PERFORMANCE OF CONCRETE GRAVITY DAMS UNDER EARTHQUAKE

EFFECTS

A THESIS SUBMITTED TO
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES
OF
MIDDLE EAST TECHNICAL UNIVERSITY

BY

BERAT FEYZA SOYSAL

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR
THE DEGREE OF MASTER OF SCIENCE
IN
CIVIL ENGINEERING

JANUARY 2014
 Approval of the thesis:

PERFORMANCE OF CONCRETE GRAVITY DAMS UNDER

EARTHQUAKE EFFECTS

submitted by BERAT FEYZA SOYSAL in partial fulfillment of the requirements

for the degree of Master of Science in Civil Engineering Department, Middle
East Technical University by,

Prof. Dr. Canan Özgen

Dean, Graduate School of Natural and Applied Sciences

Prof. Dr. Ahmet Cevdet Yalçıner

Head of Department, Civil Engineering

Assoc. Prof. Dr. Yalın Arıcı

Supervisor, Civil Engineering Dept., METU

Examining Committee Members:

Prof. Dr. Mehmet Utku

Civil Engineering Dept., METU

Assoc. Prof. Dr. Yalın Arıcı

Civil Engineering Dept., METU

Prof. Dr. Barış Binici

Civil Engineering Dept., METU

Assoc. Prof. Dr. Afşin Sarıtaş

Civil Engineering Dept., METU

M.Sc Altuğ Akman

ES Project Engineering and Consultancy

DATE: 29.01.2014
I hereby declare that all information in this document has been obtained and
presented in accordance with academic rules and ethical conduct. I also declare
that, as required by these rules and conduct, I have fully cited and referenced
all material and results that are not original to this work.

Name, Last name : BERAT FEYZA SOYSAL

Signature :

iv
 ABSTRACT

PERFORMANCE OF CONCRETE GRAVITY DAMS UNDER EARTHQUAKE

EFFECTS

Soysal, Berat Feyza

M.Sc., Department of Civil Engineering

Supervisor: Assoc. Prof. Dr. Yalın Arıcı
January 2014, 83 pages

Concerns about the seismic safety of gravity dams have increased with the
construction of numerous new dams in the developing world as well as the need for
the evaluation of existing dam stock in the developed countries. The common
procedure for the design of concrete gravity dams has been linear analyses: there is
scant experience on the design of such systems using nonlinear analyses as the post-
linear behavior and performance limits of gravity dam monoliths are not well known.
This study is focused on the investigation of the behavior of a concrete gravity dam
monolith using the incremental dynamic analysis (IDA) procedure with the goal of
determining the development of damage on the monolith and the corresponding
cracking patterns. The constitutive model used was first calibrated to experimental
data in order to verify the ability of the model to simulate the fracture in this
material. IDA was then conducted using 21 different ground motions in order to
assess the state of damage on the monolith for increasing intensity measures (IM).
The results show that in lieu of spectral acceleration, velocity based IMs are more
effective in determining the damage process on the monolith. A simple equation for
the prediction of the damage development was proposed based on the velocity
characteristics of the ground motion. Finally, the results of the IDA were compared

v
to the static pushover analysis results in order to evaluate the scope of possible use of
such analyses for determining the performance levels of dams.

Keywords: Concrete Gravity Dam, Smeared Crack Model, Nonlinear Analysis,

Incremental Dynamic Analysis, Static Pushover Analysis.

vi
 ÖZ

BETON AĞIRLIK BARAJLARIN DEPREM YÜKLERİ ALTINDA

PERFORMANSI

Soysal, Berat Feyza

Yüksek Lisans, İnşaat Mühendisliği Bölümü

Tez Yöneticisi: Doç. Dr. Yalın Arıcı

Ocak 2014, 83 sayfa

Gelişmekte olan ülkelerde inşa edilen birçok baraj yanında, gelişmiş ülkelerin de
baraj stoklarının güvenliğinin kontrol edilmesi gerekliliği beton barajların sismik
güvenilirliği hakkında çalışmaları arttırmaktadır. Beton ağırlık barajlarının
tasarımında genelde doğrusal analizler kullanılmaktadır. Doğrusal olmayan
analizlerle bu tip yapıların analizinde tecrübe çok sınırlıdır: zira bu yapıların doğrusal
olmayan durumda davranışı ve performans limitleri hakkında gözlem ve bilgi
oldukça azdır. Bu çalışmada beton ağırlık baraj davranışı monolitte oluşan hasarın
gelişimi ve buna bağlı çatlama düzenlerinin elde edilmesi amacı ile artımlı dinamik
analiz yöntemi ile incelenmiştir. Öncelikle malzeme modeli literatürden elde edilen
deney sonuçları ile karşılaştırılarak kullanılan çatlak modelinin performansı
gerçeklenmiştir. Bir sonraki adımda 21 değişik deprem kaydı kullanılarak yapılan
artımlı dinamik analizlerle barajdaki hasarın gelişimi ve deprem kayıtlarının çeşitli
şiddet ölçümleri arasındaki ilişki incelenmiştir. Baraj monolitindeki hasarı
belirlemede spektral ivme değeri yerine hız ile ilgili şiddet değerlerinin daha başarılı
olduğu görülmüştür. Monolitteki hasarı öngörmek için deprem kayıtlarının hız
karakterlerine dayanan basit bir denklem önerilmiştir. Son olarak, statik itme

vii
analizlerinin monolit performansını belirlemek için kullanılma olanağını incelemek
üzere artımlı dinamik analiz sonuçları ile statik analiz sonuçları karşılaştırılmıştır.

Anahtar Kelimeler: Beton Ağırlık Baraj, Yayılı Çatlak Modeli, Doğrusal Olmayan
Analiz, Artımlı Dinamik Analiz, Statik İtme Analizi.

viii
To my parents…

ix
 ACKNOWLEDGEMENTS

Above all others, I would like to express my gratitude to my supervisor Assoc. Prof.
Dr. Yalın Arıcı for his guidance, advice, criticism, encouragement and insight
throughout the research. He has been a brilliant supervisor and mentor. It was an
honor to work with him.

Prof. Dr. Barış Binici is also sincerely acknowledged for his valuable support.

I would like to thank my co-workers Alper Aldemir and Okan Koçkaya for their help
whenever I needed.

I would like to express my gratitude to my friends Elif Gökçe Öz, Ezgi Berberoğlu-
Yılmaz, Melis Aysun Ekici, Arzu İpek Yılmaz, Can Özbay, Görkem Deniz Köksoy
and Deniz Yılmaz-Karaeren. Finally, I would like to express special thanks to Utku
Albostan, for being there for me with his continuous support.

I express my sincere appreciation to my dear family for their understanding and their
endless support: my mother Zehra Soysal, my father Fatih Soysal and my brother and
his wife Furkan-Simten Soysal.

This study has been conducted with the funding provided by The Scientific and
Technological Research Council of Turkey (TUBITAK) under the grant
MAG111M712.

x
 TABLE OF CONTENTS

ABSTRACT..................................................................................................................v
ÖZ...............................................................................................................................vii
ACKNOWLEDGEMENTS..........................................................................................x
TABLE OF CONTENTS.............................................................................................xi
LIST OF FIGURES...................................................................................................xiii
LIST OF TABLES......................................................................................................xv
CHAPTERS
1. INTRODUCTION.................................................................................................1
1.1 General ....................................................................................................... 1
1.2 Literature Review ....................................................................................... 2
1.3 Concrete Constitutive Modeling .................................................................. 6
1.3.1 Multi-Directional Fixed Crack Model ................................................. 8
1.3.2 Total Strain Crack Model .................................................................... 9
1.3.3 Post-Peak Response ............................................................................ 9
1.4 Finite Element Model and the Element Library ......................................... 10
1.4.1 Q8AXI Element .................................................................................10
1.4.2 CT12M Element ................................................................................11
1.4.3 L6BEN Element ................................................................................12
1.4.4 CQ16M Element ................................................................................12
1.5 Objectives and Scope ................................................................................ 13
2. CALIBRATION TESTS.....................................................................................15
2.1 Introduction .............................................................................................. 15
2.2 Analytical Simulation of Direct Tension Tests .......................................... 16
2.2.1 Investigation of the Effect of Mesh Density .......................................18
2.3 Analytical Simulation of a Notched Model Concrete Dam ........................ 20
2.3.1 Investigation of CMOD Displacement ...............................................21
2.3.2 Investigation of Cracking Pattern .......................................................25

xi
 2.4 Pseudo-Dynamic Earthquake Simulation and Analytical Pushover
Analysis .............................................................................................................. 27
2.4.1 Static Earthquake Simulation ............................................................. 29
2.4.2 Analytical Simulation of the Pushover Experiment ............................ 31
2.5 Analytical Simulation of Shake Table Testing on a Scaled Gravity Dam .. 32
2.5.1 Shake Table Test and Specimen ........................................................ 32
2.5.2 Analysis Results ................................................................................ 34
2.5.3 Investigation of the Effect of Modeling Parameters on the Crack
Location and Propagation ................................................................................ 37
2.5.3.1 The Effect of Tension Softening Functions and Fracture Energy 38
2.5.3.2 The Effect of Mesh Density ........................................................ 40
2.5.4 The Effect of the Time Stepping Factor ............................................. 43
2.5.5 Detailed Investigation of the Simulation of Crack Opening ............... 45
2.5.5.1 The Effect of the Material Modeling........................................... 45
2.5.5.2 The Effect of the Shear Retention Factor .................................... 46
2.5.5.3 The Effect of Rayleigh Damping ................................................ 48
2.5.5.4 Second Cracking Test ................................................................. 49
2.6 Summary .................................................................................................. 50
3. INCREMENTAL DYNAMIC ANALYSIS OF A CONCRETE GRAVITY
DAM...........................................................................................................................53
3.1 Introduction .............................................................................................. 53
3.2 Gravity Dam Model.................................................................................. 53
3.2.1 Incremental Dynamic Analysis .......................................................... 56
3.3 Selection of Ground Motions .................................................................... 57
3.4 Results of IDA ......................................................................................... 59
3.5 Pushover Analysis .................................................................................... 69
4. CONCLUSION AND FUTURE PLANS...........................................................75
4.1 Conclusion ............................................................................................... 75
4.2 Future Plans ............................................................................................. 77
REFERENCES...........................................................................................................79

xii
 LIST OF FIGURES

FIGURES
Figure 1-1 A Schematic Drawing for the Dam Cross-Sections, (the dotted lines show
the cracking, not in scale) ......................................................................................... 4
Figure 1-2 Concrete Fracture Models ....................................................................... 8
Figure 1-3 Concrete Tension Softening Models ......................................................10
Figure 1-4 Q8AXI Element (TNO DIANA, 2010) ..................................................11
Figure 1-5 CT12M Element (TNO DIANA, 2010) ..................................................11
Figure 1-6 L6BEN Element (TNO DIANA, 2010) ..................................................12
Figure 1-7 CQ16M Element (TNO DIANA, 2010) .................................................12
Figure 2-1 Sketch of the Direct Tension Test (Li et al., 2002) .................................16
Figure 2-2 Secant Method (TNO DIANA, 2010).....................................................17
Figure 2-3 Analytical Model of the Direct Tension Test in DIANA ........................18
Figure 2-4 Load-Displacement Response, Mesh I vs. Mesh II .................................19
Figure 2-5 Sketch of the Experiment by (Carpinteri et al., 1992) .............................21
Figure 2-6 Arc-Length Method (TNO DIANA, 2010) .............................................22
Figure 2-7 Analytical Model of the Specimen in DIANA ........................................23
Figure 2-8 Comparison of the Results Obtained using Different Softening Functions
with the Experiment Results....................................................................................23
Figure 2-9 Comparison of the Results from Mesh-I and Mesh-II with the Experiment
for Exponential Softening .......................................................................................24
Figure 2-10 Cracking Patterns for Mesh I and Mesh II with Different Softening
Functions ................................................................................................................25
Figure 2-11 Effect of the Young’s Modulus (E) and the Fracture Energy (G f) on the
Behavior .................................................................................................................26
Figure 2-12 Melen Dam and the Dam Model (Binici et al., 2012) ...........................28
Figure 2-13 Analytical Model in DIANA ................................................................29
Figure 2-14 Comparison of the Base Shear Time Histories......................................30

xiii
Figure 2-15 Comparison of the Base Shear Crest Displacement Graphs .................. 31
Figure 2-16 Cracking Schemes after Pushover Analyses ......................................... 32
Figure 2-17 Base Shear-Top Displacement Graph .................................................. 32
Figure 2-18 Dam Model (Tinawi et al., 2000) ......................................................... 33
Figure 2-19 Input Acceleration (Tinawi et al., 2000) ............................................... 34
Figure 2-20 DIANA Dam Model ............................................................................ 35
Figure 2-21 Cracking Scheme for FPAs of 0.94g and 0.98g .................................... 37
Figure 2-22 DIANA Models with Different Mesh Densities ................................... 38
Figure 2-23 Final Crack Length at the Downstream Notch for Fixed ft or Gf........... 39
Figure 2-24 Cracking Scheme for Different ft-Gf Combinations for Model 1 .......... 40
Figure 2-25 Cracking Scheme for Different Gf for Model 2 .................................... 42
Figure 2-26 Cracking Scheme for Model 3 ............................................................. 43
Figure 2-27 Comparison of the Final Crack Length Obtained with Different Time
Steps....................................................................................................................... 44
Figure 2-28 Comparison of Response Quantities with Total Strain Fixed & Rotating
Crack and Multi-Direction Fixed Crack Models ..................................................... 46
Figure 2-29 The Variation of the Final Crack Length and the Crack Opening with
Shear Retention Factor ........................................................................................... 48
Figure 2-30 Response Quantities Obtained Using Different Damping Ratios .......... 49
Figure 3-1 Concrete Gravity Dam Model ................................................................ 54
Figure 3-2 Mode Shapes ......................................................................................... 55
Figure 3-3 Acceleration Response Spectra (Scaled at T 1) for the Selected Ground
Motions .................................................................................................................. 59
Figure 3-4 Cracking Schemes from the IDA ........................................................... 60
Figure 3-5 Crack Zones and Cracking Schemes ...................................................... 65
Figure 3-6 First Cracked Period of the Dam............................................................ 66
Figure 3-7 Total Crack Length of the Dam vs. Different Intensity Measures ........... 67
Figure 3-8 Total Crack Length of the Dam with the Proposed Intensity Measure .... 68
Figure 3-9 Total Lateral Load Distribution for the Pushover Analysis..................... 71
Figure 3-10 A Schematic Drawing of the Pushover Loading Directions .................. 71
Figure 3-11 Cracking Scheme of the Pushover Analyses ........................................ 72
Figure 3-12 IDA vs. Pushover Analyses Results ..................................................... 73

xiv
 LIST OF TABLES

TABLES
Table 2-1 Crack Length Variation for Exponential Softening with Different E and G f
...............................................................................................................................26
Table 2-2 First Five Natural Frequencies (Hz).........................................................36
Table 2-3 Maximum Stresses of the Models ............................................................41
Table 3-1 Selected Ground Motions ........................................................................58

xv
xvi
1.
CHAPTER 1

INTRODUCTION

1.1 General

Throughout the history, dams have been constructed for various reasons. In the
ancient times, they were built for water supply and irrigation purposes. Today, many
dams are being built for water supply, flood control, irrigation, navigation, sediment
control and hydropower (ICOLD, 2013). Dams can be divided into two categories
according to the type of the material which they are constructed: namely concrete
and embankment dams. Concrete dams can further be categorized by their structural
systems, i.e. concrete gravity dams, concrete arch dams and concrete buttress dams.
Earth and rock fill dams and concrete faced rock fill dams make up the embankment
dams (Yanmaz, 2006). In Turkey, before 2000’s most of the dams were built as earth
or rock fill dams; however, this trend has reversed in the last decade. Very few
embankment dams have been built in the last 10 years; concrete dams are generally
preferred over embankment dams (Yılmaztürk, 2012).

A concrete gravity dam, by definition, is a wide concrete berm that resists the
hydrostatic forces by its geometry, weight and material strength. They are commonly
preferred because of the fast construction and cost savings associated with smaller
geometries. They can be built on a site where there is a strong foundation to bear the
weight of the dam. In plan, most of the time, these dams are straight. If conditions
are suitable, concrete gravity dams can be very tall, sometimes as high as 285 m
(Grande Dixence Dam) (Ali et. al., 2012).

1
Çubuk 1 Dam is the first concrete gravity dam built in Turkey (1936). Some other
examples of the concrete gravity and the concrete arch/gravity dams are Porsuk
(1948), Sarıyar (1956), Karakaya (1987) and Berke Dams (1999). (DSİ, 2013)

As explained before, one of the reasons for the building dams is to obtain
hydropower. In some countries such as Norway, Democratic Republic of the Congo
and Brazil, over 90% of the electric power are supplied by the hydroelectric
resources by year 1998 (ICOLD, 2013). However, at the end of 2009, less than 20%
of the energy production in Turkey is provided by hydroelectric power. Therefore, in
order to increase the hydroelectric power production in Turkey, investments have
been made since 2006, which significantly accelerated the design and construction
process for many hydropower projects (Yılmaztürk, 2010).

1.2 Literature Review

The concerns about the seismic safety of concrete dams have been increasing in the
last decade with the design and construction of new dams as well as the need for the
evaluation the existing dam stock which were built inadequate from the seismic point
of view. Shih Kang Dam failed in Taiwan due to earthquake and two dams were
severely damaged due to ground excitations: Hsinfengkiang Dam in China and
Koyna Dam in India (Nuss et al., 2012). Therefore, all existing dams in seismic
regions and the new dams to be built should be seismically analyzed to satisfy the
required performance level under severe earthquake excitations (NRC, 1990).

In order to investigate the performance of concrete gravity dams under ground

excitations, linear elastic analyses are generally preferred in the literature, based as
much on the historical development of the analyses methods as well as the frequency
dependent nature of the dam-rock-reservoir interaction behavior. The pioneering
works in this regard were conducted by Chopra and his coworkers by analyzing
concrete gravity dams using two dimensional linear elastic methods (Chopra and
Gupta, 1981; Hall and Chopra, 1982; Fenves and Chopra, 1985). The behavior of the

2
dam under earthquake excitation was determined to be significantly affected by the
dam-foundation-reservoir interaction, compressibility of the water and the
hydrodynamic pressure wave absorption. These analyses were conducted in the
frequency domain due to the importance of the simulation of the ingoing-outgoing
waves in the reservoir as well as the rock foundation.

When a dam is subjected to a severe earthquake, the stresses occurring in the dam
can exceed the strength, which will cause the dam to go beyond its elastic limit,
leading to significant structural damage (Asteris and Tzamtzis, 2003). This
phenomenon can be investigated by nonlinear analysis procedures. The nonlinear
analyses account for the tensile cracking in concrete and the resultant change in the
stiffness of the system (Araujo and Awruch, 1998).

In order to examine the cracking of concrete gravity dams, many studies were made
using nonlinear analyses. The two most commonly studied dams are the Koyna Dam
and the Pine Flat dam (Figure 1-1). Koyna Dam was severely damaged during the
Earthquake of December 11, 1967 (Çalayır and Karaton, 2005): the cracking scheme
obtained from the nonlinear analyses was compared to the cracking scheme that had
occurred so that the capability of the crack models to simulate the real cracking was
investigated. The first study to analyze the Koyna Dam by nonlinear analysis is Pal
(1976). In this study, the effect of the reservoir was neglected and the foundation was
assumed to be rigid. The analyses were made using the smeared crack model, which
is explained in the proceeding section. Further studying of the cracking scheme in
this work shows that the results were biased due to the mesh selection, a common
pitfall of using cracking models in FE analyses (Pal, 1976). Bhattacharjee and Leger
(1994) studied the cracking scheme of the Koyna Dam by nonlinear analyses with
smeared cracking model and rigid foundation assumption taking the reservoir effect
into account. The full reservoir condition and the overflow pressure were considered.
The cracking obtained in their study was in good correlation with the real cracking of
the dam. Three dimensional models were also used for investigation of the behavior
of the Koyna Dam. Using a smeared cracking model and a rigid foundation, along
with modeling the dam-reservoir interaction, Mirzabozorg and Ghaemian (2005)
obtained a cracking scheme which was in a good agreement with the real cracking of

3
the dam. The cracking profile for the Pine flat dam was studied by El-Aidi and Hall
(1989), and Fenves and Vargas-Loli (1988) (Mansouri et al., 2011). It was observed
that in a similar pattern to the Koyna Dam, the cracking occurred at the base of the
dam and at the part where there is a change in slope profile.

As given above, a vast number of studies have been made to assess the cracking
scheme of dams having a slope-change in their profile (Figure 1-1). The change in
the slope profile leads to a stress concentration at that location prescribing the start of
a crack, mostly near the crest of the dam. However, this phenomenon also leads to a
bias in the analyses: this crack and its propagation always takes over the analyses and
defines the final failure state. In this study, a cross section bereft of any slope
changes were examined in order to understand the performance of such a dam
without being biased by such a geometric detail assuring a failure at that location
(Figure 1-1).

(a) Koyna Dam (b) Pine Flat Dam (c) A Regular

Hall (1988) El-Aidi and Hall (1989) Cross-Section Dam

Figure 1-1 A Schematic Drawing for the Dam Cross-Sections, (the dotted lines show
the cracking, not in scale)

4
The research on the cracking schemes of the concrete gravity dams has mostly been
focused on 1) the prediction of the performance of Koyna or Pine Flat Dams and 2)
the validation of cracking models for monolithic concrete, of which these structures
as well as some test specimens are typical examples. However, the extreme limit
states of gravity dams have not been studied, in contrast to the significant research
effort for buildings in the last 3 decades for the determination of the limit states to
enable performance based design (Freeman et al., 1975; Fajfar 2000; Hasan et al.,
2002; Chopra, 2004). Jiang and Du (2012) investigated the cracked Koyna Dam to
assess the stability of the block detached from the top of the dam finding out a large
safety margin for the rigid body motion of the block. Depending on the orientation of
the crack, five to eight times of the experienced Koyna earthquake could render the
sliding of the top block. Pekau and Zhu (2006) performed a study on the sliding of
blocks within a cracked dam. Sliding at the base and a crack at some height of the
dam was found to reach 4 cm: however, it is very hard to decide on a performance
limit based on this value even in the analysis methodology and earthquake
uncertainty was assumed to be treated “correctly”. Incremental dynamic analysis can
also be used in order to determine the limit states of the dams similar to buildings.
Alembaghari and Ghaemian (2012) performed IDA on the Pine flat dam to assess the
limit states and the capacity of the dam, proposing an adhoc lateral force distribution
for the static pushover analysis that considers not only the inertial loads but also the
hydrodynamic loads. The relation between the peak crest displacement (seismic
demand parameter) and the spectral acceleration, peak ground acceleration and peak
ground velocity (intensity measures) were obtained. The crest displacement and the
dissipated energy were chosen as damage measures, however, a clear relation
between these parameters and intensity levels were not provided. A prediction about
the level of cracking does not appear to be one of the goals of this study.

For the nonlinear modeling of gravity dams, some aspects of the exact simulation of
the dam-rock-reservoir interaction have to be forgone. Given the inability to conduct
linear analyses, hence to obtain foundation impedance properties defining the wave
transfer within the rock, the foundation has to be modeled as massless or with a finite
element mesh having transmitting boundaries (Ghanaat and Chudgar, 2007) or

5
perfectly matched layers (Basu and Chopra, 2003). The frequency dependent dam-
reservoir interaction is usually taken into account via the added mass approach
introduced by Westergaard (1933). The alternative is to use a large mesh with
absorbing boundaries that would enable the waves within the reservoir to travel away
from the dam. Given the computational demand for these models, as well as the
numerical difficulties with absorbing boundaries, an added mass approach for
hydrodynamic effects and a massless foundation model is usually used in the time
domain for simulation of the nonlinear behavior of dams. This approach is
extensively used in the design and evaluation of dams in the literature (Leger and
Boughoufalah, 1989; USACE, 2003; Chuhan et al., 2009)

Accordingly, in this study the foundation is modeled by the simplified approach

introduced in USACE (1995). The foundation is modeled as a massless mesh, so that
the only contribution is to the stiffness matrix and the applied ground motions will
reach the dam body without any change. For hydrodynamic forces, Westergaard’s
added mass approach is used (Westergaard, 1933) neglecting the compressibility of
the reservoir water. The hydrodynamic forces in the upstream direction are
calculated by Equation 1-1.

p = ρ üg √ (1-1)

In this equation, p is the hydrodynamic pressure at elevation z, where z is measured

from the bottom of the dam, ρ is the water density, ü g is the ground acceleration and
H is the reservoir elevation.

1.3 Concrete Constitutive Modeling

Concrete fracture models can be divided into two categories as discrete crack and
smeared crack models (Figure 1-2). In the discrete crack model, the crack is modeled
as a geometrical discontinuity, while in the smeared crack model; the cracked solid is
analyzed as a continuum (Rots, 1988).

6
The discrete crack model was first introduced by Ngo and Scordelis (1967) in order
to model the dominant cracks within a concrete continuum. In this model, the cracks
are simulated by the separation of the nodes, which led to some disadvantages: The
change in the nodal connectivity of the finite element model after each crack does
not fit the nature of the finite element displacement method (Rots, 1988).
Furthermore, since the cracking has to be constrained along the element edges, it
introduces a mesh bias (de Borst et al., 2004). This crack model is suitable when the
crack location is known beforehand and the cracking forms in the form of mode I
fracture (Rots, 1988).

The smeared crack model for modeling concrete behavior was introduced by Rashid
(1968). In this model, the minor cracks occurring (which may later form dominant
cracks) because of the heterogeneous nature of the concrete were taken into account.
The cracked solid maintains its continuum (i.e. the FE mesh is not altered) and the
crack model is described via the stress-strain relationship. It is assumed that the
behavior of the uncracked concrete follows the isotropic stress-strain law and upon
cracking, it obeys the orthotropic law, where the orthotropy axes are determined
based on the crack initiation. This model has several advantages over the discrete
crack model: Since the cracked solid is still a continuum, the topology of the finite
element mesh is unaltered. Moreover, orientations of the cracks are not
predetermined since the axes of orthotropy can change. Because of these reasons, in
this study, the smeared crack model is used in the finite element simulations.
Smeared crack model is divided into two as the multi-directional fixed crack model
and the total strain crack model. Total strain crack model further decomposes into the
fixed and rotating crack models, which are explained in the following section.

7
 Concrete Crack Models

Discrete Crack Model Smeared Crack Model

Multi-Directional Total Strain

Fixed Crack Model Crack Model

Fixed Crack Rotating Crack

Model Model

Figure 1-2 Concrete Fracture Models

1.3.1 Multi-Directional Fixed Crack Model

In the multi-directional fixed crack model, the total strain of concrete is decomposed
into the uncracked concrete strain and the crack strain. Therefore, the crack strain can
further be decomposed for each individual crack and each crack is assigned a
coordinate system. This further sub-decomposition enables a number of cracks to
occur simultaneously. In this model, cracking occurs when the principle tensile stress
exceeds the tensile strength and the angle between an existing crack and the current
principle stress direction exceeds a threshold angle. Therefore, the formation of non-
orthogonal cracks is enabled. In a two dimensional setting, the orthotropic stress-
strain relationship is presented below in Equation 1-2 (Rots, 1988).

[ ] [ ] (1-2)

[ ]

8
In the above equation, E is the Young’s Modulus, υ is the Poisson’s Ratio, µ is the
reduction factor for mode I and β is the shear retention factor.

1.3.2 Total Strain Crack Model

Fixed crack and rotating crack models constitute the total strain crack model. Unlike
the multi-directional fixed crack model, the strain is decomposed only for the
cracked and uncracked concrete in the total strain crack model; the crack strain is not
further decomposed. In this model, in an integration point only two orthogonal
cracks can form. The crack is initiated based on the principle tensile stress exceeding
the tensile strength. In the total strain fixed crack model, the orientation of the crack
remains unchanged, while in the total strain rotating crack model, the orientation of
crack is updated continuously and the coaxiality between the principal stress and
strain is preserved. The tangential stress-strain relationship for the total strain
rotating crack model in the two dimensional setting is presented in Equation 1-3
(Rots, 1988).

[ ] [ ] (1-3)

[ ]

1.3.3 Post-Peak Response

The post-peak softening response of concrete can be modeled using different tension
softening functions, namely brittle, linear, exponential and Hordyk (Rots, 1988).
These functions show different behavior after reaching the tensile strength of
concrete (Figure 1-3). In the brittle tension softening function, the load carrying
capacity drops immediately after the tensile strength is reached, while in linear,

9
exponential and Hordyk models the load carrying capacity drops according to the
fracture energy (Gf) and the finite element size (h), as shown in Figure 1-3.

σ Brittle Tension Softening

σ
Linear Tension Softening
ft Exponential Tension Softening ft
Hordyk Tension Softening

Gf/h
ε
ε εmax

Figure 1-3 Concrete Tension Softening Models

1.4 Finite Element Model and the Element Library

The general purpose finite element program DIANA (Displacement Analyzer) was
used as the simulation tool in this study. The software allows nonlinear transient
analyses with 2D, 3D, plane-strain, plane-stress as well as axisymmetric elements. A
range of elements were used within the analyses performed in this study, such as the
four-node isoparametric element Q8AXI, the six node isoparametric plane stress
element CT12M, the two-node isoparametric beam element L6BEN and the eight
node isoparametric element CQ16M (TNO DIANA, 2010). Some details on the
elements used in the study are provided below.

1.4.1 Q8AXI Element

Q8AXI is a four-node isoparametric axisymmetric solid ring element having a

quadrilateral cross-section (Figure 1-4). The unknowns are the translations in the
nodes in the global x and y directions. This element has linear shape functions (TNO
DIANA, 2010).

10
 Figure 1-4 Q8AXI Element (TNO DIANA, 2010)

The displacements, ui, at the nodes in x and y directions are approximated by

Equation 1-4 (TNO DIANA, 2010).

ui(ξ ,η) = a0 + a1ξ + a2η + a3ξη (1-4)

1.4.2 CT12M Element

CT12M is a six-node triangular isoparametric plane stress element with quadratic

shape functions (Figure 1-5). The unknown translational displacements, u i, at the
nodes are approximated by Equation 1-5 given below (TNO DIANA, 2010).

Figure 1-5 CT12M Element (TNO DIANA, 2010)

ui(ξ ,η) = a0 + a1ξ + a2η + a3ξη + a4ξ2 + a5η2 (1-5)

11
1.4.3 L6BEN Element

L6BEN is a two-node, two-dimensional beam element (Figure 1-6). The unknown

displacements in the x-direction, ux and in the y-direction, uy are calculated by
Equation 1-6 demonstrated below (TNO DIANA, 2010).

Figure 1-6 L6BEN Element (TNO DIANA, 2010)

ux(ξ) = a0 + a1ξ

uy(ξ) = b0 + b1 ξ + b2 ξ2 + b3 ξ3 (1-6)

1.4.4 CQ16M Element

CQ16M is an eight-node quadrilateral isoparametric plane stress element with

quadratic shape functions (Figure 1-7). The unknown translational displacements, ui,
at the nodes are approximated by Equation 1-7 presented below (TNO DIANA,
2010).

Figure 1-7 CQ16M Element (TNO DIANA, 2010)

12
 ui(ξ ,η) = a0 + a1ξ + a2η + a3ξη + a4ξ2 + a5η2 + a6ξ2η + a7ξη2 (1-7)

1.5 Objectives and Scope

Many dams are being built in seismically very active locations in valley ridges very
close to major fault lines. These sites, which were avoided in the past for more
preferable locations are actively being realized as the possible locations for
hydropower location are increasingly scarce. Moreover, most of the dam stocks in
the developed world were built for lower seismic risk than currently prescribed for
new structures in these countries. Many of these systems are older than 40-50 years,
in which period our knowledge about the fault systems as well as procedures for
determining/predicting earthquake levels increased significantly. Consequently,
although no gravity dam system has ever been fully tested within design ground
motion levels, such as Loma Prieta, Chi Chi and Koyna Earthquakes, significant risk
exists (Nuss et al., 2012). Such a ground motion can lead to a significant damage on
a dam, far from its elastic peak response, requiring nonlinear analyses to assess the
performance of the dam. In contrast to the experience in linear analyses, the
nonlinear behavior and performance limits of dam monoliths are not well known.

The primary goal of this study is to investigate the nonlinear behavior of a typical 80
m high concrete gravity dam monolith in this regard.

The quantification of damage on a dam monolith is a cumbersome task. In order to

properly address this issue, the following steps were taken:

 The damage wrecked on a system by earthquakes is very much time history

dependent. Therefore, using a range of ground motions, the possible
development of cracks and the typical cracking schemes in such a monolith
was investigated.
 Extensive cracking on the dam body significantly changes the period of the
structure, therefore an investigation of the cracking of the dam body on the
fundamental period of the system was conducted.

13
  Given that the cracking on the system is dependent on the ground motion
characteristics, a relationship between the cracking and the earthquake
ground motion characteristics were sought. As the cracking on a dam body
can be expressed as a dam measure, this relationship can also be expressed as
a preliminary design tool to assess expected damage on a system from an
earthquake.
 The IDA analyses conducted are computationally very expensive and time-
consuming. Therefore, similar to reinforced concrete buildings, pushover
analysis was used to investigate the performance limits of the dam body. A
comparison of pushover analyses with the IDA results was performed to
investigate the effectiveness or suitability of this approach to the response
prediction of a dam monolith.

For the sake of brevity, the study as mentioned above is presented in only three
chapters in this thesis. First, the preliminary information and the validation study on
the material models used are presented in Chapter 2. The calibration of the
constitutive model for the material to four different experimental studies focused on
gravity dams is presented. The results of the IDA analyses are presented in Chapter 3
along with pushover analysis results. The relationship between seismic demand and
intensity measures are computed and presented in this chapter. In Chapter 4, the
conclusions from this study and future avenues of research on this topic are
summarized.

14
2.
CHAPTER 2

CALIBRATION TESTS

2.1 Introduction

Calibration is the process of adjusting the properties of an analytical model in order

to match a set of experimental results obtained in the laboratory. It is a very
important process which shows that a mathematical model is capable of reproducing
physical phenomenon, and can be used henceforth as a prediction tool. In this study,
the general purpose finite element program DIANA was used in the analyses. In
order to investigate the capabilities of the program and use it as a prediction tool, a
set of calibration studies were conducted for laboratory experiments conducted with
different types of loading. Four works from the literature were selected to investigate
capabilities of DIANA for the simulation of concrete gravity dams under seismic
loading. The first of the chosen benchmark experiments was the roller compacted
concrete (RCC) direct tension test conducted by Li et al. (2002) performed under
static conditions. The second benchmark study was the experiments conducted by
Carpinteri et al. (1992) on a model concrete gravity dam under lateral loading in
static conditions. The third benchmark study is the pseudo dynamic testing of
concrete dams by Binici et al. (2012). The last benchmark study used is a detailed
shake table testing study on a scaled model of a concrete gravity dam conducted by
Tinawi et al. (2000).

15
2.2 Analytical Simulation of Direct Tension Tests

In their experimental study, Li et al. (2002) tested cylindrical RCC cores obtained
from an RCC dam under uniaxial direct tension. The RCC cores were obtained from
the RCC matrix and the interface. They tested two categories of specimens with
different compressive strengths to examine the effect of compressive strength.
Furthermore, to investigate the effect of specimen size, two groups with different
sizes of specimens were tested. A total number of 28 specimens were tested having
different characteristics such as the tensile strength, peak strain and fracture energy.
The main goal of the experiment was to obtain the complete stress-deformation
curves of the specimens (Figure 2-1). Moreover, some relationships between the
specimen characteristics and the mechanical properties were proposed by the authors.
The experiment was displacement controlled: the sketch of the test setup is presented
in Figure 2-1.

The analytical study was focused on the acquisition of complete stress-deformation

curve. Therefore, one specimen from the RCC matrix was selected for the analysis.
Another aspect of this study was to obtain the fracture energy correctly.

(a) Experimental Setup (b) Stress-Deformation Curve

Figure 2-1 Sketch of the Direct Tension Test (Li et al., 2002)

16
For the analytical modeling in DIANA, the specimen labeled as T-17 by Li et al.
(2002) was selected. The size of the specimen was 150x300 mm. The compressive
and tensile strengths were 15 MPa and 1.10 MPa, respectively. The modulus of
elasticity was given as 23.6 GPa and the fracture energy 263 N/m.

Four-node isoparametric axisymmetric elements (Q8AXI) were used in this study

(TNO DIANA, 2010). The bottom nodes were restrained in the x and y directions
while the top nodes were restrained in the y direction for the application of loading.
In order to prescribe the location of cracking, tensile strengths of elements 5 and 13
were reduced by 1%. An initial displacement was applied to the top nodes of the
model and increased incrementally throughout the analysis. Total strain rotating
crack model with the predefined Hordyk tension softening (TNO DIANA, 2010) was
used in the analyses. The analyses were conducted satisfying both the force and
displacement converge criteria (at 0.01) using 1000 load steps. Secant method was
used in the analyses, in which the secant stiffness matrix is derived for each loading
stage (Figure 2-2 TNO DIANA, 2010).

Figure 2-2 Secant Method (TNO DIANA, 2010)

The 16 element model used for the simulation and the results of the analyses are
presented in Figure 2-3. The area under the axial stress deformation curve yields the
fracture energy (Li et al. 2002). In order to obtain the prescribed fracture energy, 263

17
N/m, the “effective” element size, h, in the DIANA model had to be prescribed since
this variable is assumed as 1 by default for axisymmetric elements in DIANA. The
area under the ‘DIANA Model’ curve in Figure 2-3 was calculated as 254 N/m,
which is very close to the stated fracture energy 263 N/m. The effective element size
to be used in the material model was determined to be 0.4a for square elements with
side length a. The comparison of the axial stress-deformation response from Li et al.
(2002) and the DIANA model, as shown in Figure 2-3, show that the post-peak
response of the model, as modeled with Hordyk softening, was satisfactory for this
test, albeit unable to catch the initial sudden drop in the resistance of the specimen.

13 5

(a) Finite Element Mesh (b) Stress-Displacement Curve

Figure 2-3 Analytical Model of the Direct Tension Test in DIANA

2.2.1 Investigation of the Effect of Mesh Density

Mesh density is regarded as a critical part of the modeling of the behavior of concrete
structures given that the early analyses in this field were observed to be largely
biased by mesh dependence. The concentration of the cracking in concrete to singled
out locations renders the modeling of displacements by continuous displacement
functions, such as in the finite elements, a significant challenge. Mesh dependence is
usually addressed using the crack energy concept (Bazant, 1986): a formulation used

18
in DIANA as well. However, even with this formulation, the modeling and accurate
prediction of displacements for bulk concrete structures is still an issue with the post-
peak response determining the displacement behavior and crack propagation.

In order to study the effect of mesh density on the analyses, the analyses given above
(16 elements, Mesh I) were re-conducted with a different finite element mesh
consisting of 64 elements (Mesh II, Figure 2-4). For the second mesh (Mesh-II), a
higher load step of 4000 was used and the element size was reduced to 0.007. Similar
to Mesh I, the effective element size was found as 0.37a. The fracture energy was
calculated as 270 N/m. The comparison between Mesh-I, Mesh-II and the test by Li
et al. (2002) is presented in Figure 2-4.

(a) Analytical Model with 64 (b) Comparison, Mesh I-II

Elements

Figure 2-4 Load-Displacement Response, Mesh I vs. Mesh II

It can be concluded from Figure 2-4 that increasing the mesh density did not affect
the response significantly. The results obtained were not biased by the mesh chosen
for the analyses. For this case, dividing the model into more elements necessitated a

19
significant increase in the number of load steps leading to much higher
computational cost.

2.3 Analytical Simulation of a Notched Model Concrete Dam

Carpinteri et al. (1992) tested two small scale (1:40) concrete gravity dams under
hydrostatic loading. The height of the specimens was 2.4 m with 0.03 upstream and
0.7 downstream slopes. A horizontal notch was prepared in the upstream face at 60
cm from the base of the dam in order to determine the initiation point for cracking:
the depths of the notches were 15 and 30 cm for the first and second models,
respectively. Three tests were conducted on these two specimens. The experiments
were displacement controlled using the crack mouth opening displacement (CMOD)
at the notch as the controlling parameter. The first two tests were conducted on the
same specimen due to an unexpected cracking at the bottom of the specimen
observed in the first test during the application of additional loads before hydrostatic
loading. The experiment was repeated (the second test) after the repair of the model
and the test was conducted under the aforementioned hydrostatic loading. Finally, the
third test was conducted on the second specimen only under hydrostatic loading; the
self-weight simulation was not conducted. The relationship between the total applied
force and CMOD was obtained for these experiments along with the cracking profile
on the specimens.

In their study, Carpinteri et al. (1992) also performed finite element analyses of the
experiments using the discrete cracking model. In the simulations, they assumed the
modulus of elasticity as 35.7 GPa, Poisson’s Ratio as 0.1, tensile strength as 3.6 MPa
and the fracture energy as 184 N/m. The comparison of the analytical and
experimental results from Carpinteri et al. (1992) is given in Figure 2-5 for the third
test.

20
 (a) Force vs. CMOD Graph (b) Cracking Profile (Carpinteri et
(Carpinteri et al., 1992) al., 1992)

Figure 2-5 Sketch of the Experiment by (Carpinteri et al., 1992)

The experiment of Carpinteri et al. (1992) was used in this section in order to
validate the performance of the utilized material model.

2.3.1 Investigation of CMOD Displacement

The modeling of the experiment (third test only) was conducted using a plane stress
model. Eight-node quadrilateral elements, CQ16M and six-node triangular elements,
CT12M (TNO DIANA, 2010) were used in the analyses. The notch in the specimen
was predefined in the model by removing the necessary elements at its location.
Similar to the experiment conducted by Carpinteri et al. (1992), the analyses were
displacement controlled and arc-length method formulated in (TNO DIANA, 2010)
was used (Figure 2-6) with CMOD control at the crack mouth (displacements of the
nodes at the opposite sides of the notch). The arc-length method constrains the norm
of the incremental displacements in a load-stepping analysis and hence the post peak
behavior and unloading can be simulated correctly (TNO DIANA, 2010). Two
different mesh structures, namely, medium (Mesh-I) and dense (Mesh-II) were

21
utilized to investigate the dependence of the results on the finite element mesh. Four
different tensile softening functions, namely brittle, exponential, Hordyk and linear
(TNO DIANA, 2010) functions were used in order to investigate the effect of the
shape of the post-peak concrete response on the analyses results.

Figure 2-6 Arc-Length Method (TNO DIANA, 2010)

Mesh I consisted of 669 CQ16M elements and 26 CT12M as given in Figure 2-7.
Total strain rotating crack model was utilized in the analyses. In order to overcome
the initial low stiffness and capacity of the model, the modulus of elasticity was
increased by 1.58 and the fracture energy by 1.36.

22
 0.248 m 0.248 m

2.4 m 2.4 m

0.15 m 0.15 m
2.0 m 2.0 m

(a) Mesh-I (b) Mesh-II

Figure 2-7 Analytical Model of the Specimen in DIANA

For mesh I, the behavior of the models with different post-peak softening functions is
presented in Figure 2-8 along with the experimental data. Linear tension softening
function appears to over predict the peak response while the Hordyk model under
predicts the peak. Brittle response results in a very significant under prediction of the
capacity of the monolith since no fracture energy is specified. It can be concluded
from Figure 2-8 that the behavior is best represented by the linear and exponential
tension softening functions.

(a) Mesh-I (b) Mesh-II

Figure 2-8 Comparison of the Results Obtained using Different Softening Functions
with the Experiment Results

23
A much denser mesh was used for the second set of analyses simulating the
experimental test of Carpinteri et al. (1992). This mesh (Figure 2-7) was composed
of 1464 CQ16M and 39 CT12M elements. The comparison of tension softening
functions together with the experimental data is given in Figure 2-8. The results
obtained from Mesh II appear very similar to their counterparts from Mesh I (Figure
2-8). Exponential and linear tension softening functions yielded the closest result to
the experiment.

The most significant effect of the increase in the mesh density appears to be the
better representation of the initial stiffness for the second model. A direct comparison
of the analyses conducted with exponential tension softening models for the different
meshes are presented in Figure 2-9, showing that initial stiffness is better estimated
for the denser mesh and the capacity of the specimen is incrementally better
estimated.

Figure 2-9 Comparison of the Results from Mesh-I and Mesh-II with the Experiment
for Exponential Softening

24
2.3.2 Investigation of Cracking Pattern

The effect of the material model on the development and propagation pattern of the
cracking within the analytical model was also investigated. For the above mentioned
four softening functions and two different mesh structures, Figure 2-10 shows the
cracking pattern of the model dam, where the red line shows the cracking observed in
the experiment of Carpinteri et al. (1992). The crack length was estimated very
similarly for the exponential, Hordyk and linear softening functions, while the brittle
softening function yielded significantly larger cracking in the specimen. However,
this softening function failed to simulate the peak load for the CMOD behavior of the
model as mentioned above.

(a) Mesh I, Brittle (b) Mesh I, (c) Mesh I, Hordyk (d) Mesh I, Linear
Exponential

(e) Mesh II, Brittle (f) Mesh II, (g) Mesh II, (h) Mesh II, Linear
Exponential Hordyk
Figure 2-10 Cracking Patterns for Mesh I and Mesh II with Different Softening
Functions

25
As given above, the crack length predicted by the models (~1.0 m) were less than the
actual cracking observed in the experiment by Carpinteri et al. (1992) (1.6 m),
although the CMOD vs. load was estimated well. The results were similar to those
obtained in an analytical model by Bhattacharjee and Leger (1994) using two
different cracking models reporting crack lengths of 1.1 and 1.2 m. As explained
before, the analyses given above were conducted increasing the Young’s Modulus
and the fracture energy. The Young’s Modulus and the fracture energy both effect
the propagation of the crack significantly. A set of analyses were then conducted in
order to investigate the effects of these two parameters on the crack length and the
dam behavior. The variation of the crack length and the load-CMOD graph with
these parameters are presented in Table 2-1 and Figure 2-11, respectively.

Table 2-1 Crack Length Variation for Exponential Softening with Different E and G f
Young's Fracture Crack Length
Modulus Energy (m)
1.58x 1.36x 0.9
Mesh I 1.00x 1.00x 1.12
1.58x 1.00x 0.95
1.58x 1.36x 0.84
Mesh II 1.00x 1.00x 1.03
1.58x 1.00x 0.91

(a) Mesh I, Different E and Gf (b) Mesh II, Different E and Gf

Figure 2-11 Effect of the Young’s Modulus (E) and the Fracture Energy (Gf) on the
Behavior

26
The crack length was best estimated using the original modulus and fracture energy,
which failed to reproduce the load-CMOD behavior of the experiment. An amplified
Young’s Modulus and fracture energy was the best combination of parameters to
reproduce the load-CMOD response and to obtain the crack length similar to the
work done by Bhattacharjee and Leger (1994). Given the similarity of the analytical
results from all different studies, it is admissible that a factor that was not considered
in these analyses resulted in the significant extension of cracking during the
experiment.

2.4 Pseudo-Dynamic Earthquake Simulation and Analytical Pushover

Analysis

A set of experiments were conducted at Middle East Technical University to

investigate the seismic behavior of RCC dams under the supervision of Prof. Dr.
Barış Binici supported by TUBITAK under the grant 111M712. The experiments
were performed on a 1/75 scaled model of the 120 m high Melen Dam which was
designed for water supply to the city of İstanbul. The upstream and downstream
slopes were 0.15 and 0.75, respectively (Figure 2-12). Three different specimens
were built in order to investigate the effect of construction material on the
performance. The special setup of the test enabled the use of only the bottom half of
the dam section while the inertial and load effects of the remaining part were
included in the test using special loading apparatus (Figure 2-12). The first specimen
was built out of regular concrete while the other two specimens were built imitating
RCC construction practices with different target strengths. The thicknesses of the
specimens were 20 cm, representing a plane stress setting. These models were tested
under three levels of scaled earthquakes by pseudo-dynamic analysis, followed by
pushover analyses leading to failure (Binici et al., 2012).

27
 (a) Melen Dam (b) Sketch of the Experimental Setup

Figure 2-12 Melen Dam and the Dam Model (Binici et al., 2012)

An important point in the experiments was to obtain the base shear and the
overturning moment for the scaled specimen close to the unscaled dam. Additional
axial load had to be prescribed to the system in order to simulate its own weight and
the hydrostatic effects. Similarly, additional mass had to be placed to simulate the
inertial effect.

The first experiment conducted in the laboratory was chosen for calibration purposes
in this study. The analytical dam model is presented in Figure 2-13. The dam body
consisted of 500 CQ16M elements and the steel plate on top of the dam model was
modeled using L6BEN beam elements. The Young’s Modulus of steel plate was
assumed as 200 GPa and the Poisson’s Ratio as 0.3. For the dam body, the Poisson’s
Ratio was assumed as 0.2 and the density as 2400 kg/m3. In order to obtain the first
natural frequency of dam model at 14.3 Hz, the Young’s Modulus of concrete was
assumed as 12.35 GPa. In accordance with the experimental results, the tensile and
compressive strengths of concrete were used as 1.95 MPa and 25 MPa, respectively
(Aldemir et al., 2013). Softening functions used in the analyses were chosen as the
parabolic compression softening and the exponential tensile softening. The
compressive and tensile fracture energy values were assumed as 20000 N/m and 100
N/m, respectively. Analyses were conducted using the total strain rotating crack
model. Simulation of the pseudo-dynamic model was conducted by applying the
crest displacements measured on the specimen for three levels of earthquakes, OBE,

28
MDE and MCE levels, statically. Static pushover analysis was then conducted in
order to determine the failure state.

Upstream Downstream

0.95 m

1.36 m

Figure 2-13 Analytical Model in DIANA

2.4.1 Static Earthquake Simulation

The time histories of the base shear are compared with the experimental results for
the three earthquakes in Figure 2-14. The application of the crest displacements from
the experiment as static loading to the crest enables eliminating the uncertainty in
such tests due to the hard to prescribe damping ratio.

29
 (a) OBE Level (b) MDE Level

(c) MCE Level

Figure 2-14 Comparison of the Base Shear Time Histories

The results of the simulation were quite close for the OBE and MDE levels. For the
MCE level event, the DIANA model estimated the base shear with a 10% error.

The base shear top displacement graphs were also compared with the experimental
data (Figure 2-15). For OBE level, due to some problems in the experiment, different
initial stiffness was estimated. The simulation results for MDE level were quite close
to the experiment. For MCE level, however, since the cracking pattern could not be
simulated correctly, the behavior was estimated different.

30
 (a) OBE Level (b) MDE Level

(c) MCE Level

Figure 2-15 Comparison of the Base Shear Crest Displacement Graphs

2.4.2 Analytical Simulation of the Pushover Experiment

Two pushover analyses were conducted by applying displacement to the crest nodes
that was increased incrementally. The first analysis, referred to as Analysis 1, was
conducted on the uncracked dam, while in the second analysis, referred to as
Analysis 2, the crack lengths after the MCE level were taken into account. In
Analysis 2, the tensile strength of elements that should be cracked due to MCE event
was reduced so that cracking occurred at the beginning of the pushover analysis. The
cracking schemes are presented in Figure 2-16 together with the upstream crack
photograph of the experiment (Binici et al., 2012). The length of crack at the
upstream face of the model was obtained as 30 cm in Analysis 2, which was in a
good correlation with the experimental results, since the crack propagated in the
experiment during pushover. Given the initial cracking scheme was not prescribed
correctly, Analysis 1 only yielded cracking at the bottom of the specimen.

31
 (a) Analysis 1 (b) Analysis 2 (c) Upstream Cracking
Binici et al. (2012)
Figure 2-16 Cracking Schemes after Pushover Analyses

The base shear-top displacement behavior from the analytical study was compared to
the experiment results in Figure 2-17. As can be seen from Figure 2-17, capacities
were estimated similarly for the Analysis 1 and 2, governed by the base crack.
However, for the Analysis 1, the top displacement capacity was underestimated.

Figure 2-17 Base Shear-Top Displacement Graph

2.5 Analytical Simulation of Shake Table Testing on a Scaled Gravity Dam

2.5.1 Shake Table Test and Specimen

The experimental results of Tinawi et al. (2000) was reproduced in this study using a
finite element model in DIANA in order to calibrate the modeling assumptions for

32
the analytical simulation of the behavior of concrete gravity dams. In their
experimental study, Tinawi et al. (2000) tested a plain concrete gravity dam monolith
having a height of 3.4 m. The upstream side of the monolith with a thickness of 0.25
m was vertical while the downstream side was built 1V/0.7H. Two notches having a
thickness of 10 mm and a length of 250 mm were made in order to prescribe the
location of the crack on the monolith. A 2700 kg mass was placed at the crest so that
the natural frequency of the system was reduced. The photograph and a sketch of the
specimen that was tested are presented in Figure 2-18.

2700 kg

3.4 m
1.0 m

0.25 m

(a) Photograph of the Dam Model (b) Sketch of the Dam Model
Figure 2-18 Dam Model (Tinawi et al., 2000)

In the experiment, the specimen was subjected to triangular acceleration pulses,

which was reflected as the excitation function given in Figure 2-19 due to the trouble
with the control of the shake table. This pulse was scaled with respect to its first peak
acceleration (FPA) to three different values: 0.87, 0.94 and 0.98g. It should be noted
that the second peak was greater than the first peak acceleration. The tests with the
FPA of 0.94g and 0.98 are referred as the first and second cracking tests, respectively
(Tinawi et al., 2000). The first test with the FPA of 0.87 did not lead to any cracking
on the specimen.

33
 Figure 2-19 Input Acceleration (Tinawi et al., 2000)

2.5.2 Analysis Results

The specimen was modeled in a plane stress setting with a mesh size of
approximately 9 cm, corresponding to 695 CQ16M elements and 14 CT12M
elements, as given in Figure 2-20. The attached masses and the beam system for
attaching the mass were accurately reflected in the model in order to simulate the
mass distribution of the system properly. The notches in the specimen were
predefined in the model by removing the necessary elements at its location. Stiffness
proportional Rayleigh damping was used in the study. In accordance with Tinawi et
al. (2000), the damping ratio in the analyses was assumed as 1% and 10% in the first
mode of the structure, respectively, for the first and second cracking tests.

34
 Figure 2-20 DIANA Dam Model

As mentioned above, during the testing, some problems with the shaking table were
observed caused by the inherent flexibility of the shake table in the vertical axis. In
order to consider the effect of the flexibility of the table, 7 individual springs were
modeled at the base of the DIANA model, similar to the analytical Model M2 used
by Tinawi et al. (2000). The stiffness values of the springs were set to 230 kN/mm so
as to obtain the first natural frequency of the test setup correctly. The foundation
stiffness selected as such was higher than the selection by Tinawi et al. (2000). The
natural frequencies obtained henceforth are compared to the empirical results and
analytical model M2 from Tinawi et al. (2000) in Table 2-2. DIANA model was
more successful in yielding a close estimation of the first five modes compared to the
model by Tinawi et al. (2000). Both models had difficulty in reproducing the vertical
vibration mode of the system (mode 2).

35
 Table 2-2 First Five Natural Frequencies (Hz)

Laboratory Test Model M2 DIANA

Tinawi et al. (2000) Tinawi et al. (2000) Model
f1 16.4 16.4 16.33
f2 30.0 - -
f3 60.0 57.5 63.23
f4 105.0 72.0 103.14
f5 150.0 145.3 146.93

For the nonlinear analyses, the modulus of elasticity and the compressive strength
values were assumed in accordance with Tinawi et al. (2000) as 14.8 MPa and 14.8
GPa, respectively. In order to enable cracking in the first cracking test, tensile
strength was set to 3.25 MPa and the brittle tension softening model was used.
Tensile strength was amplified by 1.53x in order to consider the strength increase
due to dynamic loading, in contrast to 1.75x amplification used by Tinawi et al.
(2000). For the shake table test with an FPA of 0.87g, no cracking was obtained in
the model. Increasing the FPA incrementally, in accordance with Tinawi et al.
(2000), a partial crack in the downstream notch was obtained for the first cracking
test for FPA= 0.94g. A complete crack joining the upstream and downstream notches
were obtained for the second cracking test with FPA= 0.98g. The cracking schemes
for both tests are compared with the experimental results of Tinawi et al. (2000) in
Figure 2-21.

36
 FPA 0.94g FPA 0.98g

Figure 2-21 Cracking Scheme for FPAs of 0.94g and 0.98g

2.5.3 Investigation of the Effect of Modeling Parameters on the Crack

Location and Propagation

Mesh density and the material parameters used are the primary factors affecting the
development of the cracking on the specimen. In order to consider the effect of mesh
density, the dam was modeled with three different element sizes of approximately 5
cm, 10 cm and 15 cm (Figure 2-22), referred to as Model 1, Model 2 and Model 3,
respectively. Models 1, 2 and 3 were composed of 1671, 717 and 520 CQ16M
elements, respectively. The material parameters affecting the propagation of the
crack can be summarized as the properties of the softening curve, the fracture energy
and the tensile strength of the material.

37
 Figure 2-22 DIANA Models with Different Mesh Densities

2.5.3.1 The Effect of Tension Softening Functions and Fracture Energy

The final crack length for the first and second cracking test were studied considering
the effect of the softening curve of the material in this section using model 1 with a
dense mesh structure. As before, a set of seven springs were modeled at the base of
the dam models to obtain the first frequency of the system at 16.4 Hz. The
foundation stiffness of the model was the same as the one modeled by Tinawi et al.
(2000).

The post-peak response as well as the fracture energy of the concrete was observed
to significantly change the propagation of the cracking during the shaking of the
specimen. The effect of the prescribed fracture energy on different post-peak
response models were investigated first keeping the tensile strength of the model at
3.73 MPa and modifying the fracture energy. As shown in Figure 2-23a, the fracture
energy significantly changes the propagation behavior with the linear softening
model displaying the highest sensitivity to this parameter. An increase tensile
strength keeping the fracture energy constant also significantly reduced the amount
of crack propagation in the analytical model, as shown in Figure 2-23b. Increase in
any one of these parameters changed the length of crack propagation preventing the
crack to reach the experimental observation of 30 cm.

38
 (a) ft=3.73 MPa (b) Gf=105 kN/mm

Figure 2-23 Final Crack Length at the Downstream Notch for Fixed ft or Gf

The effect of the softening curve can also be observed from Figure 2-23. For both
combinations, Figure 2-23a and b, the crack length is obtained the smallest when
linear tensile softening function was used, followed by the exponential model.
Hordyk tensile softening led to the highest crack propagation among the models. The
post-peak slope of the Hordyk model appears to be very effective in defining the
length of the propagation of the crack.

The crack propagation and scheme were in good correlation with the test results by
Tinawi et al. (2000). However, the correct cracking scheme was obtained using two
different combinations of the fracture energy and the tensile strength, such as
combination 1 with the tensile strength 4.5 MPa and the fracture energy 60 kN/mm
and combination 2, with the tensile strength 2.4 MPa and the fracture energy 105
kN/mm. The cracking schemes for both combinations are presented in Figure 2-24.
The inverse problem obviously has more than one solution and the shape of the
softening curve and the tensile strength can be varied with respect to each other to
obtain similar crack propagation between the upstream and downstream notches.
However, the cracking behavior for the first cracking test was different: the first
combination reproduced the results of that test as well as the second one.

39
 (a) ft=4.5 MPa and Gf=60 kN/mm (b) ft=2.4 MPa and Gf=105 kN/mm

Figure 2-24 Cracking Scheme for Different ft-Gf Combinations for Model 1

Finally, it can be concluded that cracking was significantly affected by both the
tensile strength and the fracture energy defining the softening curve of the concrete.
The propagation of the crack observed in the second test can be reproduced by
applying dynamic application factor to either the tensile strength or the fracture
energy. Amplifying both of the values led to less crack penetration.

2.5.3.2 The Effect of Mesh Density

The different mesh structures as given in Figure 2-22, fine (Model 1), medium
(Model 2) and coarse (Model 3) meshes, were utilized to determine the effect of a
selected mesh density on the crack propagation within the analytical model. The
spring stiffness values were assumed as 193 kN/mm, 205 kN/mm and 210 kN/mm
each, for the Models 1, 2 and 3, respectively. Elastic analyses were conducted to
obtain the maximum tensile stress at the tips of upstream and downstream notches.
The results are tabulated in Table 2-3.

40
 Table 2-3 Maximum Stresses of the Models

Model 1 Model 2 Model 3

Max. Stress (MPa) Max. Stress (MPa) Max. Stress (MPa)
Upstream Downstream Upstream Downstream Upstream Downstream
Notch Notch Notch Notch Notch Notch
4.04 5.54 2.81 3.81 2.76 3.42

As given in Table 2-3, the stress generated near the edge of the notch increased as the
element size decreased. When the mesh became denser, there was a significant
increase in the stress concentration at the notch leading to an increase in the stress
demand. The issue was significantly more problematic in model 1; due to the nature
of the very fine mesh, very large stress concentrations were forming near the
discontinuity (the notch), leading to unexpected cracking, such as the cracking
observed in the upstream notch for the first cracking test. Therefore, the model had to
be biased towards large tensile strengths due to this issue, which is a common, well-
known problem for solution of continuum problems with finite elements. Analyses
were also conducted with high values of tensile strength and fracture energy of 4.5
MPa and 300 N/m, respectively, which still led to cracking near the notches that did
not propagate into the rest of the model.

Given that the stress near a discontinuity is theoretically near infinity, the finite
element model tries to approach to the theoretical solution when the element size is
decreased. However, the cracking occurs due to effect of stresses on an effective
area, not the point of discontinuity. This is a limitation of the macro-modeling of the
concrete which is very heterogeneous in nature. As well as setting a limit on the
maximum element size, a limit should also be abided on the minimum element size.
Significantly more information on the limitations on the macro modeling of concrete
are provided in Bazant (1986). The minimum element size is a valid limitation to the
modeling of this specimen, because of its small size and the presence of notches
yielding stress concentrations. When dynamic analyses were repeated on a model
without a notch, a similar mesh bias was not observed. Within the limits of the

41
current computational power, however, the minimum element size, which is
specified as three times the maximum aggregate size (Bazant, 1986), is not a limit for
the large dam monoliths as we cannot use as fine a density of mesh for these systems.

The analyses of the model 2 and 3 were performed using exponential tension
softening with 1% and 4% stiffness proportional Rayleigh Damping for first and
second cracking tests, respectively. Total strain rotating crack model was used in the
analyses. For Model 2, the crack penetration was reproduced with a tensile strength
of 3.73 MPa and a fracture energy of 80 kN/mm (Figure 2-25). These values were
very close to their counterparts assumed in the analytical study by Tinawi et al.
(2000). Reducing the fracture energy slightly (60 kN/mm), a similar cracking scheme
was obtained as given in the same figure, leading to the overestimation of the
cracking in the first test. Crack mouth opening was estimated larger in this case, and
the crest response changed significantly.

(a) ft=3.73 MPa and Gf=80 kN/mm (b) ft=3.73 MPa and Gf=60 kN/mm

Figure 2-25 Cracking Scheme for Different Gf for Model 2

For model 3, the cracking scheme was reproduced with a tensile strength of 3.3 MPa
and fracture energy of 80 kN/mm as shown in Figure 2-26. Notably, the strength of
the material had to be assumed somewhat smaller in this model, as due to the
coarseness of the mesh, smaller stress demand were obtained in the finite element

42
model. In the second crack test crack mouth opening displacement reached to a value
of nearly 1 mm. It can be concluded that when crack is localized, high values of
crack mouth opening displacement can be reached.

Figure 2-26 Cracking Scheme for Model 3

The three different meshes used in this section showed that the stress variation due to
the finite element model is an effective quantity in determining the distribution of the
cracking on concrete elements. Notably, similar fracture energies was used for all
models to reproduce the cracking given in both cracking tests simultaneously,
however, to correctly predict the initiation of cracking, the tensile strength has to be
changed for each model, i.e. from 4.5, 3.7 to 3.3 for fine, medium and coarse meshes,
respectively. These represent 2.11 times, 1.74 times and 1.55 times the static strength
for the specimen. The fracture energy was used as 1.3x the static value as well.
Given that models for large dam monoliths will be considered on the coarse side, the
use of 1.5x the tensile strength as suggested in USACE (1995) appear reasonable for
nonlinear modeling to obtain the cracking on these systems.

2.5.4 The Effect of the Time Stepping Factor

Given that the duration of the pulse is very important for determining the crack
propagation, the effect of time stepping on the analyses results are considered in this
section. In order to investigate the effect of the time stepping factor, analyses were

43
conducted with five different time steps Δt of 0.00100, 0.00060, 0.00040, 0.00025
and 0.00005 seconds corresponding to 250, 450, 625, 1000 and 5000 steps in the
analyses for reaching the total duration of 0.25 seconds. While the maximum tensile
stresses generated in the dam were same for these five different analyses, for a
decrease in the Δt, the effective duration of the maximum stress increased, leading to
more cracking on the specimen. The final crack length did not change with further
reduction of Δt from Δt=0.0004 seconds. Reducing Δt to 0.00005, there was only a
slight difference in the response of the dam. The effect of the time step Δt on the
final crack length is summarized in Figure 2-27. The Δt used for the analyses
changed the time period in which the maximum stress was effective on the system,
hence for larger Δt values, with the reduction of effective time, the crack propagation
was reduced. Consequently, the use of 1000 time steps was adequate in order to
optimally represent the behavior of the dam and decrease the computational cost. It
should be noted that Tinawi et al. (2000) used 5000 time steps in their analytical
study. The use of Δt=0.00025 (1000 time steps) falls towards the smaller of the limits
prescribed for nonlinear transient analyses in Chopra (2007). For the fourth mode at
105 Hz, Tn/20 equals 0.0005 seconds.

Figure 2-27 Comparison of the Final Crack Length Obtained with Different Time
Steps

44
2.5.5 Detailed Investigation of the Simulation of Crack Opening

The ability of the analytical models to reproduce the displacements from the two
cracking tests was investigated in this section.

2.5.5.1 The Effect of the Material Modeling

Using the total strain rotating/fixed crack and the multi-directional fixed crack
model, the prediction of the displacement response of the specimen at the crack
mouth was investigated in this section. The shear retention factor is nominally equal
to 1 in the rotating crack model, while it was formally specified as 1.0 for the total
strain fixed crack and multi-direction fixed crack model. The results obtained for the
crack mouth opening at the downstream notch, the crest relative displacement and
the crest relative acceleration for the total strain models are presented in Figure 2-28
for the first test. The total strain crack models yield very similar results as given in
the figure. Given the test setup, which prescribed the plane of maximum principal
stress well near the prebuilt notches, the models yielded very similar results.
However, it can be seen these models failed to simulate the downstream notch crack
mouth opening displacement after the first peak of the CMOD. A better response was
obtained when the multi-directional fixed crack model was used. These results were
closer to the results obtained in the analytical model by Tinawi et al (2000). In this
model, the shear retention factor within the crack plays an important role, the effect
of which is investigated in the following section.

45
 (a) Downstream Notch CMOD (b) Crest Relative Displacement

(c) Crest Relative Acceleration

Figure 2-28 Comparison of Response Quantities with Total Strain Fixed & Rotating
Crack and Multi-Direction Fixed Crack Models

2.5.5.2 The Effect of the Shear Retention Factor

As given in Figure 2-28, the notch opening was not reproduced correctly with the
total strain models for a shear retention factor of 1, prescribing full shear transfer at
the crack plane after the closing of the cracks. According to Rots and Borst (1987),
when there is an unsymmetrical loading, the crack will both open and slide,
damaging the crack plane. A reduction in shear stiffness in terms of a shear retention
factor should be taken into account. In order to model this loss in the crack plane,

46
fixed crack model, in a total strain or multi-direction formulation, is more
appropriate. The crack opening displacement, crest displacement and the crest
acceleration of the specimen are presented in Figure 2-28 for both models using a
shear retention factor of β=0.2. The inclusion of a lower shear retention factor did not
significantly affect the displacement and acceleration response for the total strain
crack model, on the other hand, a satisfactory prediction of the displacement
response was obtained with the multi-direction fixed crack model as shown above.
However, a further complication arose for these models: reproducing the CMOD
values could only be obtained with crack elongation twice as much as Tinawi et al.
(2000).

The effect of shear retention factor on the crack length and the CMOD was studied
using a range of shear retention factors and computing the final crack length and the
CMOD at 0.15 seconds. As can be seen from Figure 2-29, the final crack length
increased with a decreasing shear retention factor. The increase in the crack length
also corresponded to an increase in the CMOD at t=0.15 sec. These two graphs
suggest the use of a shear retention factor around 0.2 in order to best represent the
behavior of the specimen for the first cracking test. Although Tinawi et al. (2000)
used a varying shear retention factor in the analytical modeling of the experiment in
accordance with the suggestion from Rots and Borst (1987) in the form of a shear
softening function, a constant shear retention factor assumed as in the DIANA
analyses led to satisfactory results for matching the final crack length and the CMOD
at 0.15 seconds. It can be concluded that as a general trend, when the shear capacity
of the concrete decreases, the crack propagates more and CMOD increases.

47
 (a) Final Crack Length with Shear (b) CMOD at t=0.15 sec
Retention Factor

Figure 2-29 The Variation of the Final Crack Length and the Crack Opening with
Shear Retention Factor

2.5.5.3 The Effect of Rayleigh Damping

The displacements obtained in the analyses were also seen to be sensitive to the
damping prescribed to the model. Very little of the damping in the specimen occurs
due to the viscous damping sources naturally prescribed in analytical solutions;
majority of the damping takes place as a result of the crushing and shearing at the
crack interface which significantly changes the displacement response of the
monolith. A single damping ratio, at 1%, was used above, in accordance with Tinawi
et al. (2000). The effect of higher Rayleigh Damping on the analysis results were
investigated using 1%, 5%, 10%, 15% and 23% in a set of analyses. The maximum
tensile stress generated at the tip of the downstream notch decreased with increasing
damping ratio as given in Figure 2-30. A linear relationship to this reduction is
discernible: the stress dropped as much as the damping was increased. The damping
ratio affects the crest acceleration directly as shown in Figure 2-30.

48
 (a) Maximum Stress at Downstream (b) Crest Acceleration
Notch
Figure 2-30 Response Quantities Obtained Using Different Damping Ratios

In order to obtain the CMOD displacements more correctly, a more detailed

formulation for the damping at the crack is obviously necessary. The treatment of
damping at the crack plane requires a more detailed concrete constitutive model
which is left for a future study.

2.5.5.4 Second Cracking Test

In the second cracking test, the results of the analytical study reproduced the crack
penetrating between the two notches as found by Tinawi et al. (2000); however, the
crack mouth opening displacement was obtained very small compared to the
laboratory test and the crest response was obtained out-of-phase with the test. The
use of different shear retention factors had no significant effect on final crack length
and crack mouth opening displacement. Analyses were conducted with different
damping ratios from 1% to 23% to model the second cracking test; however,
damping ratio did not increase the CMOD significantly. According to Yamaguchi et
al. (n.d.), the abrupt decline of the response can be modeled with transient damping
only, not with Rayleigh damping for this phase of the test. Using a transient damping
proportional to the instantaneous secant stiffness matrix, the decline of the response
was modeled correctly, while the dam continued its harmonic motion using Rayleigh

49
damping. In the analytical models, Tinawi et al. (2000) used instantaneous damping
as well. Kimata et al. (2008) modeled the specimen with stiffness proportional
damping as a function of time, reaching similar results. As the formulation currently
used does not include step by step determination of the damping component of the
resisting force based on the material stiffness properties, like the secant stiffness, the
displacement response of the specimen could not be reproduced properly. As
mentioned above, a more detailed concrete constitutive model should be formalized
for this purpose.

2.6 Summary

The results from the four different calibration studies are summarized as follows:

 The shape of the post-peak curve and the strength determine the crack
propagation process. For all the three test, it was observed that softening
functions with faster declining curves after the peak were more appropriate
for use, such as Hordyk or exponential models.
 For the finer meshes approaching the limits of macro-modeling of concrete,
concentration of the stresses led to significant stress demands, and required
the assumption of higher tensile strength values in order to reproduce test
behavior. However, within current limitations of computational power,
modeling of monoliths is still well within macro modeling limits. The 1.5x
dynamic strength, as prescribed by USACE (1995) appears appropriate for
dynamic modeling of concrete monoliths.
 Both the rotating and fixed crack models yielded similar cracking patterns
predicting the crack propagation correctly for the experiments. However, the
displacement predictions were rather off the mark. Further study in
constitutive modeling is required to predict accurate displacements or crack
openings for monolithic concrete structures.
 Shear retention factor and the damping behavior within the crack appear to be
important factors in the determination of correct crack mouth displacements.

50
However, these are very complex issues significantly related to aggregate
interlocking, crushing within the crack, and possibly the number of cycles the
cracks go through. Further testing is required in order to address these issues
with a more detailed constitutive model.

51
52
3.
CHAPTER 3

INCREMENTAL DYNAMIC ANALYSIS OF A CONCRETE GRAVITY

DAM

3.1 Introduction

In Chapter 2, it was shown that the crack length and its location in a dam can be
simulated accurately in a finite element model. The analytical model as well as the
total strain crack models can be used to investigate the distribution of cracking within
a dam body. To this end, the nonlinear behavior of an 80 m high dam monolith was
investigated in this section using incremental dynamic analysis (IDA) with the
primary purpose of understanding the nonlinear performance of this structure as well
as its performance limits. This chapter of the thesis is organized as follows. First, the
finite element model used in the analyses is outlined, followed by a brief definition
of the incremental dynamic analysis procedure used. The set of 21 ground motions
used for the analyses are presented next. The analysis results are then presented, in
which the relationship between key seismic intensity parameters and the seismic
demand results are investigated. Finally, the results of the IDA are compared to the
static pushover analysis results in order to evaluate the scope of possible use of
pushover analyses for determining the performance levels of dams.

3.2 Gravity Dam Model

The dam investigated in this chapter is an arbitrary concrete system, with a typical 80
m high monolith with a vertical upstream and a 0.8H/1V downstream face. The crest

53
length of the monolith was assumed as 6 m (Figure 3-1). As explained before, the
foundation was modeled according to the simplified approach introduced in USACE
(1995) and the hydrodynamic forces were accounted for by Westergaard’s added
mass approach (Westergaard, 1933). The reservoir elevation was assumed at 78 m.

6m

80 m

Figure 3-1 Concrete Gravity Dam Model

The analyses in this section were performed using the total strain rotating crack
model. The dam body was modeled with 5025 CQ16M elements and 45 CT12M
elements while the foundation was modeled with 4430 CT12M elements. The
Young’s Modulus, Poisson’s ratio, density and the tensile strength for the dam
monolith were assumed as 12.0 GPa, 0.2, 2400 kg/m3 and 1.8 MPa, respectively.
Exponential tension softening function was used with a fracture energy of 225 N/m.
The use of this value for the element size given in the dam mesh was validated so
that no strength reduction would occur during the analyses. It was assumed that
failure would occur due to tensile overstressing (Alembagheri and Ghaemian, 2012);
therefore, the crushing failure was ignored and the compressive behavior was
assumed elastic. For the foundation, a medium quality rock foundation was assumed;
Young’s Modulus was taken as 10.8 GPa and the Poisson’s Ratio as 0.3.

54
The effective damping ratio (Equation 3-1) due to the effect of the foundation and
reservoir was calculated in accordance with USACE (1995) as given below. It takes
the individual damping contributions of dam, foundation and reservoir into account.

̃= + + (3-1)

In the above equation, ̃ is the effective damping ratio, is the damping value that
is suggested to be assumed as 5% for OBE level ground motions. and are the
added damping ratios due to hydrodynamic effects and dam-foundation interaction,
respectively. Rr is the ratio of the first natural period of the structure with full
reservoir & flexible foundation case to the empty reservoir & flexible foundation
case. Similarly, Rf is the ratio of the first natural period of the structure with flexible
foundation & empty reservoir case to the rigid foundation & empty reservoir case.
The damping ratio was calculated as 10% from the above equation. Mass and
stiffness proportional Rayleigh Damping which produced 10% effective damping
ratio in the first mode of the structure was used in the analyses. The first four natural
periods together with their mode shapes are presented in Figure 3-2.

(a) Mode Shape 1, T = 0.47 sec (b) Mode Shape 2, T = 0.20 sec

(c) Mode Shape 3, T = 0.18 sec (d) Mode Shape 4, T = 0.10 sec

Figure 3-2 Mode Shapes

55
A constant time increment of 0.005 was used in the transient analyses. The
convergence norm for both the residual displacement and force criteria was selected
as 0.001. After the application of self-weight of the dam, hydrostatic load was
applied and the dam model was subjected to several ground motions with different
intensities.

3.2.1 Incremental Dynamic Analysis

In incremental dynamic analysis (IDA), the structure is subjected to a set of ground

motions which are scaled to multiple levels of intensity so that the relationship
between the damage and intensity measures is obtained. The natural (unscaled)
ground motions are scaled with a positive scaling factor that increases incrementally
(Vamvatsikos and Cornell, 2002). The intensity measure (IM) is a positive scalar that
depends on the intensity of a ground motion and increases incrementally with the
scaling factor. Some common examples of IMs for the design or evaluation of
buildings are the peak ground acceleration (PGA), peak ground velocity (PGV) and
the 5% damped spectral acceleration at the structure’s fist-mode period (Sa(T1,5%))
(Vamvatsikos and Cornell, 2002).

The damage measure (DM) is a positive observable scalar, which can be found from
the output of the non-linear dynamic analysis results. The quantification can be
subjective based on the user’s goals (design or evaluation) and constraints (capability
of the analyses tools and time constraints). Some common response parameters used
are the maximum base shear or maximum crest displacement of a structure, which
are sometimes used as damage measures for ductile structures. However, the use of
these parameters as damage measures includes the intrinsic assumption of elastic-
perfectly plastic behavior, such as the perceived behavior in well-reinforced ductile
buildings. The behavior of a dam monolith is different from such a well-defined
response, as such, a seismic demand parameter, such as the crest displacement, can
hardly be termed as a damage measure for this monolith. A more practical damage

56
measure, in terms of total cracking of the monolith, is investigated in this section
along with the usual seismic demand parameters presented above.

3.3 Selection of Ground Motions

In order to perform IDA of the 80 m high concrete gravity dam monolith, a set of 21
ground motions were selected (Table 3-1) so that the effect of different
characteristics of ground motions could be investigated with the effect of different
scaled levels. The motions were selected from Pacific Earthquake Engineering
Research (PEER) strong motion database which were recorded on rock or stiff soil
(type A or B of USGS) that had large magnitudes ranging from 6 to 7.6 and
epicentral distances of 2 to 39 km, in accordance with FEMA440. To perform IDA,
selected ground motions were scaled such that the spectral acceleration at the dam’s
first natural period (0.47 sec) was scaled from 0.4g to 0.8g with 0.05g increments.
The spectral accelerations of these ground motions are presented in Figure 3-3. These
target levels led to the use of scale factors ranging from 0.22 to 6.68.

57
 Table 3-1 Selected Ground Motions
No Earthquake Name Component Moment PGA PGV R USGS Sa(5%) at
Magnitude (g) (cm/s) (km) Soil Type T1 (g)
1 Imperial Valley, 1979 315 degree 6.5 0.204 16.1 14.2 B 0.307
2 San Fernando, 1971 90 degree 6.6 0.110 13.3 31.7 B 0.318
3 San Fernando, 1971 270 degree 6.6 0.136 5.6 38.9 B 0.120
4 Landers, 1992 0 degree 7.3 0.171 20.2 23.2 B 0.290
5 Loma Prieta, 1989 0 degree 6.9 0.512 41.2 13.0 B 0.543
6 Loma Prieta, 1989 67 degree 6.9 0.357 28.6 11.6 B 0.842
7 Loma Prieta, 1989 270 degree 6.9 0.244 20.3 21.4 B 0.720
8 Morgan Hill, 1984 90 degree 6.2 0.292 36.7 11.8 B 0.484
9 Northridge, 1994 360 degree 6.7 0.514 52.2 22.6 B 1.153
10 Düzce, 1061 Lamont E 7.1 0.134 13.7 15.6 B 0.246
11 Manjil, Abbar - ABBAR Longitudinal 7.37 0.515 43.8 12.6 B 0.703
58

12 Tabas Transverse 7.35 0.852 98.2 2.1 A 1.860

13 Loma Prieta, Gilroy 337 degree 6.9 0.325 22.3 11.6 B 0.566
14 Northridge, Burbank – Hw Rd. 330 degree 6.7 0.163 8.5 20 B 0.275
15 Northridge, 24088 Pacoima 360 degree 6.7 0.433 51.5 8.2 B 1.091
16 Spitak 0 degree 6.8 0.199 28.6 30 B 0.391
17 Cape Mendocino 1992 0 degree 7.1 1.497 127.4 8.5 A 1.582
18 Chi-Chi, Taiwan, 1999 E 7.6 0.133 39.8 14.3 A 0.196
19 Kocaeli – Gebze, 1999 0 degree 7.4 0.244 50.3 17 A 0.431
20 Kocaeli – İzmit, 1999 90 degree 7.4 0.22 29.8 4.8 A 0.341
21 Palm Springs, 1986 270 degree 6.0 0.612 31.5 7.3 A 0.922

58
 Figure 3-3 Acceleration Response Spectra (Scaled at T1) for the Selected Ground
Motions

3.4 Results of IDA

A total number of 189 nonlinear transient analyses were conducted for the IDA. The
cracking (both open and closed) at the end of the time history for all scales of
different ground motions are presented in Figure 3-4. It can be seen from this figure
that, the cracking initiates from the base of the dam at the upstream face. Later, a
partial crack occurs at the base of the dam at the downstream face and as a general
trend, cracking at the dam body first initiates from the downstream face. When the
intensity of the motion increases, downstream cracks join the upstream cracks.

59
0.40g
0.45g
0.50g
0.55g
0.60g
0.65g
0.70g
0.75g
0.80g Motion 1 Motion 2 Motion 3 Motion 4 Motion 5 Motion 6

Figure 3-4 Cracking Schemes from the IDA

60
 Motion 7 Motion 8 Motion 9 Motion 10 Motion 11

0.40g
0.45g
0.50g
0.55g
0.60g
0.65g
0.70g
0.75g
0.80g

Figure 3-4 (Continued) Cracking Schemes from the IDA

61
0.40g
0.45g
0.50g
0.55g
0.60g
0.65g
0.70g
0.75g
0.80g Motion 12 Motion 13 Motion 14 Motion 15 Motion 16

Figure 3-4 (Continued) Cracking Schemes from the IDA

62
 Motion 17 Motion 18 Motion 19 Motion 20 Motion 21

0.40g
0.45g
0.50g
0.55g
0.60g
0.65g
0.70g
0.75g
0.80g

Figure 3-4 (Continued) Cracking Schemes from the IDA

63
The selection of a damage measure for a gravity dam monolith, similar to the use of a
parameter like story drift ratio in buildings, is a significant challenge. First of all,
field or experimental data on dams that has undergone significant damage is
extremely scarce. Secondly, a numerical index should to be tied to some qualitative
performance level which is still subjective. Performance levels are still widely
debated within the dam engineering community. Therefore, a relatively simpler
choice is made in this study and the total crack length within the dam body was
chosen as the damage measure (Kanenawa et al., 2004). Cracks may start at various
heights on the dam body, and propagating perpendicular to the upstream (horizontal
cracks) and the downstream (inclined cracks) face. Accordingly, the dam body is
divided into 17 crack initiation zones with 5 m increments, the first one
corresponding to the cracking that would occur at the base, and the sequential ones
corresponding to cracks that will form in these zones within the height of the
increment (Figure 3-5). In accordance with El-Aidi and Hall (1989), if more than one
crack occurred in one of the crack zones, the longest crack was taken into account
and the other cracks were neglected due to the well-known smearing of damage to
the nearby elements along the main crack in the constitutive model. Some minor
cracks in the upstream direction (that did not propagate towards the downstream
cracks) were also neglected. As a demonstration of the process, the individual cracks
taken into account for motion 18, scaled to 0.80g are presented in Figure 3-5 with red
lines. The total crack length was calculated as 174 m for this example; 117 m in the
upstream direction and 57 m in the downstream direction.

64
(a) A schematic drawing of the (b) Mildly-damaged (c) Severely-damaged
crack zones cracked scheme (motion cracked scheme (motion
7, Sa scaled to 0.80g) 18, Sa scaled to 0.80g)

Figure 3-5 Crack Zones and Cracking Schemes

The effect of damage on the structure in terms of extending the fundamental period
was also recognized and the elongated natural period and its effect were investigated.
By simulating the discontinuities, with decreasing the moduli of the cracked
elements, in the upstream and downstream sides separately in Eigen analyses, the
change in the fundamental period was compiled, for each motion, as given in Figure
3-6. The obtained range was an upper bound on the damaged period considering the
facts that not every crack is open during a cycle of motion and the cracks open and
close (upstream and downstream side usually alternating) during a given earthquake
shaking. When the elongated period due to the upstream and downstream cracking
was investigated separately, it was observed that the influence of the upstream
cracking on the period elongation was very significant, unlike the downstream
cracking. With the increase in the length and number of upstream cracks, the period
of the dam elongates considerably, by as much as 3 times of the initial period
(motion 18, scaled to 0.80g). The effect of downstream cracking was significantly
less; the elongation was limited to 20%.

At average, the period was determined to increase by 45% due to the damage

65
occurring on the system, an increase of natural period from 0.47 sec to 0.68 sec.
Even if the damaging effect of motion 18 is disregarded, the period increase is 42%.
A second degree polynomial can be fitted to the mean of the estimates for each
design level with a correlation coefficient of 0.68 (shown in magenta in Figure 3-6);
however, the correlation coefficient was increased to 0.81 when motion 18 is
disregarded.

Figure 3-6 First Cracked Period of the Dam

For the design of such dams or large infrastructure projects, seismic hazard analyses
are commonly conducted with the purpose of obtaining a target design spectrum that
is utilized for generation of ground motions. Hence, spectral acceleration is most of
the time the sole IM used in the design of such structures. In this study, apart from
the spectral values Sa, the peak ground velocity (PGV) and spectral velocity
Sv(T1,5%) was also considered. In Figure 3-7, the variation of damage on this typical
monolith with different damage measures are presented together with their
correlation coefficients (R2), namely different spectral acceleration levels (i.e.
different targets), peak ground acceleration (PGA), spectral velocity at first natural
period (Sv), peak ground velocity (PGV), arias intensity and specific energy density.
The spectral velocity as a prediction parameter yielded the highest R2=0.81 value
compared to other intensity measures. However, even for this parameter, there is a
noticeably wide spread of the data points (subjectively) around the best fit. Spectral
acceleration, the traditional parameter used in the design and evaluation of structures,
did a very poor job in predicting the damage occurring on the dam monolith. Given

66
the poor performance of these commonly used parameters, prediction using a more
involved formula using different IM values was investigated.

(a) Spectral Acceleration at First Natural (b) PGA, R2=0.71

Period, R2=0.73

(c) Spectral Velocity at First Natural (d) PGV, R2=0.57

Period, R2=0.81

(e) Arias Intensity, R2=0.67 (f) Specific Energy Density, R2=0.45

Figure 3-7 Total Crack Length of the Dam vs. Different Intensity Measures

67
A preliminary relationship between the total crack length and the considered
intensity measures, PGV and Sv(T1,5%), is proposed in Equation 3-2. The predictions
on the crack length obtained using the proposed formulation is compared to the
analyses results in Figure 3-8. In this equation (3-2), PGV represents the peak ground
velocity of the motion and Sv(T1,5%) represent the spectral velocity at the structure’s
first natural period. From Figure 3-8, it can be observed that the chosen measure
successfully correlates with the behavior of the dam, with a correlation coefficient of
0.90. When the IM is above 0.45, it can be seen that the total crack length increases
rapidly (Figure 3-8); therefore, it is suggested that in design, the IM should be kept
below 0.45.
IM = || | | (3-2)

Figure 3-8 Total Crack Length of the Dam with the Proposed Intensity Measure

Overall, in lieu of the spectral acceleration quantity that is extensively used in

earthquake engineering, the velocity characteristics of the ground motion and the
spectral velocity appear to be the significant factors that affect the damage occurring
on the system. With the consideration of the velocity characteristics of the ground
motion, it was possible to estimate a bound on the damage level on the monolith for
the chosen target level. A similar estimation of damage on the system using spectral
acceleration of the motion can be widely off the mark, as shown in Figure 3-7. This
issue presents a significant problem with the current methodology of assessing
seismic hazard for such structures, which is conducted in terms of defining a target
response spectrum. The crack propagation on the monolith composed of the brittle

68
concrete material was determined to be very much dependent on the velocity
characteristics of the ground motion indicating that the duration spent at certain force
levels are as important as the force levels.

Following the tradition of decades of spectrum use along with the much accepted
procedure of reduction for ductility, a general understanding is accepted to govern
the performance of buildings: some damage is expected for a given target level with
a significant reduction (1/8th to 1/4th) of the design force level. However for a very
rigid, massive structure like a dam monolith comprised of the brittle material
concrete, the spectral level governs only the level in which initial cracking will
occur: the duration and velocity characteristics of the motion is much more important
in the final performance of the given section. Therefore the use of spectral levels is
correct only for designs within the elastic range: much more information about the
earthquake time history is required in order to design or evaluate the same structure
in the nonlinear range.

In order to quantify the uncertainty in the velocity and duration characteristics of the
ground motion, the use of many motions and scaling factors such as suggested in
FEMA440 appears to be a reasonable option, given that provision of target ground
velocity or duration information for seismic hazard at a local site is still quite far
away. The use of the variability in recorded motions to address the variability in the
velocity and duration for a target level motion at a site is more feasible at this stage,
as this method typically requires the use of more than 10 motions in the analyses.
Use of a very few (typically 3) spectrum fitted motions, in this sense, is not adequate
for the determination of the level of performance of a system given the large
variability in the performance level for different motions.

3.5 Pushover Analysis

A pushover analysis (or nonlinear static procedure) can be defined as applying a

static lateral force distribution on a given structure in order to estimate the seismic

69
structural deformations (FEMA, 1997). It is a widely used method to determine the
performance limits of structures. Pushover analyses for buildings are widely popular
given the relative ease of use of the methodology along with the well-defined
performance limits generally obtained from similar analyses.

Given the significant cost of IDA analyses, the pushover analysis is an attractive tool
also for dam monoliths in order to understand the behavior better using a much faster
analyses approach. However, there are two significant problems in applying such
analyses to monoliths: First of all, some component of the lateral force on a dam
system (i.e. hydrodynamic force) is frequency dependent, therefore it is very hard to
ascertain the shape of the hydrodynamic force on a dam monolith given that it
depends on the nature of the ground motion as well as structural properties. On the
other hand, the lateral force on the structure, which is typically applied proportional
to the first mode shape on such a structure is hardly reflective of the distribution of
the force in later stages of a damaging motion. In other words, it is very hard to
reflect a realistic force distribution on such a structure given the complexity of the
damaged system unlike frame systems.

Given the complexities of applying pushover analysis technique to a dam monolith,

the analysis given here was conducted with some basic assumptions simplifying the
problem. The lateral load distribution was assumed as given in Equation 3-3a
(Alembagheri and Ghaemian, 2012).

P(y) = F(y) + H(y) (3-3a)

where P(y) is the total lateral load distribution that accounts for the contributions
coming from the inertial effects, F(y) and hydrodynamic effects, H(y). y is the height
measured from the base of the dam. The hydrodynamic force distribution is derived
from the Westergaard’s added mass approach assuming incompressible reservoir
fluid. F(y) and H(y) are defined in Equations 3-3b and 3-3c, respectively.

F(y) = α b(y) ψ1(y) ρc (3-3b)

H(y) = ρw √ for y h (3-3c)

70
where α is defined as the weight factor, b is the dam width, ψ1 is the normalized first-
mode shape of the dam, ρc is the density of concrete, ρw is the water density and h is
the reservoir height. The weight factor, α, was taken as 4 so that the total inertial load
be 3 times the hydrodynamic load (Alembagheri and Ghaemian, 2012). The lateral
load distribution is presented in Figure 3-9.

Figure 3-9 Total Lateral Load Distribution for the Pushover Analysis

Pushover analyses were conducted for a full reservoir and an empty reservoir case.
Also, the analyses were conducted with the load application into the downstream
(referred to as direction 1) and the upstream faces (referred to as direction 2) of the
dam. A schematic drawing of the loading directions is presented in Figure 3-10.

(a) Pushover Loading, Direction 1 (b) Pushover Loading, Direction 2

Figure 3-10 A Schematic Drawing of the Pushover Loading Directions

71
The cracking schemes obtained using these simple load distributions are shown in
Figure 3-11.

(a) Full Reservoir, Dir 1 Loading (b) Full Reservoir, Dir 2 Loading

(c) Empty Reservoir, Dir 1 Loading (d) Empty Reservoir, Dir 2 Loading
Figure 3-11 Cracking Scheme of the Pushover Analyses

As given in Figure 3-11, the pushover analyses resulted in significantly different

cracking distribution on the dam compared to the IDA analyses results. For the full
and empty reservoir cases, the upstream-downstream direction loading damaged the
base of the dam completely: the system failed when the crack starting at the upstream
face reached the downstream face. For the other direction loading, the crack occurred
parallel to the downstream face propagating towards the base for the full reservoir
case. Without the reservoir, this crack was coupled by a crack forming at the base on
the downstream side.

Notably, the distribution of cracking from the pushover analyses was much different
from the IDA analyses, much more concentrated on the base of the dam. None of the
IDA analyses showed a base crack traversing between the upstream and downstream

72
face, or even the monolith base having a crack traversing more than 50% of its
length. The load distribution in a given time history analysis was obviously much
more different than the one assumed in pushover analyses as the failure behavior was
distinctly different for these two methods. The total crack length for full reservoir
and empty reservoir case was obtained as 104 m and 145 m, respectively, compared
to the maximum crack length obtained from IDA analyses, which was 174 m. The
results are reasonable only if the damage measure specified at crack length at face
value. Although the total crack length could be said to be estimated close to the IDA
results for the empty reservoir case, the resulting cracking scheme does not match
with IDA. In the IDA analyses, due to redistribution of the loading, the cracking was
spread over the dam height; however, in the pushover analysis such a redistribution
could not be obtained.

The base shear versus the crest displacement for the full reservoir case is compared
for IDA and pushover analysis for both loading directions in Figure 3-12. The mean
base shear obtained for all the motions at a given displacement is presented. The
pushover curve and the IDA curve were very similar in the initial, elastic part. On the
other hand, the curves started to diverge from each other around 10 cm of crest
displacement. Base shear values obtained from the transient analyses were as high as
50% from the pushover analyses, clearly showing the different load carrying pattern
for both analyses. Pushover analyses significantly underestimated the base shear as
well as lateral displacement capacity of the dam for direction 2 as well.

(a) Displacement vs Base Shear, Dir 1 (b) Displacement vs Base Shear, Dir 2
Figure 3-12 IDA vs. Pushover Analyses Results

73
Given the significant difference between the results of the IDA and pushover
analyses, it is quite evident that the pushover analysis methodology should be
significantly changed for its proper application to dam monoliths. The distribution of
loading is a significant issue both for determining the hydrodynamic contribution as
well as reflecting the change of the inertial force to the system. From the IDA
analyses, it was observed very clearly that once the base of the dam cracks to some
extent at the upstream and downstream faces, the crack initiation is directly
transferred to points higher up on the dam body. For the remaining time of the
motion, the majority of the cracks occur further away from the base and propagate
towards inside the dam body. An adaptive methodology reflecting the effect of
damage on the load distribution on the dam body is required in order to obtain a
more realistic damage distribution similar to the IDA analyses results.

74
4.
CHAPTER 4

CONCLUSION AND FUTURE PLANS

4.1 Conclusion

In this study, the behavior of a concrete gravity dam monolith was investigated using
the incremental dynamic analysis (IDA) to determine the development of damage on
the monolith and the corresponding cracking patterns.

First, the constitutive model was calibrated to the experimental data so that the
ability of the model to simulate the fracture in this material was verified. The
following conclusions were drawn based on the results of the simulations.

 The shape of the post-peak curve and the strength determine the crack
propagation process. For all the three tests, it was observed that softening
functions with faster declining curves after the peak were more appropriate
for use, such as Hordyk or exponential models.
 For the finer meshes approaching the limits of macro-modeling of concrete,
concentration of the stresses led to significant stress demands, and required
the assumption of higher tensile strength values in order to reproduce test
behavior. However, within current limitations of computational power,
modeling of monoliths is still well within macro modeling limits. The 1.5x
dynamic strength, as prescribed by USACE (1995) appears appropriate for
dynamic modeling of concrete monoliths.
 Both rotating and fixed crack models yielded similar cracking patterns
predicting the crack propagation correctly for the experiments. However, the
displacement predictions were rather off the mark. Further study in

75
 constitutive modeling is required to reproduce accurate displacement of
cracked concrete models.
 Shear retention factor and the damping behavior within the crack appear to be
important factors in the determination of correct crack mouth displacements.
However, these are very complex issues significantly related to aggregate
interlocking, crushing within the crack, and possibly the number of cycles the
cracks go through. Further testing is required in order to address these issues
with a more complicated constitutive model.

Next, IDA was performed on an 80 m high concrete gravity dam monolith. The
following conclusions were drawn based on the results from this study.

 The damage development on the monolith is significantly affected by the

velocity characteristics of the ground motion and the spectral velocity, in lieu
of the spectral acceleration values commonly used in the design of these
structures. The spectral quantities such as PGV and S v(T1,5%), affected the
cracking behavior of dam significantly.
 Upstream cracks change the behavior of the monolith much more
comprehensively compared to the inclined downstream cracks. The natural
period increased to as high as 3 times the initial period with upstream
cracking compared to a maximum increase of 1.2 times for the downstream
cracking. An intensity measure based on PGV and Sv(T1,5%) was proposed
for the estimation of the damage level of the monolith.
 The use of spectral acceleration to prescribe hazard on a site is only
satisfactory if the design/evaluation of the dam structure is carried out in the
linear range. Spectral acceleration only defines the initiation of cracking for a
particular system: very different performance may be observed for a system
for three motions with identical spectral acceleration response, such as
motions fitted to the same target spectrum. Use of three motions for
determination of nonlinear performance could be highly off the mark.
 In order to quantify the uncertainty in the velocity and duration characteristics
of the ground motion (defining the damage on the system), the use of the
methodology as suggested in FEMA440 appears to be a reasonable option.

76
 The use of the variability in recorded motions to address the variability in the
velocity and duration for a target level motion at a site is more feasible given
that seismic hazard at a site could hardly be presented in the form of velocity
and duration response at this stage.

Finally, the results of the IDA were compared to the static pushover analysis results
in order to evaluate the scope of possible use of such analyses for determining the
performance levels of dams. The pushover analysis with a simple loading
methodology was not able to correctly represent the damage and the load
redistribution on the system as seen in the dynamic analyses, underestimating the
cracking scheme and base shear. Significantly more effort has to be made in
developing an adaptive loading scheme for the pushover methodology to be
applicable to similar systems.

4.2 Future Plans

Some avenues of future study based on the findings in this study are given below.

 In order to successfully simulate the more complex behavior of dams, a new

constitutive model for accurately representing the concrete behavior is
required. This model needs to incorporate a transient damping with respect to
the state of a crack as well as the incremental loss of shear retention in a
crack with respect to the number of opening-closing cycles the crack
undergoes.
 The effect of the vertical component of a ground motion was not considered
in the IDA analyses. The effect of the vertical component and the interaction
between the two-directions of ground motion will be considered in a future
study.

77
 The proposed intensity measure was obtained for a selected dam geometry.
The measure can be generalized by taking the dimensions of the dam
(upstream, downstream slopes and the height) into account.
 A pushover method to more successfully predict the behavior of dams needs
to be developed. The redistribution of inertial and hydrodynamic loading
should be represented in a staged load application based on the damage state
in this procedure.

78
 REFERENCES

Aldemir, A., Binici, B., Canbay, E., Kurç, Ö., & Arıcı, Y. (2013). Seismic
Performance Evaluation of a Concrete Gravity Dam by Using Pseudo Dynamic
Testing. The 2013 World Congress on Advances in Structural Engineering and
Mechanics, Jeju.

Alembagheri, M., & Ghaemian, M. (2012). Seismic Assessment of Concrete Gravity

Dams Using Capacity Estimation and Damage Indexes. Earthquake Engineering and
Structural Dynamics, Vol. 42, pp. 123-144.

Ali, M. H., Alam, M. R., Haque, M. N., & Alam, M. J. (2012). Comparison of
Design and Analysis of Concrete Gravity Dam. Natural Resources, Vol. 3, pp. 18-28.

Araujo, J. M., & Awruch, A. M. (1998). Probabilistic Finite Element Analysis of

Concrete Gravity Dams. Advances in Engineering Software, Vol. 29, pp. 97-104.

Asteris, P. G., & Tzamtzis, A. D. (2003). Nonlinear Seismic Response Analysis of

Realistic Gravity Dam-Reservoir Systems. International Journal of Nonlinear
Sciences and Numerical Simulation, Vol. 4, pp. 329-338.

Basu, U., & Chopra, A. K. (2003). Perfectly Matched Layers for Time-Harmonic
Elastodynamics of Unbounded Domains: Theory and Finite-Element
Implementation. Computer Methods in Applied Mechanics and Engineering, Vol.
192, pp. 1337-1375.

Bhattacharjee, S. S., & Leger, P. (1994). Application of NLFM Models to Predict

Cracking in Concrete Gravity Dams. Journal of Structural Engineering, Vol. 120, pp.
1255-1271.

Bazant, Z. (1986). Mechanics of Distributed Cracking. Applied Mechanics Reviews,

Vol. 39, pp. 675-705.

Binici, B., Arıcı, Y., Kurç, Ö., Canbay, E., Aldemir, A., & Soysal, B. F. (2012).
TÜBİTAK Araştırma Projesi Gelişme Raporu. Ankara.

Carpinteri, A., Valente, S.V., Ferrara, G., & Imperato, L. (1992). Experimental and
Numerical Fracture Modelling of a Gravity Dam. Fracture Mechanics of Concrete
Structures, pp. 351-360.

Chopra, A. K. (2004). Estimating Seismic Demands for Performance-Based

Engineering of Buildings. The 13th World Conference on Earthquake Engineering,
Vancouver.

79
Chopra, A. K. Dynamics of Structures Theory and Applications to Earthquake
Engineering, 3rd Ed., Pearson Prentice Hall, New Jersey, 2007.

Chopra, A. K., & Gupta, S. (1981). Hydrodynamic and Foundation Interaction

Effects in Earthquake Response of a Concrete Gravity Dam. Journal of Structural
Division, ASCE, Vol. 107, pp. 1399-1412.

Chuhan, Z., Jianwen, P., & Jinting W. (2009). Influence of Seismic Input
Mechanisms and Radiation Damping on Arch Dam Response. Soil Dynamics and
Earthquake Engineering, Vol. 29, pp. 1282-1293.

Çalayır, Y., & Karaton, M. (2005). Seismic Fracture Analysis of Concrete Gravity
Dams Including Dam-Reservoir Interaction. Computers and Structures, Vol. 83, pp.
1595-1606.

de Borst, R., Remmers, J. J. C., Needleman, A., & Abellan, M. A. (2004). Discrete
vs Smeared Crack Models for Concrete Fracture: Bridging the Gap. International
Journal for Numerical and Analytical Methods in Geomechanics, Vol. 28, pp. 583-
607.

Devlet Su İşleri Genel Müdürlüğü, http://www.dsi.gov.tr/, last visited on December

2013.

El-Aidi, B., & Hall, J. F. (1989). Non-linear Earthquake Response of Concrete

Gravity Dams. Part 2: Behavior. Earthquake Engineering and Structural Dynamics,
Vol. 18, pp. 853-865.

Fajfar, P. (2000). A Nonlinear Analysis Method for Performance-Based Seismic

Design. Earthquake Spectra, Vol. 16, pp. 573-592.

FEMA. (2005). Improvement of Nonlinear Static Seismic Analysis Procedures.

FEMA 440: Washington, D. C.

FEMA. (1997). NEHRP Guidelines for the Seismic Rehabilitation of Buildings.

FEMA 273: Washington, D. C.

Fenves, G., & Chopra, A. K. (1985). Effects of Reservoir Bottom Absorption and
Dam-Water-Foundation Rock Interaction on Frequency Response Functions for
Concrete Gravity Dams. Earthquake Engineering and Structural Dynamics, Vol. 13,
pp. 13-31.

Freeman, S. A., Nicoletti, J. P., & Tyrell, J. V. (1975). Evaluations of Existing

Buildings for Seismic Risk – A Case Study of Puget Sound Naval Shipyard,
Washington. Proceedings of the U.S. National Conference on Earthquake Engineers,
Berkeley.

80
Ghanaat, Y., & Chudgar, A. K. (2007). Seismic Design and Evaluation of Concrete
Dams-An Engineering Manual.

Hall, J. F., & Chopra, A. K. (1980). Dynamic Response of Embankment, Concrete

Gravity and Arch Dams Including Hydrodynamic Interaction. Earthquake
Engineering Research Center, University of California, Berkeley, UCB/EERC-80/39.

Hall, J. F., & Chopra, A. K. (1982). Hydrodynamic Effects in the Dynamic Response
of Concrete Gravity Dams. Earthquake Engineering and Structural Dynamics, Vol.
10, pp. 333-345.

Hall, J.F. (1988). The Dynamic and Earthquake Behavior of Concrete Dams: Review
of Experimental Behavior and Observational Evidence. Soil Dynamics and
Earthquake Engineering, Vol. 7, pp. 57-121.

Hasan, R., Xu, L., & Grierson, D. E. (2002). Push-over Analysis for Performance-
Based Seismic Design. Computers and Structures, Vol. 80, pp. 2483-2493.

International Commission on Large Dams, http://www.icold-

cigb.org/GB/Dams/role_of_ dams.asp, last visited on December 2013.

Jiang, S., & Du, C. (2012). Seismic Stability Analysis of Concrete Gravity Dams
with Penetrated Cracks. Water Science and Engineering, Vol. 5, pp. 105-119.

Kanenawa, K., Sasaki, T., & Yamaguchi, Y. (2004). Effects of Fracture Material
Properties on Crack Propagation of Concrete Gravity Dams During Large
Earthquakes. The 13th World Conference on Earthquake Engineering, Vancouver.

Kimata, H., Fujita, Y., Niimi, K., Niyamoto, D., Nakayama, K., & Ushida, Y. (2008).
Seismic Safety of Concrete Gravity Dams Based on Dynamic Crack Propagation
Analysis During Large-Scale Earthquakes. The 14th World Conference on
Earthquake Engineering, Beijing.

Leger, P., & Boughoufalah, M. (1989). Earthquake Input Mechanisms for Time-
Domain Analysis of Dam-Foundation Systems. Engineering Structures, Vol. 11, pp.
37-46.

Li, Q., Zhang, F., Zhang, W., & Yang, L. (2002). Fracture and Tension Properties of
Roller Compacted Concrete Cores in Uniaxial Tension. Journal of Materials in Civil
Engineering, Vol. 14, pp. 366-373.

Mansouri, A., Neshaei, M. A. L., & Aghajany, R. (2011). Fracture Analysis of

Concrete Gravity Dam Under Earthquake Induced Loads. Journal of Applied
Sciences and Environmental Management, Vol. 15, pp. 317-325.

81
Mirzabozorg, H., & Ghaemian, M. (2005). Non-linear Behavior of Mass Concrete in
Three-Dimensional Problems Using a Smeared Crack Approach. Earthquake
Engineering and Structural Dynamics, Vol. 34, pp. 247-269.

National Research Council. Earthquake Engineering for Concrete Dams: Design,

Performance, and Research Needs, National Academy Press, Washington D.C.,
1990.

Ngo D., & Scordelis A. C. (1967). Finite Element Analysis of Reinforced Concrete
Beams. Journal of ACI, Vol. 64, pp. 152-163.

Nuss, L. K., Matsumoto, N., & Hansen, K. D. (2012). Shaken, but not Stirred –
Earthquake Performance of Concrete Dams. 32nd Annual USSD Conference, New
Orleans

Pal, W. (1976). Seismic Cracking of Concrete Gravity Dam. Journal of Structural

Division, Vol. 102, pp. 1827-1844.

Pekau, O. A., & Zhu, X. (2006). Seismic Behavior of Cracked Gravity Dams.
Earthquake Engineering and Structural Dynamics, Vol. 35, pp. 477-495.

Rashid, Y. R. (1968). Analysis of Prestressed Concrete Pressure Vessels. Nuclear

Engineering and Design, Vol. 7, pp. 334-344.

Rots, J. G. Computational Modeling of Concrete Fracture. Delft University of

Technology, Delft, 1988.

Rots J. G., & de Borst, R. (1988). Analysis of Mixed-Mode Fracture in Concrete.

Journal of Engineering Mechanics, Vol. 113, pp. 1739-1758.

Tinawi, R., Leger, P., Leclerc, M., & Cipolla, G. (2000). Seismic Safety of Gravity
Dams: From Shake Table Experiments to Numerical Analysis. Journal of Structural
Engineering, Vol. 126, pp. 518-529.

TNO DIANA. User’s Manual, R. 9.4.3, 2010.

US Army Corps of Engineers. (1995). Seismic Design Provisions for Roller

Compacted Concrete Dams. Washington, D. C: Engineering and Design, EP 110-2-
12.

US Army Corps of Engineers. (2003). Time-History Dynamic Analysis of Concrete

Hydraulic Structures. Washington, D. C: Engineering and Design, EM 1110-2-6051.

Vamvatiskos, D., & Cornell, C. A. (2002) Incremental Dynamic Analysis.

Earthquake Engineering and Structural Dynamics, Vol. 31, pp. 491-514.

82
Westergaard, H. M. (1933). Water Pressures on Dams during Earthquakes. American
Society of Civil Engineers, Vol. 98, pp. 418-433.

Yamaguchi, Y., Sasaki, T., Iwashita, T., Sasaki, S., & Kurahashi, H. (n.d.).
Evaluation of Tensile Cracks in Concrete Gravity Dams by Shaking Table Tests and
Non-linear FEM Analyses with Smeared Crack Model.

Yanmaz, A. M., Applied Water Resources Engineering, 3rd Ed., METU Press,
Ankara, 2006.

Yılmaztürk, M., Three Dimensional Dynamic Response of a Concrete Gravity Dam,

Middle East Technical University, Ankara, 2013.

83