – and the method is referred to as Modal Response History
Analysis (MRHA). However, if the condition is that the complex responses such as
accelerations of all floors at the centers of mass are known, the question is: “Can they be
decomposed back to their modal responses?” The answer is “Yes.” According to the
theoretical modal expansion of displacement vector in the equation (2.85) derived in general
for 2D simple structure (Chopra A. K., 2012), the modal coordinate (or modal acceleration
depending on which response type is used in the formula) of each mode can be theoretically
found once the mass matrix, all mode shapes and the associated theoretical dynamic degree
of freedoms (DOFs) are known. And the concept that is just shown is, generally, called
Modal Decomposition (MD) of acceleration responses (in theory).
Given that in reality any real building is defined in 3D and their dynamic degrees of freedom
can be simplified as those lumped, at the center of mass of each floor, including translational
DOF in x and y, and rotational DOF around z axis (if the building has J floors, the number
of its dynamic DOF is 3J), and to separate theoretical variables with those from
measurement, the formula used for Modal Decomposition of measured accelerations from
all floors (in 3D) is as follows
𝝓𝑇𝑛 𝑴𝒖̈ (𝑡)

𝑞̈̃𝑛 (𝑡) = (3.1)
𝑚𝑛
where 𝝓𝑛 is the nth mode shape in 3D (dimension = 3J ×1) obtained as a result of ambient
vibration test defined as in Eq.(2.111), M is the mass matrix of the structure corresponding
to the dynamic DOFs (the mass of each floor is lumped at the center of mass) with the
dimension of 3J x 3J containing seismic mass in x direction of each floor in its diagonal
element for J entries; in y direction for another J entries; and lastly rotational masses for J
entries explained in the section 2.3.1, 𝑚𝑛 is the nth mode modal mass defined as in Eq.(2.82),
𝒖̈ (𝑡) is the relative acceleration vector time histories associated with the dynamic DOFs of
the building (at the center of mass) obtained as briefly explained in the step B-2 in the section
3.3 and will be thoroughly explained in the subsequent section (dimension = 3J × I, where I
is the signal length in the unit of sample), and 𝑞̈̃𝑛 (𝑡) is called modal acceleration estimated
based on measured accelerations from all floors which is made outstanding with the accent
mark Tilde. It should be noted that this formula is theoretically supported within the elastic
range of the building.
The thing is, the equation above is derived for the theoretical sake and may not be usable in
practice. Since firstly, in reality how can one measure the rotational response or even mere
translational ones yet exactly at the center of mass (where normally, measurement of
responses cannot be conducted at the that very exact location); secondly, none of any tall
building ever has permanent full set up of measurement tools (i.e., accelerometers attached
66
on all floors), the above recently cited equation becomes most likely obsolete in practice.
That is why right after this, the formulation of useful and practical modal decomposition
formula is to be proposed in the next section of this chapter; however, Modal Decomposition
based on measured accelerations from all floors will be used a lot in this study as the
reference modal decomposition approach.
3.4 Formulation of the Modal Decomposition Formula Based on Measured Limited

Number of Acceleration Records
In this subsection, the Modal Decomposition approach based on measured limited number
of floor acceleration records (MD-Lim) will be formulated. As mentioned in the section 3.2,
the scenario is that there are only 4 levels where accelerometers are placed. Through the
detailed steps to be explained in the next section, relative accelerations associated with the
building’s dynamic DOFs of some certain floors can be obtained.
Thus, let us start with the fact that relative acceleration records associated with their DOFs
from three floors have already been obtained, and the derivation begins with the following
equation, theoretically valid within the elastic range
3×𝑁
𝒖̈ (𝑡) = ∑ 𝑞̈ 𝑛 (𝑡)𝝓𝑛 (3.2)

n=1
Where 𝒖̈ (𝑡) is the relative acceleration vector time histories associated with the dynamic
DOFs of the building (at the center of mass); the dimension of it is equal to 3J × I, where I
is the signal length in the unit of sample), 𝝓𝑛 is the nth mode shape in 3D (dimension =
3J ×1) obtained as a result of ambient vibration test defined as in Eq.(2.111), 𝑞̈ 𝑛 (𝑡) is the
exact modal acceleration time histories of the nth mode (dimension = 1 × I), and 𝑁 is the
important vibrational modes of the building.
Given that the results of the past literatures done on seismic response prediction using
simplified analysis methods (Najam, 2017) signifies that the contribution from the first 3
dominant vibrational modes is sufficient to predict the complex responses of tall buildings
specified in the literature. Thus, Eq. (3.2) is simplified into the equation down below
3×3
𝒖̈ (𝑡) = ∑ 𝑞̈ 𝑛 (𝑡)𝝓𝑛 (3.3)

n=1
In comparison with Eq. (3.2), “3×N” is changed into “3 × 3” for the dynamic behavior of
any tall building under a ground motion can be expressed with the movement in 3 directions,
and with the contribution from 3 modes required, the number of necessary modes become 9.
If the equation above is valid, the reduced form of it is expressed as follows
3×3
𝒖̈ 𝑅 (𝑡) = ∑ 𝑞̈ 𝑛 (𝑡)𝝓𝑛𝑅 (3.4)

n=1
𝒖̈ 𝑅 is the reduced acceleration vector (dimension = 9 × I) and equal to 〈𝒖̈ 𝑥𝑅 𝒖̈ 𝑦𝑅 𝒖̈ 𝜃𝑅 〉𝑇 ,

where each sub vector keeps three relative floor acceleration time histories at three centers
of mass of their own directions; 𝑞̈ 𝑛 is still the exact modal acceleration time histories of the
67
nth mode (dimension = 1 × I); 𝝓𝑛𝑅 is defined as reduced mode shape, (dimension = 9 × 1)
and equal to 〈𝝓𝑛𝑥𝑅 𝝓𝑛𝑦𝑅 𝝓𝑛𝜃𝑅 〉𝑇 , where each sub vector keeps three mode shape values
at the centers of mass corresponding to the floors (the accelerometers are attached) in their
own directions. With this equation, the derivation from equation (2.82) to (2.85) cannot be
directly applied with since the orthogonality between each reduced mode shape with respect
to the mass matrix does not exist, or mathematically, 𝝓𝑻𝒏𝑹 𝑴𝝓𝒓𝑹 is not even valid. So, in
order to start with equation (3.4) , there must be a matrix (denoted here in this study to as A)
with the dimension of 9 × 9 being able to satisfy the following expression, and where n≠r.
𝝓𝑻𝑛𝑅 𝑨𝝓𝑟𝑅 = 𝟎 (3.5)
The above equation says that all mode shapes are orthogonal to each other with respect to
the matrix A, which is that 𝝓𝑻𝒏𝑹 𝑨𝝓𝒏𝑹 will only yield non-zero values where n=1,2,3…9.
And in line with the equation (2.80), the expression leads to another constant scalar variable,
𝐴𝑛 = 𝝓𝑇𝑛𝑅 𝑨𝝓𝑛𝑅 . (3.6)
The expression in both Eq.(3.5) and Eq.(3.6) are the main conditions that the matrix A must
satisfy, yet some more properties are to be added for the sake of, mathematically speaking,
simplicity. Firstly, the parameters are to be set up such that the matrix A becomes symmetric
(𝑥𝑛𝑚 = 𝑥𝑚𝑛 where n and m = 1,2,3…,9) - this is done to reduce the number of variables.
Secondly, diagonal elements in the matrix is to be assigned as non-zero number. Thirdly, set
𝑥11 = 1 which signifies that all other elements are normalized with this entry – these comes
from experience in developing the matrix A because, in some certain cases, the entire matrix
may turn out to be zero matrix without these additional properties. To start determining the
matrix A, let us set A to be as follows
𝑥11 𝑥12 𝑥13 𝑥14 𝑥15 𝑥16 𝑥17 𝑥18 𝑥19
𝑥12 𝑥22 𝑥23 𝑥24 𝑥25 𝑥26 𝑥27 𝑥28 𝑥29
𝑥13 𝑥23 𝑥33 𝑥34 𝑥35 𝑥36 𝑥37 𝑥38 𝑥39
𝑥14 𝑥24 𝑥34 𝑥44 𝑥45 𝑥46 𝑥47 𝑥48 𝑥49
𝑨 = 𝑥15 𝑥25 𝑥35 𝑥45 𝑥55 𝑥56 𝑥57 𝑥58 𝑥59 (3.7)
𝑥16 𝑥26 𝑥36 𝑥46 𝑥56 𝑥66 𝑥67 𝑥68 𝑥69
𝑥17 𝑥27 𝑥37 𝑥47 𝑥57 𝑥67 𝑥77 𝑥78 𝑥79
𝑥18 𝑥28 𝑥38 𝑥48 𝑥58 𝑥68 𝑥78 𝑥88 𝑥89
[𝑥19 𝑥29 𝑥39 𝑥49 𝑥59 𝑥69 𝑥79 𝑥89 𝑥99 ]
These variables in the matrix are later to be solved such that they will satisfy Eq. (3.5), and
if not, they are going to make the expression erroneous. So, let us set up 𝑒𝑖 to be any ith error
equation from any expression 𝝓𝑻𝑛𝑅 𝑨𝝓𝑟𝑅 where n≠r and i is a variable indicating which error
equation it is, ranging from 1 to 36 since 𝝓1𝑅 must be orthogonal to the other eight reduced
mode shapes, 𝝓2𝑅 must be orthogonal to the other seven ones, and so on. So far, 𝑒𝑖 ,
containing 44 variables, is a linear function of 𝑥𝑚𝑛 . Then, let us define 𝐸𝑖 as square of error
or 𝑒𝑖2 , and is mathematically shown as
𝐸𝑖 = (𝝓𝑻𝑛𝑅 𝑨𝝓𝑟𝑅 )2 . (3.8)
68
At this point, 𝐸𝑖 is the second-degree polynomial with 44 variables. In order that these
variables shall not be solved individually and just satisfy each individual equation. The
following equation is the key.
36
𝐸 = ∑ 𝑒𝑖 (3.9)
𝑖=1
Thus, all variables to be solved have to simultaneously satisfy the above equation at the same
time. And to determine them, least square of error scheme is chosen. Take partial derivative
of the equation (3.9) with respect to every single variable and equate each of them to zero.
And typically, it can be shown as
𝜕𝐸
=0 (3.10)
𝜕𝑥𝑚𝑛
Now with these 44 linear equations, one can solve all 44 variables at the same time and
substitute them back to the equation (3.7) to acquire the matrix A. Then using the Eq.(3.6),
𝐴𝑛 can be determined. In analogy to the original formula in the equation (2.85), the modal
decomposition formula based on measured limited number of acceleration records as the
input is expressed as
𝝓𝑇𝑛𝑅 𝑨𝒖̈ 𝑅 (𝑡)

𝑞̈̂𝑛 (𝑡) = (3.11)
𝐴𝑛
where 𝑞̈̂𝑛 (t) is called modal acceleration estimated based on measured limited number of
acceleration records which is made outstanding with the accent mark Circumflex. Do not be
confused with 𝑞̈̃𝑛 (𝑡) defined in Eq.(3.1). The matrix A can be called Orthogonal Filter matrix
since its duty is to filter out the accelerations contributed from other modes out, so the modal
decomposition approach based on Eq.(3.11) can also be referred to as Orthogonal Filter
technique.
3.5 How to Convert Measured Accelerations to the Relative Accelerations

Associated with the Dynamic Degrees of Freedom of the building (for Modal
Decomposition Approach)
In the previous section, the practical modal decomposition formula is formulated, and
despite the fact that the above equation does not require as many relative accelerations
associated with dynamic DOFs as shown in Figure 3.2 (a) as of the original formula does, it
is necessary to be remined that reduced acceleration vector, 𝒖̈ 𝑹 still requires relative
accelerations associated with some certain dynamic DOFs, schematically shown in Figure
3.2 (b). However, these theoretical signals cannot be obtained directly from the measurement
but to be obtained indirectly by, at first, following the practice. Accelerometers have to be
attached on the considered building as mentioned earlier in section 3.2, and the absolute
acceleration signals corresponding to DOFs shown in Figure 3.2 (c) are recorded under any
seismic event. The signals at the base (or the lowest basement), recorded by a biaxial
accelerometer are assumed to be the measured ground acceleration in x direction 𝑢̈ 𝑔𝑥 (𝑡) and
y direction 𝑢̈ 𝑔𝑦 (𝑡). For the rest three floors on top, sensor locations are similar to that in the
study of Karunkritkul (2019). In a floor, two biaxial accelerometers are to be, preferably,
mounted as far as possible from the center of mass for the sake of capturing torsional
response of motion under the seismic event. Then, all signals must be dealt with noise
filtering caused by filed measurement using any filtering technique, after that relative
69
accelerations associated with DOFs of sensors can be obtained by subtracting the measured
ground motion (only the component of the same direction) out.
Figure 3.2 Degrees of Freedom of any tall building stated in this study: (a) theoretical
dynamic DOFs of all floors, (b) theoretical dynamic DOFs of some floors, (c) typical DOFs
of some certain floors required for the measurement set up
Figure 3.3 Movement of a floor relative to its center of mass

Predicated upon the assumption of rigid floor movement, the displacements at and rotation
around the center of mass 𝑋𝑐𝑚 , 𝑌𝑐𝑚 and 𝜃𝑐𝑛 can be determined from the displacements at
any two locations. Following the schematic representation of floor movement in Figure 3.3,
the example of relationships between top left corner displacements (denoted by 𝑋𝑙𝑐 and 𝑌𝑙𝑐 ),
70
bottom-right corner displacements (denoted by 𝑋𝑟𝑐 and 𝑌𝑟𝑐 ) and movement of the center of
mass are as follows:
𝑋𝑙𝑐 (𝑡) = 𝑋𝑐𝑚 (𝑡) − 𝑟𝑦𝑙𝑐 · 𝜃𝑐𝑚 (𝑡)

𝑌𝑙𝑐 (𝑡) = 𝑌𝑐𝑚 (𝑡) − 𝑟𝑥𝑙𝑐 · 𝜃𝑐𝑚 (𝑡) (3.12)
𝑋𝑟𝑐 (𝑡) = 𝑋𝑐𝑚 (𝑡) + 𝑟𝑦𝑟𝑐 · 𝜃𝑐𝑚 (𝑡)
𝑌𝑟𝑐 (𝑡) = 𝑌𝑐𝑚 (𝑡) + 𝑟𝑥𝑟𝑐 · 𝜃𝑐𝑚 (𝑡)
where 𝑟𝑦𝑙𝑐 , 𝑟𝑥𝑙𝑐 , 𝑟𝑦𝑟𝑐 and 𝑟𝑥𝑟𝑐 are self-explanatory (by looking at Figure 3.3), and be noted
that all signs in the equations are depending on location of the considered sensors which
means that if one of the sensors is, for example, at the top right, some equations will be
invalid for the case; to readers, do not use them blindly. By applying second derivatives to
both sides of all relationships individually to the equation (3.12) , the following relationships
regarding accelerations become
𝑋̈𝑙𝑐 (𝑡) = 𝑋̈𝑐𝑚 (𝑡) − 𝑟𝑦𝑙𝑐 · 𝜃̈𝑐𝑚 (𝑡)

𝑋̈𝑟𝑐 (𝑡) = 𝑋̈𝑐𝑚 (𝑡) + 𝑟𝑦𝑟𝑐 · 𝜃̈𝑐𝑚 (𝑡) (3.13)
̈ (𝑡) − 𝑟𝑥𝑙𝑐 · 𝜃̈𝑐𝑚 (𝑡)
̈ (𝑡) = 𝑌𝑐𝑚
𝑌𝑙𝑐
̈ (𝑡) = 𝑌𝑐𝑚
𝑌𝑟𝑐 ̈ (𝑡) + 𝑟𝑥𝑟𝑐 · 𝜃̈𝑐𝑚 (𝑡).
Thus, if the floor acceleration records at both corners in both x and y directions plus location
of the center of mass in each floor are known, one can determine the theoretical dynamic
DOF time histories at the center of mass by solving the first and second pairs of relationships
noted in the equation (3.13).
3.6 How to Relate Estimated Modal Accelerations to the Global Seismic Demands
(for Modal Decomposition Approach)
After, the estimated modal accelerations (no matters either 𝑞̈̃𝑛 (𝑡) or 𝑞̈̂𝑛 (𝑡)) are obtained.
They can be used to calculate various kinds of responses under the seismic event, for each
one of them represents the contribution of its own vibrational mode under the seismic
excitation. But before jumping into equations, one step is required that is to determine
estimated modal coordinates first, for they will also be used in the following equations. A
number of numerical double integration processes must be performed. If the nth mode modal
acceleration time history estimated based on measured accelerations from all floors, 𝑞̈̃𝑛 (𝑡)
is used in the double integration process, the result is the time history of the estimated modal
coordinate symbolized as 𝑞̃𝑛 (𝑡) while if the estimation is based on measured accelerations
of some floors, the result time history is also the estimated modal coordinate, yet symbolized
as 𝑞̂𝑛 (𝑡). As mentioned, the accent marks on the variables signify which modal
decomposition formula they originally are derived from, and will be consistently used
throughout this thesis.
3.6.1 Demand Estimation Under the Elastic Range
In the elastic range of a tall building, the seismic responses can be expressed in terms of
modal contribution, for the concept of mode of vibration is still theoretically valid. Thus, if
somehow the considered building is known to be within the linearly elastic range, the
response prediction of various types of the existing building under a ground motion can be
done using the equations similar to those used in Modal Response History Analysis
71
(MRHA), with some adjustments in notation to signify that these equations are to be used in
the practical application. Those listed below are for the case of MD-Lim, and for the case of
MD-All these equations can be substituted with associated variables.
Estimation of the displacement time-history vector

3×3
̂ (t) = ∑ 𝝓𝑛 𝑞̂𝑛 (𝑡)

𝒖 (3.14)
𝑛 =1
̂ (t) is the estimated displacement time history vector (dimension = 3J × I).
where 𝒖
Estimation of the acceleration time-history vector
3×3
𝒖̈̂ (𝑡) = ∑ 𝝓𝑛 𝒒̈̂𝑛 (𝑡) (3.15)

𝑛 =1
where 𝒖̈̂ (𝑡) is the estimated acceleration time-history vector (dimension = 3J × I). It should
be noted here that the estimated accelerations here are relative ones, and if estimated absolute
floor accelerations are needed the formula below can be used.
̂𝑥 (𝑡) = 𝒖̈̂ 𝒙 (𝑡) + 𝑢̈ 𝑔𝑥 (𝑡)𝒊

𝒂
(3.16)
̂𝑦 (𝑡) = 𝒖̈̂ 𝒚 (𝑡) + 𝑢̈ 𝑔𝑦 (𝑡)𝒊
𝒂
Where 𝒂̂𝑥 (𝑡) and 𝒂 ̂𝑦 (𝑡) are the estimated absolute floor acceleration time-history vectors in
x and y directions – these quantities are to be later used in seismic damage evaluation of
objects, 𝒖̈̂ 𝑥 (𝑡) and 𝒖̈̂ 𝑦 (𝑡) are the estimated relative floor acceleration time-history vector in
x and y directions (where 𝒖̈̂ (𝑡) = 〈𝒖̈̂ 𝑥 (𝑡) 𝒖̈̂ 𝑦 (𝑡) 𝒖̈̂ 𝜃 (𝑡)〉𝑇 ), 𝒊 is the influence vector
explained in section 2.3.2 (dimension = J × 1, where J is the total number of stories)
Estimation of the inter-story drift time-history of the ith floor
𝑢̂𝑖𝑥 (𝑡) − 𝑢̂(𝑖−1)𝑥 (𝑡)

Inter story drift (in x direction) =
ℎ𝑖
(3.17)
𝑢̂𝑖𝑦 (𝑡) − 𝑢̂(𝑖−1)𝑦 (𝑡)
Inter story drift (in y direction) =
ℎ𝑖
where 𝑢̂𝑖𝑥 (𝑡) and 𝑢̂𝑖𝑦 (𝑡) are the displacement time-histories of the ith floor in x and y
directions, respectively, and hi is the height of that story.
72
Estimation of the shear-force story time-history of the ith floor contributed from the nth mode
𝐽
𝑉̂𝑖𝑛𝑥 (𝑡) = 𝜔𝑛2 𝑞̂𝑛 (𝑡) ∑ 𝑚𝑗 𝜙𝑛𝑥𝑗

𝑗=𝑖
(3.18)
𝐽
𝑉̂𝑖𝑛𝑦 (𝑡) = 𝜔𝑛2 𝑞̂𝑛 (𝑡) ∑ 𝑚𝑗 𝜙𝑛𝑦𝑗

𝑗=𝑖
where 𝑉̂𝑖𝑛𝑥 (𝑡) and 𝑉̂𝑖𝑛𝑦 (𝑡) are the shear-story time-histories of the ith floor contributed from
the nth mode in x and y directions, respectively, J is the total number of stories, mj is the
lumped mass of that story, 𝜙𝑛𝑥𝑗 and 𝜙𝑛𝑦𝑗 are the mode shape values of the nth mode
associated with the ith floor level in x and y directions, respectively (obtained from ambient
vibration test), and 𝜔𝑛 is the natural frequency (rad/s) of the nth mode (obtained from ambient
vibration test). It should be noted that responses in x direction are mainly contributed from
the contribution from their 3 dominant modes in x direction, yet the contribution from the 3
dominant modes in y and the other 3 modes of torsion as well should be included as well,
especially, for the case of coupled-behavior building (but in the case of pure sway building,
only contribution from 3 dominant modes in x direction is enough, so the number in such
case is simply just “3”). But in general, 9 modes are required.
Estimation of the shear-force story time-history of the ith floor

3×3
𝑉̂𝑖𝑥 (𝑡) = ∑ 𝑉̂𝑖𝑛𝑥 (𝑡)

𝑛 =1
(3.19)
3×3
𝑉̂𝑖𝑦 (𝑡) = ∑ 𝑉̂𝑖𝑛𝑦 (𝑡)

𝑛 =1
where 𝑉̂𝑖𝑥 (𝑡) and 𝑉̂𝑖𝑦 (𝑡) are the shear-force story time-histories of the ith floor in x and y
directions.
Estimation of the moment story time-history of the ith floor contributed from the nth mode
𝐽
̂𝑖𝑛𝑥 (𝑡) =
𝑀 𝑤𝑛2 𝑞̂𝑛 (𝑡) ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑥𝑗
𝑗=𝑖
(3.20)
𝐽
̂𝑖𝑛𝑦 (𝑡) = 𝑤𝑛2 𝑞̂𝑛 (𝑡) ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑦𝑗
𝑀
𝑗=𝑖
where 𝑀 ̂𝑖𝑛𝑥 (𝑡) and 𝑀̂𝑖𝑛𝑦 (𝑡) are the moment story time-histories of the ith floor contributed
from the nth mode in x and y directions, respectively, 𝐻𝑗 and 𝐻𝑖 are the height of the jth and
ith floors, respectively.
73
Estimation of the moment story time-history of the ith floor
3×3
̂𝑖𝑥 (𝑡) = ∑ 𝑀
𝑀 ̂𝑖𝑛𝑥 (𝑡)
𝑛 =1
(3.21)
3×3
̂𝑖𝑦 (𝑡) = ∑ 𝑀
𝑀 ̂𝑖𝑛𝑦 (𝑡)
𝑛 =1
̂𝑖𝑥 (𝑡) and 𝑀

where 𝑀 ̂𝑖𝑦 (𝑡) are the moment time-histories of the ith floor in x and y directions.
3.6.2 Demand Estimation Under the Inelastic Range
Given that when the condition of a tall building is beyond its elastic limit, theoretically
speaking, the equations listed for the case of elastic range demand estimation are all invalid
and cannot give theoretically exact results anymore. However, based on the original study
of Chopra & Goel (2001), the concept of mode of vibrations can reasonably be applied for
demand prediction beyond the linearly elastic range (the study was done with a numerical
model of a steel building), and the concept is proven promising by the study of many
researchers (Najam, 2017; Mehmood, Warnitchai, & Suwansaya, Seismic Evaluation of Tall
Buildings Using a Simplified but Accurare Analysis Procedure, 2018; Panday, 2017;
Vasanthapragash, 2018) on the seismic demand prediction using a simplified analysis
method on RC-core walls tall buildings. The method is called Uncoupled Modal Response
History Analysis (UMRHA) which is formulated based on SDF governing equation and will
be referred to in this thesis as SDF-UMRHA. On the ground of SDF-UMRHA assumption
the modal deflected shapes along with modal story force patterns of the considered building
will not change significantly due to seismic damage, Eq.(3.14), (3.15) and (3.17) can still be
used for response predictions of the same types even within the inelastic range. For Modal
Decomposition approach (either MD-All or MD-Lim), however, when it comes to force
prediction estimated based on SDF-UMRHA formulation, there is one single element that is
still missing which is inelastic SDF force of the nth mode, 𝐹𝑠𝑛 (𝑡)/𝐿𝑛 . To illustrate Eq.(3.22)
is used for the calculation of shear-force of the ith story contributed from the nth mode in x
direction in SDF-UMRHA procedure (in the case of a symmetric building causing Pure-
sway behavior).
𝐽
𝑉̅𝑖𝑛𝑥 (𝑡) = 𝛤𝑛𝑥 𝐹𝑠𝑛 (𝑡)/𝐿𝑛 ∑ 𝑚𝑗 𝜙𝑛𝑥𝑗 (3.22)

𝑗=𝑖
where 𝑉̅inx (𝑡) is shear-force time-history of the ith story contributed from the nth mode
predicted using SDF-UMRHA (or SDF-MRHA) formula and 𝛤𝑛𝑥 is a modal participation
factor conceptually explained by Eq.(2.92). As can be seen that if the formula, for instance,
in Eq.(3.22) or other formulae noted elsewhere of SDF-UMRHA are to be adjusted and
implemented for story force prediction in the modal decomposition approach (either MD-
All or MD-Lim), estimated inelastic SDF force of the nth mode must be, somehow, known
first. The thing is, right now based on the flow of this chapter, the known modal responses
are only estimated modal accelerations and estimated modal coordinates. So, there are two
main possible ways to move on for force prediction. The first one is continuing prediction
scheme based on SDF-UMRHA formula (that means to take the assumption used in SDF-
74
UMRHA to be valid) and another one is to find any other methods. In the next subsections,
the author of this thesis is proposing three possible methods.
3.6.2.1 Story Force Prediction Based on Estimated Floor Acceleration Time History
Vector
The input required in this method is the estimated “relative” floor acceleration time history
vector which has already been determined through either MD-All or MD-Lim and measured
ground motion. This is one of the time-history based methods being proposed, and the
method itself does not depend on the concept of mode of vibration in the inelastic range.
The derivation shown below, to illustrate the concept, is for the simple case of Pure-sway
building being excited in x direction. Given that the governing equation of the numerical
model under a measured ground motion only having x component (defined consistently from
this point on as 𝒖̈ 𝑔𝑥 (𝑡); also, the y component ground motion is 𝒖̈ 𝑔𝑦 (𝑡)) as an input is
𝑴𝑥 𝒖̈ 𝑥 (𝑡) + 𝑪𝑥 𝒖̇ 𝑥 (𝑡) + 𝑭𝑟𝑒𝑠,𝑥 (𝑡) = −𝑴𝑥 𝒊𝒖̈ 𝑔𝑥 (𝑡) (3.23)
where 𝑴𝑥 (dimension = J × J, where J is the number of total stories) is the mass matrix in x
direction (containing the lumped mass of each floor in its diagonal element), 𝑪𝑥 is the
damping matrix in x direction (dimension = J × J), 𝑭𝑟𝑒𝑠,𝑥 (𝑡) is the resisting force time-
history vector in x direction (dimension = J × I, where I is the length of the time-history),
𝒖̈ 𝑥 (𝑡) is the acceleration time-history vector (dimension = J × I), 𝒖̇ 𝑥 (𝑡) is the velocity time-
history vector (dimension = J × I), and 𝒊 is the influence vector explained in section 2.3.2
(dimension = J × 1). By ignoring the effect of damping force term, the resisting force (of the
numerical model) can be recalculated approximately using the equation as follows
𝑭𝑟𝑒𝑠,𝑥 (𝑡) ≈ −𝑴𝑥 𝒊𝒖̈ 𝑔𝑥 − 𝑴𝑥 𝒖̈ 𝑥 (𝑡) (3.24)
Predicated upon Eq.(3.24), the similar equations used for force prediction in the modal
decomposition approach (either for MD-All or MD-Lim), illustrated for the case of MD-
Lim, is shown below
𝑭̂ 𝑟𝑒𝑠,𝑥 (𝑡) ≈ −𝑴𝑥 𝒊𝑢̈ 𝑔𝑥 (𝑡) − 𝑴𝑥 𝒖̈̂ 𝒙 (𝑡)
(3.25)
̂ 𝑟𝑒𝑠,𝑦 (𝑡) ≈ −𝑴𝑦 𝒊𝑢̈ 𝑔𝑦 (𝑡) − 𝑴𝑦 𝒖̈̂ 𝒚 (𝑡)
𝑭
where 𝑭̂ 𝑟𝑒𝑠,𝑥 (𝑡) and 𝑭

̂ 𝑟𝑒𝑠,𝑦 (𝑡) are estimated resisting force time-history vectors in x and y
directions, 𝑢̈ 𝑔𝑥 (𝑡) and 𝑢̈ 𝑔𝑦 (𝑡) are the measured ground motion (taken from the lowest
basement level) in x and y direction, 𝒖̈̂ 𝒙 (𝑡) and 𝒖̈̂ 𝒚 (𝑡) are estimated acceleration time-history
vectors (the sub-vectors are extracted from 𝒖̈̂ which composes of x, y and around z
acceleration time histories, respectively). It should be noted that Eq.(3.25) can be used for
those buildings having Coupled behavior as well, for the 3D coupled mode effect is taken
care of already in Eq.(3.15). Then, these resisting force vectors can be used to calculate
estimated shear-force and moment of the ith story. The examples of them can be found in
Eq.(3.26) and (3.27) for the case of force prediction in x direction.
𝐽
𝑉̂𝑖𝑥 (𝑡) = ∑ 𝐹̂𝑟𝑒𝑠,𝑥𝑗 (𝑡) (3.26)

j=i
where 𝑉̂𝑖𝑥 (𝑡) is estimated shear-force story time-history of the ith floor and 𝐹̂𝑟𝑒𝑠,𝑥𝑗 (𝑡) is the
resisting force time-history of the jth floor (J = total number of stories) in x direction.
75
𝐽
̂𝑖𝑥 (𝑡) = ∑ 𝐹̂𝑟𝑒𝑠,𝑥𝑗 (𝑡)(𝐻𝑖 − 𝐻𝑗 )
𝑀 (3.27)
𝑖=𝑗
̂𝑖𝑥 (𝑡) is estimated moment story time-history of the ith floor in x direction.
where 𝑀
3.6.2.2 Story Force Prediction Based on Inelastic SDF Relationships

The input of this method is estimated modal accelerations (either from MD-All or MD-Lim)
and modal velocities (through the process of the integration of their own estimated modal
accelerations), and the measured ground motion. This is also one of the time-history based
methods being proposed, and the method itself assumes that the concept of mode of vibration
in the inelastic range exists. For this method, inelastic SDF forces will be estimated first
following inelastic SDF relationships in Eq.(2.125) or (2.127), the estimated inelastic SDF
force, to illustrate the case where ground motion only has x direction component, is shown
in the equation below
𝐹̂𝑠𝑛 (𝑡) ̂̈ (𝑡) − 2𝜉 𝜔 𝐷

̂̇
= −𝑢̈ 𝑔𝑥 (𝑡) − 𝐷 𝑛 𝑛 𝑛 𝑛 (𝑡) (3.28)
𝐿𝑛𝑥
where 𝐿𝑛𝑥 = 𝝓𝑇𝑛𝑥 𝑴𝑥 𝒊 (the influence vector 𝒊 is the vector having all entries equal to 1 with
̂̈ (𝑡) is estimated modal acceleration in the “D” coordinate
the dimension equal to J × 1). 𝐷 𝑛
which is determined from 𝑞̈̂𝑛 (𝑡) predicated upon Eq.(2.102). 𝜔𝑛 and 𝜉𝑛 are the natural
frequency (rad/s) and damping ratio of the nth mode (obtained from ambient vibration test).
Similar to SDF-UMRHA formulae for force prediction, story forces under the case - the
ground motion only has x direction component - are calculated as follows
𝐽
𝐹̂𝑠𝑛 (𝑡)
𝑉̂𝑖𝑛𝑥 (𝑡) = 𝛤𝑛𝑥 ( ) ∑ 𝑚𝑗 𝜙𝑛𝑥𝑗
𝐿𝑛𝑥
𝑗=𝑖
𝐽 (3.29)
𝐹̂ (𝑡)
̂𝑖𝑛𝑥 (𝑡) = 𝛤𝑛𝑥 ( 𝑠𝑛 ) ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑥𝑗
𝑀
𝐿𝑛𝑥
𝑗=𝑖
Where 𝑉̂𝑖𝑛𝑥 (𝑡 ) and 𝑀 ̂𝑖𝑛𝑥 (𝑡) are estimated shear-force story and moment time-histories of
the i floor contributed from the nth mode, and 𝛤𝑛𝑥 = 𝐿𝑛𝑥 / 𝑚𝑛 (modal mass, 𝑚𝑛 =
th
𝝓𝑇𝑛 𝑴𝝓𝑛 ). Also, do not confuse M with 𝑴𝑥 , M is the global mass matrix with the dimension
of 3J × 3J while 𝑴𝑥 has the dimension of J × J. Readers should be noted that modal mass
is calculated based on “all-direction properties,” and not only a single direction one while
the calculation of 𝛤𝑛𝑥 , 𝛤𝑛𝑦 , 𝐿𝑛𝑥 and 𝐿𝑛𝑦 are opposite. For the case where the measured ground
motion has two components, i.e., x and y, the inelastic SDF relationships cannot be expressed
in “D” coordinate but “q” coordinate (please refer to section 2.3 on how MRHA under 2
components ground motion can be performed). The Eq.(3.28) is alternatively turned into a
different form as shown below
= − (𝛤𝑛𝑥 𝑢̈ 𝑔𝑦 (𝑡) + 𝛤𝑛𝑦 𝑢̈ 𝑔𝑦 (𝑡)) − 𝑞̈̂𝑛 (𝑡) − 2𝜉𝑛 𝜔𝑛 𝑞̇̂𝑛 (𝑡) (3.30)
𝑚𝑛
76
where 𝛤𝑛𝑦 = 𝐿𝑛𝑦 / 𝑚𝑛 . By analogy with Eq.(3.29), the following equations shows story
forces in x direction for the case where inelastic SDF force of the nth mode as a result of 2
components ground motion, predicted using MD-Lim, are
𝐽
𝑉̂𝑖𝑛𝑥 (𝑡) = ∑ 𝑚𝑗 𝜙𝑛𝑥𝑗
𝑚𝑛
𝑗=𝑖
𝐽
(3.31)
𝐹̂ (𝑡)
̂𝑖𝑛𝑥 (𝑡) = 𝑠𝑛 ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑥𝑗
𝑀
𝑚𝑛
𝑗=𝑖
For the calculations of estimated story forces in y direction contributed from the nth mode
are noted in Eq. (3.32)
𝐽
𝑉̂𝑖𝑛𝑦 (𝑡) = ∑ 𝑚𝑗 𝜙𝑛𝑦𝑗
𝑚𝑛
𝑗=𝑖
𝐽
(3.32)
𝐹̂ (𝑡)
̂𝑖𝑛𝑦 (𝑡) = 𝑠𝑛 ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑦𝑗
𝑀
𝑚𝑛
𝑗=𝑖
Then Eq.(3.19) and (3.21) shall be used next to calculate total story forces of the ith floor.
3.6.2.3 Story Force Prediction Based on the Updated Numerical Model

This third proposed method actually requires a big additional step in the process of seismic
evaluation based on modal composition either MD-All or MD-Lim formulation. As can be
seen that starting from the estimation of displacement, inter-story drift and acceleration of
all levels, the only needed data is simply measured acceleration data under the seismic event
and identified dynamic properties of the considered building (intact condition). If the force
prediction scheme implemented is either per section 3.6.2.1 or 3.6.2.2, nothing is
additionally required more but to just proceed with the already mentioned data, and then
story forces can be subsequently determined. Up to this point, seismic damage evaluation
can be performed utilizing the estimated demands. On the other hand, if the force prediction
scheme per this subsection is selected, the earlier mentioned additional thing is the updated
numerical model of the considered building. In general, when mentioning about updating a
numerical model, they are all about updating linear properties of the numerical model such
that through the Modal Analysis of that updated numerical model, calculated dynamic
properties (i.e., natural frequencies and mode shapes) of important modes match those
identified from the real building determined from ambient vibration test (Kaynardag &
Soyoz, 2017; Maria & Celebi, 2019). The key about model updating is that sensitive
parameters selected (after sensitivity check) are to be either deterministically or
probabilistically changed so that dynamic properties are closer to the targets. However, the
results from literature showed that they are never exactly the same; even if they were,
“equivalent” linear behavior can only be made matched, and it does not guarantee at all the
performance of the updated numerical model for inelastic demand prediction.
In this study, the concern regarding error caused by model updating will not be taken into
account, but the author’s concentration will be on the effectiveness of using “perfectly”
77
updated numerical for force prediction. The numerical model is perfect because it was
excited to generate seismic responses and used as the “already” updated numerical model.
Except additional requirement of updated numerical model, the rest of the required input of
this method is the estimated modal coordinates (either from MD-All or MD-Lim). As a
requirement of performing SDF-UMRHA procedure and also of this method, Cyclic
Pushover analyses on the updated numerical model must be conducted beforehand for a few
important of vibrational modes; as a results, Cyclic Pushover curves (see Figure 2.37 as a
reference) are obtained and turned into inelastic SDF force-deformation relationship using
Eq.(2.121). Then any hysteretic SDF model such as Modified Flag Shape (Panday, 2017)
can be employed to map the hysteretic behaviors of the inelastic SDF force-deformation
relationships of every mode caused by typical cyclically-increasing-static load (proportional
to its own modal inertia force pattern). These results in “idealized nonlinear SDF systems”
(each system is for each mode). A system can be mathematically expressed in the form of a
function as shown below
𝐹𝑠𝑛 (𝑡𝑖 ) = 𝑓(𝑞𝑛 (𝑡𝑖 ), 𝑠𝑖𝑔𝑛𝑞̇ (𝑡𝑖 ), 𝑞𝑛,𝑀𝑎𝑥 (𝑡𝑖 )) (3.33)

The parameters in Eq.(3.33) are defined as those of SDF-UMRHA, where 𝑞𝑛 (𝑡𝑖 ) is modal
coordinate at the ith time step, 𝑠𝑖𝑔𝑛𝑞̇ (𝑡𝑖 ) expresses how modal coordinate changes, i.e.,
increase or decrease compared to the previous step, and 𝑞𝑛,𝑀𝑎𝑥 (𝑡𝑖 ) is the past peak up to the
current time step i. These “idealized nonlinear SDF systems” are used in Eq.(2.112) in SDF-
UMRHA procedure for seismic demands prediction. To apply it into force prediction in
practice based on modal decomposition approach (either MD-All or MD-Lim), the notation
of the above equation is changed into Eq.(3.34) (the illustrated equation is for MD-Lim).
𝐹̂𝑠𝑛 (𝑡𝑖 ) = 𝑓(𝑞̂𝑛 (𝑡𝑖 ), 𝑠𝑖𝑔𝑛𝑞̇̂ (𝑡𝑖 ), 𝑞̂𝑛,𝑀𝑎𝑥 (𝑡𝑖 )) (3.34)

Utilizing estimated modal coordinates (either from MD-All or MD-Lim) as inputs to their
associated idealized nonlinear SDF systems, typically expressed in Eq.(3.34) (established
through performing Cyclic Pushover analyses of the updated numerical model), results in
estimated inelastic SDF forces of all considered modes. Using the same equations listed,
Eq.(3.28) to (3.32), for the case of story force prediction based on inelastic SDF relationships
in section 3.6.2.2, estimated shear force and moment story time histories of every floor can
be determined. Readers should notice that formulae used after estimated inelastic SDF forces
for both the 2nd and the 3rd methods are the same. The only mainly different part is how
inelastic SDF forces are determined.
As can be seen that this proposed force prediction method is time-history based standing on
SDF-UMRHA procedure. Another version, force prediction can be done utilizing updated
numerical mode, is based on the concept of Modal Pushover Analyses (MPA) (Chopra &
Goel, 2001; Najam, 2017) in which target displacements are replaced with maximum
estimated modal roof displacements. Instead of Cyclic Pushover analyses, only Monotonic
Pushover analyses (MP) are sufficient. By following the same procedure on how to obtain
idealized nonlinear SDF “systems” except pushover curves from MP is used, the counterpart
systems in which the inputs are only maximum estimated modal coordinates can be
established. These systems can only predict envelope values of inelastic SDF force only. The
mathematical expression of a typical system is shown as follows
𝐹̂𝑠𝑛,𝑀𝑎𝑥 = 𝑓(𝑞̂𝑛,𝑀𝑎𝑥 ) (3.35)
78
where 𝐹̂𝑠𝑛,𝑀𝑎𝑥 is the maximum inelastic SDF force of the nth mode, and 𝑞̂𝑛,𝑀𝑎𝑥 is the
maximum value of estimated modal coordinate of the nth mode. Using Eq.(3.31) for story
force estimation in x direction as an example, but substituting 𝑞̂𝑛,𝑀𝑎𝑥 in place of 𝑞̂𝑛 (𝑡), and
the result becomes 𝐹̂𝑠𝑛,𝑀𝑎𝑥 instead of 𝐹̂𝑠𝑛 (𝑡). Then following the guidelines in Eq.(3.31) and
(3.32), the equations will results in 𝑉̂𝑖𝑛𝑥,𝑚𝑎𝑥 and 𝑀
̂𝑖𝑛𝑥,𝑚𝑎𝑥 (maximum estimated shear-force
and moment story of the i floor contributed from the nth mode in x direction). Then, the
th
equation down below shall be used to determine maximum estimated shear force and
moment story of the ith floor in x direction.
3×3
2
𝑉̂𝑖𝑥,𝑚𝑎𝑥 = √( ∑ 𝑉̂𝑖𝑛𝑥,𝑚𝑎𝑥 )
𝑛 =1
(3.36)
3×3
̂𝑖𝑛𝑥,𝑚𝑎𝑥 2 )
̂𝑖𝑥,𝑚𝑎𝑥 = √( ∑ 𝑀
𝑀
𝑛 =1
Eq.(3.36) is based on the well-known modal combination rule – SRSS, for the maximum
responses contributed do not, in general, occur at the same time. And the summary of the
proposed force prediction process based on Modal decomposition beyond elastic limits is
shown in Figure 3.4. And this completes the section 3.6.
79
Figure 3.4 The flow chart summarizing the proposed force prediction process
beyond elastic limits based on Modal Decomposition approach
80
3.7 Case Study Building
The numerical model used in this study represents a symmetric 34-story building. Its gravity
load resisting system is RC column-slab frame, and has RC-core walls to resist the lateral
load. The first to the fifth floor has floor height comparatively higher than the rest and has
no infill walls for they represent podium levels commonly existing in tall buildings; at the
other floors, infill walls are located symmetrically. More general information about the case
study numerical model is noted in Table 3.1. Modeling from the beginning in the software
Perform3D (CSI., 2006) is quite time consuming; thus, the practice is to “draw” the model
in SAP2000 or ETABS (CSI., 2017) first and import elements and nodes of the numerical
model to Perform3D. Also, locations of center of masses of each floor have to be calculated
in ETABS as well in order that seismic masses used in Perform3D can be input at the nodes
associated with those identified locations. The typical plan view of the numerical model and
its 3D finite element model in ETABS is shown in Figure 3.5. After all necessary nodes and
elements are imported to the model in Perform3D. Fixed base and rigid floor diaphragm
(translational and rotational about vertical axis movement of the entire floor is the same)
assumptions are made.
Table 3.1 Properties and dimensions of case study building
Property Remarks Quantity/Dimension
No. of Stories 34
Height (m) 117
Typical story height (m) 3.5
Height of podium (m) 22.6
RC shear wall thickness (cm) uniform 250
RC column dimensions (cm x cm), type A uniform 90 x 120
RC column dimensions (cm x cm), type B uniform 40 x 240
Base - 12th floor 1.27
Longitudinal reinforcement ratio in RC
13th - 19th floor 0.81
shear wall (%)
20th - roof 0.46
Longitudinal reinforcement in typical RC
uniform 0.73
columns (%), type A
Longitudinal reinforcement in typical RC uniform 2.25
columns (%), type B
RC walls 36
Specified compressive strength of RC columns 36
concrete 𝑓𝑐′ (Mpa) Post-tensioned
28
slabs
Specified yield strength of longitudinal
reinforcement steel bars in RC walls and 500
RC column 𝑓𝑦 (Mpa)
81
Figure 3.5 Finite Element modeling: (a) 3D view, (b) plan view
Slabs are modeled using elastic thin shell element with cracked stiffness property. Seismic
masses calculated as 100% of service dead load and 25% of unreduced live load of each
floor (LATBSDC, 2017) are lumped at the centers of masses. For nonlinear idealizations,
fiber modeling (PEER/ATC, 2010) is employed. Only at both ends of columns, fiber cross-
sections are used to idealize axial-flexural behavior of the columns over the range equal to
the smaller dimension between two sides of the cross-section. In the middle, elastic element
is used, and this form a compound element used for modeling of columns. For RC walls, an
element of Multi Vertical Line Element Model (Orakcal & Wallace, 2004) is used to idealize
average axial-flexural behavior of continuous RC walls over one floor height. Unconfined
concrete of well-known Mander’s model approximated by a tri-linear backbone curve and
bilinear hysteretic model of steel with strain hardening behavior, with expected material
properties are used in those fiber sections. Shear demand in columns and RC walls is
assumed to be within elastic range (with some stiffness reduction (CSI., 2006)), and to check
this assumption shear strengths of each element are to be compared with the corresponding
demands to verify the assumption made. P-delta effect is to be considered in the analysis. A
pair of two equivalent, compression only, diagonal struts is used to represent an infill panel
with strut properties taken from literatures (Niyompanitpatana, 2017; Suthasit & Warnitchai,
2018). Constant 2.5% modal damping (LATBSDC, 2017) is used for each mode to account
for the energy dissipation within the elastic range, and let the nonlinear hysteretic behavior
idealization deals with inelastic dissipated energy. Additionally, Rayleigh damping is added
to stabilize unimportant modes of vibration greater than interested modes. It should be noted
here that type A columns are modeled as the compound frame element, but type B columns
which look like walls are modeled like RC walls. (Niyompanitpatana, 2017)
82
3.8 Selection of Ground motions
In the past studies (Najam, 2017; Vasanthapragash, 2018), there were attempts to study the
performance of a simplified method to estimate seismic demands of RC core walls tall
buildings under various classification of ground motions including Shallow Crustal – Near
Fault – Pulse like, Shallow Crustal – Far Field, Subduction with Basin effect and Subduction
without Basin effect. However, in this study, selection of ground motions without the main
consideration on ground motion classification, but based on the characteristics of seismic
responses of the selected building to those ground motions is the main goal. To illustrate,
different ground motions are chosen to excite the numerical model such that proportions of
contributions from each mode to seismic responses, indicated by proportions of maximum
modal coordinates of each mode (predetermined through MD approach), are different. These
ground motions were also scaled such that they caused about the almost the same level of
nonlinearity to both RC walls and columns (indicated by D/C ratios of many limit states set).
The general details of the selected ground motions are shown in Table 3.2, and their time
histories along with their 2.5%-damped response spectrum are shown in Figure 3.6.
Table 3.2 Ground motion details
Earthquake Event
Station Year PGA (g) Scaling factor
(component)
El-Centro (NS*) El-Centro 1940 0.3129 1

San Fernando (NS) Castaic Old Ridge Route 1971 0.2752 3
Northridge (NS) 14145 Mulholland 1994 0.4158 1
*NS = North-South component
3.9 Reference Methods
In this section, readers should notice with care that even though some analysis methods are
capable of predicting seismic responses in the manner of mode by mode contributions, and
Modal decomposition approaches (either MD-All mentioned in section 3.3 or MD-Lim
formulated in section 3.4) can be employed to decompose the complex responses of
structures into mode by mode contributions, they are fundamentally different on usage
purposes and on how they are utilized to obtain modal (mode by mode) responses. Anyway,
all of the methods to be mentioned including NLRHA itself will be referred to as reference
methods in this study.
3.9.1 Reference Analysis Methods
Despite the fact that NLRHA is the most accurate analysis method, the results from NLRHA
are only combined complex responses. Thus, in this study, two more analysis methods based
on mode by mode calculation will be used as other reference analysis methods, namely,
SDF-UMRHA and MDF-UMRHA to calculate seismic demands contributed from each
mode.
83
Figure 3.6 Ground motions: (a) 2.5%-damped response spectra, (b) El-Centro (NS) x 1
time-history, (c) San Fernando (NS) x 3 time-history, (d) Northridge (NS) x 1 time-history
84
3.9.1.1 Single Degree of Freedom – Uncoupled Modal Responses History Analysis
In the past, many literatures (Najam, 2017; Tahir, Pennung, & Phichaya, 2018) have used
this method to simplify the way one can predict nonlinear seismic demands of tall buildings
under input ground motions. The name of the method is actually known as just Uncoupled
Modal Response Histories Analysis (UMRHA) – originally developed in the study of
Chopra and Goel (2001). But in this study, one more similar analysis method based on some
common assumptions, yet performed in the manner of Multi Degree of Freedom is to be
proposed, and named as MDF-UMRHA. Thus, this original UMRHA is after this referred
to as SDF-UMRHA. This method will predict the same results as those predicted by Modal
Response History Analysis (MRHA) mentioned in subsection 2.3.1.3, if the responses are
within the elastic range. With the same aforementioned reason, MRHA will be referred to as
SDF-MRHA. The governing equation used in SDF-UMRHA under the case of two
components ground motions is as follows
𝐹𝑠𝑛
𝑞̈ 𝑛 + 2𝜉𝑛 𝜔𝑛 𝑞̇ 𝑛 + = −Γ𝑛𝑥 𝑢̈ 𝑔𝑥 (𝑡) − Γ𝑛𝑦 𝑢̈ 𝑔𝑦 (𝑡) (3.37)
𝑚𝑛
where Γ𝑛𝑥 , Γ𝑛𝑦 , 𝑢̈ 𝑔𝑥 (𝑡) and 𝑢̈ 𝑔𝑦 (𝑡) have already been defined in the section 3.6. and 𝜉𝑛 is
taken equal to modal damping ratio of the nth mode of the numerical mode. 𝜔𝑛 is the
equivalent natural-frequency of the nth mode determined by performing Monotonic Pushover
analysis using modal inertia force pattern to push the numerical within the elastic range;
Pushover curve, as a result, are turned into the relationship between 𝐹𝑠𝑛 /𝑚𝑛 and 𝑞𝑛 , where
the slope of the relationship is 𝜔𝑛2 . Readers should note that if in the numerical model, infill
walls are modelled using compression-only-diagonal struts, dynamic properties (i.e., mode
shapes) and modal inertia force patterns should be taken/calculated from/based on the
adjusted numerical model (please see the postscript of step A-2 of subsection 3.10.1 for the
reason and further detail). The formulae used for seismic demand estimation for SDF-
UMRHA procedure are noted elsewhere (Chopra A. K., 2012; Tahir, Pennung, & Phichaya,
2018). In this study, the software Opensees is used to perform these nonlinear SDF analyses.
3.9.1.2 Multi Degree of Freedom – Uncoupled Modal Responses History Analysis

The idea behind this method is actually based on the concept of modal response contribution
(mentioned in subsection 2.3.1.4); the numerical model will be excited with the dynamic
force vector time-history proportional to modal inertia force pattern (“Dynamic Forces” is
the load type used in Perform3D to apply this type of loading to the center of masses of the
numerical model), and this will only excite responses of the nth mode only if the responses
are within elastic limits. However, other mode contributions will slightly come into play
when the elastic limit is exceeded, yet those portions are assumed to be negligible in this
proposed analysis method. The governing equation of NLRHA under one component ground
motion in x direction is
𝑴𝒖̈ + C𝒖̇ + 𝑭𝑟𝑒𝑠 = −𝑴𝒙 𝒊𝑢̈ 𝑔𝑥 (𝑡) (3.38)

The following equation is the associated governing equation of MDF-UMRHA.
𝑴𝒖̈ + C𝒖̇ + 𝑭𝑟𝑒𝑠 = −𝒔𝑛 𝑢̈ 𝑔𝑥 (𝑡) (3.39)
where 𝒔𝑛 is the vector proportional to the modal inertia force pattern (𝒔𝑛 = 𝛤𝑛𝑥 𝑴𝝓𝒏 ); The
dimension of 𝒔𝑛 is 3J×1, for this vector is 3D force vector (Chopra & Goel, 2004) calculated
85
using 3D mass matrix and mode shape. 𝛤𝑛𝑥 is the modal participation factor for the case that
ground motion has only one component (𝛤𝑛𝑥 = 𝝓𝑇𝑛 𝑴𝒙 𝒊/𝑚𝑛 ). Anyway, for the case that
ground motion have two components, readers should get back to basic idea laid in subsection
2.3.1.4. In this study, Perform3D are employed to perform these nonlinear MDF analyses.
Indeed, this method is so time-consuming and it is selected to be used specially in this study
only. To illustrate, if 3 dominant modes in x direction are considered to be sufficient in
seismic response prediction, this analysis method require thrice time as much as that of
NLRHA; nevertheless, this method does not require Cyclic Pushover analyses as in the case
of SDF-UMRHA in order to realize the nonlinear hysteretic behaviors of the numerical
model. Lastly, 𝒔𝑛 , in the case where infill walls are modelled using compression-only
diagonal struts, must be calculated based on the correct mode shapes (see subsection 3.9.1.1).
3.9.2 Reference Modal Decomposition Approach
In this study, Modal Decomposition based measured accelerations from all floors (MD-All)
will be used as reference modal decomposition approach, for it is also worthy investigating
the comparison between the estimated responses using MD-Lim (Modal Decomposition
based on measured limited number of acceleration records - Orthogonal Filter) and MD-All.
For further details on how this method works to decompose acceleration responses into each
mode contribution and then utilized to estimate various kinds of seismic responses, please
check section 3.3 to 3.6.
3.9.3 Verification of Reference Methods
The only reference method that is not needed to be verified is NLRHA itself instead results
from NLRHA will be used as reference seismic demands used for the verification of the rest
of the methods.
3.9.3.1 Verification within linearly elastic range
In the linear range, scaled ground motions in section 3.8 will be scaled down to one-tenth of
the original time-histories (the results are examined already to be within the first nonlinearity
of the numerical model in all cases) and used to excite the nonlinear numerical model (not
elastic model) described in section 3.7 in y direction, and through this NLRHA the seismic
responses are obtained. On the one hand, SDF-UMRHA and MDF-UMRHA are employed
to calculate seismic responses contributed from each mode within elastic range, and their
combined responses can be calculated subsequently. On the other hand, MD-All is applied
to decompose the required seismic responses, i.e., accelerations from NLRHA, and seismic
responses contributed from each mode and their associated combined responses within
elastic range can be calculated. Now, combined responses from all reference methods can
be verified with those from NLRHA while modal responses can be compared with each other
among the three methods. Within this range all of analysis methods are expected to yield the
same results.
3.9.3.2 Verification beyond linearly elastic range

To cause nonlinearity to the numerical model, scaled ground motions in section 3.8 will be
used to excite the nonlinear numerical model described in section 3.7 in y direction, and
through this NLRHA the seismic responses are obtained. The rest of the procedure is the
same as of the section 3.9.3.1 despite the fact that responses predicted from different methods
86
Figure 3.7 Flow chart of typical procedure for verifying the reference methods in this study
may not be the same. The overview of the typical reference-method-verification procedure
is (applicable to both before and beyond elastic limits) demonstrated in Figure 3.7.
3.9.4 Special Study on Determination of How Many Modes of Vibration Required to

Accurately Estimate Various Kinds of Seismic Responses
In this subsection, the special study on how many modes of vibration needed in order to
predict the seismic responses accurately is explained. Predicated upon the results shown in
subsection 4.4.2, it suggested that MD-All is the most accurate reference method to
determine mode by mode responses beyond the inelastic range of the numerical model. Since
the numerical model has Pure-Sway behavior and the ground motion is only used to excite
the numerical model in y direction, contribution from higher modes of vibration in y
direction only will be more decomposed and then added back for the estimation. The
Eq.(3.15) is repeated here with some change for the illustration (only consideration of modes
of vibration in y direction).
87
𝑁
𝒖̈̃ 𝒏 (𝑡) = ∑ 𝝓𝑛 𝒒̈̃𝑛 (𝑡) (3.40)

𝑛 =1
As can be seen that at first N = 3, which means that it requires only contribution from 3
dominant modes of vibration, especially from y direction. But in this subsection, the number
of N will be increased from 3 to be 4, 6 and 9. And since it can be seen that the force
prediction method explained in the section 3.6.2.1 and 3.6.2.2 can accurately predict the
story forces in the inelastic range, so these two methods are chosen for this study. Regarding
the first method for force prediction, once the estimated acceleration time-history vector is
estimated with more modes of vibration included, the resisting force time-history vector in
Eq.(3.25) will then be more accurate to predict story forces following the guidelines shown
in Eq.(3.26) and (3.27). For the second method for force prediction, the Eq.(3.28) will be
used to estimate inelastic SDF force of more modes, and they will be related to story forces
contributed from those modes in y direction using Eq.(3.32). Finally, Eq.(3.19) and (3.21)
with the number of N changed from 3 to 4, 6 and 9 will be used for the calculation of
estimated story forces, and comparison of all results will be done subsequently.
3.10 Procedure for Verifying the Proposed Seismic Damage Evaluation Scheme
Under the Simplified Predetermined Scenario
In this section, the procedure on how to verify the proposed seismic damage evaluation
scheme mainly predicated upon Modal Decomposition approach based on limited number
of measured acceleration records (section 3.2, 3.4 to 3.6) using a simplified predetermined
scenario will be explained.
3.10.1 Overview of the Simplified Predetermined Scenario of the Verification Procedure
The effectiveness of the proposed scheme of this study will be checked based on the scenario
set here is this subsection. The numerical model described in section 3.7 which represents a
Pure-sway behavior tall building, is taken as the “real building” in this study. It will be
excited only in H2 direction (H2 – y direction in Perform3D) with three different ground
motions described in section 3.8 in order to generate seismic responses under those ground
motions, performed using Nonlinear Response History Analyses procedure (NLRHA) of the
finite element software Perform3D. Various kinds of seismic responses as a result of
NLRHA will be taken as main reference for the comparison of “combined” responses, and
also “absolute” acceleration time-histories as per section 3.2 will be extracted from the
results, representing “clean” measurement data.
The proposed method is to be applied to evaluate seismic damage. Before the seismic events,
the following steps (section 3.2.1) are completed
A-1 The numerical model has to be created to represent the considered building. But in this
study the numerical described in section 3.7 (which has earlier been thought of as “real
building”) will also be taken as the already created numerical model.
A-2 Obtaining dynamic properties from full ambience vibration test. But in this study, to
minimize the error that might cause by field measurement, dynamic properties of the “real
building” are taken directly from Modal Analysis results of the numerical model described
88
in section 3.7 except that the properties of struts in that model must be reduced to half of the
normal strength values, only to obtain the “supposed-to-be” dynamic properties.
PS: This is because of the fact that while Perform3D calculates dynamic properties of the
numerical model, it takes the stiffness matrix of the no load condition as input. At the point,
even though the strength of struts in tension side is zero – compression only struts, the
software uses the input initial stiffness (expected to be of compression side only) to calculate
stiffness of tension side. And this causes twice the values of expected stiffness of struts to be
used in Modal Analysis of the original model. That is why only to be able to obtain the
correct dynamic properties out (to be used in some parts of Orthogonal Filter and reference
methods (section 3.9)), one additional model with properties of struts reduced in half of the
original value must be created. But Do not use the adjusted model more than just obtaining
dynamic properties. The original model will be used for the case of NLRHA and MDF-
UMRHA, for once the load is applied, the tension property of struts shows zero in number
immediately.
A-3 Updating the created numerical model. But in this study, the numerical model is already
perfect, so that means possible error due to model updating process is not taken into
consideration.
A-4 Realizing the hysteretic behavior of the updated numerical model. For this step, as
mentioned in section 3.2.2, Cyclic Pushover analyses using (“correct”) modal inertia force
patterns will be performed on the updated numerical model with the typical gradually-
cyclically-roof-drift-controlled manner. Then, Cyclic Pushover curves can be obtained and
turned to each mode inelastic SDF relationship. After that, Modified Flag Shape parameters
(Panday, 2017) will be determined (mainly) by trial and error to map the so-called
“Nonlinear SDF system” to each mode inelastic SDF relationship.
Then, it follows the steps of part B (section 3.2.2).
B-1 Proceeding with the measured acceleration signals. As mentioned, that time-histories
will be extracted from the “real building.” Everything will be done following section 3.5
except that there is no need to deal with noisy signal in this study, so no filtering technique
will be applied.
B-2 Applying Modal Decomposition technique based on section 3.4 to obtain estimated
modal acceleration of each mode of vibration which indicate how much contribution each
mode has to the overall seismic response.
B-3 Relating estimate modal accelerations from B-2 to estimate seismic demands of various
response types of the “real building” under the excitation of the selected seismic ground
motions. Results from this step will be verified with those from reference methods
(explained in subsection 3.9.2).
B-4 Utilizing envelopes of calculated quantities as indicators in the seismic damage

evaluation. In this study, this step will be dismissed; for if the estimate responses from the
step B-3 are accurately predicted, results of evaluation of seismic damage based on those
results will not be affected by the scheme itself.
89
3.10.2 Verification of the Seismic Demands Estimated from the Proposed Scheme
Nonlinear seismic demands estimated as a result of the application of the proposed seismic
damage evaluation scheme based on MD-Lim to the simplified predetermined scenario of
seismic events happening to the “real building” in this study, specifically, including floor
displacement, floor acceleration (in this study, since it will not cause any wrong meaning,
relative accelerations will only be demonstrated for the sake of simplicity), inter-story drift,
story shear and story moment in terms of height-wise envelope (mainly focused on) and
time-history will be verified with those calculated by NLRHA. The mentioned responses in
terms of mode by mode contribution will also be compared with those results determined
and already verified from other reference methods (section 3.9). And discrepancies, if any,
between results determined using MD-Lim and MD-All shall be recognized and emphasized.
3.11 Additional Step to Enhance the Accuracy of the Proposed Scheme on Account
of Higher Mode Effects
3.11.1 Background behind the proposed correction procedure
Predicated upon the results and discussion in chapter 4, demands estimated employing MD-
All (when the contribution of three modes are only assumed to be significant) shows that
there are outstanding discrepancies between both shear and acceleration envelopes
calculated / estimated by NLRHA and MD-All. Based on section 4.5, it implies that too less
considered important modes of vibration are included. Even knowing this, at the highest
capability of this proposed scheme, only three modes can decompose. To obtain the fourth
mode contribution, another set of accelerometers (with new vertical accelerometer layout)
must be mounted on one more floor, and the Orthogonal Filter based on four floors (not
including the base one) has to be formulated; the thing is, even though that could possibly
be done, only four mode included still does not totally solve satisfy the need. Up to six modes
are necessary to reasonably account for the higher mode effects. In this section, the proposed
enhancement method based on SDF-MRHA and the associated verification process will be
explained / described.
3.11.2 The proposed enhancement procedure
According to the idea that seismic damage evaluation must be done rapidly along with the
finding that even though the condition of tall buildings is beyond elastic range, higher modes
generally response linearly based on the past study on simplified analysis method– SDF-
UMRHA for seismic demand prediction (Tahir, Pennung, & Phichaya, 2018), this suggests
that higher mode effects can be accounted for by simple additional calculation through linear
analysis. In chapter 4, estimated floor accelerations of some floors are found to be slightly
overestimated already because of the error caused by MD-Lim itself, and floor displacements
as well as inter-story drifts are nicely predicted using only 3 dominant modes in y direction,
so only estimated story forces will, for the reasons, be further corrected.
The analysis method to be used for the simple calculation is Modal Response History
Analysis (“to be referred in this study as SDF-MRHA”). The following equation is the
governing equation of SDF-MRHA procedure under 2 component ground motions.
𝑞̈ 𝑛 + 2𝜉𝑛 𝜔𝑛 𝑞̇ 𝑛 + 𝜔𝑛2 𝑞𝑛 = −Γ𝑛𝑥 𝑢̈ 𝑔𝑥 (𝑡)−Γ𝑛𝑦 𝑢̈ 𝑔𝑦 (𝑡) (3.41)
90
The only input to Eq.(3.41) is the measured ground motion (𝑢̈ 𝑔𝑥 (𝑡) and 𝑢̈ 𝑔𝑦 (𝑡)), and the rest
of the parameters have already been identified through the full ambience vibration test. 𝑞̈ 𝑛 ,
𝑞̇ 𝑛 and 𝑞𝑛 are modal acceleration, modal velocity and modal coordinate of the nth mode in
SDF-MRHA procedure. Below are the formulae used in SDF-MRHA.
Shear-force story time-history of the ith floor contributed from the nth mode
𝐽
𝑉𝑖𝑛𝑥 (𝑡) = 𝜔𝑛2 𝑞𝑛 (𝑡) ∑ 𝑚𝑗 𝜙𝑛𝑥𝑗

𝑗=𝑖
𝐽 (3.42)
𝑉𝑖𝑛𝑦 (𝑡) = 𝜔𝑛2 𝑞𝑛 (𝑡) ∑ 𝑚𝑗 𝜙𝑛𝑦𝑗
𝑗=𝑖
where 𝑉𝑖𝑛𝑥 (𝑡) and 𝑉𝑖𝑛𝑦 (𝑡) are shear force time histories of the nth floor contributed from
the nth mode in x and y direction additionally calculated using SDF-MRHA procedure.
Moment story time history of the ith floor contributed from the nth mode
𝐽
𝑀𝑖𝑛𝑥 (𝑡) = 𝑤𝑛2 q𝑛 (𝑡) ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑥𝑗

𝑗=𝑖
𝐽 (3.43)
𝑀𝑖𝑛𝑦 (𝑡) = 𝑤𝑛2 𝑞𝑛 (𝑡) ∑(𝐻𝑗 − 𝐻𝑖 )𝑚𝑗 𝜙𝑛𝑦𝑗
𝑗=𝑖
where 𝑀𝑖𝑛𝑥 (𝑡) and 𝑀𝑖𝑛𝑦 (𝑡) are moment time-history of the nth floor contributed from the
nth mode in x and y direction additionally calculated using SDF-MRHA procedure.
Contribution of story forces from mode 1 to 3 are estimated based on section 3.6, and the
correction ones are calculated per Eq.(3.42) and (3.43).
And this completes chapter 3 - Methodology.
91
CHAPTER 4
4 RESULTS AND DISCUSSION
4.1 Overview
This chapter is dedicated to documentation of results and discussion of the study:

“Evaluation of Seismic Damage of Tall Buildings from Acceleration Response Time
Histories by a Modal Decomposition Approach”. It features Modal Analysis results, Cyclic
Pushover analysis results realized as idealized Nonlinear SDF system, results from the
verification of reference methods, special study on number of required necessary mode
results, verification of the proposed seismic scheme results, and finally, the results of the
higher mode correction process by simple calculations.
4.2 Modal Analysis Results
As stated in chapter 3, in order to obtain the dynamic characteristics of the nonlinear

numerical model with infill walls modelled as compression only diagonal struts, the
nonlinear numerical model must be specially adjusted by reducing the properties, i.e.,
strength of struts, by half. Modal Analysis was conducted by Perform3D. Results of three
dominant modes of each direction only are to be presented in Table 4.1. For nomenclature,
Y = y-translational direction, X = x-translational direction, and T = torsional direction. The
table presents mass participation ratio of each dominant mode. 3D mode shapes of them
obtained from Perform3D are presented in Figure 4.1. And since the numerical model in
this study, and the ground motion was applied in only one direction (y),
Table 4.2 is then only used to show the calculated dynamic properties of modes of vibration
in y direction, not only three modes, but up to 9 modes. These modal properties determined
as a result of this section were actually used in place of the identified dynamic properties
from full ambience vibration test in this study.
4.3 Monotonic and Cyclic Pushover Analysis Results
4.3.1 Monotonic Pushover Analyses Within Elastic range
As stated in subsection 3.9.1.2, in order to perform SDF-UMRHA, 𝜔𝑛 - equivalent natural-

frequency of the nth mode of every considered mode is required. After the analyses were
done, all of the pushover curves of nine modes in y direction are turned into inelastic force
– modal coordinate relationships in the “q” coordinate (𝐹𝑠𝑛 /𝑚𝑛 - 𝑞𝑛 ) following Eq.(2.121)
as guidelines. And the slopes of all relationship were taken as equal to 𝜔𝑛2 , then the associated
natural periods were found to be, as expected, equal to those listed in
Table 4.2 except that of the first mode which is equal to 4.45 second (supposed to be 4.385
second). Theoretically, this should not be the case. But following the guidelines of the recent
literature (Vasanthapragash, 2018), the natural period of the first mode shall be taken as that
determined from MP.
4.3.2 Hysteretic Behavior of the Numerical Model
Monotonic and Cyclic Pushover Analyses were conducted in Perform3D to obtain the
hysteretic behaviors of three modes in y direction. MP of the nth mode was done by pushing
92
the numerical model by the force proportional to the modal inertia force pattern to the
expected modal roof displacement (predetermined by MD-All approach) in the gradually
increasing manner while CP of the nth mode was also done by pushing the numerical model
using the same force pattern but with cyclically-gradually-increasing manner to the very
same modal roof displacement. Pushover Curves of each mode are shown in Figure 4.2.
After that they are then turned into their corresponding inelastic SDF force – deformation
relationships in the “D” coordinate displayed in Figure 4.3. In this study, SDF hysteresis
behavior model named Modified Flag Shape (MFS) model (Panday, 2017) was used to map
the hysteretic behavior of the determined inelastic SDF force-deformation relationship of
each mode. Important parameters of MFS model were identified using trial and error method
(verbally suggested by the creator of the MFS model that this way is more effective than the
systematic way proposed by him in his literature) for each mode. After each process had
been done, “idealized nonlinear SDF system” of each mode was obtained. Before utilizing
them in this study, the calibration was done by attempting to regenerate the cyclic force-
deformation relationships as shown in Figure 4.3 by inputting the similar modal coordinates
of the plots into the associated idealized nonlinear SDF systems to predict the forces. The
comparison results shown in Figure 4.4 are reasonably accurate, so idealized nonlinear SDF
systems were later used for SDF-UMRHA (subsection 3.9.1.1) and in force prediction in
subsection 3.6.2.3.
Table 4.1 Mass participation Ratio

Period Mass Participation Ratio Cumulative Mass participation Ratio
Mode
(s) H1 (x direction) H2 (y direction) H1 (x direction) H2 (y direction)
Y1 4.385 0.000 0.787 0.000 0.787
T1 3.268 0.000 0.000 0.000 0.787
X1 2.631 0.709 0.000 0.709 0.787
Y2 1.295 0.000 0.109 0.709 0.896
T2 1.036 0.000 0.000 0.709 0.896
Y3 0.639 0.000 0.029 0.709 0.925
X2 0.631 0.170 0.000 0.879 0.925
T3 0.576 0.000 0.000 0.879 0.925
X3 0.277 0.047 0.000 0.926 0.925
Table 4.2 Calculated dynamic properties

Mode Period (s) 𝝓𝑇𝑛 𝑴𝝓𝑛 𝝓𝑇𝑛 𝑴𝒊 𝝓𝑇𝑛 𝑴𝒊 / 𝝓𝑇𝑛 𝑴𝝓𝑛
Y1 4.385 13796.095 18817.648 1.364
Y2 1.295 12150.472 -6556.836 -0.540
Y3 0.639 11180.207 3251.039 0.291
Y4 0.398 12114.364 -2465.634 -0.204
Y5 0.278 12113.789 2046.566 0.169
Y6 0.212 11328.508 1894.085 0.167
Y7 0.167 10224.876 1569.253 0.153
Y8 0.137 12843.421 1443.387 0.112
Y9 0.113 15650.454 1345.122 0.086
93
X translational direction
𝑇𝑥1 = 2.631 𝑇𝑥2 = 0.631 𝑇𝑥3 = 0.277
𝑇𝑥1 = 2.631
Y translational direction
𝑇𝑦1 = 4.385 𝑇𝑦2 = 1.295 𝑇𝑦3 = 0.639
𝑇𝑥1 = 2.631 𝑇𝑥1 = 2.631 𝑇𝑥1 = 2.631

Torsional direction
𝑇𝑇1 = 3.268 𝑇𝑇1 =1.036 𝑇𝑇1 = 0.576

𝑇𝑥1 = 2.631 𝑇𝑥1 = 2.631 𝑇𝑥1 = 2.631
Figure 4.1 Mode shapes of three dominant modes in each direction
94
0.06
Base shear Coefficient 0.04
0.02
0
-0.80% -0.60% -0.40% -0.20% 0.00% 0.20% 0.40% 0.60% 0.80%
Roof drift (%)
-0.02
Monotonic Pushover (positive)
Monotonic Pushover (negative)
-0.04 Cyclic Pushover
-0.06
0.09
0.06
0.04
Base shear Coefficient
0.01
-0.30% -0.20% -0.10% -0.020.00% 0.10% 0.20% 0.30%

Roof drift (%)
-0.04 Monotonic Pushover (positive)
Monotonic pushover (negative)
-0.07 Cyclic Pushover
-0.09
0.08
0.06
0.04
Base shear Coefficient
0.02
0
-0.15% -0.10% -0.05% 0.00% 0.05% 0.10% 0.15%
-0.02 Roof drift (%)
-0.04
Monotonic Pushover (positive)
-0.06 Monotonic Pushover (negative)
Cyclic Pushover
-0.08
Figure 4.2 Pushover curves of modes of vibration in y direction:

(a) 1st mode, (b) 2nd mode, (3) 3rd mode
95
800
Fs1/L1
600
400
200
0
-600 -400 -200 0 200 400 600
-200 D1
-400
-600
-800
8,000
Fs2/L2
6,000
4,000
2,000
0
-800 -600 -400 -200 0 200 400 600 800
-2,000 D2
-4,000
-6,000
-8,000
30,000
Fs3/L3
20,000
10,000
0
-500 -400 -300 -200 -100 0 100 200 300 400 500
D3
-10,000
-20,000
-30,000
Figure 4.3 Inelastic SDF force-deformation relationships of modes of

vibration in y direction: (a) 1st mode, (b) 2nd mode, (c) 3rd mode
96
Figure 4.4 Comparison between SDF force-deformation relationship - derived from Cyclic
Pushover analysis and that predicted by idealized SDF nonlinear system:
(a) 1st mode, (b) 2nd mode, (c) 3rd mode
97
4.4 Verification of Reference methods
In this section, the verification results of reference methods within both linear and nonlinear
range following the procedure noted in subsection 3.9.3 is documented. It should be noted
here that for all analysis methods the problem regarding period elongation preliminarily
examined by running the same type of analysis method under the same ground motion but
with different time step. If the results show discrepancies between time-history plots, the
time step should be reduced more to the point where no more lacking between the results of
two analyses with the current time step and the previous one. That is why all analyses from
all methods were conducted with the time step equal to 0.005 second. However, the time
step must not be larger than that of ground motion records, for it may miss the essential part
of that ground motion (CSI., 2006). Also, the time step must not be larger than the period of
the highest considered important mode over a constant value 12 (CSI., 2006) (in this study,
it is the period of the 9th mode in y direction / 12, equal to 0.113/12 = 0.0094 second).
4.4.1 Verification results within linearly elastic range
4.4.1.1 Results Under 0.1× El-Centro Ground Motion (NS)

0.1 × El-Centro ground motion (NS) is the ground motion used in this study that can cause
comparatively high contribution from the first mode of vibration with respect to the rest of
selected ground motions to this numerical model. According to the results shown in Figure
4.5, Figure 4.6, Figure 4.7, Figure 4.8, and Figure 4.9, the summary can be made as follows.
The combined results from all reference methods can nicely predict the complex responses
calculated from NLRHA within linearly elastic range except those of estimated Shear-force
story and estimated relative floor acceleration in terms of both time-history and envelope
plots. This is because of the effect of higher modes which will be shown later. Anyway,
within the scope of elastic range, the conclusion from the previous study implied that up to
six modes is needed to estimate the acceleration and Shear-force response type accurately
(Vasanthapragash, 2018).
Regarding each mode comparison, the modal responses predicted by MDF-UMRHA are
exactly the same as those from MD-All while SDF-UMRHA results are not that different
from the other two reference methods. It should be noted here that for the comparison of
each mode contribution, the reason why both modal accelerations and modal coordinates are
not plotted here is MDF-UMRHA can only predict responses in normal coordinate. Thus,
the results in modal coordinate from SDF-UMRHA and MD-All are then turned into normal
coordinate first, and then comparison can be made. However, since mode shape values at the
roof-top level are equal to unity for all three modes, so recklessly speaking, plots one to three
of both Figure 4.5 and Figure 4.6 actually represent modal coordinates and modal
accelerations of three dominant modes (see Eq.(3.14) and (3.15)).
As stated in section 3.6.1, under elastic range, the story forces contributed from each mode
can be related to its own modal coordinate and known stiffness of SDF - 𝜔𝑛2 - as illustrated
in Eq.(3.18) and (3.20), and then Eq. (3.19) and (3.21) were utilized to combine the
contribution from all modes together. Thus, there was no need of applying any proposed
force prediction procedure explained in section 3.6.2. Also, to note the ground motion
stopped at 30 seconds.
98
Figure 4.5 Verification of estimated roof-top displacement time-history
under 0.1× El-Centro ground motion (NS)
99
Figure 4.6 Verification of estimated roof-top (relative) acceleration time-history
under 0.1 × El-Centro ground motion (NS)
100
Figure 4.7 Verification of estimated Base-Shear time-history
101
102
Figure 4.9 Verification of estimated response envelopes
under 0.1 × El-Centro ground motion (NS): (a) estimated floor displacement envelope,
(b) estimated inter-story drift envelope, (c) estimated relative floor acceleration envelope,
(d) estimated Moment story envelope and (e) estimated Shear-force story envelope
103
4.4.1.2 Results Under 0.3 × San Fernando Ground Motion (NS)
The seismic responses of the numerical model to 0.3 × San Fernando ground motion (NS)
are greatly influenced by the higher modes (2nd and 3rd modes highly contribute) with respect
to the responses of the first ground motion. As can be seen that the rough ratio between
maximum roof top displacement caused by the first to the second mode is about six times
for the case of 0.1 × El-Centro (NS) (see Figure 4.5) while it becomes only slightly more
than one in this case as will be shown soon. The results of the verification of reference
methods employed for the case of this ground motion are displayed in Figure 4.10, Figure
4.11, Figure 4.12, Figure 4.13 and Figure 4.14. The same conclusions can also be made in
this case: good verification results, negligible discrepancies among the mode by mode
responses, and three modes are not good enough to accurately predict relative floor
acceleration and Shear-force story.
It should be noted here the reason why Moment story time-history of the 20th floor was
plotted is because the author wanted to show the composition of Moment having contribution
form many modes (if at the base were chosen, it would most likely only show the
contribution from the 1st mode, especially the results of section 4.4.1.1). This is also the same
reason why roof-top relative floor acceleration and floor displacement time-histories were,
intentionally, not plotted in the same scale for all modes, for the aim of this section is to
verify the reference methods. Thus, they were plotted to be seen clearly. Also, to note, the
ground motion stopped at about 60 seconds.
4.4.1.3 Results Under 0.1 × Northridge Ground Motion (NS)
This ground motion causes comparatively high contributions from the 1st and 2nd modes of
vibration with respect to the rest of the selected ground motions. Results for the verification
using this ground motion can be found in Figure 4.15, Figure 4.16, Figure 4.17, Figure 4.18
and Figure 4.19. The summary of what is shown is generally the same except that in this
case only 3 modes seems to be sufficient for estimating Shear-force story in terms of both
time-history and envelop plots. Also, to note, the ground motion stopped at 25 second.
4.4.1.4 Conclusion of the Verification Results Within Linearly Elastic Range
In conclusion, based on the results shown in subsection 4.4.1.1 to 4.4.1.3 with the small
amplitude ground motions employed, they do not cause any nonlinearity to the numerical
model. Both additional reference analysis methods i.e., SDF-UMRHA and MDF-UMRHA
and the reference modal decomposition approach i.e., MD-All can reasonably predict the
complex seismic responses using only contribution from three modes. However, it would
have been better if there had had more modes of vibration to be included, especially, for
floor acceleration and Shear-story response prediction.
104
under 0.3 × San Fernando ground motion (NS)
105
106
107
108
under 0.3 × San Fernando ground motion (NS): (a) estimated floor displacement envelope,
109
Figure 4.15 Verification of roof-top displacement time-history
under 0.1 × Northridge ground motion (NS)
110
Figure 4.16 Verification of estimated roof-top acceleration time-history
111
112
113
under 0.1 × Northridge ground motion (NS): (a) estimated floor displacement envelope,
114
4.4.2 Verification results beyond linearly elastic range
4.4.2.1 Results Under 1.0 × El-Centro Ground Motion (NS)

It should be reminded that beyond the elastic limits, all of the additional reference methods
are not expected to yield theoretically exact results anymore. Figure 4.20, Figure 4.21 and
Figure 4.22 shows the verification results of displacement and acceleration related responses.
In general, based on the comparisons between mode by mode estimated responses, each
reference method did not yield the same results anymore. Anyway, the verification results
of estimated floor displacement by MD-All in terms of both time-history and envelope can
accurately predict the responses. Acceleration responses cannot be well estimated by any
method as in the previous cases of elastic demands.
Regarding force predictions in the reference modal decomposition approach, as explained in
subsection 3.6.2, it could not be done in the same way as in the case within elastic range
anymore, so three main methods were proposed for force prediction of MD-All. It should be
noted that SDF-UMRHA and MDF-UMRHA results were kept the same (the way both
reference analysis methods calculated forces are certain) despite the fact that the results
estimated by each proposed method for force prediction for the reference modal
decomposition approach may be changed.
The verification results when the method in subsection 3.6.2.1 was employed for MD-All
are shown in Figure 4.23 and Figure 4.24. MD-All accurately predicted the story forces in
terms of time-history (yet the peaks of Shear-force stories were not well predicted). SDF-
UMRHA and MDF-UMRHA slightly overestimated the moment story, and they estimated
Shear-force envelope to be higher than those of MD-All.
Figure 4.25, Figure 4.26, Figure 4.27 and Figure 4.28 are the results when subsection 3.6.2.2
was employed as force prediction method for MD-All. It should be noted that the previous
method did not predict the mode by mode story-forces first and summed them up in time
domain to be the total forces as were done in this method, yet in general, this proposed force
prediction method for MD-All yields the very similar combined results as in the previous
case. With regard to comparison of mode by mode story forces, both reference analysis
methods yield closer results to each other with respect to those of MD-All. The two reference
analysis methods predict the estimated Base-Shear – roof displacement relationships of the
2nd and 3rd modes to have linear relations (see Figure 4.27). Nevertheless, according to MD-
All being employed with the force prediction method of subsection 3.6.2.2, it seems there
was some influence from the stiffness degradation of the first mode, obviously indicated by
loops being formed in the first plot of Figure 4.27, to the estimated Base-Shear against roof-
displacement relationships of the 2nd and 3rd modes. This could not be captured by the two
analysis methods.
Based on the 1st version (time-history based) of the proposed force prediction method in
subsection 3.6.2.3 for MD-All, inelastic SDF forces were predicted first through the
idealized nonlinear SDF systems and then later used to calculate mode by mode forces. The
estimated responses using this method for MD-All shown in Figure 4.29, Figure 4.30, Figure
4.31 and Figure 4.32 are not that different (there are some as will be pointed out soon) from
the previous one in terms of both time-history and envelope, but readers should note that
115
since each idealized nonlinear SDF system was established based on the Cyclic Pushover
analysis of each mode in which the cyclic responses were generated without the intentional
interference of other modes of vibration (some negligible coupling effect might have
occurred), being opposite to the case of ground motions used as force excitation, the system
can only predict linear relationships between the estimated Base-Shear and roof-
displacement of the 2nd and 3rd mode. The findings in Figure 4.31 are, that is why, similar to
those of reference methods (for their formulations are based on the very same concept of no
intentional interference of other modes). Furthermore, the proposed method for MD-All
overestimated envelope of Shear-force stories near the base. This did not happen with the
other two force prediction methods shown before.
Finally, the envelope-based version (the 2nd version) of the method in subsection 3.6.2.3 was
used to predict the SRSS envelope of story forces. Each mode envelope separately estimated
for MD-All is shown in Figure 4.33. The verification results of story force envelopes are
displayed in Figure 4.34. This proposed method took one more step further assumption
which was to use the SRSS modal combination rule in combining modal envelope responses,
and this affects the results: the other proposed methods say that three modes are enough to
predict Moment story envelope while this proposed method used in MD-All displayed the
moment envelope in one side to match the NLRHA envelope (agree with the results of the
previous methods), but the other side did not. This uncertain trend for this force prediction
method was also found in the responses of the numerical model to the other ground motions
as well (as will later be illustrated).
116
117
118
(b) estimated inter-story drift envelope, (c) estimated relative floor acceleration envelope
119
Figure 4.23 Verification of estimated Story force time-histories
(at the base / of the 20th floor) under 1.0 × El-Centro ground motion (NS) /
MD-All – story forces predicted utilizing the method stated in section 3.6.2.1:
“Story force prediction based on Estimated Floor Acceleration Time History Vector”
Figure 4.24 Verification of estimated Story force envelopes

under 1.0 × El-Centro ground motion (NS) /
MD-All– story forces predicted utilizing the method stated in section 3.6.2.1:
120
“Story force prediction based on inelastic SDF relationships”
121
Figure 4.26 Verification of estimated Moment time-history of the 20th floor
122
Figure 4.27 Comparison of estimated Base-Shear against estimated roof-displacement of
each mode under 1.0 × El-Centro ground motion (NS) /

123
“Story force prediction based on the updated numerical model (version 1)”
124
125
Figure 4.31 Comparison of estimated Base-Shear against estimated roof-displacement of each
mode under 1.0 × El-Centro ground motion (NS) /

126
Figure 4.33 Mode by mode story force envelopes
predicted by MD-All utilizing the method stated in section 3.6.2.3:

127
4.4.2.2 Results Under 3.0 × San Fernando Ground Motion (NS)
In Figure 4.35, the verification result shows that only MD-All can nicely predict the roof-
top displacement time-history. Readers may have noticed the discrepancy in the period about
25 – 35 seconds. That will be later shown to be because of the process of error correction in
double integration process and not because of the essence of the method itself. Figure 4.36
and Figure 4.37 show that when the ground motion excited the higher modes more, not only
acceleration type responses cannot be well predicted, but also the case of inter-story drifts.
Verification results for force prediction under 3.0 × San Fernando ground motion (NS) are
displayed in Figure 4.38 to Figure 4.49.
Regarding force estimation using reference analysis methods, in terms of envelopes, they
overestimate the Moment stories (despite the fact that as compared to the last case where
higher mode should come into play more, it was expected to underestimate the results since
only three modes were used). They also overestimated Shear-force story of some floors, yet
left those of some floors underestimated. This is assumed to be the combined effect of the
inefficiency of higher modes and the fact that the methods themselves caused
overestimation.
For the first two methods for force prediction of MD-All (subsection 3.6.2.1 and 3.6.2.1),
they were employed and nicely estimated the Moment stories in terms of both envelope and
time-history while both methods agree with each other that only three dominant modes are
not sufficient to predict Shear-force stories. On the other hand, the third method (1st version)
which was formulated based on idealized nonlinear SDF system realized using no-
intentional-interference-of-other-mode method like Cyclic Pushover analysis, fairly
resembled the results of both reference analysis methods in terms of envelope plots. But the
important thing to be noted is that based on the time-history plot, one can see that the plots
of MD-All and NLRHA actually went together periodically in Figure 4.45 (This is how MD-
All utilizing the method in 3.6.2.3 is different from both analysis methods) except that the
peaks determined from this method for MD-All also overestimated. The trend was correct
because they were controlled by the estimated input modal coordinates (directly taken from
double integration of estimated “decomposed” modal accelerations) to the idealized
nonlinear SDF systems, yet as stated the way the “systems” were taken is environmentally
different from the real scenarios where all modes are altogether contributing. And this is
inferred from the results to be the cause of error caused by this force prediction approach.
The evidence is also clearly shown in Figure 4.42 that though the 2nd and 3rd modes
contributed a lot to the overall seismic demands such that only the two modes formed “loops”
based on SDF-UMRHA and MDF-UMRHA, MD-All based on subsection 3.6.2.2 implies
that the presence of intense nonlinearity of the 2nd and 3rd mode caused stiffness degradation
to the 1st mode. This could not be captured once the force prediction based on 3.6.2.2 (version
1) was employed for MD-All as shown in Figure 4.46.
The other evidence showing that the 1st mode contribution to the seismic excitation is low is
also illustrated in Figure 4.48 as a result of employing the 2nd version of the method of
subsection 3.6.2.2 for force prediction of MD-All. Also as discussed earlier in subsection
4.4.2.1, the performance of the approach is uncertain. In this case, it shows that Shear-force
stories are not sufficient to be predicted by three modes (as expected to be reasonable), yet
overestimated the Moment stories (see Figure 4.49).
128
Lastly, it should be additionally noted that in Figure 4.42, the plot between estimated Base-
Shear and estimated roof-displacement, especially for the 2nd and 3rd modes, predicted by
MDF-UMRHA seemed to be affected by the influence of other modes “spilled” into. As
mentioned before, SDF-UMRHA assumes that the coupling effect is negligible even within
the inelastic range, so the effect is ignored by not being accounted once formulating the
idealized nonlinear SDF system. However, in the case of MDF-UMRHA, with the same
assumption, coupling effect are allowed to occur during the analysis of each individual mode
for they should be so small and negligible and the responses are assumed be dominated by
that mode, yet the results in the figure shows the messiness of the plots caused by ignoring
the presence of other modes getting involved by spillover effect. Again, do not be confused.
In SDF-UMRHA we forced all contribution solved by the SDF governing equation to be
from one individual mode, but in MDF-UMRHA, responses are expected to be dominated
by that mode.
129
130
131
under 3.0 × San Fernando ground motion (NS):
(a) estimated floor displacement envelope,
(b) estimated inter-story drift envelope,
(c) estimated relative floor acceleration envelope
132
(at the base / of the 20th floor) under 3.0 × San Fernando ground motion (NS) /

under 3.0 × San Fernando ground motion (NS) /
133
134
135
each mode under 3.0 × San Fernando ground motion (NS) /

136
137
138
each mode under 3.0 × San Fernando ground motion (NS) /

139

140
4.4.2.3 Results Under 1.0 × Northridge Ground Motion (NS)
The verification results for this case are shown in Figure 4.50 to Figure 4.64. According to
the results, floor displacements and inter-story drifts were nicely estimated by all reference
methods in terms of envelope while floor accelerations were not the case.
For force predictions according to subsection 3.6.2.1 and 3.6.2.2 for MD-All, Moment
stories can be well predicted in terms of both time-history and envelope, yet slightly
underestimated Shear stories of some floors (but this was because of too less modes of
vibration included). The two methods seem promising as the force prediction methods for
MD-All.
Conversely, the force prediction method explained in subsection 3.6.2.3 (1st version)
overestimated the story forces for both Shear-force and Moment story response types. The
results are also similar for the cases of both reference analysis methods.
Under this ground motion, the force prediction method of subsection 3.6.2.3 (2nd version)
for MD-All nicely estimated the story force envelopes (they, up to some extent, agreed with
those from the two proposed methods for MD-All).
4.4.2.4 Conclusions of the Verification Results Beyond Linearly Elastic Range

In conclusion, based on the results obtained from subsection 4.4.2.1 to 4.4.2.3, when the
seismic demands exceed the elastic limits, the estimated results from different reference
methods which were very close to each other in the case of elastic demands were no longer
equal to each other anymore. At least the common results implied that only including
contribution from three modes are not enough to estimate floor acceleration responses; floor
displacements and inter-story drifts can be estimated with reasonable accuracy using only
contribution from three modes no matters which reference method was employed. Regarding
story force predictions, both analysis methods along with MD-All employing the force
prediction method in subsection 3.6.2.3 (1st version) are observed in this study to
overestimate the story forces, especially in the cases where higher modes had a lot of
contribution. On the other hand, once MD-All being equipped with the force prediction
procedures mentioned either in subsection 3.6.2.1 or 3.6.2.2, it can reasonably estimate the
story forces. To illustrate, in all cases, the Moment stories predicted (contributed from three
modes) closely match those from NLRHA, and consistently underestimate Shear-force
stories. Furthermore, though one method (subsection 3.6.2.1) was not itself based on
calculation of mode by mode contribution, and another one (subsection 3.6.2.2) was, the
story forces in terms of both time-history and envelope estimated by each method yielded
the very similar results. Therefore, MD-All utilizing in one case the method in subsection
3.6.2.1 and another case that in subsection 3.6.2.2 was chosen as the reference method to
study the higher mode effect following the procedure stated in subsection 3.9.4: “Special
study on determination of how many modes of vibration required to accurately estimate
various kinds of seismic responses.”
141
142
under 1.0 × Northridge motion (NS)
143
under 1.0 × Northridge ground motion (NS): (a) estimated floor displacement envelope,
144
(at the base / of the 20th floor) under 1.0 × Northridge ground motion (NS) /

under 1.0 × Northridge ground motion (NS) /
145
146
147
each mode under 1.0 × Northridge ground motion (NS) /

148
149
150
each mode under 1.0 × Northridge (NS) /

151

152
4.5 Results of the Special Study on Determination of How Many Modes of
Vibration Required to Accurately Estimate Various Kinds of Seismic Responses
As mentioned in the conclusion of subsection 4.4.2, among the reference methods, the most
promising, mode by mode manner, reference method is Modal Decomposition based on
measured accelerations from all floors (MD-All). Since the method itself does not suggest
anything about force prediction beyond inelastic range, that is why three main methods in
subsection 3.6.2 were proposed. And based on the results shown in the previous section, the
first method: “Story force prediction based on estimated floor acceleration time-history
vector” and the second method: “Story force based on inelastic SDF relationships” were
observed to be consistent and reasonable in terms of their performance to estimate story
forces. However, MD-All itself along with both force prediction methods could not in
general accurately predict floor acceleration and Shear-force story time-histories. It was
anyway inferred to be due to lack of higher modes included, but there was a need to prove
the statement. Thereby, in this section, the results of the procedure explained in subsection
3.9.4 to study the effect of higher modes are displayed.
4.5.1 Results of Special Study Under 1.0 × El-Centro Ground Motion (NS)
The results of the case under this ground motion is shown in Figure 4.65 to Figure 4.70. In
general, once using higher number of modes of vibration, the estimated responses improved.
It requires up to nine modes which finally made estimation of the acceleration type responses
reasonably accurate while only four modes are good enough for Moment story type response,
and about six modes for Shear-force story type responses.
4.5.2 Results of Special Study Under 3.0 × San Fernando Ground Motion (NS)
The results of the case under this ground motion is shown in Figure 4.71 to Figure 4.76.
Since this ground motion excited more contribution from higher modes compared to the
other cases, it is observed that in general, higher number of modes were required in order to
estimate seismic responses. Acceleration type responses needed more than nine modes for
the floor acceleration estimation of some floors. While four modes were still enough to well
estimate Moment story type responses, up to nine modes were needed for Shear-force story
type responses, specifically only for Shear-force story estimation of some floors.
4.5.3 Results of the Special Study Under 1.0 × Northridge Ground Motion (NS)
The results of the case under this ground motion is shown in Figure 4.77 to Figure 4.82. The
same discussion as written for subsection 4.5.1 can also be made here as well; that is Nine,
four and six modes were sufficient for the estimations of acceleration type, Moment story
type and Shear-force story responses, respectively.
153
Figure 4.65 Verification of estimated relative floor acceleration envelope from MD-All
under 1.0 × El-Centro ground motion (NS) with different number of considered modes (N):
(a) N = 3, (b) N=4, (c) N=6, (d) N=9
Figure 4.66 Verification of estimated roof-top (relative) floor acceleration time-history

from MD-All when N = 9 under 1.0 × El-Centro ground motion (NS)
154
Figure 4.67 Verification of estimated Moment story envelope from MD-All
(a) N = 3, (b) N=4, (c) N=6, (d) N=9
155
Figure 4.68 Verification of estimated Shear-force story envelope from MD-All
(a) N = 3, (b) N=4, (c) N=6, (d) N=9
156
Figure 4.69 Verification of estimated force time-histories at the base from MD-All
utilizing the force prediction method mentioned in subsection 3.6.2.1 when N = 9
157
under 3.0 × San Fernando ground motion (NS) with different number of considered modes
(N): (a) N = 3, (b) N=4, (c) N=6, (d) N=9
F

from MD-All when N = 9 under 3.0 × San Fernando ground motion (NS)
158
under 3.0 × San Fernando ground motion (NS) with different number of
considered modes (N): (a) N = 3, (b) N=4, (c) N=6, (d) N=9
159
under 3.0 × San Fernando ground motion (NS) with different number of
160
161
under 1.0 × Northridge ground motion (NS) with different number of considered modes
(N): (a) N = 3, (b) N=4, (c) N=6, (d) N=9

from MD-All when N = 9 under 1.0 × Northridge ground motion (NS)
162
under 1.0 × Northridge ground motion (NS) with different number of
163
under 1.0 × Northridge ground motion (NS) with different number of
164
under 1.0 × Northridge motion (NS)
165
4.6 Performance of the Proposed Seismic Damage Evaluation Scheme Under the
Simplified Predetermined Scenario
There is a need to verify the proposed seismic damage evaluation scheme (section 3.2)
mainly based on the newly formulated modal decomposition approach: “Modal
Decomposition based on measured limited number of acceleration records” or “MD-Lim.”
But since in this study all these things were just proposed, the simplified predetermined
scenario explained section 3.10 was then adopted. As earlier explained, the most important
expected findings are to realize the inherent error (if any) along with the performance of the
proposed method without accounting for the error that will occur in practice. Previously, it
was shown in section 4.4 that among all, mode by mode, reference methods MD-All is the
most promising method utilized to estimate seismic response in terms of each mode
contribution. Despite the fact that only three modes were not enough to estimate all kinds of
seismic responses as shown in section 4.5, but for comparison with MD-Lim that can at most
estimate up to three mode contributions, the number of considered modes of vibration used
in MD-All was only set to be three (N = 3). Seismic demands calculated from NLRHA and
combined seismic demands estimated using MD-All (N=3), employing every proposed force
prediction method for any modal decomposition approach (either for MD-All or MD-Lim)
in subsection 3.6.2, were mainly used as reference seismic demands. Also, to show some
insights, those of SDF-UMRHA and MDF-UMRHA were occasionally used as reference
seismic demands.
Here is the preparation before the coming earthquake events based on the proposed seismic
damage evaluation scheme following the steps in subsection 3.2.1,
• step A-1 was done and explained in section 3.7

• step A-2 was completed by obtaining dynamic properties from section 4.2
• step A-3 was not needed to be done, for the numerical model in section 3.7 is already
perfect. It is the same as the “real building” of this study,
• and step A-4 was also finished in section 4.3.
Then three ground motions defined in section 3.8 come to shake the “real building”
separately in only y direction; the conditions of the “real building” before each seismic event
are set to be in the intact condition (no past histories of nonlinearity in any element occurring
before applying each ground motion to the numerical model).
4.6.1 Estimation of Seismic Demands Under 1.0 × El-Centro Ground Motion (NS)
After the “real building” was shaken by 1.0 × El-Centro ground motion (NS), the absolute
acceleration records (they are calculated by NLRHA of Perform3D as explained in section
3.10) were taken from two specified nodes of each floor belonging to the numerical mode in
Perform3D. The typical nodes of each floor and vertical arrangement of assumed-to-be-
sensor locations in this study are displayed in Figure 4.83 using the 2D plan view and 3D
view of the numerical model in ETABS. Now let us follow step B-1 mentioned in subsection
3.2.2 and 3.10.1. As displayed in the figure, floor no.11, no.23 and the roof-top floor were
chosen in this study that the absolute acceleration signals would be obtained from. Those
floors were chosen predicated upon the concept of observability mentioned in subsection
2.2.4.
166
Station 1, jth floor
Station 2, jth floor
Station 1,
Station 2,
34th floor
34th floor
Station 1, Station 2,
23rd floor 23rd floor
Station 2,
11th floor
Station 1,
11th floor
Station 1,
at the base
Figure 4.83 Identification of nodes from which absolute acceleration time-histories

were extracted out: (a) 2D plan view, (b) 3D view
167
𝑢̈ 𝑗𝑦 (𝑡)
𝑢̈ 𝑗𝑥 (𝑡)
𝑢̈ 𝑗𝜃 (𝑡)
Figure 4.84 Schematic representation of relative floor accelerations

associated with the center of mass of the jth floor
These records assumed to be from the accelerometers in practice were actually without noise,
for they were generated from the NLRHA analysis method. Thus, they were thought of as
perfectly filtered signals. Per the proposed scheme, at each station (each node) of each floor,
absolute acceleration signals in both x and y directions were extracted out. As a result, in
total not including the two signals at the base station, there were six signals in x direction
and the other six signals in y direction taken from the superstructure part. To make them
relative acceleration responses to be later utilized as the input of the next step, the individual
signal was subtracted in time-history manner by the absolute acceleration signal taken from
the base associated with its own direction. Up to this point, there were four filtered relative
acceleration signals associated with the DOFs of the stations (nodes) per floor i.e., two
signals in x direction and the other two in y direction. Then the final step of the main step B-
1 was to use those processed acceleration signals to determine the relative acceleration
signals associated with the DOFs of the center of mass of each floor (based on the mentioned
simplified dynamic DOFs of a 3D building mentioned in subsection 2.3.1.5). Following the
thoroughly explained procedure written in section 3.5, three signals per floor were
determined including relative translational acceleration signals in x and y directions and
(conceptual) rotational acceleration signal around z axis of the center of mass. To verify this
procedure, the relative accelerations of the nodes corresponding to the center of mass of each
floor were also additionally extracted as a result of NLRHA procedure in Perform3D and
used to verify the determined relative accelerations signals (since the numerical model has
Pure-sway behavior and the ground motion was only used to excite only in y direction, only
y components would be in comparison). The results in Figure 4.85 shows that the procedure
explained in section 3.5 had achieved the goal.
Let us denote 𝑢̈ 𝑗𝑥 (𝑡), 𝑢̈ 𝑗𝑦 (𝑡), and 𝑢̈ 𝑗𝜃 (𝑡) (schematically shown in Figure 4.84) as the relative
acceleration signals at the center of mass of the jth floor of the x and y translational directions,
and rotational around z axis direction. The reduced acceleration time-history vector, 𝒖̈ 𝑅 (𝑡)
to be used in Eq.(3.11) was expressed in this study as follows
168
𝒖̈ 𝑅 (𝑡) = 〈𝒖̈ 𝑥𝑅 (𝑡) 𝒖̈ 𝑦𝑅 (𝑡) 𝒖̈ 𝜃𝑅 (𝑡)〉𝑇
𝒖̈ 𝑥𝑅 (𝑡) = 〈𝑢̈ 34𝑥 (𝑡) 𝑢̈ 23𝑥 (𝑡) 𝑢̈ 11𝑥 (𝑡)〉𝑇
(4.1)
𝒖̈ 𝑦𝑅 (𝑡) = 〈𝑢̈ 34𝑦 (𝑡) 𝑢̈ 23𝑦 (𝑡) 𝑢̈ 11𝑦 (𝑡)〉𝑇
𝒖̈ 𝜃𝑅 (𝑡) = 〈𝑢̈ 34𝜃 (𝑡) 𝑢̈ 23𝜃 (𝑡) 𝑢̈ 11𝜃 (𝑡)〉𝑇
Next, it followed by the main step B-2: applying Modal Decomposition based on measured
limited numbers of acceleration records (MD-Lim) explained in section 3.4 to obtain
estimated modal acceleration based on MD-Lim, 𝑞̈̂𝑛 (𝑡) which as noted it represents the
contributions of the nth mode to the overall seismic responses. To elaborate on how it was
actually completed, firstly, the mode shape vectors, 𝝓𝑛 taken as a result of section 4.2 were
made to be in the form of reduced mode shape vectors, 𝝓𝑛𝑅 . It should be noted here that
since the original mode shapes required to be used in this study are the deflected shape of
the center of masses once moving freely (free vibration) under the contribution of one
particular mode, the reduced form of mode shape vectors also contain the values associated
with the centers of masses of the selected floors. The reduced mode shape of the nth mode,
𝝓𝑛𝑅 , expressed in this study is
𝝓𝑛𝑅 = 〈𝝓𝑛𝑥𝑅 𝝓𝑛𝑦𝑅 𝝓𝑛𝜃𝑅 〉𝑇

𝝓𝑛𝑥𝑅 = 〈𝜙𝑛𝑥34 𝜙𝑛𝑥23 𝜙𝑛𝑥11 〉𝑇
(4.2)
𝝓𝑛𝑦𝑅 = 〈𝜙𝑛𝑦34 𝜙𝑛𝑦23 𝜙𝑛𝑦11 〉𝑇
𝝓𝑛𝜃𝑅 = 〈𝜙𝑛𝜃34 𝜙𝑛𝜃23 𝜙𝑛𝜃11 〉𝑇
where 𝜙𝑛𝑥𝑗 , 𝜙𝑛𝑥𝑗 and 𝜙𝑛𝜃𝑗 are the mode shape values at the center of mass of the jth floor
of the nth mode in x direction, y direction and rotational around z axis direction, respectively.
Let us define reduced mode shape matrix,𝜱𝑅 , to be the matrix in which each column
contains one reduced mode shape vector, and starting from the first three mode (ranging
from the 1st to the 3rd mode of that direction) of x translational direction, and then y
translation direction, and lastly rotational around z axis direction. The matrix 𝜱𝑅 of this
study is
*e-3 *e-3 *e-3
1.000 1.000 1.000 0 0 0 0 0 0
0.658 −0.268 −0.772 0 0 0 0 0 0
0.261 −0.769 0.662 0 0 0 0 0 0
0 0 0 1.000 1.000 1.000 0 0 0
0 0 0 0.779 −0.100 −0.753 0 0 0
0 0 0 0.401 −0.836 0.550 0 0 0
0 0 0 0 0 0 −2.120 0.800 0.540
0 0 0 0 0 0 −1.860 0.060 −0.410
[ 0 0 0 0 0 0 −1.150 −0.760 0.230 ]
As can be confirmed that the behavior of the numerical model described in section 3.7 is
Pure-Sway, there is no movement in any other direction except that of the dominant
direction. The matrix A generated by following the procedure in section 3.4 of “real
building” in this study is as follows
169
1.000 1.158 1.075 0 0 0 0 0 0
1.158 4.002 0 0 0 0 0 0 0
1.075 0 1.0 0 0 0 0 0 0
0 0 0 0.390 0.016 −0.026 0 0 0
0 0 0 0.016 1.000 0 0 0 0
0 0 0 −0.026 0 1.000 0 0 0
0 0 0 0 0 0 0.450 −0.006 −0.014
0 0 0 0 0 0 −0.006 1.000 0.000
[ 0 0 0 0 0 0 −0.014 0.000 1.000 ]
To verify the matrix A, the expressions in Eq.(3.5) and (3.6) must be satisfied, or to put it
simply, all reduced mode shape vectors must be orthogonal to each other with respect to the
Orthogonal Filter matrix, A. The expression 𝜱𝑇𝑅 𝑨𝜱𝑅 was used for the verification; the
results of the expression is shown as follows
*e-3 *e-3 *e-3
4.880 0 0 0 0 0 0 0 0
0 −0.395 0 0 0 0 0 0 0
0 0 3.459 0 0 0 0 0 0
0 0 0 1.161 0 0 0 0 0
0 0 0 0 1.139 0 0 0 0
0 0 0 0 0 1.206 0 0 0
0 0 0 0 0 0 6.684 0 0
0 0 0 0 0 0 0 8.858 0
[ 0 0 0 0 0 0 0 0 3.514]
The matrix above implies that the matrix A was generated to have the expected property as
a filter matrix, for only diagonal elements contain non-zero value. The nth diagonal element
was defined as 𝐴𝑛 per section 3.4. It should be noted that the figures in all matrices shown
in this document were once rounded for the sake of display. Then Eq.(3.11) was used for the
determination of each mode estimated modal acceleration based on MD-Lim, 𝑞̈̂𝑛 (𝑡).
Estimated modal accelerations based on MD-Lim, 𝑞̈̂𝑛 (𝑡) of three important modes in y
direction along with those from MD-All, 𝑞̈̃𝑛 (𝑡) as a result of subsection 4.4.2.1 were plotted
together against time for the verification as displayed in Figure 4.86. The results show that
estimated modal accelerations based on MD-Lim followed the same trend of those from
MD-All, yet they were not exactly equal. Those from MD-Lim were likely to be
contaminated by higher frequency signals based on the observation. To ensure the inference,
FFT (Fast Fourier Transform) was used to turn estimated modal accelerations based on MD-
Lim into their frequency domain counterparts. The normalized Fourier amplitude against
frequency of each signal were plotted along with the associated reference from MD-All and
SDF-UMRHA in Figure 4.87. “Bell shapes” shown in the plots suggested that those
frequency components occurred as resonant responses. However, instead of only showing
the outstanding bell shape only around the natural frequency of the considered mode of
vibration which happened in the case of both MD-All and SDF-UMRHA, signals determined
from MD-Lim had a lot of outstanding bell shapes, compared to local frequency responses.
Furthermore, the expected resonant bell shape responses (of the 2nd and 3rd modes around
170
their resonant frequencies) existing in the plot of the 2nd and 3rd modes did not repeat in the
plot of the 1st mode, and vice versa; all of the evidence implied that the Orthogonal Filter
method (MD-Lim) formulated based on the fact that reduced mode shapes of the 1st, 2nd and
3rd modes are orthogonal to each other with respect to Orthogonal Filter matrix, A, can only
pull, for example, the 2nd and 3rd mode contribution (possibly some other modes in which
their reduced mode shapes were similar to those modes) out of the overall response (if the
interested mode is the 1st mode) instead of being able to purely decompose the 1st mode
contribution out. And that was why the responses in both time domain and frequency domain
of estimated modal accelerations showed the sign of contamination from higher frequency
signal. Moreover, to further ensure the idea, estimated acceleration time-histories of the
floors assumed to have mounted accelerometers predicted using MD-All (3 modes) were
used as the input of MD-Lim. It should be noted that since the results of MD-All present the
movement at the centers of masses already, so the step was directly skipped to B-2. The key
finding of MD-Lim having combined three mode contributions of acceleration records was
that MD-Lim could perfectly decompose the combined responses into each mode
contribution with respect to those from MD-All; however, the result was not documented
here in this thesis.
Knowing the main issue happening with the estimated modal accelerations based on MD-
Lim, 𝑞̈̂𝑛 (𝑡), let us move to the next main step B-3: “Relating estimated modal accelerations
to other type seismic demands”. It started with determining estimated modal coordinate of
each mode by double integrating the associated estimated modal acceleration. In this study,
a developed software for signal processing of strong-motion data (Seismosoft, 2013) was
utilized as a tool only for double integration of signals. The software uses “Trapezoidal rule”
for numerical double integration process, and has the potential to filter the processed / input
signals and deal with the error due to double integration process. However, with some certain
limitations, in this study built-in functions in MATLAB (MathWorks, 2018) was used to
deal with the two mentioned issues instead. Anyway, for the double integration process of
estimated modal accelerations under this ground motion did not leave any outstanding
cumulative error, so for this ground motion, no remedy of double integration error was
applied. The verification results of estimated modal coordinates in time domain are shown
in Figure 4.88 and the ones in frequency domain are in Figure 4.89. Fortunately, estimated
modal coordinates through the process of double integration were very slightly affected by
the contamination of higher frequency, especially the fundamental mode of vibration (1st
mode). Following section 3.6, estimated floor displacements, estimated inter-story drifts,
and estimated relative floor accelerations were calculated and their envelopes were shown
in Figure 4.90. It is observed that response envelopes of floor displacement and inter-story
drift were not significantly affected by the issue related to Orthogonal Filter approach while
the estimated relative floor acceleration envelope were fairly overestimated with respect to
the assumed-to-be-exact-three-mode-combined envelope estimated by MD-All. And one
more interesting observation is that relative acceleration envelopes, of the three floors
accelerometers assumed to be mounted, estimated by MD-Lim were exactly equal to those
of NLRHA. To prove the claim to be correct as well in terms of time-history, the verification
results of relative floor acceleration time-histories of those floors due to the ground motion
were plotted in Figure 4.91. This gave more information on how the Orthogonal Filter matrix
worked. When it decomposed the first mode out, it may have taken some other modes which
had similar mode shape vectors (speaking only about reduced vector) out as well causing the
so-called 1st mode estimated modal acceleration to be contaminated with those additional
171
modes incorrectly decomposed out together with the 1st mode. However, since the results
showed that MD-Lim did not overestimate the acceleration responses of those floors
compared with NLRHA results, this inference can be made that: the modes incorrectly
decomposed to be part of the 1st mode estimated modal acceleration would not be repeatedly
decomposed to be part of other modes again i.e., the second and third modes unless those
relative floor acceleration would have been overestimated.
Then, four ways force prediction of MD-Lim could be done following three main proposed
methods in subsection 3.6.2 to determine estimated story forces. It should be noted that the
estimated story forces determined by the same method for MD-All were also chosen to be
the reference responses of the associated estimated story forces from MD-Lim. Verification
results of base-force time-histories and story force envelopes of each method are displayed
as follows: results per story force prediction based on estimated floor acceleration time-
history vector (subsection 3.6.2.1) shown in Figure 4.92 and Figure 4.93; results per story
force prediction based on inelastic SDF relationships (subsection 3.6.2.2) in Figure 4.94 and
Figure 4.95; results per story force prediction based on the updated numerical model – 1st
version (subsection 3.6.2.3) in Figure 4.96 and Figure 4.97; results (only envelopes) per story
force prediction based on the updated numerical model – 2nd version (subsection 3.6.2.3) in
Figure 4.98 . Due to the findings in section 4.5, if there had been more numbers of modes,
modal decomposition based on measured accelerations from all floors (MD-All) would have
been utilized to estimate story forces very accurately. This inferred that at least even though
three modes were not sufficient, combined story forces determined by MD-All utilizing the
1st and 2nd force prediction methods should represent almost exact combined contributions
from three modes. On the other hand, combined story forces (contributed from three modes)
from both versions of the 3rd method employed for MD-All at least can be utilized as the
benchmarks for the comparisons with those being predicted by the same methods for MD-
Lim.
The results turned out that only the 3rd method of force prediction for both versions explained
in subsection 3.6.2.3 employed for MD-Lim were also the same as those from MD-All. This
is because of the fact that inelastic SDF forces which were mainly used to calculate story
force contributed from the nth mode predictions require estimated modal coordinates. Despite
the error caused by Orthogonal Filter approach (MD-Lim), estimated modal coordinates
from MD-Lim, 𝑞̂𝑛 , verified in Figure 4.88 seemed not to be greatly affected. And even if
they were affected more than this which will be shown to be the case for the other two cases
of ground motions written in the subsequent subsections, this method proved itself to be
“robust” enough unlike the first two methods explained in subsection 3.6.2.1 and 3.6.2.2.
The results showed that story forces predicted by those two methods for MD-Lim
overestimated the envelope of exact complex responses at the base calculated by NLRHA
by about four times (see Figure 4.93 and Figure 4.95).
The rest of this subsection was dedicated to the explanation on how these happened. Let us
start with the first method. Figure 4.99(a) shows the force vectors of Eq.(3.24) and (3.25) at
the time where base-shear predicted by this force prediction method for MD-Lim is the
highest. Eq.(3.24) is for the estimation of the resisting force vector of the numerical model
in x direction at time t, yet can be replaced for y direction for the sake of explanation.
Eq.(3.25) is for the estimation of the resisting force vector based on MD-Lim. Both equations
have the common vector which is -1 × force excitation vector, −𝑴𝑦 𝒊𝑢̈ 𝑔𝑦 (𝑡); this vector is
plotted in black color. Then, moving to -1 × inertia force vectors, the -1 × inertia force
172
vector of the numerical model in y direction, −𝑴𝑦 𝒖̈ 𝒚 (𝑡), was plotted in cyan per Eq.(3.24);
for Eq.(3.25), the -1 × inertia force vector, −𝑴𝑦 𝒖̈̂ 𝒚 (𝑡), was plotted in pink. As can be seen
that, since MD-Lim predicted wrongly the relative floor acceleration vectors at the time step,
the inertia force vector calculated at the time step was also wrong (yet those inertia forces of
the roof top, 23rd and 11th floors were equal to those of the reference). Then both -1 × inertia
force vectors were added with the -1 × excitation force vector, and they equated to the
̂ 𝑟𝑒𝑠,𝑦 (𝑡) of Eq.(3.25)
resisting force vectors. 𝑭𝑟𝑒𝑠,𝑦 (𝑡) of Eq.(3.24) was plotted in blue while 𝑭
was plotted in red. Now for the simplicity of explanation, those two resisting force vectors
were also plotted in Figure 4.99(b) with the same color yet in dashed line manner while the
solid ones represent the associated maximum resisting force envelopes calculated by each
equation i.e., Eq.(3.24) and (3.25). It can be seen that even though the estimated resisting
force of MD-Lim at the time (dashed red) were calculated wrongly, it was within the
envelope not only of itself (solid red) but also the envelope of its reference as well (solid
blue), and how come it made the base shear the highest value? It happened because at this
time step, the estimated resisting force of each floor predicted by MD-Lim were all in
positive direction while the reference ones (solid blue) were in both negative and positive
directions. In order to calculate Shear-force at the base, the summation of resisting forces of
all floors must be done, and that is why the error was accumulated and finally overestimated
up to four time of the reference value. Figure 4.99(c) shows the solid black lines as the
envelope of shear-force story vector taken from NLRHA and dashed black line as the shear-
force story vector at that time while the one estimated wrongly by MD-Lim employing the
first method was plotted in dashed red line. Even though the 1st force prediction method was
not based on the concept of mode of vibration, estimated relative floor accelerations wrongly
estimated by MD-Lim (mainly due to Orthogonal Filter matrix) resulted in erroneous
estimation of story forces.
Regarding the error caused by the force prediction method explained in subsection 3.6.2.2
employed for MD-Lim, overestimated combined story forces were originally due to
overestimated modal story forces, and the contribution of each mode story forces was indeed
calculated from inelastic SDF force in modal coordinate (see Eq.(3.29) as an example). This
means that the error displayed in Figure 4.94 and Figure 4.95 were most likely because of
these quantities. Thus, after this, the calculations of the inelastic SDF forces predicated upon
Eq.(3.28) would be examined on how error occurred for MD-Lim while they will be verified
with those from MD-All. Figure 4.100(a) showed the time-histories of forces in modal
coordinate (including excitation force, inertia force and inelastic SDF force) of the first mode
predicted by MD-Lim employing the second force prediction method while Figure 4.100(b)
displayed the reference time-histories of forces in modal coordinate of the same mode from
MD-All employing the same method. Figure 4.101 and Figure 4.102 belonged to the second
and third modes, respectively. For all plots, −𝑢̈ 𝑔𝑦 (𝑡) representing -1 × excitation force of
the SDF was plotted in black color; −𝐷̂̈ 𝑛 (𝑡) representing -1 × inertia force of the SDF was
plotted in red color; 𝐹̂𝑠𝑛 (𝑡)/𝐿𝑛𝑦 being inelastic SDF force was plotted in cyan. It should be
noted that damping forces of the SDFs were ignored since they were so small comparatively
as will be shown later. Based on the results, in general it seems that inelastic SDF force time-
history (plotted in cyan) of the first mode was mostly affected. The way to derive a cyan line
was approximately to sum the black line (-1 × excitation force) with the red line (-1 × inertia
force) together. Since the black line was the same in both subplot (a) and subplot (b), the red
line was the one causing difference between both cyan lines. Red lines were actually -1 ×
estimated modal acceleration of the nth mode either predicted by MD-Lim or MD-All.
173
Specifically, since the red lines of the first mode has the magnitude around that of the black
lines as shown in Figure 4.100. A little prediction error of red line for MD-Lim would have
caused a relatively big error to the summation (cyan line = inelastic SDF force). And it
happened; for example, result in Figure 4.100(a) always shows that the black line and red
line are in opposite direction while the green window in Figure 4.100(b) displays the time
period the red line, wrongly predicted by MD-Lim, had the same direction as the black line.
An inelastic SDF force (cyan line) value within that green window turned out to cause the
highest estimated inelastic SDF force of the 1st mode for MD-Lim employing this method
≈ 1600, and that was around four time higher than the envelope of inelastic SDF force of
the same mode for MD-All employing the same method ≈ 400. However, when considering
the verification results of the 2nd and 3rd modes in Figure 4.101 and Figure 4.102, the effect
was found to be mitigated. Since -1 × the inertia forces (red lines) were larger in value
compared with the -1 × excitation force (black line), the small error due to Orthogonal Filter
matrix inherently existing in red lines lesser affected the summation (cyan). That was clearly
shown in Figure 4.102 for the verification result of the 3rd mode. To clearly see how MD-
Lim causes error to estimated inelastic SDF force of each mode with respect to that of MD-
All, Table 4.3 to Table 4.5 were made. The tables showed the root mean square of force
time-histories in modal coordinate of Eq.(3.28) normalized with that of excitation force of
every mode. Interesting findings were found. Firstly, damping force in all cases were
considered relatively small. Secondly, MD-Lim predicted that inertia forces in all modes to
be relatively higher than those of MD-All for only about 11%, yet these discrepancies caused
different effect to the estimated resisting forces – or inelastic SDF force to be consistent. As
can be seen that the first mode “average” A/B ratio of resisting force was the highest, and
this support the idea already mentioned to be the source of error of the force prediction
method explained in subsection 3.6.2.2. The results could have been even worse if the
fundamental mode of the numerical mode had its natural frequency even lower than this, for
ground motions in section 3.8 seemed to have relatively high frequency components causing
the fundamental mode to response within the Inertial range.
Table 4.3 Root mean square of force time-histories in modal coordinate of the first mode
Inertia force Damping force Resisting force Excitation force Remark
A 1.214 0.026 0.683 1.000 MD-Lim
B 1.087 0.026 0.386 1.000 MD-All
A/B 1.117 1.000 1.769 1.000 -
Table 4.4 Root mean square of force time-histories in modal coordinate of the second mode
A 1.948 0.081 1.691 1.000 MD-Lim
B 1.732 0.081 1.311 1.000 MD-All
A/B 1.125 1.000 1.290 1.000 -
Table 4.5 Root mean square of force time-histories in modal coordinate of the third mode
A 4.107 0.106 4.057 1.000 MD-Lim
B 3.693 0.102 3.560 1.000 MD-All
A/B 1.112 1.039 1.140 1.000 -
174
Based on what have been shown up to now on seismic demands under 1.0 × El-Centro
ground motion (NS) estimated by Modal Decomposition based on measured limited number
of acceleration records (MD-Lim), the following “local” conclusions can be made on the
performance of utilizing MD-Lim as the modal decomposition approach in order to obtain
the contribution of each mode represented by its modal acceleration and finally relating them
to various kinds of seismic responses. First, due to the issue regarding how Orthogonal Filter
matrix unexpectedly worked, modal acceleration of each mode was slightly estimated
wrongly, specially being contaminated by higher frequency responses which could not be
filtered out by Orthogonal Filter matrix. Thus, relative floor acceleration responses were
overestimated compared with the assumed-to-be-combined-three-mode-contribution
acceleration responses (from MD-All), yet those determined through MD-Lim were not that
bad with respect to those from NLRHA. Once estimated modal accelerations were doubly
integrated, their estimated modal coordinates were shown not to be significantly affected.
Floor displacements and inter-story drifts estimated by MD-Lim were then shown to be
similar to those from MD-All. To predict story forces, four ways from three main methods
were employed. The results turned out that the first two methods (stated in subsection 3.6.2.1
and 3.6.2.2) employed for MD-Lim gave the unacceptable results despite the fact that the
performance of them shown in section 4.4.2 in the case of MD-All were ranked as the best
two methods to predict story forces beyond elastic range of the numerical model. Anyway,
it was shown that only slight error in estimated modal accelerations of MD-Lim made the
performance of them upside down. Regarding the last method in subsection 3.6.2.3 (having
two versions), both of them require estimated modal coordinates as inputs. As mentioned
earlier, they were not greatly affected by Orthogonal Filter approach (MD-Lim); thus,
despite the fact that they could not predict the estimated combined story forces to be equal
to the assumed-to-be-combined-three-mode-contribution story forces (equal to those
estimated by MD-All employing either one of the first two methods as shown in subsection
4.4.2), the method was robust enough to not be easily affected by slight error as in the first
two methods. This completes the main step B-3. And as earlier mentioned that in this study
the focus was the performance of the proposed scheme under the simplified predetermined
scenario in terms of seismic demands estimation. Step B-4 was just to compare the estimated
demands with the available resources (for example, documented in the literatures or
accessible experimental data) in order to evaluate seismic damage of the building. Since if
the results from step B-3 was accurate, the seismic damage evaluation results would also be
accurate. Thus, step B-4 was intentionally dismissed in this study.
175
Figure 4.85 Comparison between relative floor accelerations
at the centers of masses determined based on section 3.5
and reference associated accelerations obtained from NLRHA
176
Figure 4.86 Verification of estimated modal acceleration time-histories obtained as a
result of employing Modal Decomposition based on measured limited numbers
of acceleration records, (MD-Lim) under 1.0 × El-Centro ground motion (NS)
177
Figure 4.87 Verification of estimated modal accelerations in frequency domain obtained as
a result of employing Modal Decomposition based on measured limited numbers
178
Figure 4.88 Verification of estimated modal coordinate time-histories obtained as a result
of employing Modal Decomposition based on measured limited numbers
179
Figure 4.89 Verification of estimated modal coordinates in frequency domain obtained as a
180
Figure 4.90 Verification of estimated response envelopes based on MD-Lim
181
Figure 4.91 Estimated relative floor accelerations associated with the floors, having
accelerometers mounted to, predicted by MD-Lim
182
Figure 4.92 Verification of estimated force time-histories at the base from MD-Lim
utilizing the force prediction method mentioned in subsection 3.6.2.1
results from MD-All also being predicted by the same method in subsection 3.6.2.1
Figure 4.93 Verification of estimated story force envelopes from MD-Lim

183

184
(1st version, time-history based) under 1.0 × El-Centro ground motion (NS) / results from
MD-All also being predicted by the same method in subsection 3.6.2.3

st
(1 version, time-history based) under 1.0 × El-Centro ground motion (NS) /
185
(2nd version, envelope based) under 1.0 × El-Centro ground motion (NS) /
Figure 4.99 Plots dedicated to explanation on how the cumulative error occurred once
employing the force prediction method in subsection 3.6.2.1 for MD-Lim to predict Shear-
force story envelope under 1.0 × El-Centro ground motion (NS):
(a) and (b) story force vectors, (c) Shear-force story vectors
186
Figure 4.100 Verification of estimated force time-histories in modal coordinate
under 1.0 × El-Centro ground motion (NS) of the 1st mode predicted using the method in
subsection 3.6.2.2: (a) MD-All (reference), (b) MD-Lim

under 1.0 × El-Centro ground motion (NS) of the 2nd mode predicted using the method in
187
under 1.0 × El-Centro ground motion (NS) of the 3rd mode predicted using the method in
188
4.6.2 Estimation of Seismic Demands Under 3.0 × San Fernando Ground Motion (NS)
All of the steps and explanations on how they were completed in order to obtain the seismic
demands under this ground motion, staring from B-1 to B-3, were exactly the same and
thoroughly explained in subsection 4.6.1, so those things were note repeated here in this
subsection. But the very important thing that was not shown in subsection 4.6.1 was how to
deal with error caused by double integration process of estimated modal accelerations (step
B-3); however, that was because it did not happen in the case of 1.0 × El-Centro ground
motion (NS). It should be noted that the error due to double integration process did not just
happen only in the case of MD-Lim but also that of MD-All (but intentionally, not earlier
reported). Figure 4.103 showed the cumulative error of estimated modal coordinates caused
by double integration process. Through the observation of the results obtained, the
cumulative error implies that there was some interruption from low frequency signal. To
correct it the common way was to the find the average tend of the error and subtract it out
from the signal. The technique is commonly known as “Baseline correct.” However, the past
results from literatures show in common that this method cannot perfectly get rid of the error.
Therefore, another way from the very same basic idea is implemented instead. That is to get
rid of the error caused by low frequency signal in frequency domain by applying digital
Highpass filter designed by the built-in function in MATLAB (MathWorks, 2018) called
“filterDesigner.”
FFT were used to turn estimated modal accelerations and estimated modal coordinates into
frequency domain signals, and then, Fourier magnitude of them were determined. The key
is to plot estimated modal acceleration and modal coordinate of the same mode together in
the same plot to observe that at which frequency Highpass filter should be applied so that
the erroneous low frequency signal caused by double integration can be taken out. Figure
4.105 showed the plots mentioned. Consider the 3rd mode, since its natural frequency is 1.566
Hz (section 4.2), the resonant response as expected showed bell shape around that frequency
for both estimated modal acceleration and modal coordinate. For the modal acceleration one,
even though the “background response smaller than 0.5 Hz was also shown in the plot, yet
they were so small relatively. It was also confirmed by the plot of modal coordinate. So, for
this mode a Highpass filter designed with “Equiripple” method (FIR type), having passing-
frequency set to be 0.5 Hz and stopping frequency set to be 0.01 Hz was utilized as the digital
filter. The characteristic of the filter was shown in Figure 4.104. It was used to filter the 3rd
mode estimated modal coordinate. Regarding the 2nd mode, its natural frequency is 0.772 Hz
(section 4.2), the resonant response as expected showed bell shape around that frequency for
both signals. With the same idea, about 0.5 Hz could be set as passing-frequency, so the filter
in Figure 4.104 was used to filter the estimated modal coordinate of the 2nd mode as well.
Consider the 1st mode plot, the bell shape behavior displayed to be around 0.22 Hz (near the
natural frequency of the 1st mode). The same filter could not be used because for this mode
the passing-frequency was around 0.16 Hz. Unfortunately, for the passing-frequency as low
as 0.16 Hz, the filter must be designed to attenuate very fast which is not practical. Thus, the
traditional method would be employed instead. But let us finish mentioning about those
signals being filtered. After the signals were filtered, the delay effect happened. These
filtered signals were corrected by being shifted equal to “filter order”/2 samples. An example
of the delay effect of the 2nd mode estimated modal coordinate after being filtered and the
one that the delay effect had already been taken out is shown in Figure 4.106. The thing is,
because of the delay effect some parts of the records would be lost and this should be noted.
To deal with the first mode, another alternative traditional method was to find the baseline
189
(trend line) of the error and subtract them out of the erroneous part instead. The erroneous
part of the signal (after the trend seemed to be interrupted with the error) was fitted with a
3rd degree polynomial function. After the equation of the trend was obtained, the trend of
error was subtracted out of the erroneous part. This completed the how error due to double
integration was dealt with in this study.
Verification results of estimated modal accelerations and estimated modal coordinates (after
being corrected), having those from MD-All as their references are shown in Figure 4.107
and Figure 4.108. Estimated modal accelerations based on MD-Lim were contaminated by
higher modes responses as in the previous case while estimated modal coordinates were only
slightly affected as also similar to the previous case. Then, they were used to determine
relative floor accelerations, floor displacements and inter-story drifts. The envelopes of the
estimated responses are displayed in Figure 4.109. Story forces along with the story force
envelopes of each proposed force prediction method were subsequently determined. The
base forces and story force envelopes of the method explained in subsection 3.6.2.1 and
3.6.2.2 were displayed from Figure 4.110 to Figure 4.113. As expected, the two methods
were too sensitive to error and could not be employed as force prediction of MD-Lim. Figure
4.114 and Figure 4.115 showed that the negative effect caused by Orthogonal Filter approach
(MD-Lim) did not significantly affect estimated story forces predicted by the method of
subsection 3.6.2.3 (1st version) in terms of both envelope and time-history compared with
MD-All results. However, this force prediction method itself overestimated the story forces.
And since the method adopted the very same concept that SDF-UMRHA (directly) and
MDF-UMRHA (indirectly) had, the comparison results between story force envelopes of
MD-Lim and of them were fairly similar as shown in Figure 4.116. The concept was that
“the modal hysteretic behavior of the numerical model of each mode created through Cyclic
Pushover analysis using modal inertia force pattern of that mode is independent to those of
the other modes of vibration.” By ignoring the coupling effect mentioned, despite combining
only contributions from three dominant modes, the estimated story forces were typically
overestimated under this ground motion. Lastly verification results of the force prediction
method per the 2nd version of subsection 3.6.2.3 employed for MD-Lim as shown in Figure
4.117, ensured that the 3rd force prediction method based on estimated modal coordinate
were not affected by Orthogonal Filter approach (MD-Lim). And as being mentioned
consistently, the method itself was uncertain as in this case it reasonably estimated Shear-
force story envelope, yet overestimated Moment story envelope under this ground motion.
190
Figure 4.103 Cumulative error, caused by double integration process,
of modal coordinates under 3.0 × San Fernando ground motion (NS)
Figure 4.104 Magnitude response of the Highpass filter employed to deal with low
frequency signals caused by double integration process
191
Figure 4.105 Esimated modal accelerations and erroneous estimated modal coordinates
caused by double integraton process in frequency domain
Figure 4.106 Delay effect as a result of Highpass filter
192
of acceleration records, (MD-Lim) under 3.0 × San Fernando ground motion (NS)
193
of acceleration records, (MD-Lim) under 3.0 × San Fernando ground motion (NS)
194
under 3.0 × San Fernando ground motion (NS):
(a) estimated floor displacement envelope, (b) estimated inter-story drift envelope,
195
under 3.0 × San Fernando ground motion (NS):/

196

197
(1st version, time-history based) under 3.0 × San Fernando ground motion (NS):/

198
both reference analyses methods also included for the sake of insightful comparison

nd
(2 version, envelope based) under 3.0 × San Fernando ground motion (NS):/
199
4.6.3 Estimation of Seismic Demands Under 1.0 × Northridge Ground Motion (NS)
Following what had been done in the previous subsections, the estimated modal
accelerations and estimated modal coordinates were determined, verified and displayed in
Figure 4.118 and Figure 4.119. Then, they were used to determine relative floor
accelerations, floor displacements and inter-story drifts. The envelopes of the estimated
responses are verified and displayed in Figure 4.120. Then force prediction methods for MD-
Lim in section 3.6 were employed in order to obtain story forces. The verification results of
all proposed force prediction method employed for MD-Lim are shown from Figure 4.121
to Figure 4.128. The general conclusions of the verification of estimated seismic responses
based on MD-Lim under this ground motion was the same as those from the previous
subsection.
It should be noted here that even though the force prediction method based on the updated
numerical model from subsection 3.6.2.3 (no matters time-history or envelope based
version) was used to estimate story force demands under all selected ground motions for
both MD-Lim and MD-All to be almost similar to each other, it did not mean that those story
force envelopes displayed in subsection 4.6.1, 4.6.2 and 4.6.3 predicted by them were
accurate. It meant that the method was robust enough that the error due to Orthogonal Filter
approach (MD-Lim) did not significantly affect the results. For the accurate combined-three-
mode-contribution story forces were shown in section 4.5, and they were not equal to those
predicted by the force prediction method per subsection 3.6.2.3. Nevertheless, the estimated
story force envelopes under this ground motion in Figure 4.128 predicted by the envelope
based of the 3rd method were shown to be surprisingly accurate for both Shear-force and
Moment story envelopes.
200
of acceleration records, (MD-Lim) under 1.0 × Northridge ground motion (NS)
201
of acceleration records, (MD-Lim) under 1.0 × Northridge ground motion (NS)
202
under 1.0 × Northridge ground motion (NS):
(a) estimated floor displacement envelope, (b) estimated inter-story drift envelope,
203
under 1.0 × Northridge ground motion (NS)/

204

205
st
(1 version, time-history based) under 1.0 × Northridge ground motion (NS)/

(1st version, time-history based) under 1.0 × Northridge ground motion (NS)/
206
(1st version, time-history based) under 1.0 × Northridge ground motion (NS)/
both reference analyses methods also included for the sake of insightful comparison

nd
(2 version, envelope based) under 1.0 × Northridge ground motion (NS)/
207
4.7 The Performance of the Proposed Additional Step to Enhance the Accuracy of
the Proposed Scheme on Account of Higher Mode Effects
It was explained in section 3.11 for the need why further simplified calculations are needed
to account for the higher mode effects. In this section, the results of the story forces estimated
based on Orthogonal Filter approach per section 4.6 being corrected for their higher mode
effects are to be examined. According to section 4.6, the proposed force prediction methods
explained in subsection 3.6.2.1 and 3.6.2.2 were not robust enough to be employed for MD-
Lim, and that was the reason why the story forces predicted by them for MD-Lim would not
be further corrected by the procedure explained in section 3.11. Thus, let us focus of the 3rd
method. The correction process of the time-history based version of subsection 3.6.2.3 was
that the contribution from the fourth to the sixth mode of story forces of each floor calculated
by Modal Response History Analysis (SDF-MRHA), having the measured ground motion
as the only input, would be added up to the combined contributions from three modes in
time-history manner. On the other hand, if the envelope based was utilized, the envelope of
the considered story force contributed by each additional mode (through SDF-MRHA)
would be included in the combined envelope of the story force through SRSS modal
combination rule.
4.7.1 Verification of Corrected Estimated Story Forces Under 1.0 × El-Centro Ground
Motion (NS)
After three more mode contributions calculated by SDF-MRHA were added, the verification
results of MD-Lim employing the time-history-based version in Figure 4.129 showed that
correcting Moment story contributed by those higher modes were not necessary while
correcting Shear-force story contributed by those higher modes indeed helped fulfill the
missing envelope values as can be seen that the correct envelope lines were closer to the true
envelopes of NLRHA. The results were very satisfactory. Regarding the verification results
of MD-Lim employing the envelope-based version in Figure 4.130, Figure 4.131 and Figure
4.132, Moment story contributions added from higher modes were only slight; this agreed
with the previous results. However, Shear-force story contributions calculated from SDF-
MRHA once included by SRSS method instead of by time-history manner did not give good
results compared with those from time-history based one, yet it was better than not being
included.
4.7.2 Verification of Corrected Estimated Story Forces Under 3.0 × San Fernando Ground
Motion (NS)
The verification results were displayed from Figure 4.133 to Figure 4.136. Under this ground
motion which specifically caused a lot of contributions from higher modes of vibration of
the numerical model compared with the case of the first ground motion, both versions of the
method employed for MD-Lim already obviously overestimated Moment stories of each
floor; this was even without the correcting part of higher modes added. Thus, the corrected
envelopes of them did not show any meaningful information. Similarly, the envelope of
Shear-force story predicted based on the time-history version of the method in subsection
3.6.2.3 overestimated the responses of some floors already without being corrected, so once
it was corrected, it further overestimated the responses. On the other hand, since the
208
envelope-based method did originally not overestimate Shear-force story, once corrected,
the envelope was reasonably estimated with slight overestimation of some floors.
4.7.3 Verification of Corrected Estimated Story Forces Under 1.0 × Northridge Ground
Motion (NS)
The verification results were displayed from Figure 4.137 to Figure 4.140. As earlier
discussed, the time-history based method in subsection 3.6.2.3 in general seemed to
overestimate story forces either for the case of MD-All or MD-Lim and the effect of
overestimation was the highest in the previous case which had a lot of contributions from
higher modes (i.e., the second and third modes). Under this ground motion, even though the
effect seemed to be slightly mitigated (only the first and second modes contributed a lot),
yet the original estimated story forces estimated by time-history based version of this method
for MD-Lim overestimated the results. This was why the corrected story forces did not give
any meaningful result. Conversely, the verification results of those estimated by envelope-
based version seemed to outperform the counterparts before the correction. And once they
were corrected, the results suggested that higher modes contributions did play important
roles here under this ground motion.
209
Figure 4.129 Verification of the estimated story forces predicted by MD-Lim employing the
1st version (time-history based) force prediction method in subsection 3.6.2.3 after being
further corrected by higher mode contributions calculated by SDF-MRHA analysis
procedure under 1.0 × El-Centro ground motion (NS)
2nd version (envelope based) force prediction method in subsection 3.6.2.3 after being further
corrected by higher mode contributions calculated by SDF-MRHA analysis procedure
210
Figure 4.131 SRSS envelopes of estimated Moment story envelopes along with
their mode by mode envelopes under 1.0 × El-Centro ground motion (NS):
(a) three modes from MD-Lim only,
(b) three modes from MD-Lim plus three modes from SDF-MRHA
their mode by mode envelopes under 1.0 × El-Centro ground motion (NS):
211
Figure 4.133 Verification of the estimated story forces predicted by MD-Lim employing
the 1st version (time-history based) force prediction method in subsection 3.6.2.3 after
being further corrected by higher mode contributions calculated by SDF-MRHA analysis
procedure under 3.0 × San Fernando ground motion (NS)
2nd version (envelope based) force prediction method in subsection 3.6.2.3 after being
procedure under 3.0 × San Fernando ground motion (NS)
212
their mode by mode envelopes under 3.0 × San Fernando ground motion (NS):
their mode by mode envelopes under 3.0 × San Fernando ground motion (NS):
213
1st version (time-history based) force prediction method in subsection 3.6.2.3 after being
procedure under 1.0 × Northridge ground motion (NS)
2nd version (envelope based) force prediction method in subsection 3.6.2.3 after being further
corrected by higher mode contributions calculated by SDF-MRHA analysis procedure under
1.0 × Northridge ground motion (NS)
214
their mode by mode envelopes under 1.0 × Northridge ground motion (NS):
their mode by mode envelopes under 1.0 × Northridge ground motion (NS):
215
CHAPTER 5
5 CONCLUSIONS AND RECOMMENDATIONS
5.1 Conclusions
According to the fact that there were acceleration records measured by accelerometers
mounted on top of buildings during seismic events supposed to be utilized for the sake of
Structural Health Monitoring of those considered buildings after the seismic events, these
records should have been more beneficially utilized. Therefore, the focus of this study was
on the proposed seismic damage evaluation scheme utilizing only limited number of
acceleration records measured under the potential ground motion. The key of the proposed
scheme was the newly developed modal decomposition technique called Orthogonal Filter
technique (Modal Decomposition based on measured limited number of acceleration
records) formulated in this study. With this approach, only those measured acceleration
records from four levels, various seismic demands due to the earthquake event can be rapidly
estimated and further used to evaluate seismic damage of the considered building after the
seismic event, and this helps practicing engineers appraise the need for any repair of
damaged components in the building.
In order to verify and identify the effectiveness of Orthogonal Filter technique utilized in the
proposed seismic damage evaluation scheme, the simplified predetermined scenario was
purposefully set. It was expected that the inherent error caused by this approach (if any)
could be found and realized in this study; the scenario was set such that practical error due
to measurement and modal updating process were not to be accounted in this study. The
numerical model in this study was used as both “real building” and perfectly updated
numerical model while the time-histories extracted from the updated numerical model under
selected ground motions being applied only in one direction (through NLRHA) were
assumed to be perfectly filtered absolute acceleration signals. The reference seismic
demands used in this study were determined from SDF-UMRHA, MDF-UMRHA and MD-
All (Modal Decomposition based on measured accelerations from all floors). However, since
both MD-All and MD-Lim could not be directly used to relate to story forces, three force
prediction methods based on estimated floors accelerations, inelastic SDF relationships and
updated numerical model were proposed in this study to satisfy the issue.
Furthermore, to ensure the source of error, the special study on how many modes of vibration
are required to accurately estimate seismic demands under a ground motion. Based on the
insights gained from the special study, the additional step to correct the insufficiency of the
higher modes effect in order to enhance the accuracy of estimated story forces predicted by
Orthogonal Filter technique (MD-Lim) was proposed. The further steps were only to perform
SDF-MRHA procedure for some higher modes and add the calculated contributions to the
already estimated story forces as a result of MD-Lim employing any of the proposed force
prediction methods.
216
The following conclusions predicated upon the results of this study are as follows:
Regarding reference methods in this study
1. Within the linearly elastic range of the numerical model, SDF-UMRHA, MDF-
UMHA and MD-All could estimate the modal contribution of any response type to
be almost equal to each other. Even though the way each reference method was
formulated was different, they were based on the same concept of mode of vibration
within elastic range. Also, the results from all reference methods agreed with each
other that, only contributions from three modes of vibration (of the same direction)
were not enough to accurately estimate Shear-force story and relative floor
acceleration response types.
2. Beyond the linearly elastic range of the numerical model, each reference method
started to estimate different modal responses, yet those from SDF-UMRHA and
MDF-UMRHA were closer to each other with respect to those of MD-All. Regarding
combined responses of estimated floor displacements and inter-story drifts, those of
MD-All were the most accurate in terms of both time-history and envelope while all
methods suggested still that only three modes were not enough to estimate relative
floor accelerations.
Concerning story force prediction, MD-All employing the proposed force prediction
methods based on estimated accelerations and inelastic SDF relationships estimated
Moment stories accurately, and estimated that Shear-story envelope could not be
reasonably predicted with only contributions from three modes which agreed with
the results in linearly elastic range.
Once MD-All employed the time-history based version of the force prediction based
on updated numerical model, the story force envelopes estimated by it typically
overestimated those calculated by NLRHA. Despite the fact that these results
followed the same trend of those from both of the reference analysis methods, they
and the previous results within the elastic range were shown to be contradictory.
Finally, once the envelope-based method of the force prediction method based on the
updated numerical model was employed by MD-All, in general, it overestimated the
Moment stories as in the same way the time-history version does, yet surprisingly
predicted Shear-story response type with reasonably accuracy.
To finalize, MD-All was observed to be the most accurate method concerning non
story-force response types and MD-All employing either the 1st or 2nd proposed force
prediction methods to be the most accurate method with regard to story-force
prediction, in spite of the fact that only three dominant modes in one direction could
not reasonably predict Shear-story and relative floor acceleration response types.
Thus, to ensure that the error of the estimation was because of the inefficiency of
high modes of vibration included, MD-All along with the two force prediction
methods were further used in the special study.
217
Regarding special study on how many modes of vibration are required to accurately estimate
various seismic demands
1. Since it was earlier found that three modes were enough to reasonably estimate floor
displacements and inter-story drifts, only the rest of response types were considered
on this special study.
2. Three modes were actually sufficient to reasonably estimate Moment stories; the
results showed that contributions up to only four modes could accurately estimate
this response type.
3. It typically required up to six modes to estimate Shear-force stories. However, in the
rare case of such ground motion specifically exciting higher modes, up to nines mode
were actually needed.
4. It typically required up to nine modes to reasonably estimate relative floor
acceleration responses.
Regarding the performance of Orthogonal Filter approach (MD-Lim) within the context of
the simplified predetermined scenario
1. Estimated modal accelerations based on MD-Lim were contaminated with higher

mode responses due to the fact that Orthogonal Filter approach could not perfectly
decompose only the interested mode of vibration, yet estimated modal coordinates
determined through double integration process were not significantly affected by this
issue
2. Since estimated floor displacements and inter-story drifts were calculated using only
estimated modal coordinates of three modes from MD-Lim, the results were
satisfactory.
3. On the other hand, relative acceleration responses required estimated modal
accelerations from MD-Lim which were not perfectly decomposed, the envelope of
the response was overestimated compared with the combined contributions from
three important modes based on MD-All. However, that overestimation of estimated
relative floor accelerations of some floors seemed to be fairly reasonable with respect
to the calculated demands from NLRHA. In general, it was not that satisfactory.
4. Once employing the force prediction methods for MD-Lim based on estimated floor
accelerations and inelastic SDF relationships, instead of resulting in very accurate
combined contributions of three modes as in the case of MD-All, they unacceptably
overestimated the story forces. The two proposed force prediction methods were then
concluded to not be usable as force prediction methods for MD-Lim despite its best
performance for MD-All. This was due to the contamination of higher mode
responses in the decomposed modal acceleration of each mode.
5. About the performance of both versions of the third force prediction method based
on the updated numerical model, compared with those associated story forces
estimated by MD-All, the results estimated by MD-Lim employing the same version
yielded similar results. This means that the method was robust enough to not be
significantly affected by Orthogonal Filter approach (MD-Lim). Despite the fact that
the method typically overestimated story forces of some floors, the story forces,
especially Shear-force stories, estimated by three modes contributions were not
enough to predict the responses of other floors.
218
Regarding the performance of the proposed additional step to enhance the accuracy of the
proposed seismic scheme on account of higher mode effects
1. Concerning the responses due to the ground motion that excite mainly only the
contribution of the 1st mode, this correction method could really enhance the
accuracy of estimated story forces predicted using any version of the force prediction
method based on the updated numerical model.
2. When the responses were caused by the ground motions that excite the higher modes
more, time-history based on the 3rd method employed for MD-Lim originally
overestimated story forces in general, so the correction process did not pay off.
Similarly, Moment stories predicted by the envelope-based version under the same
characteristic of ground motion led to the same conclusion. However, Shear-force
stories predicted by the envelope-based version of the 3rd method employed for MD-
Lim along with the correcting step surprisingly gave reasonably accurate results.
5.2 Recommendations for the Future Studies
In this study, Orthogonal Filter approach was formulated based on acceleration records of
four levels which can only be used to estimate the combined responses based on the
contribution of three modes, and those responses were affected by the scheme itself. The
future research could implement even other ways how each mode contribution should be
decomposed so that error would not happen as in the case of this study. Furthermore, if the
same path is chosen to be followed, the performance of Orthogonal Filter approach within
more complicated scenarios should be checked. To illustrate, the numerical model chosen
may have asymmetric plans causing Coupled behavior, or the exciting ground motions may
have two components in direction, or the error in practice may be taken into consideration
in the future study including the error due to model updating and noisy signals caused by the
field measurement.
219
REFERENCES
Antonella, S., Carmelo, G., & Marco, G. (2015). Post-earthquake Continuous Dynamic
Monitoring of the Gabbia Tower in Mantua, Italy. Construction and Building
Materials, 101-112.
ASCE. (2013). Minimum Design Loads for Buildings and Other Structures. ASCE/SEI 7-10.
Brincker, R., Lingmi, Z., & Palle, A. (2001). Modal Identification of Output-only Systems
Using Frequency Domain Decomposition. Smart Materials and Structures, 441-445.
Carlo, R., & Giovanni, F. (2014). Operational Modal Analysis of Civil Engineering
Structures: An Introduction and Guide for Applications. New York.
Celebi, M., Huang, M., Shakal, A., Hooper, J., & Klemencic, R. (2013). Ambient Response
of a Unique Performance-based Design Tall Building with Dynamic Response
Modification Features. The Strucural Design of Tall and Special Buildings, 816-829.
Chopra, A. K. (2012). Dynamics of Structures: Theory and Applications to Earthquake

Engineering, Fourth Edition. Boston: Prentice Hall.
Chopra, A. K., & Goel, R. K. (2001). A Modal Pushover Analysis Procedure to Estimate
Seismic Demands for Buildings : Theory and Preliminary Evaluation. University of
California, Berkeley: Pacific Earthquake Engineering Research (PEER), PEER
Report 2001/03.
Chopra, A., & Goel, R. (2004). A Modal Pushover Analysis Procedure to Estimate Seismic
Demands for Unsymmetric-plan Buildings. Earthquake Engineering and Structural
Dynamics, 33:903-927. doi:10.1002/eqe.380
Cimellaro, G. P., Pianta, S., & Stefano, A. D. (2012). Output-Only Modal Identification of
Ancient L'Aquila City Hall and Civic Tower. Journal of Strucutral Engineering, 481-
491.
CSI. (2006). Perform3D - Nonlinear Analysis, Design and Drafting of Building Systems.
California: Computers and Structures, Inc.
220
CSI. (2017). ETABS - Integrated Analysis, Design and Drafting of Builidng Systems.
California: Computers and Structures, Inc.
Department of Public Works and Highways . (2015). Guidelines and Implementing Rules on
Earthquake Recording Instrumentation for Buildings. Manila.
Dionysius, M. S., & Fujino, Y. (2014). Long-term Seismic Monitoring of Base-isolated

Building with Emphasis on Serviceability Assessment. Earthquake Engineering and
Structural Dynamics. doi:10.1002/eqe.2538
Doebling, S., Charles, R., & Prime, M. (1996). A Summary of Vibration Based Damage
Identification. Los Alamos: Los Alamos National Laboratory.
Dora, F., Mariella, D., Nicola, I. G., & Michele, M. (2012). Ambient Vibration Testing,
Dynamic Identification and Model Updating of a historic tower. NDT&E
International, 88-95.
Erwin, K. (2011). Advanced Engineering Mathematics. John Wiley & Sons, Inc.
Frigui, F., J, P. F., C, M., O, D., F, P., & S, J. (2018). Global Methodology for Damage
Detection and localization in Civil Engineering Structures. Engineering Structures,
686-695.
Gentile, C., & Saisi, A. (2007). Ambient Vibration Testing of Historic Masonry Towers for
Structural Identification and Damage Assessment. Construction and Building
Materials, 1311-1321.
Hyo, S. P., & Byung, K. O. (2018). Real-time Strucutural Health Monitoring of a Supertall
Building Under Construction Based on Visual Modal Identication Strategy.
Automation in Construction 85, 273-289.
James, M. W., & Pan, T.-C. (2008). Identifying Loading and Response Mechanisms from Ten
Years of Performance Monitoring of a Tall Building. Performance of Constructed
Facilities, 24-34.
221
Jia-Zhan, S., Yong, X., Lu, C., Xin, Z., Qi-Lin, Z., You-Lin, X., . . . Ai-Rong, C. (2013). Long-
term Structural Performance Monitoring System. Civil Structure and Health
Monitoring, 3:49-61.
Karbhari, V., Guan, H., & Sikorsky, C. (2009). 7 - Operational Modal Analysis for Vibration-
based Structural Health Monitoring of Civil Structures. The Journal of Structural
Health Monitoring of Civil Infrastructure Systems, 213-259.
Karunkritkul, K. (2019). An Intergrated Study of Structural Health Monitoring by

Measurements of Acceleration and Strain caused by Ambient Effects. Thailand :
Asian Institute of Technology.
Kato, M., & Shimada, S. (1986). Vibration of PC Bridge During Failure Process. The Jounal
of Strucural Engineering, 1692-1703.
Kaynardag, K., & Soyoz, S. (2017). Effect of Identification on Seismic Performance

Assessment of a Tall Building. Bull Earthquake Engineering, 3227-3243.
Kijewski, T., & Kareem, A. (2002). On the Reliability of a Class of System Identification
Techniques: Insights from Bootstrap Theory. Structural Safety 24, 261-280.
LATBSDC. (2017). An Alternative Procedure for Seismic Analysis and Design of Tall
Buildings Located in the Los Angeles Region. Los Angeles Tall Buildings Structural
Design Council.
Maria, P. L., & Celebi, M. (2019). Seismic Structural Health Monitoring From Theory to
Successful Applications. Cham, Switzerland: Springer Tracts in Civil Engineering
(STCE).
MathWorks. (2018). MATLAB 2018a. Natick, Massachusetts, United States: The

MathWorks, Inc.
Mehmood, T. (2015). Investigation of Nonlinear Seismic Response of High-rise Wall

Structures Using Modal Decomposition Technique. Pathum Thani: Asian Institute of
Technology.
222
Mehmood, T., Warnitchai, P., & Suwansaya, P. (2018). Seismic Evaluation of Tall Buildings
Using a Simplified but Accurare Analysis Procedure. Journal of Earthquake
Engineering 22, 356-381.
Najam, F. (2017). Evaluation of Nonlinear Seismic Demands of High-Rise RC Shear Wall

Buildings Using Simplified Analysis Procedures. Asian Institute of Technology.
Niyompanitpatana, S. (2017). Effects of Masonry Infill Walls with Different Opening

Configurations on Seismic Behavior of Long-span Gravity-load-designed RC
Frames. Pathum Thani: Asian Institute of Technology.
Orakcal, K., & Wallace, J. W. (2004). Flexural Modeling of Reinforced Concrete Walls -
Experimental Verification. ACI Structural Journal. doi:10.1061/(ASCE)0733-
9445(2004)130:4(618)
Panday, S. (2017). Developement of Modal Hysterectic Model for the Seismic Response
Analysis of Tall Building with RC Shear Walls. Pathum thani: Asian Institute of
Technology.
PEER/ATC. (2010). Modeling and Acceptance Criteria for Seimsic Design and Analysis of
Tall Buildings. Applied Technology Council and Pafific Earthquake Engineering
Research Center.
Rytter, A. (1993). Vibrational Based Inspection of Civil Engineering Structures. Technology

and Structural Engineering, Aalborg University.: Aalborg: Dept. of Building.
Saisi, A., Gentile, C., & Guidobaldi, M. (2015). Post-earthquake continuous dynamic
monitoring of the Gabbia Tower. Construction and Building Materials, 101-112.
Salawu, O. (1997). Detection of Structural Damage Through Changes in Frequency: a

Review. The journal of Engineering Structures, 718-723.
Seismosoft. (2013). SeismoSignal v5.1 - A Computer Program for Signal Processing of

Strong-motion Data. Seismosoft Ltd.
223
Suthasit, M., & Warnitchai, P. (2018). Seismic Performance of US-code Conforming RC
Moment Frames Constructed with Regularly Distributed Masonry Infills.
Earthquake Engineering.
Tahir, M., Pennung, W., & Phichaya, S. (2018). Seismic evaluation of tall buildings using a
simplified but accurate analysis procedure. Earthquake Engineering, 22:356-381.
Tsuchihashi, T., & Yasuda, M. (2017). Rapid Diagnosis Systems Using Accelerometers in
Seismic Damage of Tall Buildings. Internaltional Journal of High-Rise Buidlings,
207-216.
Vasanthapragash, N. (2018). Assessment of Nonlinear Seismic Demand of High-Rise RC

Shear Wall Buildings by Modal Approach. Pathum Thani : Asian Institute of
Technology .
Yong, X., Yi-qing, N., Peng, Z., Wei-yang, L., & Jan-ming, K. (2011). Stress Development
of a Supertall Structure During Construcution: Field Monitoring and Numerical
Analysis. Computer-Aided Civil and Infrastructure Engineering, 542-559.
Yukio, T., Akihito, Y., Lingmi, Z., Takayoshi, I., Shinji, N., & Kazuhiro, S. (n.d.). Examples
of Modal Identification of Structures in Japan by FDD and MRD Techniques.
224

Evaluation of Seismic Damage of Tall Buildings From Acceleration Response Time Histories by A Modal Decomposition Approach

Uploaded by

Copyright:

Available Formats

Evaluation of Seismic Damage of Tall Buildings From Acceleration Response Time Histories by A Modal Decomposition Approach

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

Evaluation of Seismic Damage of Tall Buildings From Acceleration Response Time Histories by A Modal Decomposition Approach

Uploaded by

Copyright:

Available Formats

Evaluation of Seismic Damage of Tall Buildings from Acceleration

Response Time Histories by a Modal Decomposition Approach

A thesis submitted in partial fulfillment of the requirements for the

Examination Committee: Prof. Pennung Warnitchai (Chairperson)

Scholarship Donor: HM King’s Scholarships

Asian Institute of Technology

CHAPTER TITLE PAGE

Figure 2.2 Sensor layout 5

Figure 2.5 Transfer functions of acceleration responses at the top of both

Figure 2.6 Influence of the shaking intensity on isolator's relative displacement

Figure 2.11 Variation of identified natural frequencies (global modes) versus

Figure 2.12 Variation of natural frequency between 01/06/2013 and 10/07/2013:

Figure 2.13 Change in frequency-temperature correlation induced by the seismic

Figure 2.16 Sensor layout of the monitoring system 16

Figure 2.23 Sensor layout on Roppongi Tower 23

Figure 2.24 Acceleration and displacement at the top of the building 24

Figure 2.26 Schematic representation of LTI system 26

Figure 2.27 Discrete Fourier Transform of the rectangular window 35

Figure 2.28 Typical sensor layout for building-like structures 37

Figure 2.29 Aliasing in digitized wave form 37

Figure 2.30 (a, b) APSD of signals in orthogonal directions, (c, d) coherence

Figure 2.31 Schematic chart of HPBW 41

Figure 2.32 Typical Singular Value plots 42

Figure 2.35 Typical plan of multi-story building 51

Figure 2.36 Modal decomposition of roof displacement due to 𝑝𝑒𝑓𝑓, 𝑛𝑡, n = 1,

Figure 2.40 Summary of parameters in MFS 60

Figure 2.41 Construction of idealized tri-linear backbone curve 60

Figure 3.3 Movement of a floor relative to its center of mass 70

Figure 4.1 Mode shapes of three dominant modes in each direction 94

Figure 4.2 Pushover curves of modes of vibration in y direction: 95

Figure 4.4 Comparison between SDF force-deformation relationship - derived

Figure 4.5 Verification of estimated roof-top displacement time-history 99

Figure 4.6 Verification of estimated roof-top (relative) acceleration time-

Figure 4.7 Verification of estimated Base-Shear time-history 101

Figure 4.9 Verification of estimated response envelopes 103

Figure 4.10 Verification of estimated roof-top displacement time-history 105

Figure 4.11 Verification of estimated roof-top (relative) acceleration time-

Figure 4.12 Verification of estimated Base-Shear time-history 107

Figure 4.14 Verification of estimated response envelopes 109

Figure 4.15 Verification of roof-top displacement time-history 110

Figure 4.16 Verification of estimated roof-top acceleration time-history 111

Figure 4.17 Verification of estimated Base-Shear time-history 112

Figure 4.19 Verification of estimated response envelopes 114

Figure 4.20 Verification of estimated roof-top displacement time-history 117

Figure 4.21 Verification of estimated roof-top (relative) acceleration time-

Figure 4.22 Verification of estimated response envelopes 119

Figure 4.23 Verification of estimated Story force time-histories 120

Figure 4.25 Verification of estimated Base-Shear time-history 121

Figure 4.27 Comparison of estimated Base-Shear against estimated roof-

Figure 4.28 Verification of estimated Story force envelopes 123

Figure 4.29 Verification of estimated Base-Shear time-history 124

Figure 4.31 Comparison of estimated Base-Shear against estimated roof-

Figure 4.32 Verification of estimated Story force envelopes 126

Figure 4.33 Mode by mode story force envelopes 127

Figure 4.34 Verification of estimated Story force envelopes 127

Figure 4.35 Verification of estimated roof-top displacement time-history 130

Figure 4.36 Verification of estimated roof-top (relative) acceleration time-

Figure 4.37 Verification of estimated response envelopes 132

Figure 4.38 Verification of estimated Story force time-histories 133