You are on page 1of 426

12

th
International
BENCHMARK WORKSHOP
ON NUMERICAL ANALYSIS OF DAMS, 2
nd
- 4
th
October, 2013, Graz AUSTRIA
Edited by
Gerald Zenz and Markus Goldgruber
Published by the Austrian National Committee on Large Dams
with the support of the Austrian Reservoir Commission (Staubeckenkommission)
I C O L D
proceedings
Austrian National Committee on Large Dams
Stremayrgasse 10/II, A-8010 Graz, AUSTRIA, Phone: ++43/316/8861, secretary@atcold.at, www.atcold.at
H
e
f
t

/

V
o
l
u
m
e

3
9







p
r
o
c
e
e
d
i
n
g
s

I
C
O
L
D

1
2
t
h

b
e
n
c
h
m
a
r
k

w
o
r
k
s
h
o
p

PROCEEDINGS OF THE
ICOLD - 12
th
INTERNATIONAL
BENCHMARK WORKSHOP
ON NUMERICAL ANALYSIS OF DAMS

2
nd
- 4
th
OCTOBER, 2013, GRAZ AUSTRIA







Edited by
Gerald Zenz
Institute of Hydraulic Engineering and Water Resources Management
Graz University of Technology
Markus Goldgruber
Institute of Hydraulic Engineering and Water Resources Management
Graz University of Technology





































Published by

ATCOLD AUSTRIAN NATIONAL COMMITTEE ON LARGE DAMS

2014 All rights reserved

The authors are responsible for the content of their contributions. The texts of the various
papers in this volume were set individually by typists under the supervision of each of the
authors concerned.

Cover pictures copyright by
Graz University of Technology

Layout, Design, Cover Artwork by
Institute of Hydraulic Engineering and Water Resources Management

Printed by Graz University of Technology

Preface

For the International Commission on Large Dams (ICOLD) the Committee on
Computational Aspects of Analysis and Design of Dams is the general Organizer of
Benchmark Workshops. This 12
th
Benchmark Workshop is held in the city of Graz, situated in
the South of the Alps on both sides of the river Mur. Six universities with over 40,000
students are in addition responsible for the young and dynamic charm of our city. Since over
200 years Graz University of Technology is one of them.

Advanced numerical tools with user friendly interfaces are available and widely used for
structural analyses. Such numerical analyses require a solid theoretical background of the
applicability of methods to be used. On the other hand, the results gained need a careful
interpretation with respect to the underlying assumptions and their practical relevance.
ICOLD Benchmark examples of generalized engineering problems are devoted to bridge the
gap between numerical analyses, the interpretation of results and their theoretical as well as
practical relevance. Since 1991 eleven benchmark workshops were organized for different
numerical problems in the field of concrete and fill dams under static and dynamic loading
conditions. The results of these benchmark workshops are made available to the dam
engineering community on the internet and in proceeding. Results are published in ICOLD
bulletins:

Bulletin N 94 Computer Software for Dams
Bulletin N 122 Computational Procedures for Dam Engineering
Bulletin N 155 Guidelines for Use of Numerical Models in Dam Engineering

This 12
th
Benchmark Workshop provides an excellent opportunity for engineers, scientists and
operators to present and exchange their experiences and the latest developments related to the
design, performance and monitoring of dams. There are three example topics and an open
theme formulated and discussed:

Theme A: Fluid Structure Interaction, Arch Dam - Reservoir at Seismic Loading
Formulators: Gerald Zenz, Markus Goldgruber

Theme B: Long Term Behavior of Rockfill Dams
Formulators: Camilo Marulanda, Joan Manuel Larrahondo

Theme C: Computational Challenges in Consequence Estimation for Risk Assessment
Formulators: Yazmin Seda-Sanabria, Enrique E. Matheu, Timothy N. McPherson

The support to this Benchmark Workshop from the Members of the Committee on
Computational Aspects of Analysis and Design of Dams and especially from the Formulators
is gratefully acknowledged. The contributions from the Core Organizing Team Markus
Goldgruber and Harald Breinhlter are very much appreciated. Finally, I want to thank the
participants for their scientific contributions herein.


Gerald Zenz
Chairman



5
Table of Content

Conference Organizations 9

Acknowledgements 11

Overview of Contributions
Theme A 13
Fluid Structure Interaction: Arch Dam Reservoir at Seismic Loading
- Benchmark Problem Description 15
- Result Comparison of the Participants 27
- Papers 65
G. Maltidis and L. Stempniewski 67
W. Kikstra, F. Sirumbal and G. Schreppers 77
G. Faggiani and P. Masarati 87
A. Tzenkov, A. Abati and G. Gatto 99
M. Chambart, T. Menouillard, N. Richart, J.-F. Molinari and R. M. Gunn 111
A. Popovici, C. Ilinca and R. Vrvorea 123
R. Malm, C. Pi Rito, M. Hassanzadeh, C. Rydell and T. Gasch 139
M. Brusin, J. Brommundt and H. Stahl 149
S. Shahriari 165
A. Frigerio and G. Mazz 167
A. Diallo and E. Robbe 177

Theme B 189
Long Term Behavior of Rockfill Dams
- Benchmark Problem Description 191
- Papers 201
F. Ezbakhe and I. Escuder-Bueno 203


7
Table of Content

Theme C 211
Computational Challenges in Consequence Estimation for Risk Assessment
- Benchmark Problem Description 213
- Result Comparison of the Participants 227
- Papers 239
M. Davison, M. Hassan, O. Gimeno, M. van Damme and C. Goff 241
M.S. Altinakar, M.Z. McGrath, V.P. Ramalingam, D. Shen, Y. Seda Sanabria
and E.E. Matheu 255
L. Mancusi, L. Giosa, A. Cantisani, A. Sole and R. Albano 271
O. Saberi, C. Dorfmann and G.Zenz 283
D. McVan, J. Ellis, G. Savant and M. Jourdan 293
B. A. Thames and A. J. Kalyanapu 309
D. Williams and K. Buchanan 325

Open Theme 339
Choice of Contributor
- Papers 339
Behavior of an arch dam under the influence of creep, AAR and opening of
the dam/foundation contact
E. Robbe 341
Need for transient thermal models, with daily inputs, to explain the
displacements of arch gravity dams
I. Escuder, D. Galn and A. Serrano 349
The rehabilitation of Beauregard Dam: the contribution of the numerical
modeling A. Frigerio and G. Mazz 359
Earthquake Assessment of Slab and Buttress Dams
H. B. Smith and L. Lia 369
Solution of dam-fluid interaction using ADAD-IZIIS software
V. Mircevska, M. Garevski, I. Bulajic and S. Schnabl 379
Influence of Surface Roughness on Sliding Stability Tests and numerical
modeling
. Eltervaag, G. Sas and L. Lia 389
Seismic Analysis of a Concrete Arch Dam
P. Dakoulas 399
Earthquake safety assessment of arch dams based on nonlinear dynamic
analyses
S. Malla 411
9
Conference Organizations

The conference was organized by
Graz University of Technology, Graz, Austria
Conference Chairman
Gerald Zenz (Graz University of Technology, Austria)
Committee on Computational Aspects of Analysis and Design of Dams
Chairman
I. EscuderBueno (Spain)
VizeChairman
G. Mazza (Italy)
Technical Advisory Team
RESTELLI, F. (Argentina) MEGHELLA, M. (Italy)
ZENZ, G. (Austria) UCHITA, Y. (Japan)
CURTIS, D. (Canada) ANDERSEN, R. (Norway)
CHEN, S. (China) POPOVICI, A. (Romania)
MARULANDA, C. (Colombia) GLAGOVSKY, V. (Russia)
VARPASUO, P. (Finland) MINARIK, M. (Slovakia)
TANCEV, L. (Form. Yug. Rep. of Macedonia) HASSANZADEH, M. (Sweden)
FROSSARD, E. (France) GUNN, R. (Switzerland)
BEETZ, U. (Germany) MATHEU, E. (United States)
DAKOULAS, P. (Greece) CARRERE, A. (Honorary Member)
(France)
NOORZAD, A. (Iran) FANELLI, M. (Honorary Member)
(Italy)

Local Organizing Committee
Graz University of Technology
Institute of Hydraulic Engineering and Water Resources Management
Head
Gerald Zenz
Organizer
Markus Goldgruber


11
Acknowledgements

The editors are grateful to the members of the following organizations for their support:

ATCOLD

ICOLD

Benchmark Workshop Problem Formulation Teams

International Water Power & Dam Construction



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
13



THEME A



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
15



Theme A
Fluid Structure Interaction
Arch Dam Reservoir at Seismic loading



Formulators:
Dipl. Ing. Markus Goldgruber
Univ. Prof. Dipl. Ing. Dr. techn. Gerald Zenz

Institute of Hydraulic Engineering and Water Resources Management
Graz University of Technology



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
16
Introduction
Advanced numerical tools with user friendly interfaces are available for structural analyses.
Such numerical analyses require a solid theoretical background of the applicability of methods
to be used. On the other hand, the results gained need a careful interpretation with respect to
the underlying assumptions and their practical relevance. ICOLD Benchmark examples of
generalized engineering problems are devoted to bridge the gap between numerical analyses,
the interpretation of results and their theoretical as well as practical relevance. Challenges of
the analyses of concrete dams are always the definition of material parameters, the spatial
discretization and the appropriate simulation of loading sequences. Additionally, specific
attention is paid on the structural integrity and entire safety under seismic loading conditions.
To account for this problem, the interaction of the dam and the reservoir is topic of this theme
A.
By means of the Finite Element Method linear and nonlinear analyses under dynamic
excitation are carried out. However, for the required and appropriate simulation of the dam
reservoir interaction different approaches are used. With respect to future nonlinear dynamic
analyses, these simulations herein shall be in the time domain only.
A common approach to take the dynamic water interaction into account is to use an added
mass approach. A more sophisticated possibility is the use of Acoustic or Fluid Elements. The
simulations of earthquake excitation of arch dams have shown that the analyzed stresses in the
structure could vary significantly based on the interaction modeling. The added mass
approach is still a widely used technique but tends to overestimate the stresses and therefore it
is conservative in contrary to other techniques.
This benchmark now intends to compare different modeling techniques and will show the
amount of deviations. All investigations are carried out for an artificially generated symmetric
arch dam and simplified loading and boundary conditions.
Universities, engineering companies and regulatory bodies are invited to contribute to the
benchmark and take part in the discussion of results gained.
Focus of this benchmark example
The focus of this benchmark is to carry out the Dynamic Fluid Structure Interaction for a large
arch dam. Every participant may choose his own order of details in modeling.
The main goal of this example is the application of different approaches like:
- Added mass technique (Westergaard, Zangar,)
- Acoustic Elements (compressible, incompressible)
- Fluid Elements (compressible, incompressible)
Further on, the usage of different Boundary Conditions is possible for:
- Reservoir - Foundation
o Reflecting (on the bottom and the sides)
o Non-reflecting (at the end of the reservoir)
The modeling of the block joint opening due to tensile stresses and nonlinear effects - is not
focus of this benchmark example. However, to carry out this analysis in the time domain will
provide the opportunity for further non-linear analyses.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
17
General basic assumptions
The following general basic assumptions and boundary conditions for the investigations
should be used:
- Same spatial discretization (Model/Mesh) of the Structure, Foundation and Reservoir
- Same Material Parameters
- Acceleration-Time-History in X-,Y-,Z-Direction
- Reservoir is infinite in length (non-reflecting)
- Rayleigh Damping
- Results to be compared Visualization

Based on these basic assumptions and results gained the contributors are encouraged to
intensify and focus their effort to achieve results with higher profound physical justification
and explain the differences. (E.g.: different spatial discretization, more appropriate modeling
of the interaction; different length of the reservoir; need for nonlinear effects).
An interpretation of the evaluated results from an engineering point of view should be given.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
18
Model and Geometry
An Arch Dam, Foundation and Reservoir Model layout for the benchmark has been generated
and is available for downloading.
Arch Dam Model
- Symmetric Geometry
- Total Height: 220 Meters
- Valley width (crest): ~ 430 Meters
- Valley width (bottom): ~ 80 Meters
Arch Dam Geometry
The Arch Dam Geometry has been generated with the Program Arch Dam Design, which
was developed as part of the Master-Thesis by DI Manuel Pagitsch.


Arch Dam Model

Plan View

View from the upstream

Main Section
Figure 1: Arch dam geometry
Foundation Model
Symmetry is used for the foundation too.
- Height: 500 Meters
- Length: 1000 Meters
- Width 1000 Meters
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
19



Figure 2: Foundation geometry
Reservoir Model
- Length: assumed minimum of 460 Meters (> 2x Height of the Dam)
- Modeling the interaction with Acoustic- or Fluid Elements

Figure 3: Reservoir geometry
500m
1000m
1000m
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
20
Acceleration Time History
- Transient Acceleration (a
max
0.1g)
- X-,Y-,Z- Direction
- Artificially generated time history

Figure 4: Reservoir geometry
Material Parameters
The Material properties are defined for isotropic and homogenous conditions.

Rock mass
- Density: 0 kg/m
3

- Poisson - ratio: 0,2
- Youngs - modulus: 25000 MPa

Water
- Density: 1000 kg/m
3

- Bulk - modulus: 2200 MPa

Dam
- Density: 2400 kg/m
3

- Poisson - ratio: 0,167
- Youngs - modulus: 27000 MPa

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
21
Mesh Properties
Two different Meshes of the entire system are provided for investigations, as these are a
coarse and a fine mesh. If desired, the parts can also be provided as ABAQUS/CAE Model
File, ACIS- or IGES-Files, if a specific mesh is intended to be discretized.
Coarse Mesh
Arch Dam
- Total number of nodes: 2083
- Total number of elements: 356
- 312 quadratic hexahedral elements of type C3D20R (ABAQUS CAE)
- 44 quadratic wedge elements of type C3D15 (ABAQUS CAE)

Foundation
- Total number of nodes: 11608
- Total number of elements: 2340
- quadratic hexahedral elements of type C3D20R (ABAQUS CAE)

Reservoir
- Total number of nodes: 12493
- Total number of elements: 2640
- quadratic hexahedral elements of type C3D20R (ABAQUS CAE)



Figure 5: Coarse mesh of the dam, foundation and reservoir
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
22
Fine Mesh

Arch Dam
- Total number of nodes: 13733
- Total number of elements: 2736
- quadratic hexahedral elements of type C3D20R (ABAQUS CAE)

Foundation
- Total number of nodes: 13298
- Total number of elements: 2700
- quadratic hexahedral elements of type C3D20R (ABAQUS CAE)

Reservoir
- Total number of nodes: 12493
- Total number of elements: 2640
- quadratic hexahedral elements of type C3D20R (ABAQUS CAE)



Figure 6: Fine mesh of the dam, foundation and reservoir
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
23
Elements and Node Numbering in ABAQUS/CAE
The provided input-files are containing a list of the nodes and elements of the mesh and also
predefined node sets for the different sections which should be investigated. The Node
numbering of ABAQUS/CAE is plotted in the following figures.

Figure 7: Node numbering of wedge and brick elements in ABAQUS/CAE

These figures are showing the node numbering for the two different element types which are
used in the provided input-files.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
24
Loading
The following loading sequence is intended to be used.
- Gravity
- Hydrostatic Water Pressure (full supply water level = Crest Height)
- Seismic Loading
o Modal Superposition or
o Direct time integration (Implicit/Explicit)
Results
Following results should be evaluated and plotted.
Eigenfrequencies (1 10)
The evaluation of the first 10 Eigenfrequencies of the structure, including the interaction with
the reservoir, should be provided.
Mode Shapes (1 10)
The evaluation and plotting of the first 10 Mode-Shapes of the structure, including the
interaction with the reservoir, should be provided.
Hoop Stresses, Vertical Stresses and Min./Max. Principal Stresses
Evaluation of the different stresses should be done for
- Static Loads
- Seismic Loads (Min., Max.)
- 3 different sections (Main Section and ~45 degrees on the left and right hand side)

Figure 8: Evaluation sections of the arch dam


Left Section
Main Section
Right Section
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
25



Figure 9: Evaluation examples for the stresses
Radial Deformation
Evaluation of the Radial Deformation should be done for
- Static Loads
- Seismic Loads (Min., Max.)
- Main section

Figure 10: Evaluation examples for the radial deformation



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
27



RESULT COMPARISON OF THE
PARTICIPANTS

THEME A



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
29
Results Comparison

Theme A
Fluid Structure Interaction
Arch Dam Reservoir at Seismic loading

Participants, Programs and Approaches
Overall there are 11 participants from 9 different countries (Swiss, Netherlands, France,
Germany, Sweden, Italy, Iran, Romania and Austria) who were contributing to the workshop
and decided to solve the problem. Each participant had the opportunity to choose his preferred
numerical program and modelling technique to account for the fluid structure interaction. The
prevailing boundary conditions, which are the same for each participant, are defined in the
section Benchmark Problem Description Theme A.

The Reference Solution (REF) in the diagrams and tables doesnt claim to be the optimum
solution. It shows the results of the simulations done at the Institute of Hydraulic Engineering
and Water Resources Management by Markus Goldgruber.

The following table lists the participants and their used programs and approaches. The
informations in the last column should point out some specific differences between the
participants which may influence the results and are worth mentioning. Some of the
participants have provided results of more than just one simulation, but for the comparison
just one of these has been used. All the other results of the approaches and models can be
found in their papers in the following section (Papers Theme A)

All participants had to evaluate eigenfrequencies, mode shapes, deformations and stresses. In
the case of dynamic simulations one will get minimum and maximum values. Therefore, in
the diagrams in the results section every participant has three lines, the minimum (left line)
and the maximum line (right line) which indicate the minimum and maximum values out of
the time history records and the line for static loading (middle line). This middle line indicates
the static value out of the sum of the two load cases, gravity and hydrostatic water load. To
retain the overview in the diagrams, the minimum, maximum and static values are not
explicitly mentioned in the legend. The diagrams for the deformations and the upstream hoop
stresses in the main section were plotted additionally for static and dynamic loading
separately.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
30
Table 1: Participants, Programs and Approaches


Finite Element
Program
Method Mesh
Some Additional
Informations
A G. Maltidis ABAQUS
Acoustic
Elements
Coarse
7.5% critical
damping
B
W.Kikstra,
F. Sirumbal,
G. Schreppers
DIANA
Acoustic
Elements
Coarse
Compressible Fluid,
Hybrid Frequency-
Time Domain
(HFTD) method
C
G. Faggiani,
P. Masarati
CANT-SD
Acoustic
Elements
Coarse

D
A. Tzenkov,
A. Abiati,
G. Gatto
DIANA
Acoustic
Elements
Coarse
Same as participant
B, Construction
steps for loadcase
deadweight
E M. Chambart DIANA
Added mass
(Westergaard)
Fine E
dyn
= E
sta
* 1.25
F
A. Popovici,
C. Ilinca,
R. Vrvorea
ABAQUS
Added mass
(Westergaard)
Coarse

G
R. Malm,
C. Pi Rito,
C. Hassanzadeh,
C. Rydell,
T. Gasch
ABAQUS
Acoustic
Elements
Fine
Infinite Elements at
the boundaries,
Acceleration-Time-
History applied on
the bottom of the
model
H M. Brusin
FENAS
ECCON IPP
Added mass
(Westergaard)
Fine
Construction steps
for loadcase
deadweight
I S. Shahriari ANSYS
Added mass
(Westergaard)
Coarse
Use of the full
Westergaard
formula
(Period/Frequency
dependent)
J
A. Frigerio,
G. Mazz
COMSOL
Acoustic
Elements
Coarse

K
A.Diallo,
E. Robbe
CODE_ASTER
Incompressible
Finite Element
added mass
Coarse
Method to calculate
the added mass
matrices
representing the
fluid-structure
interaction with a
potential approach
REF M. Goldgruber ABAQUS
Acoustic
Elements
Coarse


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
31
Results
Eigenfrequencies


Figure 1: Eigenfrequencies 1 10(Column Chart)

Table 2: Eigenfrequencies 1 10 (Table)
Participant
Mode
1 2 3 4 5 6 7 8 9 10
A 1.47 1.54 1.55 2.11 2.33 2.46 2.61 2.97 3.25 3.37
B 1.57 1.60 2.36 2.94 3.04 3.72 3.88 4.56 4.78 4.80
C 1.54 1.55 2.05 2.22 2.41 2.83 2.98 3.37 3.40 3.79
D 1.57 1.62 2.36 2.94 3.04 3.72 3.87 4.56 4.76 4.80
E 1.43 1.47 2.21 2.61 2.81 3.27 3.56 4.09 4.37 4.37
F 1.54 1.56 1.93 2.30 2.48 3.04 3.12 3.29 3.61 3.71
G 1.51 1.54 1.90 2.22 2.42 2.96 3.01 3.28 3.59 3.76
H 1.26 1.32 2.01 2.36 2.50 3.00 3.17 3.65 3.70 3.88
I 1.28 1.33 1.91 2.37 2.38 2.91 2.98 3.61 3.62 3.85
J 1.54 1.55 2.09 2.22 2.33 2.51 2.83 2.96 3.19 3.37
K 1.57 1.62 2.35 2.95 3.03 3.72 3.85 4.56 4.88 5.13
REF 1.54 1.54 2.05 2.29 2.54 2.96 3.21 3.36 3.76 3.91


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
32
Mode Shapes

Figure 2: Mode Shapes 1 10; Participants A E
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
33

Figure 3: Mode Shapes 1 10; Participants F K

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
34
Deformations Main Section




Figure 4: Deformation Main Section (A E)


Figure 5: Deformation Main Section (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-0.06 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-0.06 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
35
Deformations (Only Static Load) Main Section




Figure 6: Deformation (Only Static Load) Main Section (A E)


Figure 7: Deformation (Only Static Load) Main Section (F K)

0
20
40
60
80
100
120
140
160
180
200
220
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
36
Deformations (Only Dynamic Load) Main Section




Figure 8: Deformation (Only Dynamic Load) Main Section (A E)


Figure 9: Deformation (Only Dynamic Load) Main Section (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-0.15 -0.13 -0.11 -0.09 -0.07 -0.05 -0.03 -0.01 0.01 0.03 0.05 0.07 0.09 0.11
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-0.15 -0.13 -0.11 -0.09 -0.07 -0.05 -0.03 -0.01 0.01 0.03 0.05 0.07 0.09 0.11
H
E
I
G
H
T

(
m
)

DEFORMATION (m)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
37
Hoop Stresses Main Section Upstream




Figure 10: Hoop Stresses Main Section Upstream (A E)


Figure 11: Hoop Stresses Main Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
38
Hoop Stresses (Only Static Load) Main Section Upstream




Figure 12: Hoop Stresses (Only Static Load) Main Section Upstream (A E)


Figure 13: Hoop Stresses (Only Static Load) Main Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-8.00 -7.00 -6.00 -5.00 -4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-8.00 -7.00 -6.00 -5.00 -4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
39
Hoop Stresses (Only Dynamic Load) Main Section Upstream




Figure 14: Hoop Stresses (Only Dynamic Load) Main Section Upstream (A E)


Figure 15: Hoop Stresses (Only Dynamic Load) Main Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 8.00 10.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 8.00 10.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
40
Hoop Stresses Main Section Downstream




Figure 16: Hoop Stresses Main Section Downstream (A E)


Figure 17: Hoop Stresses Main Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -7.50 -5.00 -2.50 0.00 2.50 5.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -7.50 -5.00 -2.50 0.00 2.50 5.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
41
Hoop Stresses Left Section Upstream




Figure 18: Hoop Stresses Left Section Upstream (A E)


Figure 19: Hoop Stresses Left Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
42
Hoop Stresses Left Section Downstream




Figure 20: Hoop Stresses Left Section Downstream (A E)


Figure 21: Hoop Stresses Left Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -9.50 -7.00 -4.50 -2.00 0.50 3.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -9.50 -7.00 -4.50 -2.00 0.50 3.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
43
Hoop Stresses Right Section Upstream




Figure 22: Hoop Stresses Right Section Upstream (A E)


Figure 23: Hoop Stresses Right Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-9.00 -7.00 -5.00 -3.00 -1.00 1.00 3.00 5.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-9.00 -7.00 -5.00 -3.00 -1.00 1.00 3.00 5.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
44
Hoop Stresses Right Section Downstream




Figure 24: Hoop Stresses Right Section Downstream (A E)


Figure 25: Hoop Stresses Right Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-9.00 -7.00 -5.00 -3.00 -1.00 1.00 3.00 5.00
H
E
I
G
H
T

(
m
)

HOOP STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
45
Vertical Stresses Main Section Upstream




Figure 26: Vertical Stresses Main Section Upstream (A E)


Figure 27: Vertical Stresses Main Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 8.00 10.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 8.00 10.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
46
Vertical Stresses Main Section Downstream




Figure 28: Vertical Stresses Main Section Downstream (A E)


Figure 29: Vertical Stresses Main Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
47
Vertical Stresses Left Section Upstream




Figure 30: Vertical Stresses Left Section Upstream (A E)


Figure 31: Vertical Stresses Left Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-4.00 -2.00 0.00 2.00 4.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-4.00 -2.00 0.00 2.00 4.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
48
Vertical Stresses Left Section Downstream




Figure 32: Vertical Stresses Left Section Downstream (A E)


Figure 33: Vertical Stresses Left Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
49
Vertical Stresses Right Section Upstream




Figure 34: Vertical Stresses Right Section Upstream (A E)


Figure 35: Vertical Stresses Right Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-4.00 -2.00 0.00 2.00 4.00 6.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
50
Vertical Stresses Right Section Downstream




Figure 36: Vertical Stresses Right Section Downstream (A E)


Figure 37: Vertical Stresses Right Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
A B C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

VERTICAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
51
Minimum Principal Stresses Main Section Upstream




Figure 38: Minimum Principal Stresses Main Section Upstream (A E)


Figure 39: Minimum Principal Stresses Main Section Upstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
52
Minimum Principal Stresses Main Section Downstream




Figure 40: Minimum Principal Stresses Main Section Downstream (A E)


Figure 41: Minimum Principal Stresses Main Section Downstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
53
Minimum Principal Stresses Left Section Upstream




Figure 42: Minimum Principal Stresses Left Section Upstream (A E)


Figure 43: Minimum Principal Stresses Left Section Upstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
54
Minimum Principal Stresses Left Section Downstream




Figure 44: Minimum Principal Stresses Left Section Downstream (A E)


Figure 45: Minimum Principal Stresses Left Section Downstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-20.00 -18.00 -16.00 -14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-20.00 -18.00 -16.00 -14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
55
Minimum Principal Stresses Right Section Upstream




Figure 46: Minimum Principal Stresses Right Section Upstream (A E)


Figure 47: Minimum Principal Stresses Right Section Upstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
56
Minimum Principal Stresses Right Section Downstream




Figure 48: Minimum Principal Stresses Right Section Downstream (A E)


Figure 49: Minimum Principal Stresses Right Section Downstream (F K)
0
20
40
60
80
100
120
140
160
180
200
220
-20.00 -18.00 -16.00 -14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-20.00 -18.00 -16.00 -14.00 -12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
E
I
G
H
T

(
m
)

MIN. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
57
Maximum Principal Stresses Main Section Upstream




Figure 50: Maximum Principal Stresses Main Section Upstream (A E)


Figure 51: Maximum Principal Stresses Main Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00 9.00 11.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00 9.00 11.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
58
Maximum Principal Stresses Main Section Downstream




Figure 52: Maximum Principal Stresses Main Section Downstream (A E)


Figure 53: Maximum Principal Stresses Main Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
59
Maximum Principal Stresses Left Section Upstream




Figure 54: Maximum Principal Stresses Left Section Upstream (A E)


Figure 55: Maximum Principal Stresses Left Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
60
Maximum Principal Stresses Left Section Downstream




Figure 56: Maximum Principal Stresses Left Section Downstream (A E)


Figure 57: Maximum Principal Stresses Left Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
61
Maximum Principal Stresses Right Section Upstream




Figure 58: Maximum Principal Stresses Right Section Upstream (A E)


Figure 59: Maximum Principal Stresses Right Section Upstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-3.00 -1.00 1.00 3.00 5.00 7.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
62
Maximum Principal Stresses Right Section Downstream




Figure 60: Maximum Principal Stresses Right Section Downstream (A E)


Figure 61: Maximum Principal Stresses Right Section Downstream (F K)

0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
A C D E REF
0
20
40
60
80
100
120
140
160
180
200
220
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00
H
E
I
G
H
T

(
m
)

MAX. PRINCIPAL STRESS (MPa)
F G H I J K REF
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
63
Discussion of Results
Eigenfrequencies
It was to calculate the first 10 Eigenfrequencies of the structural response, including the
interaction with the reservoir under full reservoir conditions. The results are summarized in
Figure 1 to Figure 3 and Table 2.
The criteria to compare the results are only the Eigenfrequency no shape deformation, nor
mass contribution to the different modes and directions are considered. For commenting in
detail, this information would be valuable.
The lower frequencies compared to the others of the first two modes of participants H and I
are due to the use of the added mass approach. The overestimation of the additionally excited
mass leads to slightly lower Eigenfrequencies, while the remaining are around 1.5 Hz.
Noticeable is that the participants who used the program Diana (B, D and E) are getting higher
frequencies starting from the third. This counts also for the participant K, who used the Open
Source software Code_Aster. All other participants are getting more or less the same
Eigenfrequencies for the first 10 modes.
Deformation
The comparison of the deformations is done for the main (middle) section only. The static
loading accounts for dead weight and water loading together. No temperature loading is
accounted for.
The static deformation reveals, that many results show almost the same behavior, except those
from participants D, E, H and K. The higher static deformation of participants D, H and K are
due to modelling of the construction stages.
The dynamic deformations are, as expected, varying in a wider range. Especially worth
mentioning is also the result by participant G, he, as the only one, used infinite elements on
the vertical boundary and applied the acceleration-time-history record on the bottom of the
model, which could be the reason for in general higher values. The lower values of radial
deformation of participant E are a result of the higher youngs modulus used.
Stresses
The comparison of the stresses of each of the participant and each of the diagrams are focused
on essential aspects. Therefore, the discussion is kept general and just the quality of some
graphs and values is discussed, but not the quantity in detail. Every participant has used his
own preferred program, modelling technique and approach, so its to await that different
results are gained.
As it was up to the participant to use immediate or stepped construction sequences, the stress
distribution differs. A 0.5[MPa] difference for the static loading, at a stress level of 6[MPa],
one might accept, but not larger (Figure 12 and Figure 13).
Worth mentioning are the results by participant G. He used, as already mentioned in the
discussion of the deformations, infinite elements and applied the acceleration time history just
on the bottom of the model. So the same as for the deformation counts here, the stresses in
contrast to the others are far the highest in almost each diagram and beyond awaited results.
Participants B and D used the program Diana with the Hybrid Frequency-Time Domain
Method (HFTD-Method), which takes frequency dependent properties, such as
compressibility of fluid, reservoir-bottom absorption and far-field reflection, into account.
Both of them got similar results compared to the others, which prove the usability of this
sophisticated analysis method on the one hand but shows the applicability of less elaborated
models, under these assumptions, too.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
64
Participant K was the only one who used the Open Source software Code_Aster. Such
software, which is mostly used at research facilities and universities, is license free, but often
more difficult to apply than commercial ones. Nevertheless, the provided results are matching
with the results from the other participants.
It is worth mentioning, that the added mass approach (according to Westergaard) is able to
provide, comparable results, under the circumstances of this benchmark. All of the three
contributors (E, H and I) using this approach are in the range of expected results.
Conclusion
The comparison of all participants shows, that despite of the same boundary conditions
(Model of the Geometry, Finite Element Mesh, Load cases, Material properties, Linear
analysis, etc.) still assumptions are taken and required to carry out the analyses. These
additional assumptions are starting with the application of the construction sequence, increase
of material properties for dynamic loading, abutment boundary properties, application of
dynamic loading and some specific assumptions based on the program used.
Best practice examples and recommendations are published in ICOLD Bulletins and are
available for engineers. However, for any specific problem to be solved, the assumptions for
an analysis applied need to be reconfirmed in the light of the entire problem. In general it is
astonishing, to see the large differences between the results of individual.

Everybody had the opportunity to choose his preferred modelling technique to account for the
fluid structure interaction, but most of the contributors used either added mass technique or
acoustic elements. In practice it is still common to use an added mass approach according to
Westergaard. Normally this assumption yields conservative results in contrary to modelling
with acoustic elements. The solution of participant e.g. I, who has used Westergaards formula
with its fully, frequency dependent extension, shows very similar results to those analyzed
using higher constitutive models.
According to results of the participants using either the coarse or fine mesh (described in the
former section) had just a marginal influence on the frequencies, deformations and stresses
within the structure.
The purpose of choosing this arch dam example (220m in height, totally symmetric) wasnt
just for evaluating the influence of different modelling techniques, but also for engineers,
scientists and operators to have a kind of reference solution. The diagrams and tables of the
results of all participants should help to quantify and compare frequencies, deformations and
stresses of such a structure.

Concluding, everybody should be aware of the fact, that results of such simulations should be
treated critically, because mistakes in modelling and application cannot be excluded. Usually,
reference solutions from former comparable projects for validation should be used to proof the
results for plausibility.




ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
65



PAPERS

THEME A



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
67
Fluid Structure Interaction
Arch Dam Reservoir at Seismic Loading
G. Maltidis
1
and L. Stempniewski
2

1
Federal Waterways Engineering and Research Institute, Kumaulstrasse 17, 76187
Karlsruhe, GERMANY
2
Karlsruhe Institute for Technology, Institute of Concrete Structures and Building Materials,
Department of Concrete Structures, Gotthard-Franz-Str. 3, 76131, Karlsruhe, GERMANY
E-mail: georgios.maltidis@baw.de
Abstract
The fluid structure interaction is an important issue that must be taken into account for the
analysis and design of hydraulic structures. Since the first attempts to calculate the
hydrodynamic pressures on structures analytically (Westergaard, von Krmn, Mononobe,
Housner, Chwang, Zangar) the engineers and researchers have the last years a very useful
tool, the finite and boundary element method, in order to analyze complicated structures
taking into account different sophisticated phenomena. However, even nowadays, the
common praxis is to use the early developed techniques, because of their simplicity and
capability of implementation in the most finite element codes.
Introduction
Since 1933, the hydrodynamic pressures on oscillating structures, which are in contact with
water, are taken into account with the simplified assumption that the water is incompressible
and the structure is star using the added mass approaches, first proposed by Westergaard for
vertical star surfaces and later extended by Zangar for inclined surfaces. Although these
approaches apply under conditions which hardly are met, they are widely used also nowadays
because of their simplicity in incorporating them in finite element codes. However, the result
of analysis with the added mass approach may come out to be very conservative leading to
wrong decisions. The modeling of the water with finite solid element around the 1980s gave
the opportunity for the analyst to take account some phenomena, as the water compressibility
but raised other numerical problems as such type of modeling of water is suffering many
times of hourglass making the analysis instable. The use of acoustic elements seems to be the
more beneficial, as there are hardly numerical problems, and most of the phenomena, which
take place for a dynamic fluid structure interaction can be modeled. With acoustic elements
the analyst can consider the water compressibility, the wave absorption at the infinite end of
the reservoir and the impedance of wave radiation at the reservoir sediments.
Hydrodynamic Pressures
Added mass approaches
The most well-known added mass approach is the one of Westergaard (1933)[1]. Westergaard
proposed the following formula for the computation of hydrodynamic pressures as added
masses under the restrictions that the reservoir is infinite, the upstream surface of the dam is
vertical and the dam is rigid:
7
8
w

m H y A
g
=
(1)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
68
where m the added mass, H and y the height and the depth of the reservoir respectevily,
w
the
density of the water, g the gravitational acceleration and A the contributing area around the
node.










Figure 1: Graphical representation of Westergaards and Zangars calculation models.

Zangar (1952)[2], using an electric analogue, extended the added mass approach of
Westergaard for inclined upstream surfaces of the dam, introducing reductive factors
dependent on the angle of inclination.
0, 5 (2 ) (2 )
w
m
y y y y
m H C A
H H H H g
(
= +
(


(2)

where C
m
a coefficient based on the angle of inclination and the other parameters as
Westergaards formula.
Fluid Elements
The fluid elements are solid elements to which the characteristics of the water are applied.
The incompressibility or the water as well as the null shear resistance are introduced with a
Poisson number equal with 0,5 or close to this value for the finite element programs. The
bulk modulus of the water is K=2,2 GPa. The modeling of the water with solid elements
causes numerical instabilities because of the introducing of zero energy modes (hourglass
modes). This effect can be mitigated with the use of hourglass control and by applying the
free surface boundary condition for the vertical node displacements [9]. Moreover, a nonlinear
material behavior with tension cut off or a contact interaction which allows only compression
to be transmitted will avoid unrealistic tension stress of the dam caused by the water.
Acoustic Elements
The acoustic elements are used to model the fluid behavior of the air. They have no shear and
tension resistance and they transmit only pressures [10]. With assignment of the water bulk
modulus they model the water behavior very good. Numerous boundary conditions can be
assigned to the acoustic elements, which model natural phenomena such as wave absorption at
the far end of the reservoir, sloshing of the free surface, wave impedance at the reservoirs
bottom due to sediments etc. For the acoustic elements no special numerical care has to be
taken except for assigning the boundary conditions.
Model aspects
For this benchmark two models (one with coarse mesh and one with fine mess) are
investigated. The mesh of the reservoir is the same for both cases. The foundation was
considered massless, so no further care was taken for wave absorption or deconvolution of the
seismic motion. Because of the massless foundation with no radiation absorption of the
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
69
seismic waves and due to the lack of further non-linearities of the dams material and of the
contact interfaces, a big enough structural damping is applied. As presented in [6] for a big
range of frequencies the total Rayleigh damping is between 8 and 10 %. Due to the linear
finite element analysis a 10% viscous damping is used by [6]. Here, because of the small peak
ground acceleration of 0,1g a value of 7,5% of structural damping was chosen in order to
determine the Rayleigh stiffness damping factor a and the Rayleigh mass factor .
For the reservoir hydrodynamic pressures, two added mass approaches and one reservoir
modeling with acoustic elements were investigated. The generalized Westergaards [11] and
the Zangars added mass approaches were used. The added masses were given via a user
subroutine which defines user elements in Abaqus [5].
The two models with acoustic elements differ only in the wave absorptions method of the far
field. The first uses acoustic infinite element whereas the latter impedance boundary
condition. The impedance condition can be given either as element based or as surface based
condition. Moreover a boundary condition is given at the reservoir free surface constraining
the dynamic acoustic pressures to be zero. The surfaces of the rock and the dam are tied with
the surfaces of the reservoir.


Figure 2: The model with two different meshes and two reservoir modeling approaches: left
fine mesh and acoustic elements and right coarse mesh and added mass elements.



Figure 3: The reservoir with its boundary conditions.

Free surface condition
Boundary impedance
or acoustic infinite
elements
Tied with the dam
Tied with the rock
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
70
Table 1: Material Parameters
Rock Water Dam
Density (kg) 0 1000 2400
Poisson Ratio 0,2 - 0,167
Young Modulus (MPa) 25000 - 27000
Bulk Modulus (MPa) - 2200 -
Analysis methods
The seismic analyses were carried out using the time history with direct integration and the
modal time history. From computational time the modal time history is a little bit faster than
the time history with direct integration. The time histories were gives as nodal acceleration to
the boundaries of the rock. A baseline correction offered by Abaqus was also applied to them.
Results
The results of the analyses are presented at the next tables and diagrams. The first table shows
the ten first modes for the dam, with empty and with full reservoir modeled by the different
methods described before. The diagrams show due to lack of space only some of the results
containing the minimal and maximal vertical, minimum principal and maximum principal
stresses of the dam for the different reservoir models and for the different dam mesh. The
results are given for the paths along the height of the dam, for the upstream and the
downstream sections. For convenience abbreviations were introduced to the diagrams (e.g.
dti refers to direct time integration, West to Westergaards added mass, ac to acoustic
element, imp to impedance boundary condition for the acoustic elements, inf to acoustic
infinite elements, modal to modal dynamic analysis). The results for the fine mesh model
are given with dashpot line in order to differ easier than the ones of the coarse mesh model.
The analysis with the infinite elements had more computational cost than the analysis with the
impedance condition. In order to obtain similar results to the impedance boundary condition
with the use of infinite acoustic elements, care must be given in the definition of the infinite
elements thickness. There are trivial differences when the analyst uses the improved rather
than the planar non-reflecting condition offered by Abaqus.
The results of the fine model with acoustic elements for the reservoir gave too conservative
results. The author believes that these results for the given meshes of dam and reservoir are
not correct due to violation of the contact condition, according to which the slave surface
nodes must be finer than the master surface nodes.

Figure 4: The frequencies for the two models (coarse left, fine right) and for the four reservoir
models.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
71
Table 2: Ten first modes for the coarse model
Mode
Nr.
Empty Acoustic Westergaard Zangar
f (Hz) T (sec)
f
(Hz)
T (sec)
f
(Hz)
T
(sec)
f
(Hz)
T (sec)
1

f=1,92 T=0,52 f=1,47 T=0,68 f=1,29 T=0,78 f=1,39 T=0,72
2

f=2,03 T=0,49 f=1,54 T=0,65 f=1,32 T=0,76 f=1,44 T=0,69
3

f=2,91 T=0,34 f=1,55 T=0,65 f=1,96 T=0,51 f=2,12 T=0,47
4

f=3,59 T=0,28 f=2,11 T=0,47 f=2,30 T=0,43 f=2,49 T=0,40
5


f=3,63 T=0,28 f=2,33 T=0,43 f=2,44 T=0,41 f=2,65 T=0,38
6

f=4,29 T=0,23 f=2,46 T=0,41 f=2,86 T=0,35 f=3,07 T=0,33
7


f=4,50 T=0,22 f=2,61 T=0,38 f=3,08 T=0,32 f=3,33 T=0,30
8

f=4,80 T=0,21 f=2,97 T=0,34 f=3,57 T=0,28 f=3,82 T=0,26
9

f=5,19 T=0,19 f=3,25 T=0,31 f=3,73 T=0,27 f=4,05 T=0,25
10


f=5,52 T=0,18 f=3,37 T=0,30 f=3,77 T=0,27 f=4,07 T=0,25
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
72
Table 3: Ten first modes for the fine model
Mode
Nr.
Empty Acoustic Westergaard Zangar
f (Hz) T (sec)
f
(Hz)
T (sec)
f
(Hz)
T
(sec)
f
(Hz)
T (sec)
1

f=1,91 T=0,52 f=1,47 T=0,68 f=1,30 T=0,77 f=1,41 T=0,71
2


f=2,03 T=0,49 f=1,54 T=0,65 f=1,33 T=0,75 f=1,45 T=0,69
3

f=2,90 T=0,35 f=1,54 T=0,65 f=1,98 T=0,50 f=2,14 T=0,47
4

f=3,57 T=0,28 f=2,00 T=0,50 f=2,32 T=0,43 f=2,50 T=0,40
5


f=3,62 T=0,28 f=2,29 T=0,44 f=2,48 T=0,40 f=2,68 T=0,37
6

f=4,27 T=0,23 f=2,46 T=0,41 f=2,91 T=0,34 f=3,11 T=0,32
7

f=4,48 T=0,22 f=2,53 T=0,40 f=3,12 T=0,32 f=3,37 T=0,30
8

4,78 T=0,21 f=2,96 T=0,34 f=3,62 T=0,28 f=3,85 T=0,26
9

f=5,17 T=0,19 f=3,13 T=0,32 f=3,81 T=0,26 f=4,12 T=0,24
10

f=5,49 T=0,18 f=3,27 T=0,31 f=3,86 T=0,26 f=4,13 T=0,24
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
73
The next diagrams give some representative comparisons between results for the different
reservoir models and analysis methods.


Figure 6: The vertical stresses for the different reservoir models at the downstream main
section.


Figure 7: The vertical stresses for the different reservoir models at the upstream main section.


Figure 8: Comparison between the coarse and fine model for the vertical stresses.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
74

Figure 9: The hoop stresses for the different reservoir models at the downstream main section.


Figure 10: The hoop stresses for the different reservoir models at the upstream main section.


Figure 11: The radial deformations for the different reservoir models at the main section.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
75

Figure 12: Comparison between the modal dynamic analysis and the direct time integration
for the coarse model with the Zangars approach at the downstream (left) and upstream (right)
main section.
Conclusion
The earthquake analysis of an arch dam-reservoir-foundation system was performed with
different modeling aspects according to the formulators directions. The results show very
near values for the two added mass approaches, with the one of Zangar to be a little bit more
favorable than the one of Westergaard. The acoustic elements models with the two non-
reflecting approaches give identical results. The coarse and fine models differ only in the base
stresses due to the coarser mess of the coarse model and some deviations are noticed at the
added mass models. Although the modal dynamic analysis is much faster than the direct time
integration, delivers conservative results.
Acknowledgements
This paper consists a part of the research project Earthquake Analysis and Design of
Hydraulic Structures, which is funded by the Federal Waterways and Research Institute of
Germany in cooperation with the Institute of Concrete Structures of the Karlsruhe Institute for
Technology. The contribution of both participated Institutes and persons involved is highly
acknowledged.
References
[1] Westergaard, H. M. (1933). Water pressures on dams during earthquakes. Transactions of
the American Society of Civil Engineers, American Society of Civil Engineers, New
York, New York, Paper 1835, 1933.
[2] Zangar, C. N. (1952). Hydrodynamic Pressures on dams due to horizontal earthquake
effects. U.S. Departmant of Interior, Bureaus of Reclamation, Engineering Monographs
No.11
[3] Chwang, A.T., Housner, G.W. (1978). Hydrodynamic pressures on sloping dams during
earthquakes. Part 1. Momentum Method. Journal of Fluid Mechanics, vol. 2, part 2, pp.
335-341
[4] Chwang, A.T. (1978). Hydrodynamic pressures on sloping dams during earthquakes. Part
2. Exact Theory. Journal of Fluid Mechanics, vol. 2, part 2, pp. 343-348
[5] ABAQUS (2011),Users manual, Version 6.11. Dassault Systmes Simulia Corporation,
Providence RI, USA.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
76
[6] US Bureau of Reclamation (2006), State-of-Practice for the NonlinearAnalysis of
Concrete Dams at the Bureau of Reclamation, USBR Report, Colorado, USA.
[7] United States Society on Dams (2008). Numerical models for seismic evaluation of
concrete dams. Review, evaluation and interpretation of results. USSD, Denver, USA.
[8] Zienkiewicz, O. C., Bettess, P. (1978). Fluid-Structure dynamic interaction and wave
forces. An introduction to numerical treatment. International Journal For Numerical
Methods In Engineering, Vol. 13, 1-16.
[9] Wilson, E.D., Khalvati,M. (1983). Finite elements for the dynamic analysis of fluid-solid
systems. International Journal For Numerical Methods In Engineering, Vol. 19, 1657-
1668.
[10] Matthew Muto, Nicolas von Gersdorff, Zee Duron, Mike Knarr (2012). Effective
Modeling of Dam-Reservoir Interaction Effects Using Acoustic Finite Elements, in
Proceedings of Innovative Dam and Levee Design and Construction for Sustainable
Water Management, 32nd Annual USSD Conference, New Orleans, Louisiana, April 23-
27, 2012, Pages 1161-1168.
[11] Kuo, James Shaw-Han, (1982). Fluid-structure interactions: added mass computations for
incompressible fluid. UCB/EERC-82/09, Earthquake Engineering Research Center,
University of California, Berkeley, 1982-08.




ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
77
HFTD Analysis of an arched dam at seismic loading
W. Kikstra
1
, F. Sirumbal
1
and G. Schreppers
1

1
TNO DIANA BV, Delftechpark 19a, 2628 XJ Delft, NETHERLANDS
E-mail: info@tnodiana.com
Abstract
Hybrid Frequency-Time Domain (HFTD) method is applied to analyze the response of a dam-
foundation-reservoir system. With this method the effect of frequency dependent properties
such as compressibility of fluid, reservoir-bottom absorption and far-field reflection can be
considered. HFTD results are compared with transient Newmark-type results and effect of
frequency dependent properties is found to reduce amplitudes and stresses in the dam for the
chosen bottom absorption in the reservoir.
Introduction and analysis procedure
The benchmark case study of the arch dam-reservoir interaction at seismic loading was
modeled with DIANA software. The formulation used by DIANA to couple the Finite
Element equations of motion for Fluid-Structure Interaction (FSI) problems is a mixed
displacement scalar potential approach, which defines the solid variables in terms of
displacement degrees of freedom (DOF), and the fluid variables in terms of pressure DOF.
This definition of the fluid domain using Acoustic Finite Elements is called the Eulerian
pressure formulation. One of the advantages of this type of formulation is the simple
description of the fluid domain using a single scalar pressure variable (). This reduces
considerably the number of variables of the system since only one DOF per node is required
to describe the motion of the fluid domain.

Taking into account that for dam-reservoir interaction problems the fluid motion is not
substantial but small, considerable simplifications can be made in its equation of motion
formulation. Those simplifications are a consequence of the following hypotheses assumed in
DIANA FSI models:

- Small displacement amplitudes
- Small velocities (convective effects are omitted)
- Inviscid (viscous effects are neglected)
- Small compressibility (variation of density is small)
- No body forces in the fluid
Based on these hypotheses, the scalar fluid wave equation of motion is defined by Eq.(1).The
wave speed (c), defined in terms of the fluid density () and bulk modulus (), is defined by
Eq.(2).



(1)


(2)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
78
In the same way, the FSI condition expressed in Eq. (3) relates the fluid gradient pressure in
the normal direction (

) to the interface surface (


I
) with the structure acceleration vector
(

).

on
I
(3)

In addition to the fluid-structure interface (
I
) condition defined in Eq. (3), Figure 1 shows
three types of boundary conditions which can be defined in DIANA FSI models.


Figure 1: Fluid and solid domains, fluid-structure interface and boundary conditions

It is possible to specify two types of boundary conditions for the free surface of the reservoir
(s). The first and simplest one is a consequence of neglecting the effect of the surface waves,
prescribing a pressure equal to zero in the horizontal top free surface, as expressed in Eq. (4).
This essential or Dirichlet type of boundary condition is the one used in the benchmark case
study. Additionally, in DIANA it is possible to define a second type of boundary condition
which takes into account the pressure caused by the free surface gravity waves.

on
s
(4)

Two types of infinite extent boundary condition (
e
) are considered in the benchmark case
study. The first one defined by Eq. (5.a) prescribes the hydrodynamic pressure equal to zero at
a distance equal to the reservoir length. The second one defined by Eq. (5.b) is a radiation
boundary theoretically located at an infinite distance from the dam, which ensures that no
incoming waves enter into the system (only outgoing waves).

on
e


(5.a)

on
e


(5.b)
Finally, two types of bottom boundary condition (
b
) are considered in the benchmark case
study. The first one is setting the gradient of the pressure in the normal direction equal to zero,
as expressed in Eq. (6.a). The second one defined by Eq. (6.b) is radiation boundary which
takes into account the energy absorption of the bottom materials in terms of the wave
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
79
reflection coefficient (), defined as the ratio between the amplitudes of the incident pressure
wave over the reflective pressure wave.

on
b


(6.a)

on
b


(6.b)
In this paper, the Hybrid Frequency-Time Domain method is applied to solve the fluid
equation of motion in the frequency domain. In this way, the hydrodynamic pressure
amplitude vector is obtained in terms of the structural displacement amplitude vector.

On the other hand, the structure non-linear equation of motion defined in the time domain is
formulated based on relative displacement. In this way, the earthquake ground acceleration
vector is introduced as external loading. The non-linear internal force of the structure is
defined as the difference between the linear internal force and an unknown pseudo force
vector, which is substituted into the non-linear equation to obtain the HFTD pseudo-linear
equation of motion of the dam in the time-domain.

Nevertheless, the HFTD method solves the pseudo-linear equation of motion in the frequency
domain, and therefore it is required to use the Discrete Fourier Transform (DFT) to pass from
one domain to the other. The fluid contribution is defined by additional mass and additional
damping which both are defined in the frequency domain, and thus can be defined as
frequency dependent properties.

As a consequence the HFTD method can account for the effects of non-linear material
behavior, compressibility of fluid, radiation at infinite extend and reservoir-bottom
absorption, whereas standard transient analysis with Newmark-type of time-integration
schemes cannot consider these effects together.
Model definition
Because in the mesh that has been provided by the benchmark formulators some node
connection incompatibilities were found, a new mesh has been defined using the provided
geometry specifications as shown in Figure 2.

Figure 2: Components and geometrical characteristics of the foundation-dam-reservoir
interaction system
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
80
Element with second order displacement interpolation were used for foundation and dam and
element with linear fluid-pressure interpolation were used for the reservoir. Mesh
characterstics are listed in Table 1 and material characteristics are listed in the Tables 2-4.

Table 1: Characteristics of the Finite Element Model Mesh
Component Type DIANA element
name
Number of Elements Number of Nodes
Dam Solid 3D CHX60 712 3601
CTP45
Foundation Solid 3D CHX60 4896 23339
CTP45
Reservoir Flow 3D CTP15H 2670 11950
CHX20H

Table 2: Material parameters for the concrete dam
Parameter DIANA variable name Value / Type Units
Modulus of elasticity YOUNG 2.7 x 10
10
N / m
2

Poison modulus POISON 1.67 x 10
-1
-
Density DENSIT 2.4 x 10
3
kg / m
3

Rayleigh damping RAYLEI 5.71199 x 10
-1
-
1.447 x 10
-3
-

Table 3: Material parameters for the reservoir fluid
Parameter DIANA variable name Value / Type Units
Conductivity CONDUC 1

-
Sonic speed CSOUND 1.483 x 10
3
m / s
Density DENSIT 1.0 x 10
3
kg / m
3

Wave reflection coefficient for
infinite extent boundary
ALPHAB 0 -
Wave reflection coefficient for
bottom absorption boundary
ALPHAB 5.0 x 10
-1
-


Table 4: Material parameters for the foundation soil
Parameter DIANA variable name Value / Type Units
Modulus of elasticity YOUNG 2.5 x 10
10
N / m
2

Poison modulus POISON 2.0 x 10
-1
-
Density DENSIT 0 kg / m
3

The transient responses of the two analysis cases shown in Figure 3 are determined and
studied. Case I corresponds to the frequency independent system, for which the fluid of the
reservoir is assumed to be incompressible ( ) and the infinite extent and bottom
boundary conditions of the reservoir are defined by Eqs. (5.a) and (6.a), respectively. This
frequency independent analysis case is solved with both HFTD and Newmark methods, with
the objective of assessing the accuracy of HFTD method. On the other hand, Case II
corresponds to a frequency dependent system in which fluid compressibility and reservoir
radiation boundary conditions are included. Reservoir bottom absorption ( ) and
radiation boundary of infinite extent, defined by Eqs. (6.b) and (5.b), respectively, are
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
81
included in the analysis. As it was previously explained, the transient analysis of this
frequency dependent system cannot be solved by the standard Newmark method, only by the
HFTD method. Non-linear material behavior of the dam-structure is not considered in these
cases.

Figure 3: Foundation-dam-reservoir interaction analysis cases
Results
Prior to the time domain analysis, eigenvalues of the dam-foundation system including
interaction with the fluid reservoir were determined. The first 10 mode-shapes and
corresponding eigenvalues are presented in Figure 4.
Case I was solved using both Newmark time integration and HFTD analysis. Figure 5 shows
the amplitude of the displacement of the crest at the main section. Agreement is so close that
the vertical stress, hoop stress and radial displacement results for Case I will only be displayed
for the Newmark time stepping analysis. The envelopes for hoop stresses, vertical stresses and
radial displacements along with their static values are given in figures 6 to 8, respectively.
On the other hand, the frequency dependent properties of the reservoir and reservoir
boundaries of Case II introduce more damping in the system which should lead to lower
responses to the earthquake loading. Figure 9 shows that the crest amplitude has considerably
lower peaks compared to Case I. The same magnitude reduction is observed in the envelopes
of hoop stresses, vertical stresses and radial displacements given in Figures 10 to 12,
respectively.

A 2D study to discriminate the effects of fluid compressibility, radiation boundary and bottom
absorption shows that reservoir fluid compressibility increases the response, that there is
minimal influence of the radiation boundary if sufficient reservoir length is modeled and that
bottom absorption damps the response of the dam-reservoir system [3].

For all the analysis results presented (Eigen-analysis, Newmark and HFTD), the effect of self-
weight gravity and hydrostatic pressure loads were taking into account as initial conditions.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
82

Figure 4: Mode-shapes and eigenvalues of the dam-foundation-reservoir interaction system


Figure 5: HFTD vs Newmark. Crest amplitude [m] at main section
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
83

Figure 6: Vertical stress [Pa] against elevation [m] for left, main and right dam sections at
upstream and downstream dam face for incompressible reservoir.

Figure 7: Hoop stress [Pa] against elevation [m] for left, main and right dam sections at
upstream and downstream dam face for incompressible reservoir.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
84

Figure 8: Radial displacement [m] against elevation [m] for left, main and right dam sections
at upstream and downstream dam face for incompressible reservoir.

Figure 9: Compressible vs Incompressible. Crest amplitude [m] at main section

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
85

Figure 10: Vertical stress [Pa] against elevation [m] for left, main and right dam sections at
upstream and downstream dam face for compressible reservoir.

Figure 11: Hoop stress [Pa] against elevation [m] for left, main and right dam sections at
upstream and downstream dam face for incompressible reservoir.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
86

Figure 12: Radial displacement [m] against elevation [m] for left, main and right dam sections
at upstream and downstream dam face for compressible reservoir.
Conclusion
The given foundation-dam-reservoir system was analyzed for the full duration of the
earthquake. With the HFTD method implemented in the standard version of DIANA for the
case of frequency independent properties the same results could be reproduced as with
implicit time stepping with Newmarks method. With HFTD the effect of frequency
dependent properties such as compressibility of fluid, reservoir bottom absorption and infinite
extend reflection have been analyzed and quantified, resulting in an interesting method to be
applied specially in the dynamic analysis of dam-reservoir interaction models.
References
[1] Veletsos, A.S., and Ventura, C.E. (1985). Dynamic analysis of structures by the DFT
method. J. Struct. Eng. ASCE, Vol.111, 2625-2642.
[2] TNO DIANA BV (2011). DIANA Users Manual Release 9.4.4, Delft, The
Netherlands.
[3] Sirumbal, F. (2013) Numerical modeling of dam-reservoir interaction seismic response
using the Hybrid Frequency-Time Domain ( HFTD ) method Masters thesis, Faculty of
Civil Engineering and Geosciences, Delft University of Technology, Delft, The
Netherlands.
[4] Darbre, G.R. (1996). Nonlinear dam-reservoir interaction analysis. Proc. Eleven World
Conf. on Earthquake Engineering, Acapulco, Paper No. 760.
[5] Zienkiewicz, O.C., and Bettes, P. (1978). Fluid-structure dynamic interaction and wave
forces. An introduction to numerical treatment. Int J Numer Meth Eng, Vol. 13, 1-16.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
87
Finite Element Modelling of Seismic Fluid-Structure
Interaction for a large Arch Dam
G. Faggiani
1
and P. Masarati
1

1
Ricerca sul Sistema Energetico - RSE SpA, via R. Rubattino 54, 20134 Milano, ITALY
E-mail: giorgia.faggiani@rse-web.it
Abstract
The linear dynamic fluid-structure interaction at seismic loading for the artificially generated
large arch dam provided for the Theme A was modelled using the approach of acoustic
compressible elements, with both the coarse and fine meshes provided by the formulators of
the 12
th
ICOLD Benchmark Workshop.
The effects of incompressible fluid (theoretical hypothesis of the added mass models) and of
the partial absorption of the hydrodynamic pressure waves at the reservoir boundary (bottom
and sides) were also investigated.
Simulations were carried out using the RSE in-house FEM code CANT-SD, specifically
designed for dynamic linear and non-linear analyses of dam-reservoir systems.
The coarse and fine meshes showed not dissimilar results, except at dam-foundation interface.
The results of the analyses confirmed that incompressible models could result relatively
conservative and highlighted the benefits of the approach of acoustic elements, mainly the
possibility to take into account the damping effect on the fluid boundary.
Introduction
Seismic safety assessment of large arch dams is actually a very important matter. Both
dam-reservoir interaction and non-linear mechanisms due to the contraction joint opening and
sliding could greatly affect the mechanical behaviour of the dam: therefore they must be
properly considered in order to obtain reliable numerical simulations under strong
earthquakes.
Theme A of the 12
th
ICOLD Benchmark Workshop on Numerical Analysis of Dams [1] is
aimed at comparing the different available approaches to model the dynamic fluid-structure
interaction: to this purpose the structural response to a seismic loading of an artificially
generated large arch dam has to be investigated.
The analyses have been carried out using the RSE in-house FEM code CANT-SD [2],
specifically designed for linear and non-linear dynamic (seismic) analyses of dam-reservoir
systems. This code is currently used at RSE for safety assessment of concrete dams, and it
was adopted to deal with some themes proposed in previous ICOLD Benchmark Workshops
[3] [4] [5] [6] [7] [8].
Regarding the main simulation options required to effectively address the present Theme A,
CANT-SD models the fluid reservoir by means of acoustic elements and solves the transient
dynamic coupled problem using an implicit direct time integration method.
Geometrical and physical model
The two FEM parabolic meshes, Coarse and Fine, considered in the simulations are reported
in Figure 1 and 2. Only the dam meshes provided in the Theme A have been adopted
unchanged. The fluid domain was obtained by extruding the upstream face of the dam mesh
for a length 3 times the total height of the dam (220 m): the resulting mesh was only joined to
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
88
the upstream dam face, not to the rock. Foundation meshes were modified in order to obtain
coincident meshes at concrete-rock interface as necessary for the proper operation of CANT-
SD.


Figure 1: Arch dam FEM model Coarse mesh


Figure 2: Arch dam FEM model Fine mesh

A monolithic behaviour of the dam body has been considered, as no construction joints have
been modelled. Dam concrete and foundation rock, assumed to behave linear-elastically, were
characterized by the physical-mechanical parameters provided by the formulators and
reported in Table 1 along with the properties of the fluid.
A 5% structural damping ratio () was assumed in the analyses: damping matrix is expressed
as linear combination of mass and stiffness matrices according to Rayleigh formulation. The
calibration of structural damping ratio was based on the frequency range resulting from modal
analysis, such that the damping was almost constant in this range (Figure 3).

Table 1: Material properties
Parameter Rock mass Water Dam concrete
Modulus of elasticity (MPa) 25000 - 27000
Bulk modulus (MPa) - 2200 -
Poissons ratio (-) 0.2 - 0.167
Density (kg/m
3
) 0 1000 2400

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
89

Figure 3: Rayleigh viscous damping

The dynamic fluid-structure interaction, modelled in CANT-SD following the classic acoustic
approach [9], is here briefly summarized.
The hydrodynamic pressure in the compressible fluid of the reservoir is governed by the wave
equation:
2
2
2 2
2
2
2
2
2
t
p
C z
p
y
p
x
p
c
c
=
c
c
+
c
c
+
c
c 1

(1)
where C = (k/) is the velocity of sound in the fluid, k the bulk modulus and the density.
The boundary conditions are (Figure 4):
- free surface 0 = p
- upstream face of the dam


- open boundary


- bottom and sides


where n is the outward normal, the normal acceleration and q the damping coefficient on
the bottom and sides.
The upstream face of the dam results the only surface of interaction between structure and
fluid: the accelerations at the dam face represent the actions of the dam on the reservoir,
which in turn reacts through the hydrodynamic pressures exerted on the dam face.
The last boundary condition accounts for the partial absorption of hydrodynamic pressure
waves [10] [11]: this damping phenomenon is mainly caused by the layer of sedimentary
material possibly deposited in the reservoir, but could be significant also in cases of few or no
accumulated sediments.

0.000
0.025
0.050
0.075
0.100
0.125
0.150
0 1 2 3 4 5 6 7 8 9 10 11
D
a
m
p
i
n
g

r
a
t
i
o

[
-
]
Frequency [Hz]
stiffness
mass
Rayleigh
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
90

Figure 4: Boundaries of reservoir

Finite element subdivision of the reservoir leads to the discretized form of the outlined
acoustic problem, i.e. a system to be coupled with the structural one (that includes the
hydrodynamic pressure loads). The fluid-structure coupled problem is governed by the
following system:

{




(2)
The first subsystem governs the mechanical behaviour of the dam and the (unknown) pressure
p represents an applied load; the second governs the acoustic behaviour of the reservoir and
the (unknown) acceleration represents an assigned boundary condition.

The seismic analysis proposed in Theme A was performed for three physical models of the
reservoir behaviour, differing in fluid compressibility and/or boundary absorption.
The first physical model (Base Case) fully respects the requirements of the formulators: the
provided value of the bulk modulus of water (2200 MPa) was adopted, boundary absorption
was neglected (q=0 on bottom and sides, i.e. reflecting condition) and the non-reflecting
condition was considered at the end of the reservoir.
The other physical models have the aim to examine, for the particular given scenario, the
following interesting aspects: the effect of the incompressibility hypothesis and the
importance of the boundary absorption. A brief hint of both aspects follows, along with the
necessary choices for the two additional simulations.
The assumption of fluid incompressibility is connected with the added mass models, which
assimilate the action of the fluid to that of some kind of mass (physical mass or a mass
matrix) attached to the upstream face of the dam. It is worth noting that the comparison
reported in this paper just concerns the incompressibility hypothesis, and can be therefore
only applied to the rigorous added mass matrix model, clearly defined in [9] and referred to
as Finite Element Added Hydrodynamic Mass Model in [12]. No investigation was made
about the effects of any possible change on this matrix (i.e. for computational convenience),
nor any comparison was made with Westergaard-type techniques, however strongly
discouraged in [12] for arch dams.
Upstream face
of the dam
Open boundary
Free surface
Bottom and sides
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
91
A key simplified parameter [13] that determines the significance of water compressibility is
the ratio
r
=f
1
r
/f
1
of the fundamental frequency of the reservoir to that of the dam-foundation
system without water: the higher this ratio, the less the importance of fluid compressibility.
Although the limit
r
=2 is valid only for gravity and not for arch dams, as clearly stated in
[13], it has often been used, at least as a broad clue, for arch dams too [12].
The fundamental frequency f
1
r
of the reservoir can be roughly evaluated as C/4H, where C is
the velocity of sound in water and H is the water depth: for the analysed system the value is
f
1
r
=1.7 Hz. The fundamental frequencies of the dam-foundation system without water were
calculated and resulted 1.9 Hz for anti-symmetric mode shape and 2.0 Hz for symmetric mode
shape. The value of the ratio
r
, about 0.9, indicates that compressibility should probably be
important.
The physical model to investigate the effects of compressibility (Incompressible Case) was
developed using a value of the bulk modulus one hundred times greater than that of the Base
Case (k=2200100 MPa) and assuming the reflecting boundary condition at the end of the
reservoir. Two modal analyses were carried out, the first with k=2200100 MPa and the
second with k=220010000 MPa to verify that the value of the bulk modulus assumed in
dynamic analysis was great enough to simulate incompressibility: no significant differences
were found in the results.

The coupled mechanical-acoustic approach allows taking into account the effect of the partial
absorption of hydrodynamic pressure waves on the boundary of the reservoir [10] [11],
impossible to simulate using any added mass model. To assign the boundary absorption, the
damping coefficient q must be quantified: to this purpose it is convenient to express it by
means of the wave reflection coefficient (the ratio of the amplitude of the reflected
hydrodynamic pressure wave to the amplitude of the incident one) [11]:
) (1
) 1 (
1
o
o
+

=
C
q
(3)
The wave reflection coefficient represents a more physically meaningful description of the
phenomenon: =1 corresponds to a reflecting (rigid) boundary, =0 corresponds to a non-
reflecting (transmitting) boundary, -1<<0 corresponds to an even major damping behaviour.
The value of can be determined on the basis of field investigations [12] [14] [15]. The
results reported in [14] indicate values of , measured at seven concrete dam sites, varying
over a range from -0.55 to 0.66: three of these values were negative, due to thick layers of soft
sediments. A value of 0.82 was determined at the dam site investigated in [15]: a rock site
with very little or no accumulated sediments.
Making reference to the reported results of field investigation, a value =0.5 was considered
sensible to examine the importance of the boundary absorption in the seismic analysis: the
corresponding physical model (Damped Case) was obtained from the Base Case using a
damping coefficient q=0.000225.
Table 2 summarizes all the simulations: Base Case (B) was performed both with the coarse
and fine meshes, while Incompressible (I) and Damped (D) Cases only with the coarse mesh.
Table 2: Summary of simulations
Case Coarse mesh Fine Mesh
Base B B
Incompressible I -
Damped D -

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
92
Loadings
The numerical analyses simulated the effects of the following loadings/actions:
1. dead weight
2. hydrostatic pressure with water level at crest height
3. seismic loading
The seismic loading was provided in the Theme by means of artificially generated
acceleration time histories in the three coordinate directions X (upstream), Y (cross-stream)
and Z (vertical): the elastic response spectra (5% damping ratio) are reported in Figure 5. Its
worth noting that the given time histories represent a quite moderate intensity earthquake.
The acceleration time histories were only assigned to the bottom and sides of the foundation,
not to the boundary of the reservoir too.
The transient dynamic coupled problem was solved using an implicit direct time integration
method (HHT) [16]; an integration step of 0.002 s was chosen, in order to well represent
frequencies up to 25 Hz.


Figure 5: Elastic response spectra

Results
The following sections report and discuss the results of Base Case with both coarse and fine
meshes and of Incompressible and Damped Cases.
The stress state is represented either by diagrams showing vertical and hoop static and
dynamic (envelope) stresses, or by contour plots showing principal stresses envelopes,
expressed in megapascal, positive if tensile.
The displacements of the dam, expressed in metres, are positive if directed downstream.
Eigenfrequencies and mode shapes
The modal analysis allowed the computing of the natural resonant frequencies of the dam-
reservoir system and the corresponding mode shapes. Table 3 reports the first four
eigenfrequencies for both the coarse and the fine mesh. The values obtained with the two
meshes were almost identical: as expected, the refined discretization exhibited a slightly more
flexible behaviour than the coarse one. These frequencies corresponded to the first anti-
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 0.5 1 1.5 2 2.5 3 3.5 4
A
c
c
e
l
e
r
a
t
i
o
n

[
g
]
Period [s]
upstream
cross-stream
vertical
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
93
symmetric and the first three symmetric mode shapes, shown in Figure 6 and 7 for the coarse
and fine mesh respectively. The first four eigenfrequencies occurred in a range of periods of
about 0.450.65 s, matching the response spectra in their descending branch.
Table 3: Natural frequencies
Mode
Eigenfrequencies [Hz]
Coarse Mesh Fine Mesh
1 1.547 1.540
2 1.551 1.549
3 2.052 2.050
4 2.229 2.222



Figure 6: Mode Shapes (Base Case, Coarse Mesh)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
94



Figure 7: Mode Shapes (Base Case, Fine Mesh)
Seismic simulation with Coarse and Fine Meshes
The transient dynamic analysis supplied the stress-strain state of the system due to the
application of the seismic loading.
The simulations developed for the Base Case with Coarse (CB) and Fine (FB) meshes showed
how the use of a more refined mesh does not lead to significant differences in the results,
though involving quite a higher computational effort (about 5 times).
Vertical and hoop stresses in dam main section are reported in Figure 8 and 9 for upstream
and downstream faces respectively, both for fine and coarse meshes: stresses were evaluated
for static and seismic loads.
Vertical stresses resulted always compressive upstream, except at dam foundation interface,
while downstream tensile vertical stresses, up to 1 MPa, occurred in the upper part of the dam,
above 610 m a.s.l., caused by the seismic loading. Due to the seismic loading, a variation of
vertical stress of about 11.5 MPa was observed both upstream and downstream. Figure 8
highlights that vertical stresses calculated with the coarse mesh are underestimated near the
dam-foundation interface, where tensile stress concentrations are usual in many arch dams.
Hoop stresses both on upstream and downstream faces were compressive, confirming that the
arch effect was activated and the dam behaved according to its monolithic scheme. The effect
of the seismic loading, greater in the upper part of the dam, involved a stress variation of
about 4 MPa upstream (at 675 m a.s.l.) and 2 MPa downstream (at about 695 m a.s.l.).
Figure 10 and Figure 11 report the tensile stress envelope and the dynamic maximum
displacement contour plots for the two analysed meshes. Tensile stress on the downstream
face was essentially vertically oriented: due to the seismic loading the upper central part of the
dam exhibited stresses up to 1.6 MPa. Figure 11 shows a downstream displacement of about
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
95
8 cm, due to the seismic loading. A similar upstream displacement was observed for the
minimum envelope.
The contour plots allowed the overall comparison between simulations with coarse and fine
meshes and confirmed that the spatial trend of the stress-strain state was comparable in the
whole dam body.


Figure 8: Base Case - Vertical (left) and hoop (right) stresses on the upstream surface


Figure 9: Base Case - Vertical (left) and hoop (right) stresses on the downstream surface


Figure 10: Base Case - Maximum principal stress Coarse (left) and Fine (right) meshes,
downstream view.
495
515
535
555
575
595
615
635
655
675
695
715
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Vertical Stresses [MPa]
C Static
F Static
CB Max
FB Max
CB Min
FB Min
495
515
535
555
575
595
615
635
655
675
695
715
-11.00 -9.00 -7.00 -5.00 -3.00 -1.00 1.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
C Static
F Static
CB Max
FB Max
CB Min
FB Min
495
515
535
555
575
595
615
635
655
675
695
715
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Vertical Stresses [MPa]
C Static
F Static
CB Max
FB Max
CB Min
FB Min
495
515
535
555
575
595
615
635
655
675
695
715
-11.00 -9.00 -7.00 -5.00 -3.00 -1.00 1.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
C Static
F Static
CB Max
FB Max
CB Min
FB Min
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
96


Figure 11: Base Case - Maximum dynamic displacement Coarse (left) and Fine (right)
meshes, upstream view
Incompressible fluid and absorption effects
Based on the comparison discussed in the previous section, the effects of incompressible fluid
(Incompressible Case - I) and of the reservoir boundary absorption (Damped Case D) on the
dynamic response of the dam were investigated only with the coarse mesh.
The results of the analyses are summarized and compared with those of the Base Case (B) in
Figure 12 and 13, where vertical and hoop stresses in dam main section are reported for
upstream and downstream face respectively. The general trend of stresses for the Base Case
was already illustrated in the previous section. The curves representing stresses for
Incompressible and Damped Case generally laid respectively outside and within those of the
Base Case, confirming that the incompressible model could result relatively conservative and
that the reservoir boundary absorption could reduce the earthquake response of the dam.
In the analysed situation, relevant to an earthquake of very moderate severity, the difference
among the three models could look worthless. However these differences could become
highly significant if a greater seismic loading were considered. Referring to the hoop stresses,
a 2 times amplified earthquake would result in the occurrence of tensile stresses in the arcs
(Figure 14) involving the transition from a monolithic to an independent cantilever separated
by vertical joints behaviour, if these stresses act in the dam for a significant height (starting
from the crest). Figure 14 illustrates that for incompressible model tensile stresses resulted
about 35 times higher than for the damped model, involving the upper 80 m of the dam, a
double height if compared with damped model.


Figure 12: Comparison among Base, Incompressible and Damped Case - Vertical (left) and
hoop (right) stresses on the upstream surface
495
515
535
555
575
595
615
635
655
675
695
715
-6.00 -4.00 -2.00 0.00 2.00 4.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Vertical Stresses [MPa]
Static
B Max
B Min
I Max
I Min
D Max
D Min
495
515
535
555
575
595
615
635
655
675
695
715
-11.00 -9.00 -7.00 -5.00 -3.00 -1.00 1.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
Static
B Max
B Min
I Max
I Min
D Max
D Min
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
97

Figure 13: Comparison among Base, Incompressible and Damped Case - Vertical (left) and
hoop (right) stresses on the downstream surface


Figure 14: Comparison among Base, Incompressible and Damped Cases for an amplified
earthquake- Hoop stresses on upstream (left) and downstream (right) surface
Conclusion
Theme A of the 12
th
ICOLD Benchmark Workshop on Numerical Analysis of Dams, dealing
with the linear dynamic fluid-structure interaction at seismic loading, has been approached by
using CANT-SD, a RSE in-house FEM code for dynamic linear and non-linear analyses of
dam-reservoir systems. The dynamic fluid-structure interaction was modelled with the
approach of compressible acoustic elements.
The analyses with the reflecting boundary condition on the bottom and sides of the fluid
domain, performed with both coarse and fine meshes to test the effects of different spatial
discretization, showed that the use of a more refined mesh does not lead to significant
differences in the results, though involving quite a higher computational effort.
Simulations considering incompressible acoustic elements or reservoir boundary absorption
were also performed: the results of these analyses confirmed that the incompressible model
could result relatively conservative and that reservoir boundary absorption could significantly
reduce the earthquake response of the dam. The use of an incompressible model instead of a
compressible one, capable to account for damping effect on the fluid boundary too, speeded
up the transition from a monolithic structural scheme to a different one with independent
cantilevers and joints.
495
515
535
555
575
595
615
635
655
675
695
715
-6.00 -4.00 -2.00 0.00 2.00 4.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Vertical Stresses [MPa]
Static
B Max
B Min
I Max
I Min
D Max
D Min
495
515
535
555
575
595
615
635
655
675
695
715
-11.00 -9.00 -7.00 -5.00 -3.00 -1.00 1.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
Static
B Max
B Min
I Max
I Min
D Max
D Min
495
515
535
555
575
595
615
635
655
675
695
715
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
B Max
I Max
D Max
B-ampli2
I-ampli2
D-ampli2
Static
495
515
535
555
575
595
615
635
655
675
695
715
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 8.00
H
e
i
g
h
t

[
m

a
.
s
.
l
.
]
Hoop Stresses [MPa]
B Max
I Max
D Max
B-ampli2
I-ampli2
D-ampli
Static
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
98
Acknowledgements
This work has been financed by the Research Fund for the Italian Electrical System under the
Contract Agreement between RSE S.p.A. and the Ministry of Economic Development -
General Directorate for Nuclear Energy, Renewable Energy and Energy Efficiency in
compliance with the Decree of March 8, 2006.
References
[1] Zenz G., Goldgruber M. (2013). Theme A formulation. Fluid Structure Interaction. Arch
Dam Reservoir at Seismic loading. 12
th
ICOLD Benchmark Workshop on Numerical
Analysis of Dams, Graz, Austria.
[2] Masarati P., Meghella M. (2000). The FEM computer code CANT-SD for non-linear
static and dynamic analysis of dams. Enel.Hydro rep. n. 6045, Milano, Italy.
[3] Faggiani G., Frigerio A., Masarati P., Meghella M.(2011). Finite element modelling of
concrete swelling effects on Kariba dam. 11
th
ICOLD Benchmark Workshop on
Numerical Analysis of Dams, Valencia, Spain.
[4] Meghella M., Masarati P. (2007). FEM analyses for the interpretation of the structural
behaviour of La Acea dam. 9
th
ICOLD Benchmark Workshop on Numerical Analysis of
Dams, St. Petersburg, Russia.
[5] Meghella M., Frigerio A., Masarati P. (2005). Evaluation of AAR on the behaviour of
Poglia dam adopting two different approaches. 8
th
ICOLD Benchmark Workshop on
Numerical Analysis of Dams, Wuhan, Hubei, P.R. China.
[6] Meghella M., Mazz, G. (2003). Safety evaluation against sliding of a gravity dam with
curved shape. 7
th
ICOLD Benchmark Workshop on Numerical Analysis of Dams,
Bucharest, Romania.
[7] Bon E., Chill F., Masarati P., Massaro C. (2001). Analysis of the effects induced by
alkali-aggregate reaction (AAR) on the structural behaviour of Pian Telessio dam. 6
th

ICOLD Benchmark Workshop on Numerical Analysis of Dams, Salzburg, Austria.
[8] Bolognini L., Masarati P., Bettinali F. Galimberti C. (1994). Non-linear analysis of joint
behaviour under thermal and hydrostatic loads for an arch dam. 3
rd
ICOLD Benchmark
Workshop on Numerical Analysis of Dams, Paris, France.
[9] Zienkiewicz O. C. (1977). The Finite Element Method, 3
rd
Edition, McGraw-Hill.
[10] Fok K.L., Chopra A.K. (1986). Earthquake analysis of arch dams including dam-water
interaction, reservoir boundary absorption and foundation flexibility. Earthquake
Engineering and Structural Dynamics, Vol. 14, pp. 155-184.
[11] Fenves G., Chopra A.K. (1983). Effects of reservoir bottom absorption on earthquake
response of concrete gravity dams. Earthquake Engineering and Structural Dynamics,
Vol. 11, pp. 809-829.
[12] US Army Corps of Engineers USACE (2003). Engineering and Design - Time-History
Dynamic Analysis of Concrete Hydraulic Structures. EM 1110-2-6051.
[13] Fok K.L., Chopra A.K. (1986). Frequency response functions for arch dams:
hydrodynamic and foundation flexibility effects. Earthquake Engineering and Structural
Dynamics, Vol. 14, pp. 769-795.
[14] Ghanaat Y., Redpath B.B. (1995). Measurement of reservoir-bottom reflection
coefficient at seven concrete dam sites. QUEST Structure Report No. QS95-01.
[15] Ghanaat Y., Hall R.L., Redpath B.B. (2000). Measurement of dynamic response of arch
dams including interaction effects. Proc. of 12WCEE2000.
[16] Hilber H.M., Hughes T.J.R., Taylor R.L. (1977). Improved numerical dissipation for time
integration algorithms in structural dynamics. Earthquake Engineering and Structural
Dynamics, Vol. 5, pp. 283-292.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
99
ICOLD 12-th International Benchmark Workshop
on Numerical Analysis of Dams
Theme A: Fluid Structure Interaction Arch Dam Reservoir at
Seismic Loading
A. Tzenkov
1
, A. Abati
1
and G. Gatto
1

1
STUCKY SA, Rue du Lac 33, PO Box, 1020 Renens, SWITZERLAND
E-mail: a.tzenkov@stucky.ch
Abstract
One of the main concerns regarding the numerical dynamic analysis of arch dams is the
proper modelling of the fluid-structure interaction between the dam and the impounded water.
There are several approaches to this, which enables accounting for the hydrodynamic
pressures on the upstream face of the dam with different precision and, respectively, with
different computing effort. This work investigates the impact of the hydrodynamic approach
opted for on the computed stresses and displacements of an example 220-m high double-
curvature arch dam. It is shown that, for this particular benchmark problem, it is important to
consider the compressibility of water.
Introduction
The hydrodynamic phenomena occurring on the interface between a dam and the impounded
water may have significant effect on the structural response of the dam. The structure (the
dam wall and its foundation) and the fluid (the impounded water) are two different physical
systems that interact with each other and thus present a coupled problem. According to the
classification given in [1], the latter is a Class I coupled problem in which the coupling occurs
on the interfaces between the domains.

A milestone procedure to account for the hydrodynamic effects on dams was established by
Westergaard in 1933, [2] by introducing the concept of added mass. Although Westergaards
approach was limited within the assumptions of rigid dams with vertical upstream faces,
infinitely long reservoirs, and incompressible fluids, it enabled accounting for the
hydrodynamic effects in the everyday engineering practice. Using the electric analog method,
Zangar [3] improved Westergaards approach by establishing a family of parabolas by means
of which it is possible to compute the hydrodynamic pressure on rigid dams with sloping
upstream face. With the advent of the computer and the increased utilization of the numerical
methods, it became possible to account for the effects due to (1) the fluid compressibility, (2)
the hydrodynamic pressure waves partial absorption by deposited sediments, (3) the
foundation inertia and damping, as well as for non-linear dam behaviour (4). Frequency
domain [4][5], time domain [6], and hybrid frequency-time domain (HFTD) [7] procedures
have been developed and used over the past decades. An inherent limitation of the frequency
domain approach is that it presupposes linear structural behaviour; on the other hand, it
enables readily considering effects (1), (2) and (3). In contrast, in the time-domain, nonlinear
structural behaviour can be accounted with reasonable computational effort, but it is more
difficult to simulate the other hydrodynamic phenomena.

This work investigates the differences of the stresses and the displacements computed by
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
100
means of the FEM in using three different approaches for modelling the hydrodynamic
effects. The analyses are performed on the current Benchmark example 220 m high double-
curvature arch dam by means of the computer program DIANA [8]. First, two linear seismic
analysis are carried out in the time domain by using added masses computed by the
Westergaard formula and by means of a fluid-structure interaction analysis in the assumption
of incompressible fluid. Next, a linear analysis is performed by means of the HFTD approach
and in modelling the fluids compressibility. In addition, a nonlinear seismic analysis is
performed in the time-domain, by means of a fluid-structure interaction analysis for
incompressible fluid. Details of the procedures utilized, the results of the computations, and a
discussion on the results are presented in the following sections.
Finite Element Model
The finite element model of the example arch dam foundation reservoir system is created
based on the geometry and the spatial discretization given by the Formulator of the current
Theme A [9]. The only difference is that the reservoir models length is 5 times the dam
height. Following the conditions of [9], the investigations have been done on two meshes: a
coarse one, and a finer one. The structural system of the coarse mesh is modelled by 2516
hexahedral and wedge isoparametric solid finite elements (356 for the dam and 2160 for the
foundation). The reservoir is modelled by 1956 3-D flow elements, while the dam-reservoir
interface is represented by 177 fluid-structure interface elements. All the elements are based
on quadratic interpolation. The total number of nodes of the coarse mesh is 21547. The fine
mesh has 78426 nodes; the dam and the foundation are modelled by 2848 and 10080 elements
respectively; however, it yields almost the same result regarding the structural response as the
coarse mesh. Translational supports in the three global directions are specified as structural
boundary conditions on the bottom and side surfaces of the foundation model. The material
parameters of the dam and the foundation are the same as the ones prescribed in [9]. The sonic
wave velocity =1483 m/s is associated with the reservoir fluid elements. Finally, water
density of 1000 kg/m
3
is specified for the fluid-structure interface elements.
Westergaard Added Mass (WG)
The added masses computed by means of the Westergaard formula [2] are applied on the
nodes of the upstream face of the dam by means of CQ24TM boundary surface elements [8].
178 such elements are defined for the coarse mesh; they are 712 for the finer model of the
system. A distributed translational mass material model is associated to the boundary surface
elements, which allows precise automatic calculation of the added masses. The total mass
assembled for the coarse mesh finite element model is TM=0.14E+11 kg (for massless rock).
Without added mass, TM=0.471E+10 kg.
Fluid - Structure Interaction with Incompressible Fluid (FSI)
As already mentioned, the solution is conducted in the time domain. To render the system
frequency independent, the fluid is specified as incompressible by setting tending to infinity,
assigning hydrodynamic pressure at the far-field and at the free surface of the reservoir,
and setting the hydrodynamic pressure gradient equal to zero in the normal direction of the
reservoir bottom. In this case, the hydrodynamic effect is represented by a consistent mass
matrix that is added to the mass matrix of structural system.
Fluid Structure Interaction with Compressible Fluid (HFTD)
In this case, the system is frequency dependent. Sonic wave velocity c=1483 m/s is associated
with the reservoir fluid elements and radiation boundary condition is specified for the fluid
far-field, Equation (1):
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
101

(1)

Fully reflecting boundary is assumed for the reservoir bottom, Equation (2):

(2)

The hydrodynamic pressure is set at the free surface of the reservoir.
Analysis Performed
The load-cases specified in [9] are the self-weight of the dam, the hydrostatic pressure and the
seismic loading. The present study is performed by means of phased analysis, which allows
modelling the loading history. Thus, first are computed the stresses due to the self-weight of
the dam. The hydrostatic pressure is applied next. Finally, the seismic loading is applied in the
three directions of the supports using the base acceleration time-histories given in [9] and
multiplied by a factor of 0.1, where is the gravity acceleration.
Static Analysis
The dam construction stages are modelled in an approximate, but realistic way by means of
phased activation of the dam elements. It is done in order to obtain correct strain and stress
state due to the self-weight of the dam body prior to the application of the hydrostatic loading.
In the linear analyses, the hydrostatic pressure is activated in a single phase as an
instantaneous loading, whereas in the nonlinear analysis it is applied at ten steps
corresponding to ten consecutive levels of filling of the dam reservoir.
Eigenvalue Analysis
The results of the computed eigenfrequencies for the cases of empty and full reservoir with
added masses defined by the Westergaard formula (coarse and fine mesh) and by a FEM
incompressible fluid-structure interaction analysis are presented in Table 1. Note that the
general coordinate system axes are as follows: X-axis is the stream direction (from u/s to d/s),
Y-axis is the cross-stream direction (from right to left), and Z-axis is the vertical direction.
Table 1: Eigenrequencies and Effective Mass Percentage (Empty, WG, WG fine mesh, FSI)
Mode Empty Reservoir Full Reservoir, WG WG FM FSI
, Hz X, % Y, % Z, % , Hz X, % Y, % Z, % , Hz , Hz
1 1.931 0.0 18.4 0.0 1.314 0.0 8.5 0.0 1.305 1.572
2 2.040 30.6 0.0 1.0 1.340 26.4 0.0 0.2 1.336 1.622
3 2.929 15.4 0.0 0.7 2.004 12.4 0.0 0.1 1.992 2.362
4 3.623 0.1 7.5 0.1 2.362 15.7 0.0 1.1 2.352 2.944
5 3.643 13.2 0.0 15.0 2.519 0.0 2.4 0.0 2.498 3.040
6 4.313 0.0 36.2 0.0 2.961 0.0 6.6 0.0 2.944 3.719
7 4.550 3.1 0.0 3.4 3.184 0.8 0.0 0.1 3.153 3.869
8 4.824 4.5 0.0 61.7 3.703 1.4 0.0 0.1 3.680 4.555
9 5.203 0.0 20.8 0.0 3.903 0.0 0.7 0.0 3.860 4.764
10 5.578 0.0 0.1 0.0 3.938 7.1 0.0 1.6 3.914 4.803
66.9 83.0 81.8 63.8 18.1 3.3

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
102

Mode 1,

1.527 Hz

Mode 2,

1.622 Hz

Mode 3,

2.362 Hz

Mode 4,

2.944 Hz

Mode 5,

3.040 Hz

Mode 6,

3.719 Hz

Mode 7,

3.869 Hz

Mode 8,

4.555 Hz

Mode 9,

4.764 Hz

Mode 10,

4.803 Hz
Figure 1: Dam Mode Shapes, Incompressible Fluid-Structure Interaction Analysis
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
103
Direct Time History Analyses
As already discussed, two linear and a non-linear direct time history analyses are performed.
The analyses are carried out by the method of Hughes, Hilbert and Taylor with -0.3.
The Rayleigh proportionality constants are computed so as to give a modal damping ratio of
5% in the first and the twelfth vibration modes. The nonlinear analysis considers only the
effects due to opening/closing at the contraction joints (it is assumed that the joints are
provided with strong shear keys). The contraction joints are modelled by structural interface
elements CQ48I to which is associated a nonlinear elastic interface material model.
Hybrid Frequency Time Domain Analysis
The HFTD analysis is performed assuming linear structural behaviour and for only one time
segment comprising the whole duration of the seismic input.
Results
The following line types are used to designate the type of analysis carried out:

Figure 2: Analysis Type Designation
Hoop Stresses and Cantilever Stresses

Figure 3: Main Section Hoop Stresses (U/S on left; D/S on right)
-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20
Radial Displacement (cm)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Static
WG with Reservoir Damping
WG w/o Reservoir Damping
FSI with Incompressible Fluid
FSI with Compressible Fluid
FSI with Incompressible Fluid and NONLIN Joint

Envelopes
of Min
Envelopes of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)

Envelopes of Min
Envelopes of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Envelopes of Min
Envelopes
of Max
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
104

Figure 4: Left Section Hoop Stresses (U/S on left; D/S on right)


Figure 5: Right Section Hoop Stresses (U/S on left; D/S on right)
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)

Envelopes of Min
Envelopes
of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)

Envelopes
of Min
Envelopes
of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)

Envelopes of Min
Envelopes
of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Hoop Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Envelopes
of Min
Envelopes
of Max
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
105

Figure 6: Main Section Cantilever Stresses


Figure 7: Left Section Cantilever Stresses
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of Min
Envelopes of
Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes of Min
Envelopes
of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of Min
Envelopes of
Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes of Min
Envelopes
of Max
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
106

Figure 8: Right Section Cantilever Stresses
Minimum and Maximum Principal Stresses


Figure 9: Main Section Minimum and Maximum Principal Stresses (U/S on left; D/S on right)
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of Min
Envelopes of
Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Cantilever Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes of Min
Envelopes
of Max
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of s
3,MIN
Envelopes of
s
1,MAX
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes of s
3,MIN
Envelopes
of s
1,MAX
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
107

Figure 10: Left Section Minimum and Maximum Principal Stresses


Figure 11: Right Section Minimum and Maximum Principal Stresses
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of s
3,MIN
Envelopes
of s
1,MAX
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes
of s
3,MIN
Envelopes
of s
1,MAX
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
Upstream
Envelopes of s
3,MIN
Envelopes
of s
1,MAX
-14 -12 -10 -8 -6 -4 -2 0 2 4 6
Principal Stress (MPa)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Downstream
Envelopes
of s
3,MIN
Envelopes
of s
1,MAX
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
108
Displacements

Figure 12: Radial Displacements (Left Section on left; Main Section, Right Section on right)

Figure 13: Time History of the Crown Displacement for Different Analyses

-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20
Radial Displacement (cm)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Static
WG with Reservoir Damping
WG w/o Reservoir Damping
FSI with Incompressible Fluid
FSI with Compressible Fluid
FSI with Incompressible Fluid and NONLIN Joint

Envelopes
of Min
Envelopes of Max
-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20
Radial Displacement (cm)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Static
WG with Reservoir Damping
WG w/o Reservoir Damping
FSI with Incompressible Fluid
FSI with Compressible Fluid
FSI with Incompressible Fluid and NONLIN Joint

Envelopes
of Min
Envelopes of Max
-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20
Radial Displacement (cm)
495
515
535
555
575
595
615
635
655
675
695
715
E
l
e
v
a
t
i
o
n

(
m

a
s
l
)
220
200
180
160
140
120
100
80
60
40
20
0
R
e
s
e
r
v
o
i
r

D
e
p
t
h

(
m
)
Static
WG with Reservoir Damping
WG w/o Reservoir Damping
FSI with Incompressible Fluid
FSI with Compressible Fluid
FSI with Incompressible Fluid and NONLIN Joint

Envelopes
of Min
Envelopes of Max
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
-2
0
2
4
6
8
10
12
14
16
18
20
C
r
e
s
t

D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
WG with Reservoir Damping
FSI with Incompressible Fluid
FSI with Compressible Fluid

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
109
Discussion and Conclusion
The maximum static compressive stress is the hoop stress at the dam height on the u/s face
of the main section. It amounts to approximately 7 MPa, which is within the admissible limits
regarding the compressive strength of concrete. The maximum static radial displacement
reaches 8.2 cm at the top of the central cantilever. The maximum compressive stresses during
the ground motion reach approximately 12 MPa in the hoop direction on the u/s face. Seismic
tensile hoop stresses varying between 1 MPa and 2 MPa are computed at the top part of u/s
face of the main section; they disappear if the contraction joint opening/closing is modelled.
The maximum seismic tensile vertical stresses exceed 2 MPA at the dam height on the
upstream face of the main section. The maximum compressive and tensile stresses occurring
during the earthquake are below the admissible limits. The maximum amplitude of the
dynamic vibrations is about 8 cm with respect to the initial displaced shape of the dam.

The hydrodynamic effect modelling approaches investigated in the present study lead to
similar results regarding the structural response of the example arch dam. In general, the
Westergaard added masses approach yields higher compressive and tensile stresses, as well as
higher radial displacements. The compressible fluid analysis results in lower stresses with
respect to the incompressible fluid assumption, which is due to the increased damping of the
coupled system. In fact, according to [11], if the natural dam frequencies are significantly
lower than this of the dam reservoir,

, the behaviour of the latter is similar to the behaviour


of incompressible fluid. In our case

1.98 Hz; this explains to a certain extent the obtained


differences in the results with compressible and incompressible fluid.
References
[1] Zienkiewicz, O.C., Taylor R.L. (2000). The Finite Element Method, Fifth edition, Vol. 1:
The Basis, Butterworth-Heinemann, Oxford.
[2] Westergaard, H.W., (1933). Water pressures on dams during earthquakes. American
Society of Civil Engineers (ASCE), Proceedings.
[3] Zangar, C.N., (1952). Hydrodynamic pressures on dam due to horizontal earthquake
effects. Engineering Monograph No.11, U.S. Bureau of Reclamation.
[4] Tan, H., and Chopra, A.K., (1995). Earthquake analysis of arch dams including dam-
water foundation rock interaction. Earthquake Eng. Struct. Dyn., 24(11), 14531474.
[5] Wang, J. T., and Chopra, A. K. (2010). Linear analysis of concrete arch dams including
dam-water-foundation rock interaction considering spatially varying ground motions.
Earthquake Eng. Struct. Dyn., 39(7), 731750.
[6] Kuo, J. (1982). Added mass computations for incompressible fluid. Report UCB/EERC-
82/09, Earthquake Engineering Research Center, University of California at Berkeley.
[7] Chavez, J.W., Fenves, J.L. (1993). Earthquake analysis and response of concrete gravity
dams including base sliding. Report UCB/EERC-93/07, Earthquake Engineering
Research Center, University of California at Berkeley.
[8] TNO DIANA (2013). DIANA Users Manual, Release 9.4.4. Delft, the Netherlands.
[9] Graz University of Technology, Institute of Hydraulic Engineering and Water Resources
Management (2013). 12-th International Benchmark Workshop on Numerical Analysis of
Dams, Theme A.
[10] Hughes, T. (1987). The Finite Element Method Linear Static and Dynamic Finite
Element Analysis. Prentice-Hall, Inc.
[11] Scurit des ouvrages daccumulation (2003). Documentation de base pour la vrification
des ouvrages daccumulation aux sismes, rapports de lOFEG, srie Eaux, version 1.2.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
111
Dynamic Analysis of an Arch dam with Fluid-
Structure Interaction
Use of an Open source code AKANTU
M. Chambart
1
, T. Menouillard
1
, N. Richart
2
, J.-F. Molinari
2
and R. M. Gunn
1

1
STUCKY SA, Rue du Lac 33, CH-1020 Renens, SWITZERLAND
2
EPFL ENAC LSMS, CH-1015 Lausanne, SWITZERLAND
E-mail: mchambart@stucky.ch, tmenouillard@stucky.ch
Abstract
In this contribution, the fluid-structure interaction is modeled using the added mass technique
and the incompressible fluid model. Results show that the added mass technique is more
conservative. Different meshes are compared with this later method, demonstrating that the
problem is not mesh dependent. Finally, the computation times obtained with two different
software are compared, showing the efficiency of the new open-source software Akantu
Introduction
One of the exercises proposed by the ICOLD for 12
TH
INTERNATIONAL BENCHMARK
WORKSHOP ON NUMERICAL ANALYSIS OF DAMS consists in the dynamic analysis of
an arch dam under a seismic loading. Since the geometry, the material properties and the
loading are imposed to the participants, the focus is put on another modelling aspect which is
the fluid-structure interaction. The methods proposed to model the fluid-structure interaction
starts with the simplest ones, the added mass technique to the most sophisticated ones where
the fluid is explicitly modeled as a compressible body. In our practice, as an engineering
company, our choice for one method versus another is often governed by the gain in accuracy
versus the loss in time. Most of the time in projects, the added mass approach is used because
it is the fastest and usually sufficient to fulfill the authorities demands.
This benchmark is an interesting opportunity to compare different methods in terms of results
but also in terms of computation time. The comparison is carried out using the finite element
software DIANA [1], simply using some of the proposed methods (added mass and
incompressible fluid). As a second step, the results obtained with DIANA using the added
mass technique are compared with the ones given by a new open-source software AKANTU
[2] developed at the Swiss Ecole Polytechnique Fdrale de Lausanne. Akantu is a quite
innovative object-oriented program written in C++. The objective using this latter software is
to optimize the computation time, in order to get reliable results in shorten time, or run
sophisticated computations in an acceptable amount of time. In order to estimate the mesh
dependency of the problem, the results with two meshes are compared with the added mass
technique. Therefore the need in future for High Performance Computing can be assessed.
Numerical Modelling
Model and Mesh
Figure 1 presents different meshes used with the Westergaard theory for dealing with the
hydrodynamic pressure, and also the mesh with the fluid. There are a coarse and a fine mesh,
whose numbers of elements are respectively 2874 and 22280.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
112

Figure 1: View of the meshes: coarse and fine, and with the fluid.
Materials
In this problem, three different materials are involved: the concrete of the dam, the rock of the
foundation, and the water of the reservoir. Their density and stiffness are the valuable data in
linear elastic problem. Table 1 presents the materials parameters. The density of the rock is
numerically taken to be zero in order to avoid wave propagation at the boundary of the rock
foundation. In addition, a dynamic Youngs modulus is used, whose value is 125% the static
one [4].
Table 1: Materials properties.
Properties Concrete Rock Water
Density 2400 kg/m3 0 kg/m3 1000 kg/m3
Poisson ratio 0.167 0.2
Youngs modulus 27 GPa 25 GPa
Loadings
Static loadings.
The self-weight of the concrete is given by the density of the concrete and the gravity.
The hydrostatic pressure on the upstream face of the dam is given by the water level, i.e. 715
m asl.
Dynamics loadings.
The acceleration history of the seismic input is given through three accelerograms presented
in Figure 2, and these three accelerograms are simultaneously applied at the boundary of the
rock foundation. Their corresponding absolute displacement histories are also presented in
Figure 2. Depending on the FEM program used, the boundary conditions may have to be
given in displacement or in acceleration.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
113

Figure 2: Acceleration and displacement time histories.
Fluid-structure interaction
Two alternatives are explored in this section. First, the most basic development to deal with
hydrodynamic pressure on a surface is the method developed by Westergaard [1]; it takes the
hydrodynamic effect into account by adding masses on the upstream surface of the structure.
There is no need to mesh the fluid, thus the size of the system is restricted to the size of the
solids. The added mass is defined according to [1] as:

(1)
Where
w
is the density of the water, h
w
the total height of water and h the distance from the
bottom of the reservoir to the current point where m
w
is evaluated. Figure 3 illustrates the
Westergaard equation.

Figure 3: Westergaard theory on the upstream face of a dam.

Second, a method [3] where the fluid, considered incompressible, is explicitly meshed as a 3D
part of the system is also considered. Solid elements are quadratic, whereas fluid elements are
only linear. Therefore special interface elements are required on the upstream face of the dam
-1.5
-1
-0.5
0
0.5
1
1.5
0 2 4 6 8 10 12 14 16 18 20
A
c
c
e
l
e
r
a
t
i
o
n

[
m
/
s
2
]
Time [s]
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0 2 4 6 8 10 12 14 16 18 20
D
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Time [s]
-1.5
-1
-0.5
0
0.5
1
1.5
0 2 4 6 8 10 12 14 16 18 20
A
c
c
e
l
e
r
a
t
i
o
n

[
m
/
s
2
]
Time [s]
-0.09
-0.08
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0 2 4 6 8 10 12 14 16 18 20
D
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Time [s]
-1.5
-1
-0.5
0
0.5
1
1.5
0 2 4 6 8 10 12 14 16 18 20
A
c
c
e
l
e
r
a
t
i
o
n

[
m
/
s
2
]
Time [s]
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0 2 4 6 8 10 12 14 16 18 20
D
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Time [s]
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
114
to link the fluid to the solid domains. In addition, special boundary elements are involved to
deal with the bottom absorption, the free surface and the far field surface of the reservoir. A
method dealing with compressible fluid is not developed in this paper.
Vibration modes
The vibration modes have been evaluated with different meshes and the Westergaard method
for describing the hydrodynamic pressure. A very good agreement between the results of the
different meshes was observed. The reason is that the linear elastic material model describes
similar dynamic behavior of the structure, not depending on the mesh size. Figure 4 presents
the deformed shape of the first 10 modes within the coarse mesh.

Figure 4: Deformed shapes of the first 10 vibrations modes.

Table 2 presents the vibration frequencies and the mass percentage of the dam in each
direction obtained with the coarse mesh and Westergaard theory. Mode 1 is a left-right motion
with 20% of the mass of the dam. Mode 2 is the major vibration with 35% of the mass in the
upstream-downstream direction.

Table 2: Frequencies and mass percentage for the coarse mesh with Westergaard.
Mode Frequency
[Hz]
Mass percentage
US/DS
Mass percentage
Vertical
Mass percentage
Left-Right
1 1.48 1% 0% 20%
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
115
2 1.50 35% 1% 0%
3 2.27 15% 0% 0%
4 2.67 20% 3% 0%
5 2.92 0% 0% 6%
6 3.38 0% 0% 15%
7 3.73 1% 0% 0%
8 4.25 2% 1% 0%
9 4.54 9% 6% 0%
10 4.64 0% 0% 3%

A damping effect of 5% is taken through Rayleigh coefficients. They are evaluated from the
knowledge of two modes (1 and 13) containing 80% of total mass. The Rayleigh damping
matrix C is defined by two parameters o and | as:
C = o M + | K (2)
where M is the mass matrix, K the stiffness.
With the selected modes (1 to 13), the coefficients are: o=0.726286 and |=0.002346. Figure 5
shows the damping as a function of the frequency; the crosses on the graphs are the different
vibration modes, and the damping of frequencies 1 and 13 are exactly 5%. As it has been
selected, 80% of the mass of the dam has a damping between 4 and 5%. The modes with high
frequency, i.e. greater than 10 Hz, are minor modes in terms of mass percentage.

Figure 5: Rayleigh Damping as a function of the frequency, with the Westergaard theory.
Results
Table 3 presents the different computations run and the corresponding mesh size. With
DIANA, two simulations with Westergaard and one with the fluid were run, whereas only two
simulations were run with AKANTU. All simulation had 2000 steps for the dynamics.
Table 3: Table of the different computations run with the two codes.
DIANA AKANTU
Westergaard With Fluid Westergaard
Mesh Coarse Fine Coarse Coarse Fine
Nodes 3614 25057 14478 3614 25057
Elements 2874 22280 3632 2874 22280
Elapsed time 1h10 7h00 4h00 0h13 5h40
CPU time [s] 13291 s 40752 s 34841 s 2819 s 74000 s

The computations are run for both codes in parallel with the same number of processors
available. However for Akantu the computation is performed on a powerful laptop while for
Diana a workstation optimized to numerical simulations is used. The results show that Akantu
0
1
2
3
4
5
6
7
8
9
10
0 5 10 15
D
a
m
p
i
n
g

[
%
]

Frquency [Hz]
Analytical Coarse Mesh 5%
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
116
is more efficient since the computation time is divided by 5 for the coarse mesh and by 1.25
with the fine mesh.
The evaluation of the hoop, vertical stresses and radial displacement are performed on the
three following sections denoted by Left, Mid and Right sections as shown in Figure 6.

Figure 6: View of the different sections for post-processing the results.
Vertical stresses
The vertical stress is a relevant result within the dam. Thus Figure 7 presents the vertical
stress on the upstream and downstream faces of the Left section of the dam: static, maximum
and minimum results are presented with the coarse and fine mesh and the Westergaard theory,
with Diana and Akantu. Both meshes give similar results too for static, maximum and
minimum vertical stresses. The vertical stress is always negative in static, but its maximum
reaches positive value in some parts during the earthquake.
All four results obtained with the Westergaard theory agree well. However the stress envelops
obtained with the most sophisticated computation, i.e. incompressible meshed fluid, is quite
smaller. The Westergaard theory seems conservative.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
117

Figure 7: Vertical stresses in the different sections.

Hoop stresses
The hoop stress is the other relevant results in terms of stress in an arch dam. Figure 8
presents the hoop stress at the different sections on the upstream and downstream faces of the
dam: static, minimum and maximum stresses are presented for the different computations.
There is a good agreement between the results obtained by both meshes and the Westergaard
theory. One observes that the hoop during the earthquake is mainly negative, so the arch dam
is in compressive stress state in static and also during the earthquake.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
118

Figure 8: Hoop stresses in the different sections.

Principal maximum stresses
Figure 9 presents the maximum principal stress at the different sections on the upstream and
downstream faces of the dam: static, minimum and maximum stresses are presented for the
different computations. There is a good agreement between the results obtained by both
meshes and the Westergaard theory. One observes that the minimum of the maximum
principal stress during the earthquake is close to the static maximum principal stress. Indeed
the maximum principal stress can only increase due to the dynamic loading from the static
case. This maximum principal stress behaves like the vertical stress in the arch dam during
dynamics.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
119

Figure 9: Principal maximum stresses in the different sections.

Principal minimum stresses
Figure 10 presents the minimum principal stress at the different sections on the upstream and
downstream faces of the dam: static, minimum and maximum stresses are presented for the
different computations. There is a good agreement between the results obtained by both
meshes and the Westergaard theory. This minimum principal stress behaves like the hoop
stress in the arch dam during dynamics.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
120

Figure 10: Principal minimum stresses in the different sections.
Radial displacements
The radial displacement quantifies the motion of the dam, in static and during the Earthquake.
Figure 11 presents the radial displacement of the different sections: static, minimum and
maximum obtained with the coarse and fine mesh and the Westergaard theory with both
software and also the displacement obtained with the incompressible fluid. There is a good
agreement between these two results in static and in dynamic. The static radial displacement
at the top of the Mid section is about 8 cm, whereas the maximum during the Earthquake is
about 16 cm, and the minimum about 3 cm.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
121

Figure 11: Radial displacements.
Conclusion
This benchmark allowed to investigate the field of how, in term of numerical method, to deal
with the hydrodynamic pressure on the upstream face of an arch dam. This has then shown
that the Westergaard theory, which consists in simply adding mass on nodes of the upstream
face of the dam, is clearly not mesh dependent. Second, the method dealing with the mesh of
the incompressible fluid, gives smaller envelops in terms of maximum and minimum of
displacement and stresses of the arch dam. In addition, the development of the Westergaard
theory in an open source code did not show any relevant difficulties and the computation time
can be significantly decreased.
References
[1] Water pressures on Dams during Earthquakes, H.M. Westergaard, Transactions, ASCE
98:418-472, 1933.
[2] Fluid-Structure Interactions: added mass computations for incompressible fluid, James
Shaw-Haw Kuo, University of California, Berkeley, August 1982.
[3] Hybrid Frequency Time Domain Validation, TNO DIANA Report, 2008-DIANA-
R003, 2010.
[4] Scurit des ouvrages daccumulation Documentation de base pour la verification des
ouvrages daccumulation aux sismes, rapports de lOFEG, srie Eaux, version 1.2, mars
2003.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
123
Study on Arch Dam Reservoir Seismic Interaction
A. Popovici
1
, C. Ilinca
1
and R. Vrvorea
1

1
Technical University of Civil Engineering, 124 Lacul Tei Bd. 020396-Bucharest,
ROMANIA
E-mail: popovici@utcb.ro
Abstract
The arch dam reservoir seismic interaction is investigated using ABAQUS 6.11 and
DESARC 3.1 software. DESARC computer code offers the advantage of simplicity and
computation speed due to the degrees of freedom based on the stresses (Ritter modified
method) being very recommended for arch dams preliminary structural analysis.
The coarse mesh given by formulator was used for investigation in ABAQUS and 12 arches
equally spaced on dam height were used in DESARC. The water effect was considered
according to added mass procedure as well as acoustic elements. All analyses were performed
in the linear elastic field.
The results are presented in compliance with formulator requests: eigenfrequencies and mode
shapes, hoop stresses, vertical stresses, min./max. principal stresses and radial displacements
in three different sections for static and seismic loads. A special attention is paid to compare
the results concerning arch dam reservoir seismic interaction in different hypotheses
applying two software.
Introduction
The effects of different hypotheses on arch dam - reservoir seismic interaction are investigated
in this paper based on data provided in Theme A by formulator, Graz University of
Technology Institute of Hydraulic Engineering and Water Resources Management [1].
The arch dam is a symmetrical structure with the followings main characteristics:
- maximum height 220 m
- chord length at dams crest 430 m
- valley width at bottom 80 m

The analyses are carried out using ABAQUS 6.11 [2] and DESARC 3.1 [3],[4] software.
The finite element mesh of the arch dam structure-foundation-reservoir system used in
ABAQUS corresponds with alternative coarse mesh given by formulator (Figure 1). The main
features of the coarse mesh are as follows:
- arch dam total number of nodes 2083
- total number of elements 356 (312 C3D20R and 44 C3D15)
- foundation total number of nodes 11608
- total number of elements 2340 C3D20R
- reservoir total number of nodes 12493
- total number of elements 2640 C3D20R

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
124

Figure 1: Finite element mesh of the arch dam structure-foundation-reservoir system used in
ABAQUS code

Other hypotheses used in analyses performed with ABAQUS were the followings:
- the dead weight were applied on monolithic structure
- the reservoir influence on dam seismic response was considered by added mass
procedure excepting the free vibration analysis for full reservoir using acoustic
elements C3D20A for reservoir mesh;
- the water was considered as incompressible with shear modulus tends to zero;
- the seismic response was evaluated for empty and full reservoir by modal
superposition and direct time integration;
- all analyses were based on assumption of linear elastic behavior of materials;
- three accelerograms in direct time integration were applied on the faces of foundation,
respectively x, y and z directions;
- the fraction of critical damping in direct time integration was 5%, and coefficients
in linear Rayleigh model were computed for
1
and
2
, resulting =0.3900 and
=0.0064.
DESARC is an interactive, fast and reliable computer code very recommended for preliminary
fast static and dynamic structural analysis of arch dams. The simplicity and computation
speed of DESARC is due to the degrees of freedom based on the stresses (Ritter modified
method and not on displacements. The program can handle symmetric arch dams both with
circular and parabolic midline, rotationally symmetric structure with liquid inside or outside,
cooling towers etc.
All the geometrical parameters (thickness, radii, half lengths and cantilever shape) are
expressed under assumption of parabolic variation with elevation. Accordingly the real arch
dam shape provided by formulator was equated as assumptions presented above (Figure 2)


Figure 2: Equating real dam shape under assumptions of DESARC

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
125
A number of 12 arches of the dam was considered under analysis (Figure 3). In may be
remark from figures 2 and 3, resulting structure under DESARC assumptions has the shape
very close to the actual dam.


Figure 3: Geometrical features of the arch dam equated under DESARC assumptions

Other hypotheses used in analyses performed with DESARC were the followings:
- the dam-foundation interaction was considered by Vogt coefficients (Er/ Ec=0.90,
Er=25000 MPa and Ec=27000 Mpa);
- the reservoir influence on dam seismic response was considered by added mass
procedure; the hydrodynamic pressures collinear with the direction of horizontal
upstream downstream earthquake (Phd,0) were evaluated with Westergaard
relationship. For other directions (Phd,) was applied the following relation:

P
hd,
= P
hd,0
. cos (1)

being the angle between earthquake direction and the normal to the surface at the point
considered;
- the dead weight were applied on isolated cantilevers and monolithic structure, too.
Some aspects concerning mathematical models
The effect of the water in the reservoir, under the assumption of an incompressible ideal fluid,
is usually calculated by use of the added mass procedure. This is a mathematical artifice used
in order to simplify the analysis of the structure-liquid seismic interaction. The added mass is
determined from the hydrodynamic forces {P
h
(t)} and is attached to the mass of the structure.
Unlike the dead mass of a structure, the added mass acts only on hydrodynamic force
direction, namely the direction of the normal to the surface it is applied. Assuming that the
directions of hydrodynamic forces, of earthquake and of degrees of freedom of the structure
are similar (Figure 4a), the added mass matrix [M
h
] is determined with relation:

{

} [

] {

(2)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
126
where {

is total acceleration response to normal direction at surface in the considered


point.
In this case the analysis of the structure-liquid seismic interaction is performed as for usual
structure according to formula:

[

] {

} [] {

} [] {

} [

] {

} (3)


In the general case when the directions of the normal to surface, of the earthquake and of the
structure degrees of freedom are different, the added masses computed according to (2)
relationship must be projected successively on the normal to surface direction and on the
degrees of freedom of the structure (Figure 4b).



Figure 4: Assessment of added masses

The dynamic equilibrium equations which include the structure-liquid seismic interaction are
written in this general case as follows:

[

] [

] [

] {

} [] {

} [] {

}
[

] [

] [

] {} (4)

where [r
c,n
] has dimensions equal with the number of the degrees of freedom of the system
and contains on diagonal the cosine directors between the normal to surface in the nodes of
the system mesh and earthquake direction
and [r
n,x,y,z
] has dimensions corresponding to the number of the degrees of freedom of the
system and contains on diagonal the cosine directors between the normal to surface in the
nodes of the system mesh and directions of the dynamic degrees of freedom of the system.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
127
Results of analyses
The results are provided as requested by the formulator.
In the Table 1 are presented the natural circular frequencies (, rad/s) and respectively natural
frequencies (f, cycles/s ) values of the first 10 mode shapes including interaction with the
reservoir modeled with acoustic elements C3D20A and added mass procedure computed with
ABAQUS and the value of the frequency of the fundamental mode shape including interaction
with the reservoir by added masses computed with DESARC. It may remark that natural
frequencies computed with ABAQUS by added mass procedure are smaller with about 221
% relative to their counterparts having reservoir interaction modeled with acoustic elements.
In figure 5 is illustrated the fundamental mode shape of the arch dam in hypothesis of the full
reservoir with water elevation at the dam crest.
Table 1: Eigenfrequencies



















Figure 5: Fundamental mode shape view from downstream

Figure 6 illustrates three accelerograms provided by formulator for earthquake analyses
(maximum acceleration 0.1g) and respectively figure 7 response spectra corresponding to
accelerograms used for modal superposition analyses (spectral analyses).

ABAQUS

DESARC
Eigenfrequency-acoustic elements Eigenfrequency-added mass
Mode no. (Rad/s) (Hz) (Rad/s) (Hz) (Rad/s) (Hz)
1 9.726 1.548 7.756 1.235 10.557 1.681
2 9.843 1.567 7.837 1.248
3 12.160 1.935 12.139 1.933
4 14.458 2.301 14.865 2.367
5 15.584 2.480 15.047 2.396
6 19.109 3.041 16.460 2.621
7 19.573 3.115 17.942 2.857
8 20.695 3.294 19.430 3.094
9 22.687 3.611 19.594 3.120
10 23.280 3.705 22.935 3.652
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
128

Figure 6: Accelerograms provided by formulator for earthquake analyses


Figure 7: Response spectra (x,y,z) computed from the accelerograms provided by formulator
and used in spectral analyses

Radial displacements on x, y and z directions in dam central sections at crest and base levels
computed with ABAQUS in direct time integration method are illustrated in figure 8.
Hoop stresses, vertical stresses and radial displacements in the dam central section
downstream/upstream faces generated by combined dead weight + hydrostatic pressures are
illustrated in figures 9a,b. The results in spectral analysis and direct time integration, in the
dam central section, downstream/upstream faces, full reservoir case are comparatively
presented in figures 10a,b. The min./max stresses in the dam central section,
downstream/upstream faces resulted during direct time integration under combined actions of
dead weight + hydrostatic pressures + three-dimensional earthquake 0.1g are illustrated in
figures 11a,b

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
129



Figure 8: Radial displacements in dam central section at crest (top) and base (bottom)
computed with ABAQUS


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
130



Figure 9a: Combined dead weight + hydrostatic pressures Hoop and vertical stresses, radial
displacements on downstream face, central section

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
131



Figure 9b: Combined dead weight + hydrostatic pressures Hoop and vertical stresses, radial
displacements on upstream face, central section

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
132




Figure 10a: Spectral analysis and direct time integration, central section, downstream face,
full reservoir

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
133



Figure 10b: Spectral analysis and direct time integration, central section, upstream face, full
reservoir.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
134



Figure 11a: Min./max stresses, direct time integration, dead weight+ hydrostatic pressure +
three-dimensional earthquake 0.1g
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
135



Figure 11b: Min./max., direct time integration, dead weight+ hydrostatic pressure + three-
dimensional earthquake 0.1g

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
136
Comments on results of analyses
In compliance with some results illustrated in the above captions and other ones resulted from
the analyses carried out the followings comments may be pointed out:
DESARC3.1 a very simply, friendly and fast computer code in all applications led to results
close to those provided by much more sophisticated ABAQUS6.11 computer code. As a result
DESARC computer code is very recommended for arch dams preliminary structural analysis.
Usually, in engineering practice dead loads are applied on arch dam isolated cantilevers,
taking into account that during dam construction the contraction joints are not grouted.
However, even under these conditions, some of the dead weight of cantilevers is transferred to
arches, this quota depending of the dam shape and valley opening. Coming back to the arch
dam provided by formulator it may remark from stresses and displacements to dead loads and
hydrostatic pressures computed with DESARC in both hypotheses of isolated cantilevers and
monolithic structure (fig.8a,b). that the effects of isolated cantilever hypothesis in relation
with monolithic structure are not important. Accordingly, the analyses with ABAQUS, for
simplicity, were carried out only in case of monolithic structure.
The dam displacements due to dead weight reach about 1 cm to upstream at the crest level.
Dead weight is mainly transferred on cantilever. At the bottom of the central section the
vertical stresses reach -7 MPa compression at upstream toe and -1.5 MPa compression at
downstream heel. The hypothesis with isolated cantilever is a conservative one. On arches, the
maximum stress reaches -2 MPa compression at the crest level.
Hydrostatic pressure as independent load generates maximum radial displacement of 7.5 cm
in central section, dam crest elevation. The vertical stresses vary between 1 MPa tension and -
4 MPa compression at downstream face and, respectively between 8 MPa tension at the dam
upstream toe and -2 MPa compression at upstream face. The hoop stresses vary between 0.5
MPa tension, -3.7 MPa compression at downstream face, respectively between 0.5 MPa
tension, -7 MPa compression at upstream face. The big vertical tensile stress generated by
hydrostatic pressure at the dam upstream toe is reduced to 1.5 MPa by vertical compressive
stress in the same point due to dead weight.
Based on results concerning natural periods values computed in full reservoir hypothesis
(Table 1) (fundamental period 0.809 s ABAQUS, added mass, 0.645 ABAQUS, acoustic
elements)it may conclude the hydrodynamic forces computed with Westergaard formula to
generate added masses are higher than hydrodynamic forces developed by acoustic elements.
A comparison between seismic responses computed in the full reservoir hypothesis by
spectral analysis and direct time integration with ABAQUS points out a good correlation
between correspondent displacements response their maximum reaching about 13 cm but
generally significant differences between correspondent stresses. For instance in the dam
central section at downstream face the principal stresses vary between 1.50 MPa (tension)
and -4 MPa (compression) in direct time integration and respectively 0,9 MPa and -9.50 MPa
in spectral analysis.
Maximum displacement obtained in direct time integration at combination static loads (dead
weight + hydrostatic pressures) + earthquake (full reservoir) reaches 18 cm to downstream at
the crest level in central section. Max/min vertical stresses on downstream face varies
between 0.8 MPa and -6.0 MPa . On upstream face at the dam upstream toe the vertical stress
reaches tension of 7 MPa. It is a vulnerable point for cracking. On horizontal direction in the
same point the hoop stresses vary between 3.5 MPa tension and -2.5 MPa compression. It may
conclude that excepting dam upstream toe where exists cracking risk, the dam withstand
earthquake action without notable incident.
As a general conclusion based on data presented above, the seismic response parameters of
arch dams, especially seismic stress state differ significantly function of method of analysis.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
137
Unfortunately, the scarcity of recordings concerning arch dams behavior during strong
earthquakes make difficult to validate a method of analysis using field recordings from.
References
[1] Graz University of Technology THEME A Fluid Structure Interaction Arch Dam
Reservoir at Seismic Loading. 12-th International Benchmark Workshop on Numerical
Analysis of Dams, Graz Austria, 2013.
[2] ABAQUS 6.11. Abaqus / CAE User's Manual. United States of America: Abaqus Inc,
2009.
[3] Fanelli M., Giuseppetti G., Rabagliati U. Il metodo di Ritter Modificato. LEnergia
Eletrica, No.3, 1978.
[4] Fanelli M., Fanelli A. Designers Guide DESARC 3.1 for Windows ISMES S.p.A.,
Bergamo, 1995.
[5] Popovici A., Popescu C. Dams for water storage (in Romanian) Editura Tehnica,
Bucharest, 1992.
[6] Priscu R., Popovici A., Stematiu D., Stere C. Earthquake Engineering for Large Dams
Editura Academiei (Romania) and John Willey & Sons Ltd. (U.K.), Chichester, 1985.
[7] Popovici A., Abdulamit A., Toma I. Moldoveanu T. .Assessment of the seismic safety of
dams based on in site measurements and back analyses. Proceedings XXI-st
International Congress on Large Dams, Montreal, 2003.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
139
Concrete arch dam at seismic loading with fluid
structure interaction
R. Malm
1,2
, C. Pi Rito
2
, M. Hassanzadeh
3,4
, C. Rydell
1,3
and T. Gasch
1,3

1
KTH Royal Institute of Technology, Concrete Structures, Stockholm, Sweden.
2
SWECO Infrastructure, Hydropower and Dams, Stockholm, Sweden
3
Vattenfall Engineering, Stockholm, Sweden.
4
Lund University, Building Materials, Lund, Sweden.
E-mail: richard.malm@byv.kth.se
Abstract
A concrete arch dam have been analyzed during seismic loading with a model based on
acoustic elements to describe the water and infinite elements as quiet boundaries to prevent
wave reflection. The results have also been compared with a simplified model based on
Westergaards added mass approach. The simplified model is only used, in this study, for
comparison with the more advanced model with acoustic elements. Therefore the results from
this simplified model are just used as a rough estimate of the induced stresses and
displacements. Despite this, the simplified Westergaard model gives similar results compared
to the more advanced model with acoustic elements for the water and infinite elements for the
boundaries. The largest difference between the models often occurs at the nodes in the base of
the arch dam, which may be due to poor discretization. Generally, the Westergaard added
mass gives higher maximum principal stresses at the base on the upstream side than the
acoustic model, while often underestimating the min principal stresses at the base on the
downstream side. Both models show high tensile stresses near the base of the arch dam that
would result in cracks.
Finite element models
The studied geometry was given by [1], and in this paper, the mesh denoted as fine mesh in
[1] has been used. All numerical analyses in this paper have been performed with the
commercial software Abaqus ver.6.12. The model consists of a foundation with dimensions
1000 x 1000 x 500 m built up of 2700 second order brick elements with reduced integration.
In one of the models presented in this paper, infinite elements have been used for the outer
surfaces of the foundation in order to remove reflecting waves on the model edges, see Figure
1 a). The arch dam is 220 m high, with a width of 430 m at the crest and 80 m at the base. The
arch dam consists of 2736 second order brick elements with reduced integration (8 integration
points). The geometry of the reservoir is 460 m long (i.e. two times the dam height). In the
following analyses, two different models to account for the water will be presented. In a
simplified model, the reservoir has been replaced with nodal masses corresponding to the
Westergaard added mass approach [2]. In the more advanced model, the reservoir has been
included by means of acoustic elements and infinite acoustic elements to account for an
infinite long reservoir.
Loads and boundary conditions
All analyses have been performed in two steps; static analysis of response due to gravity and
hydrostatic water pressure, and dynamic implicit analysis (transient dynamic analysis) to
calculate the response from induced ground accelerations due to the seismic load.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
140
In all analyses, the gravity load has only been assigned to the elements that constitute the arch
dam. The rock is according to the instructions, [1], defined with zero density. In the simplified
model, based on the Westergaard added mass approach, additional nodal masses are
introduced on the upstream surface of the arch dam. However, it is important that the nodal
masses are not assigned a gravity load since that would affect the static behavior of the dam.
The hydrostatic water pressure has been assigned to the upstream surface of the arch dam, and
to all rock surfaces subjected to the water pressure, as seen in Figure 1 b). The maximum
hydrostatic pressure is defined as 2.1582 MPa (220 m x 9.81 m/s
2
x 1000 kg/m
3
). In the more
advanced model, with acoustic elements, an additional boundary condition with zero acoustic
pressure on the free surface of the water has been defined.


a) b)
Figure 1: a) Geometry and element types of the model, b) applied hydrostatic water pressure.
The ground accelerations are illustrated in Figure 2. In the dynamic analysis, a constant time
step of t = 0.01 s have been used. The upper frequency that can be captured in the analysis,
i.e. the Nyquist frequency, is thereby 50 Hz. According to the frequency analyses that were
performed, the cumulative mass is nearly 100 % for frequencies up to 30 Hz and thereby the
chosen time increment is considered sufficient. In addition, the time history signal of the
earthquake is sampled with t = 0.01 s and thereby the highest reproducible frequency in the
sample is 50 Hz.

Figure 2: Transient time history signals of the ground accelerations.
In the static analysis, all nodes on the bottom surface of the foundation were constrained for
all translations, i.e. in x-, y- and z-direction. In the seismic analysis, these boundary conditions
are replaced, where all nodes on the bottom surface were assigned prescribed ground
accelerations according to Figure 2. The sides of the foundation have not been assigned
boundary conditions in neither the static nor the dynamic analysis. The reason for this is that
0 2 4 6 8 10 12 14 16 18 20
-0.05
0
0.05
Horizontal 1 EQ
A
c
c
e
l
e
r
a
t
i
o
n

[
G
]
Time [s]
0 2 4 6 8 10 12 14 16 18 20
-0.05
0
0.05
Horizontal 2 EQ
A
c
c
e
l
e
r
a
t
i
o
n

[
G
]
Time [s]
0 2 4 6 8 10 12 14 16 18 20
-0.05
0
0.05
Vertical EQ
A
c
c
e
l
e
r
a
t
i
o
n

[
G
]
Time [s]
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
141
these surfaces will be defined with infinite elements in the model with acoustic elements. It is
not possible to define prescribed conditions in the form of accelerations, velocities or
displacements on nodes that are part of infinite elements if the infinite elements are to
function as quiet boundaries. One approach to overcome this would for instance be to
calculate corresponding stresses from the induced accelerations. However, considering the
large geometry of the rock, it is likely that seismic wave incoherence would influence the
results. It is therefore unlikely to assume that the ground vibrations would be identical on all
sides. In addition, the rock is defined with zero mass, and therefore it will not produce mass
inertia forces. Hence, the approximation adopted here, where the boundary is applied only to
the bottom nodes, is considered to be valid for this case.
Material properties and damping
All material properties have been defined by [1]. Rayleigh damping have also been used in all
presented analyses. The damping ratio for concrete has been assumed equal to 4 % according
to Regulatory Guide 1.61, [3]. The damping of the water is usually assumed to be 0.5 %; this
has been judged to be negligible and therefore not included in the analyses.
The lower frequency of the Rayleigh damping have been defined corresponding to the
frequency where 5 % of the cumulative effective mass is active, which in this case is f
1
= 1.27
Hz. This ensures minimal underestimation of response in the low frequency range according
to [4] and [5]. The upper frequency have been defined corresponding to the frequency where
80 % of the cumulative effective mass is active, which in this case is f
2
= 9.76 Hz. The
corresponding damping values
1
and
2
have been calculated so that the minimum damping
in the interval f
1
f
2
is 1 % less than the target damping value, i.e.
min
= 3 %. Based on this
approach, the obtained Rayleigh damping coefficients are = 0.66314 and = 0.0013503 with
the corresponding curve illustrated in Figure 3. The average damping in the interval f
1
f
2
is
3.6 %, i.e. slightly lower than the target value of 4 % and thereby conservative.

Figure 3: Rayleigh damping curve ( = 0.66314 and = 0.0013503)

The Rayleigh damping coefficients have been defined in the material definition for concrete
but also for the rock. Since the rock have zero density, only the stiffness proportional part of
the Rayleigh damping (= 0.0013503) will be active. The Rayleigh damping curve has been
verified against a modal analysis with uniform damping of 4 %, where the Rayleigh damping
gave slightly conservative result.
Model 1: Westergaard added mass approach
According to Westergaard [6], the hydrodynamic forces exerted on a dam due to earthquake
ground motions are equivalent to inertia forces of a volume of water attached to the dam
moving back and forth with the dam while the rest of the reservoir water remains inactive.
The influence of the reservoir on a 2D rigid monolith with vertical upstream face is included
by introducing the impulsive mass of the water and thereby, altering the dynamic properties of
0 5 10 15 20 25 30
0
0.02
0.04
0.06
0.08
0.1
Frequency (Hz)
D
a
m
p
i
n
g

Em = 5 % Em = 80 %
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
142
the structure. Kuo [7], further developed the method to also account for the curvature of an
arch dam, this have however not been considered in this paper. The reason for this is that the
simplified added mass model is only used, here, for comparison with the more advanced
model with acoustic elements. Therefore, the results from this simplified model are just
considered as a rough estimate of the induced stresses etc. A MatLab script have been
developed to calculate the tributary surface area for each node on the upstream surface, and
based on this the nodal mass has been calculated according to [6]. The calculated nodal
masses are illustrated in Figure 4, where the colors in the figure represent the added weight (in
kg) for each node on the upstream surface of the arch dam.

Figure 4: Nodal masses based on Westergaard added masses
Model 2: Acoustic and infinite elements
In the more advanced model, the water has been included by means of 2640 second order
acoustic elements. These elements are based on an acoustic formulation with the fluid wave
velocity, i.e. the variation in pressure, as the independent variable. These elements do not
include any terms for body forces and therefore, the hydrostatic pressure needs to be included
as a pressure load on the structure, as previously shown in Figure 1 b). The finite element
mesh of the acoustic elements is stationary at all nodes except at the boundaries of the fluid
domain, i.e. the fluid-structure interface. For boundaries adjacent to a structural domain, the
nodes of the acoustic medium can be prescribed to follow the nodes of the structural domain,
giving a pressure change in the acoustic medium. A pressure boundary which prescribes zero
acoustic pressure on the free surface has been defined, due to the lack of displacement DOF.
This gives no actual displacement of the free surface but is correct in the sense of wave
propagation in the medium. In order to account for an infinite long reservoir, 264 additional
infinite acoustic elements are defined on the upstream side of the reservoir. In order to prevent
reflecting waves on the model edges, 510 infinite elements have been defined for the surfaces
on the foundation rock as illustrated in Figure 1 a). The infinite elements provide quiet
boundaries and are based on Lysmer and Kuhlemeyer [8] for dynamic response. The damping
constants for longitudinal and shear waves are calculated as
( ) 2
p p
d c G = = + (1)
s s
d c G = = (2)

where,
( )( ) 1 1 2
E v

v v

=
+
,
( ) 2 1
E
G
v
=
+
and E is the elastic modulus, is poisons ratio and
c
p
and c
s
are the longitudinal and shear wave speed.
The infinite rock elements were defined with the same properties as rock (see [1]), with the
addition of a defined density of 2600 kg/m
3
, required for the damping. The bottom surface
-200 -150 -100 -50 0 50 100 150 200
450
500
550
600
650
700
750
Westergaard added mass


710000
1.42e+006
2.13e+006
2.84e+006
3.55e+006
4.26e+006
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
143
was defined without infinite elements, due to the fact that the seismic ground motion was
applied in the base. The length of the infinite rock elements are 1000 m, i.e. chosen as the
total length of the original foundation. The infinite elements are active in static analyses, by
providing horizontal stiffness to the foundation.
Eigenfrequencies and modes
The 10 first mode shapes obtained from the acoustic model are illustrated in Figure 5.


e
)

M
o
d
e

5



j
)

M
o
d
e

1
0
)

F
i
g
u
r
e

5
:

T
h
e

1
0

f
i
r
s
t

e
i
g
e
n
m
o
d
e
s

o
b
t
a
i
n
e
d

f
r
o
m

t
h
e

a
c
o
u
s
t
i
c

m
o
d
e
l
.


d
)

M
o
d
e

4



i
)

M
o
d
e

9



c
)

M
o
d
e

3



h
)

M
o
d
e

8



b
)

M
o
d
e

2



g
)

M
o
d
e

7



a
)

M
o
d
e

1



f
)

M
o
d
e

6


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
144

In some cases, the acoustic model gives two closely spaced frequencies with the same mode
shape. In Table 1, these are therefore presented together as one mode. In addition, some
modes in the acoustic model are mainly relevant to the water, and hence these are not
presented here. It can however be mentioned that the first mode corresponding to the water is
about 0.7 Hz. The frequencies for each mode is also presented from the Westergaard added
mass approach and the original model without water for comparison in Table 1.
Table 1: Calculated eigenfrequencies for the ten first modes.
Frequency (Hz)
Mode Acoustic elements Added mass W.o. water
1 1.5114 1.3730 2.0002
2 1.5431 1.3291 1.9012
3 1.9056 1.9850
4 2.2255 2.3534 2.054 2.8785
5 2.4254 2.4665
6 2.9626 2.9997 2.5633 3.5491
7 3.0084 3.1169 2.5782 3.5998
8 3.2818 3.3150 2.8773 4.1744
9 3.5886
10 3.7748 3.2224 4.4779

As seen in the table, the acoustic model gives frequencies that are in-between the case without
water and the case with Westergaard added mass. Notable is also that the first and second
mode have change place in the acoustic model compared to the others models.
Stresses
Stresses are presented for three sections, Left section (-45), Main section (0) and Right
section (+45) from the center of the arch, as defined by [1]. The following stresses are
presented; maximum and minimum principal stresses, hoop stresses and vertical stresses. For
each of these stresses, the max and min envelope for each node on the upstream (US) or
downstream (DS) surface are presented. This means that the presented stresses are not
occurring at the same time; instead these are the maximum or minimum stresses that occur for
the whole time period for each node on the upstream or downstream side of each of the
studied sections. As seen in Figure 6 there is quite similar results between the simplified
Westergaard added mass method and the method with acoustic and infinite elements. The
minimum principal and hoop stresses show, generally, better resemblance between the two
models than the max principal and vertical stresses. The largest difference between the models
often occurs at the nodes in the base of the arch dam, which may be due to poor discretization.
Generally, the Westergaard added mass gives higher maximum principal stresses at the base
on the upstream side than the acoustic model, while underestimating the min principal stresses
at the base on the downstream side. Both models show high tensile stresses near the base of
the arch dam that would result in cracks. It can from this comparison not be drawn any
conclusion that the simplified Westergaard added mass approach would be more conservative
than the method with acoustic and infinite elements. The analyses also show that, despite the
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
145
symmetric shape of the arch dam, there is a difference between the results at the left and right
section, as seen in Figure 7.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
146
Figure 6 Stresses in the main section (0) according to the model with acoustic elements and
the simplified Westergaard model, on US and DS surface respectively.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
147
Figure 7: Stresses in the left (-45) and right section (+45) according to the model with
acoustic elements on US and DS surface respectively.
Displacements
The max and min envelope of the relative radial displacements are presented in Figure 8 for
the Main section (0). In figure, the relative displacement presented i.e. the rigid body
displacement due to the earthquake has been removed. In the figure, a comparison between
the acoustic model and the simplified Westergaard model is shown. The models give similar
results, but the acoustic model gives larger deflections.


Figure 8: Displacements in the main section (0) on US surface.
Discussion
A concrete arch dam have been analyzed during seismic loading with a model based on
acoustic elements to describe the water and infinite elements to describe quiet boundaries to
prevent wave reflection. The results from this model have also been compared to a simplified
model based on Westergaards added mass approach. The simplified model is only used for
comparison and should therefore be considered as a rough estimate of the induced stresses and
displacements. The performed analyses have showed that there are several factors that have
been assumed in this study that influence the results, such as
- Rayleigh damping the choice of damping ratio for the different materials and
especially the choice of corresponding frequencies influence the results.
- Seismic excitation in these analyses, only the bottom of the foundation is subjected
to the prescribed seismic excitation. Effects such as seismic wave incoherence have
not been considered.
- Infinite boundaries in the analyses infinite boundaries have defined for both the
foundation but also for the reservoir.

The results from the two models are quite similar, where the minimum principal and hoop
stresses generally are more similar between the two models than the maximum principal and
-0.05 0 0.05 0.1 0.15
500
520
540
560
580
600
620
640
660
680
700
x-dir (radial)
upstream
Deflection (m)
E
l
e
v
a
t
i
o
n

(
m
)


Max (Westergaard)
Min (Westergaard)
Max (Acoustic)
Min (Acoustic)
Static
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
148
vertical stresses. The largest difference between the models often occurs at the nodes in the
base of the arch dam, which may be due to poor discretization of the mesh at these points.
Generally, the Westergaard added mass gives higher maximum principal stresses at the base
on the upstream side than the acoustic model, while often underestimating the min principal
stresses at the base on the downstream side. Both models show high tensile stresses near the
base of the arch dam that would result in cracks. The analyses also show that, despite the
symmetric shape of the arch dam, there is a difference between the results at the left and right
section.
Acknowledgements
The research presented was carried out as a part of "Swedish Hydropower Centre - SVC".
SVC has been established by the Swedish Energy Agency, Elforsk and Svenska Kraftnt
together with Lule University of Technology, The Royal Institute of Technology, Chalmers
University of Technology and Uppsala University. www.svc.nu.

The presented work was financially supported by SVC, Sweco Infrastructure, KTH Royal
Institute of Technology and Vattenfall Engineering. The authors would like to thank these
organizations for making our participation in this benchmark workshop possible.
References
[1] ICOLD, 2013. 12
th
International Benchmark Workshop on Numerical Analysis of Dams.
Theme A Fluid Structure Interaction. Arch Dam Reservoir at Seismic loading. Graz
University of Technology.
[2] Westergaard, H. M. (1933). Water Pressures on Dams During Earthquakes. Transactions,
American Society of Civil Engineers, Vol 98.
[3] US NRC (2007). Damping values for seismic design of nuclear power plants. Regulatory
Guide 1.61. U.S. Nuclear Regulatory Commission.
[4] ASCE 4-98, 1998, Seismic Analysis of Safety-Related Nuclear Structures and
Commentary, American Society of Civil Engineers, Reston, Virginia.
[5] ASCE/SEI 43-05, 2005, Seismic Design Criteria for Structures, Systems, and
Components in Nuclear Facilities, American Society of Civil Engineers, Reston,
Virginia.
[6] EM1110-6051 (2003). Time-History Dynamic Analysis of Concrete Hydraulic
Structures. US Army Corps of Engineers.
[7] Kuo, J.S.-H. (1982). Fluid-structure Interactions: Addedd Mass Computations for
Incompressible Fluid. Report No. UCB/EERC-82/09. Earthquake Engineering Research
Center, University of California, Berkeley.
[8] Lysmer, J., and R. L. Kuhlemeyer, Finite Dynamic Model for Infinite Media, Journal of
the Engineering Mechanics Division of the ASCE, pp. 859877, August 1969.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
149
Fluid Structure Interaction
Arch Dam Reservoir at Seismic loading
A solution using FEnas
M. Brusin
1
, Dr. J. Brommundt
1
and H. Stahl
1

1
AF-Consult Switzerland Ltd., Tfernstrasse 26, CH-5405 Baden, SWITZERLAND
E-mail: marko.brusin@afconsult.com
Abstract
This paper represents a contribution to the ICOLD 12
th
International Benchmark Workshop
on Numerical Analysis of Dams, THEME A - Fluid Structure Interaction Arch Dam
Reservoir at Seismic loading. The formulated problem, modeling of a 220 m high and 430 m
wide arch dam under static and seismic loads, was addressed by the finite element code
FEnas. FEnas is a Swiss code originally developed at the Imperial College London. The
interaction between the dam and the water in the reservoir was implemented using the added
mass technique in the general form following Westergaard.
The simulations performed included a comparative analysis of two meshes with different
spatial discretization. The computed results show that the impact of the discretization is rather
small. Moreover, the results of the analyses have been compared to typical deformations
observed at Swiss dams and similar studies on existing dams. All results here are in good
agreement with preceding studies of this kind.
Introduction
Problem Definition
The problem to be analyzed and solved here is exhaustively described in the problem
description provided by the workshop organizers [1], who are referenced as the formulator
in the following. The stresses and deformations of a dam interacting with the water in its
completely filled reservoir under earthquake load shall be simulated. As the title of the
benchmark workshops says, this interaction is the focus of this study. Consequently, typical
further challenges in studies of this type are handled in a pragmatic way, e.g. all simulations
assume linear elastic material behavior. The earthquake is represented by three correlated time
series of acceleration in the three dimensions, lasting 20 seconds in total. Two FE meshes are
provided, one in coarse and one in fine resolution.
Finite Element Code FEnas
The finite element analysis of the dam was performed using FEnas ECCON IPP/ Version
2.9.3 2012/08/24. FEnas which was initially developed at the Imperial College London and
was continuously enhanced and integrated with a graphical user interface by Walder & Trueb
Engineering AG, Switzerland [2]. FEnas is capable of solving a wide range of static and
dynamic load problems including consideration of thermal impact. The dimension of the
problems solvable in FEnas is constrained by aspects of computation only. Nonlinearity in
terms of material behavior and geometry can be considered. The typical field of application of
FEnas is in design of buildings and concrete structures, including tunnels and dams. The
FEnas element library includes all element types necessary for this type of analyses. However,
acoustic or fluid elements are not included and cannot be modeled.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
150
Modelling Concept and Implementation
General Assumptions and Approach
The software package FEnas allows to analyze the problem of interaction of fluids and
structures under seismic impact using the added mass technique. This technique is considered
to be a conservative approach and gives satisfactory results in engineering practice [3]. The
mass of water which oscillates along with the structure is added as a discrete mass at all
element nodes of the upstream face of the arch dam as described below.
All analyses were carried out for the two meshes provided by the formulator, i.e. a coarse and
a fine mesh. For each mesh the following three calculations were performed:
- Linear static analysis for the self-weight load and hydrostatic pressure load.
- Calculation of eigenvectors of oscillation for the first 10 tones, including the added
mass of water.
- Linear dynamic analysis by using direct time integration, also including the added
mass of water.
For further analysis of the results a superposition of the results of static and dynamic analyses
was performed as described below.
All approaches to the modeling, loads, boundary conditions, and material parameters are
identical for the simulations with the two meshes. Considering this, if not stated otherwise, the
following discussions refer to the fine mesh; references to the coarse mesh are explicitly
marked.
Mesh Properties
The mesh was delivered by the formulator of the benchmark competition. Two adaptations
were performed:
4. The discretization of the terrain of the coarse mesh did not coincide with the discretisation
of the dam at their mutual interface. One line of elements had to be corrected to create a
consistently meshed interface.
5. The discretization of the terrain of the fine mesh required the same adaptation. Moreover,
the discretisation of the terrain was not adapted to the refined discretisation of the dam.
Therefore, the topmost layer of elements of the terrain mesh was refined to gain a
consistent discretisation of dam and terrain at their interface.
We consider the impact of these adaptations as minor. Their big advantage is that no code
internal treatment and definition of the interfaces with different discretisation is needed but
the meshes can be used as such. Due to the selected approach with the added mass, the
developed finite element mesh for the reservoir provided by the formulator was not taken into
account. The table below provides information about the mesh parameters. Figures of the
mesh discretization are given further below.
Table 1: Mesh parameters of the adapted meshes
Total number of
Coarse mesh Fine mesh
Dam Terrain Dam Terrain
nodes 2083 11608 13733 15062
elements 356 2340 2736 3120
Hexahedron-quadratic elements with 20 nodes 312 2340 2736 2700
wedge-quadratic elements with 15 nodes 44 0 0 420

Material Parameters and Boundary Conditions
The formulator provided the following linear elastic isotropic material characteristics.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
151
Table 2: Material parameters
Parameter Dam Rock Mass (Terrain)
Density 2400 kg/m
3
massless
Poisson ratio 0.167 0.2
Youngs modulus 27000 MPa 25000 MPa

Considering that the outer limits of the modeled terrain are sufficiently far away from the
dam, the boundary conditions at the outer surface and the bottom of the terrain were
determined as springs with large stiffness that practically prevents any movement in all three
directions (quasi fixed bearing).
Modeling approach for added mass technique
The mass and its distribution are relevant when calculating the natural frequencies of a
structure as well for the dynamic analysis during the earthquake. The specific mass is
calculated from density, and is in FEnas automatically taken into account in a physical
manner.
In order to take into account the interaction of fluids and structure, it is necessary to add a
mass of water which oscillates together with the dam. Water joined to the body dam is added
to the points of the upstream face as an additional concentrated mass according to
Westergaards approach in its generalized form [3, 4]. Westergaard showed that the
hydrodynamic pressures exerted on the face of the dam due to the earthquake ground motion
is equivalent to the inertia forces of a body of water attached to the dam and moving back and
forth with the dam while the rest of reservoir water remains inactive. He suggested a parabolic
shape for this body of water with a base width equal to 7/8 of the height. In the generalized
Westergaard method [3, 4] normal hydrodynamic pressure P
n
at any point on the curved
surface of the dam is proportional to the total normal acceleration shown in figure below, i.e.:

(1)

) Z H ( H
8
7
i i i w i
= o (2)


Figure 1: Westergaard hydrodynamic pressure [3]

Where:

- Total normal acceleration at point i


i
o - Westergaard pressure coefficient at point i
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
152
w
- Mass density of water
i
H - Depth of water at vertical section that includes point i
i
Z - Height of point i above the base of the dam
The normal pressure P
n
at each point is then converted to an equivalent normal hydrodynamic
force by multiplying by the tributary area associated with that point. This procedure is semi-
automated in FEnas, the user has to define the surface load according to Westergaard, while
the conversion from surface pressure to added mass at nodes is done automatically by the
program. Since the Westergaard coefficient depends on the depth of water in front of the
upstream face, which varies with dam side, the value H is taken in accordance with the colors
on the next figure, one average value of H correspond for each color.

Figure 2: Westergaard added masses technique in FEnas (left: zones of similar H, right,
hydrodynamic forces as vectors, colours show magnitude of force in kN/m
2
)

Modelling Approach for Self-Weight and Hydrostatic Load
Deadweight or self-weight refer to the same load case. Application of this load case generally
mimics the construction schedule. Here, a pragmatic two step modeling approach is
implemented. Superposition of the two steps represents deadweight. In the first step, every
second cantilever is modeled massless with a very low Youngs modulus, whereas all other
cantilevers are modeled by their correct properties (cf. Table 2). In the second step, this
procedure is repeated exchanging the material parameters between the two sets of cantilevers.
For the further analysis, the joints are considered as grouted and the entire construction retains
hydrostatic pressure as arch dam.
Hydrostatic load has been defined as surface load perpendicular to the surface of all elements
at the upstream surface. The magnitude of the surface load depends on the water level with
respect to the vertical position of the element.
Modelling Approach for Determining the Eigenfrequencies
The formulator required calculation of the first 10 eigenfrequencies of the structure; this
includes the interaction with the reservoir, i.e. the eigenfrequencies of the dam together with
the added mass of water are to be calculated.
The lowest natural frequencies are determined by resolving the characteristic value or
eigenvalue problem. Here, only iterative methods can be used to solve eigenvalues problems.
FEnas uses the subspace iteration method. The modes of oscillation are estimated for a
number of start vectors (the subspace) and then iteration is used for continuous improvements.
The minimum number of start vector required is equal to the number of requested
eigenvalues; the maximum number is equal to the number of degrees of freedom of the entire
FE structure.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
153
Modelling Approach for Transient Seismic Loads
Linear dynamic analysis was performed using direct time integration (time history analysis).
As method for integration of the dynamic equations of motion we have used the Newmark
method. The accelerograms given by the formulator were used as such and are plotted in the
following figure.

Figure 3: Accelerograms in x, y, z direction

The acceleration values are given on the interval of 0.01 s. All performed simulation used
time steps with the same duration. With a total of 2000 time steps of 0.01 s, the total duration
of the earthquake is 20 s.
The dynamic analysis is taking into account the Rayleigh damping effect. Raleigh dumping is
based on assumption that damping is proportional to the stiffness and mass matrix of the
structure. Here, a damping of 5% for the first two eigenfrequencies was considered.
Results
Eigenfrequencies
Table 3: Eigenfrequencies
Mode No.
Coarse mesh Fine mesh
Frequency [Hz] Period [s] Frequency [Hz] Period [s]
1 1.25 0.8000 1.26 0.7937
2 1.34 0.7463 1.32 0.7576
3 2.05 0.4878 2.01 0.4975
4 2.33 0.4292 2.36 0.4237
5 2.47 0.4049 2.50 0.4000
6 3.01 0.3322 3.00 0.3333
7 3.15 0.3175 3.17 0.3155
8 3.66 0.2732 3.65 0.2740
9 3.70 0.2703 3.70 0.2703
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
154
10 3.88 0.2577 3.88 0.2577
Mode Shape

Figure 4: Radial deformation [mm] with mode shapes of the eigenfrequencies exaggerated by
a factor of 17000
Stress Analysis
All evaluations of results of stresses were done for vertical stresses, hoop stresses, and where
considered useful for principal stresses o
1
and o
3
. Note that the vertical stresses shown are not
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
155
vertical with respect to a global coordinate system but vertical with respect to the element
coordinate system. This means that the stresses shown are parallel to the element and dam
surface.
Stresses due to Static loads
Stresses of static loads are presented separately for the load of self-weight and for the
combination of self-weight and hydrostatic pressure load. They are presented for the upstream
and downstream surface allowing checking if the model results are plausible.

Figure 5: Vertical and hoop stresses [MPa] due to self-weight


Figure 6: Vertical and hoop stresses [MPa] due to self-weight and hydrostatic pressure
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
156
Seismic stress-analysis depending on the time
Monitoring was performed for the vertical and hoop stresses for selected elements on the
upstream and downstream face of the dam. There were selected three elements of both
surfaces in three different heights at main cross section defined by the formulator, which is
considered later (cf. Figure 10). On the figures below you can see the position of the elements
at which stresses are monitored. Each of the figures shown below includes minimum (MIN)
and maximum (MAX) stress observed at the selected elements (each of the selected elements
includes 20 nodes). Note that these are the stress due to seismic loading only, self-weight and
water load are not directly considered, water to a certain extend as the seismic loading affects
the Westergaard masses as well.

Figure 7: Monitored elements on main cross section


Figure 8: Time series of stresses observed at monitored elements on upstream surface

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
157

Figure 9: Time series of stresses observed at monitored elements on downstream surface
Evaluation of the stresses for three different cross sections
For three cross-sections, one in the center and two at each side halfway between the center
and the right respectively left end of the dam, stresses over dam height are shown for the
upstream and downstream surface. The three examined cross-sections are marked on the
figure below.


Figure 10: Layout of analyzed cross sections

The following loads respectively load combinations are plotted:
- Self-weight load - [SW]
- Self-weight + hydrostatic pressure load Normal operating conditions [NOC]
- Maximum of self-weight + hydrostatic pressure + seismic load [MAX]
- Minimum of self-weight + hydrostatic pressure + seismic load [MIN]
Main section Right section
Left section
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
158
It should be noted that the last two evaluations, maximum and minimum, represent envelopes
of stresses, i.e. this state of stress is artificial, because the maximum / minimum do not occur
at all nodes at the same time.

Figure 11: Stresses for main cross section

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
159

Figure 12: Stresses for left cross section

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
160

Figure 13: Stresses for right cross section
Radial Deformation
The evaluation of results of radial deformation was done considering the following sequences:
- Radial deformation due to the static loads
- Seismic radial deformation-analysis depending on the time
- Radial deformation analysis for three cross-section (cf. above)
Radial deformation of the static loads
Radial deformations of the static loads are presented separately for the load of self-weight and
for the combination of self-weight and hydrostatic pressure load.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
161

Figure 14: Radial deformation [mm] self-weight (left), self-weight + hydrostatic pressure
(right)
Seismic radial deformation-analysis depending on the time
Monitoring was performed for the radial deformation for selected nodes on the upstream face
of the dam. There were selected three nodes of upstream surfaces in three different heights at
the three cross section defined by the formulator (see Figure 10). On the figures below you
can see the position of the nodes at which radial deformation are monitored. Note that these
are the radial deformations due to seismic loading only, self-weight and water load are not
directly considered, water to a certain extend as the seismic loading affects the Westergaard
masses as well.

Figure 15: Monitored nodes on upstream surface


Figure 16: Radial deformation for left, main and right cross section respectively
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
162
Radial deformation at the three cross sections
For the three cross-sections the radial deformations over height are plotted. Evaluation of the
radial deformation is given for the following loads respectively combination of loads:
- Self-weight load - [SW]
- hydrostatic pressure load Normal operating conditions [NOC]
- Maximum of hydrostatic pressure + seismic load [MAX]
- Minimum of hydrostatic pressure + seismic load [MIN]
These combinations mimic the observations of a pendulum.

Figure 17: Radial deformation main cross section


Figure 18: Radial deformation left cross section


Figure 19: Radial deformation right cross section
Discussion and Summary
Stress-strain analysis of arch dam during seismic activity including reservoir-structure
interaction using added masses technique according to generalized Westergaard approach is
presented. Interaction of an arch dam with the impounded water leads to an increase in the
dam vibration periods. The fact that water moves with the dam increases the total mass that is
in motion. This added mass increases the natural periods of the dam, which in turn affects on
the effective earthquake inertia forces. The Westergaard method usually gives the largest
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
163
added-mass values, which is evident by its increasing the vibration periods the most.
However, this does not automatically give the largest stresses, because response of the dam
also depends on the characteristics of the earthquake ground motion. Although Westergaard
approach is widespread in practice, for different variants of quasi-static and dynamic analysis
of seismic effects, the above-mentioned facts lead to the conclusion that the results must be
taken with caution.
After comparison of two different sizes of spatial discretization of meshes we can conclude,
that the coarser mesh also gives satisfactory results which can be used in engineering practice.
In fact, the differences are small. Whereas the performed simulations with linear material
behavior are not very costly in terms of computation time and required computer power for
both meshes, the coarser mesh could be very useful as soon as more complex processes, e.g.
non-linear material behavior, are to be considered.
Maximum earthquake stresses are located in the central upper portion of the dam as well
along the dam-foundation contact zone. Maximum radial deformation occurs in the crest, and
decreasing towards lateral sides, as well as towards fixed end.
The vertical stresses observed at the upstream surface show some artefacts for the load cases
with the earthquake as they are supposed to be zero at the top of the dam like it is observed for
self-weight and self-weight with hydrostatic pressure. These non-zero values are considered to
be artefacts. The same artefact in a much smaller magnitude is seen at the top of the
downstream surface.
Stress time histories curves shows several significant tensile stress cycles that can lead to open
the contraction joints, especially in crown zone, while on the time histories of radial
deformation in addition to timing of deformation could be noted as dominant periods of the
first few natural frequencies.
The results that are shown have been compared to typical deformations observed at Swiss
dams. The overall behavior of the dam here is in good agreement with what AF has seen in
reality. The comparison of the results of the earthquake simulations with previous studies of
this type performed at AF was positive.
We can conclude that the added mass technique in combination with the direct dynamic
analysis relatively quickly and easy produce results useful in engineering practice.
Acknowledgements
The FEnas developer and support team around Dr. Urs Treb provided valuable support in the
conversion of the meshes into the FEnas mesh format.
References
[1] Graz University of Technology, Institute of Hydraulic Engineering and Water Resources
Management, 12
th
International Benchmark Workshop on Numerical Analysis of Dams,
2.-4. OCTOBER, 2013, GRAZ AUSTRIA, THEME A - Fluid Structure Interaction
Arch Dam Reservoir at Seismic loading. Download from
http://portal.tugraz.at/portal/page/portal/Files/i2130/Icold_bmws2013/Theme_A/Files_an
d_Information_Theme_A.zip, 23.07.2013
[2] FENAS, Walder + Treb Engineering AG. (2013). Walder + Treb Engineering AG.
http://www.waldertrueb.ch/index.php?id=23&L=2%3Fcomply, 23.07.2013
[3] Ghanaat, Y. (1993). Theoretical Manual for Analysis of Arch Dams. US Army Corps of
Engineers. http://www.dtic.mil/dtic/tr/fulltext/u2/a269682.pdf, 24.07.2013
[4] Kuo, J. (1982). Fluid-structure interactions: Added mass computations for incompressible
fluid. Earthquake Engineering Research Center. Berkeley: University of California.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
165
Fluid Structure Interaction
Arch Dam-Reservoir at Seismic Loading
S. Shahriari
1

1
Institute of Hydraulic Engineering and Water Resources Management, Stremayrgasse 10/2,
A-8010 Graz, AUSTRIA
E-mail: shervin.shahriari@student.tugraz.at
Abstract
In the present paper, a direct time domain procedure is used for dynamic linear analysis of the
coupled arch dam-reservoir-foundation system. The hydrodynamic force on the upstream face
of the dam is modeled by added mass method and compressible fluid elements for
comparison. The concrete and massless foundation rock was assumed to be linear elastic.
Connection between dam and foundation is modeled by coupling all DOF of the
corresponding nodes. Also, viscous damping was applied to the materials by using Rayleigh-
Damping. Numerical results showed that the different modeling techniques of the interaction
lead to different response of the system and stress distribution in the dams body. Modeling
the interaction by added mass method, increases the period of the system as well as
overestimating or underestimating the stresses compare to the model with fluid elements.
Conclusion
The following conclusions are drawn based on the numerical experiments conducted herein:

- Choosing appropriate damping parameters is a very important step in the dynamic
analysis of the dam-water-foundation system. Underestimating the systems damping
by choosing higher than 7
th
angular frequency as the second frequency for calculation
of Rayleigh-Damping parameters can lead to overestimating the stresses in the model.

- Modal responses of the dam-water-foundation system are calculated using the finite
element software, ANSYS. The results showed that the eigenfrequencies from the
model with added mass approach were lower than those calculated from the model
which utilized acoustic element to model the reservoir.

- Direct time integration procedure is used for dynamic analysis of the system and
deformations and stresses are calculated for static and dynamic load cases. Maximum
deformation due to static loading was 7.4 cm at the crest level followed by the
maximum deformation of 13.6 cm after dynamic analysis. The tensions in arch dams
are not desirable; therefore, the tension stresses were studied. Evaluation of hoop
stresses indicated that there is no significant tension developed in the main section
with maximum value of 0.6 MPa. Furthermore, the maximum tension stress value in
the U/S face of the right and left sections is higher than the main section(S=3.6 MPa).
Vertical stresses evaluation showed that the maximum tension was occurred at the
base of the U/S face of the main section and found to be 6.69 MPa. Also, the
maximum compression stress is developed (S=12.5 MPa) at 180 m from the
foundation level of the U/S face of the main section.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
167
The Seismic behaviour of an Arch Dam-Reservoir-
Foundation System
A. Frigerio
1
and G. Mazz
2

1
Ricerca sul Sistema Energetico RSE S.p.A., via R. Rubattino 54, Milan, ITALY
E-mail: antonella.frigerio@rse-web.it
Abstract
The paper summarizes the main features and the related results of the linear static, modal and
seismic analyses carried out on the concrete arch dam proposed as Theme A of the 12
th

ICOLD International Benchmark Workshop.
The analyses have been performed by means of the COMSOL Multiphysics software, making
reference to the coarse finite elements mesh provided by the Formulators, but some slightly
changes.
In the modal and dynamic analysis, the water-dam-foundation interaction has been taken into
account. The impounded water has been modeled by means of acoustic finite elements,
assuming appropriate boundary conditions to simulate the wave reflecting conditions between
the fluid and the foundation, as well as the fluid and the upstream face of the dam, and the
non-reflecting condition at the end of the reservoir channel.
Results have been provided according to the requests of the Formulators in terms of radial
displacements, hoop and vertical stresses on three vertical sections of the dam.
Introduction
Seismic analyses have been proposed and discussed during some of the first Benchmark
Workshops organized by the Committee on Computational Aspects of Analysis and Design of
Dams [1,2]. Hence, it appears appropriate the proposal of the present seismic analysis of a
large arch dam, [3], in order to verify the possible improvements both on the modeling and on
the computational aspects.
In the present paper, all the analyses have been carried out with COMSOL Multiphysics, a
software that allows modeling and simulating any physics-based system. In particular, the
COMSOL Acoustic, [4], and Structural Mechanics Modules, [5], have been used.
In the following paragraphs the geometrical and physical model will be described and the
results will be provided according to the requests of the Formulators.
Geometrical and physical model
The coarse Finite Element mesh provided by the Formulators, [3], has been slightly modified
in order to have an exact geometrical match at the interface between the nodes of the fluid and
those of the foundation rock.
The reservoir and the foundation domains have been extended upstream to avoid possible
numerical influences of the nonreflecting water surface at the end of the reservoir on the
dynamic response of the dam under seismic loading conditions.
Finally, the finite elements of the central vertical sections of the dam have been refined to
discretize with a more regular grid the whole mesh of the dam (Figure 1).
The displacement field of each finite element of the numerical model has been discretized
with quadratic shape functions.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
168

Figure 1: The coarse mesh used to study the dam-reservoir-foundation system
Material properties
Linear elastic constitutive models have been assigned to the dam and the foundation rock.
The physical-mechanical properties are summarized in Table 1.
Table 1: Concrete and rock properties
Domain Material
Density
[kg/m
3
]
Poisson ratio
Young modulus
[MPa]
Dam Concrete 2400 0.167 27000
Foundation Rock 0 0.200 25000

The Rayleight damping model has been taken into account to define the dam behavior during
seismic loading conditions. Assuming a 5% structural damping ratio, the mass and stiffness
damping parameters of the Rayleight formulation are as follows:

o = 0.94; | = 2.65E-03

The reservoir has been discretized by means of acoustic finite elements whose properties are
reported in Table 2.
Table 2: Water properties
Domain Material
Density
[kg/m
3
]
Speed of sound
in water [m/s]
Bulk modulus
[MPa]
Reservoir Water 1000 1500 2200

Loading and boundary conditions
According to the formulation of the Theme A, the following loadings sequence has been
considered:
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
169
- Dead loads
- Hydrostatic water pressure with the maximum water level equal to the dam crest
height (i.e. 715 m a.s.l.)
- Seismic loads, provided by the Formulators in terms of accelerations along the three
Cartesian directions.

Different boundary conditions have been used for the foundation domain depending on the
type of the analysis performed. In case of static and modal analyses, symmetric conditions
have been applied to the lateral rock walls whereas fixed constraints have been assigned to the
basement rock. In seismic analysis, the accelerations - or the equivalent related displacement
varying in time - have been applied to the lateral and bottom surfaces of the foundation.
In modal and seismic analyses, the fluid domain has been modeled by means of acoustic
elements, assigning the following boundary conditions:
- Dam-reservoir interface (refer to the COMSOL node acoustic-structure boundary,
[4]):


(

(1)
(2)

where n is the normal to the interface, the water density, p the fluid acoustic pressure,
q
d
the dipole source (equal to zero in the present case), u
tt
the acceleration field of the
structural domain at the fluid interface and o the stresses tensor.

- Foundation-reservoir interface (refer to the COMSOL node impedance, [4]):


(

(3)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
170
where n is the normal to the reservoir-foundation interface, the water density, p the fluid
acoustic pressure, q
d
the dipole source (equal to zero in the present case), Z
i
the acoustic
input impedance assumed equal to:

/q (4)

being q a damping coefficient that characterizes the effects of absorption of the hydrodynamic
pressure waves at the boundary, according to the following equation [6]:





(5)

o is the wave reflection coefficient that accounts for the behavior of the absorption of
hydrodynamic pressure waves at the boundary, whereas c is the speed of sound in water.
According to some literature case studies, o has been considered equal to 0.75, [7].

- Upstream-reservoir surface (refer to the COMSOL node plane wave radiation, [4]):


(

(6)

where n is the normal to the reservoir-foundation interface, the water density, p the fluid
acoustic pressure, q
d
the dipole source (equal to zero in the present case), c is the speed of
sound in water and Q
i
the monopole source (equal to zero in the present case).

- Free surface (refer to the COMSOL node sound soft boundary, [4]):


(7)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
171
Numerical analyses
In the following paragraphs the main features of each numerical analysis will be provided,
specifying the assumed hypotheses.
Static analysis
A linear static analysis has been carried out in order to apply the dead load and the hydrostatic
water pressure, thus the water domain has not been included in the geometrical model.
The applied boundary conditions are reported in the previous paragraph.
Modal analysis
The first 10 eigenfrequencies of the dam have been computed taking into account the dam-
reservoir interaction.
The applied boundary conditions are reported in the previous paragraph.
Seismic analysis
Seismic analyses in time domain are generally carried out considering a massless foundation
and applying a spatially-uniform ground motion directly at the basement rock.
In the present paper, at first an ordinary differential equations problem has been solved,
making reference to the foundation domain only, in order to compute the displacements
associated to the earthquake motions (Figure 2). In the subsequent dynamic analysis, these
displacements have been applied to the bottom and lateral rock walls to calculate the dynamic
response of the dam-reservoir-foundation system.
This procedure has been chosen because it allows computing easily the relative displacements
of the system; anyway, as an alternative, the same results could be attained applying directly
the accelerations to the foundation boundaries.
The boundary conditions taken into account during the seismic analysis are reported in the
previous paragraph.


Figure 2: Acceleration time history (a
max
~0.1g)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
172
Results
With reference to the results of modal analysis, the first 10 eigenfrequencies and the related
modal shapes are reported in Table 3.
Table 3: Eigenfrequencies and modal shapes
1 1.54247 [Hz]

6 2.51800 [Hz]

2 1.55194 [Hz]

7 2.83905 [Hz]

3 2.09799 [Hz]

8 2.96151 [Hz]

4 2.22299 [Hz]

9 3.19079 [Hz]

5 2.33002 [Hz]

10 3.37707 [Hz]


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
173
According to the requests specified by the Formulators, the results of the static and seismic
analyses are provided in terms of radial displacements [m], hoop and vertical stresses [MPa]
on the upstream and downstream faces of three vertical sections of the dam.
Table 4: Upstream face Central section (US-C) Radial displacements, hoop and vertical
stresses




Table 5: Downstream face Central section (DS-C) Radial displacements, hoop and vertical
stresses




ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
174
Table 6: Upstream face Right section (US-R) Radial displacements, hoop and vertical
stresses




Table 7: Downstream face Right section (DS-R) Radial displacements, hoop and vertical
stresses





ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
175
Table 8: Upstream face Left section (US-L) Radial displacements, hoop and vertical
stresses




Table 9: Downstream face Left section (DS-L) Radial displacements, hoop and vertical
stresses





Conclusion
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
176
The paper presents the dynamic response of the dam-reservoir-foundation system according to
the basic requests of the Formulators. Only the coarse mesh has been considered as some
preliminary analyses have demonstrated that, compared to a high computational effort, results
showed only slight differences.
A second aspect to be emphasized refers to the boundary conditions assumed for the fluid
domain at the rock interface: if a fully reflecting condition is assumed, the stresses computed
on the dam are considerably higher than those attained in the case discussed in this paper
where a partial absorption conditions has been assumed. This means that experimental
investigations should be undertaken in order to define more realistic boundary conditions.
A suggestion for the Formulators, in the synthesis phase, refers to the opportunity to make
some comparisons between the main outcomes of Theme A and those of the previous
Benchmark Workshops, cited above.
Acknowledgements
This work has been financed by the Research Fund for the Italian Electrical System under the
Contract Agreement between RSE S.p.A. and the Ministry of Economic Development -
General Directorate for Nuclear Energy, Renewable Energy and Energy Efficiency in
compliance with the Decree of March 8, 2006.
References
[1] Theme A Seismic analysis of the Talvacchia dam. (1992). 2
nd
ICOLD International
Benchmark Workshop on Numerical Analysis of Dams. Committee on Computational
Aspects of Dam Analysis and Design. Bergamo, Italy.
[2] Theme A1 Earthquake of an arch dam including the nonlinear effects of contraction
joint opening. (1996). 4
th
ICOLD International Benchmark Workshop on Numerical
Analysis of Dams. Committee on Computational Aspects of Dam Analysis and Design.
Madrid, Spain.
[3] Zenz G., Goldgruber M. (2013). Theme A formulation. Fluid Structure Interaction. Arch
Dam Reservoir at Seismic loading. 12
th
ICOLD Benchmark Workshop on Numerical
Analysis of Dams, Graz, Austria.
[4] COMSOL Multiphysics. (2013). Acoustic Module Users Guide. Version 4.3b
[5] COMSOL Multiphysics. (2013). Structural Mechanics Module Users Guide. Version
4.3b
[6] Ka-Lun Fok, Anil K. Chopra. (1986). Earthquake analysis of arch dams including dam-
water interaction, reservoir boundary absorption and foundation flexibility. Earthquake
engineering and structural dynamics, Vol.1, pp 155-184.
[7] Ghanaat Y., Redpath B.B. (1995). Measurement of reservoir-bottom reflection
coefficient at seven concrete dam sites. QUEST Structure Report No. QS95-01.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
177
12
th
International Benchmark Workshop on
Numerical Analysis of Dams
Theme A: Fluid Structure Interaction Arch Dam Reservoir at
Seismic Loading
A. Diallo
1
and E. Robbe
1
,
1
EDF-CIH, 15 Avenue Lac du Bourget Savoie Technolac, 73370 Le Bourget du Lac,
FRANCE
E-mail: abdoul-salam.diallo@edf.fr, emmanuel.robbe@edf.fr
Abstract
Dynamic response of arch dams to earthquake ground motion is significantly affected by the
interaction between the dam and the impounded water. There are several approaches to take
into account this dynamic dam-water interaction. The purpose of this benchmark is to
compare different modelling techniques with different levels of precision and therefore with
different computing efforts. This paper presents the results of the 3 approaches investigated
and the fundamental hypotheses adopted for each them. The first approach is a generalized
Westergaard added mass, the second approach is an incompressible finite-element added mass
and the third approach is based on a sub structuring method where the fluid is compressible.
All investigations are carried out for an artificially generated symmetrical arch dam and
simplified loading and boundary conditions. In general, Westergaard added mass yields
higher compressive and tensile stresses, as well as higher radial displacement. The sub
structuring approach, where the compressibility of the water and the impedance of the
foundation are taken into account, yields lower stresses.
Introduction
The objective of this paper is to present the seismic analysis of a 220 m high double curvature
arch dam undertaken as part of the 12
th
benchmark study. The results of the analysis are
presented for the 3 approaches used to take into account dam-water interaction in accordance
with the general assumptions made by the benchmark organizing committee. The analyses of
the results focus on the impact of the hydrodynamic approach used on the computed stresses
and displacements of the dam.
The following approaches are presented:
- Generalized Westergaard added mass
- Incompressible finite element added mass
- Sub structuring method where water compressibility is taken into account.
For each approach, the fundamental hypotheses are presented and the physical justification is
given.
Description of arch dam analysis - modelling methods
The analyses were performed with Code_Aster. A computer program developed by EDF; it
offers a full range of multiphysical analysis and modelling methods.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
178
Materiel parameters
Because a modal superposition method is used for the three approaches, the Rayleigh
damping cannot be used, rather, a value of modal damping is specified for each mode. This
value is fixed at 5% which is the value usually considered in real projects.
The mass density of rock is not specified by the benchmark organizing committee because the
traditional method assumes that the foundation is massless, however the sub structuring
method investigated takes into account the impedance matrix of the foundation rock. It is
therefore necessary to specify the mass density of rock; a value of 2400 kg/m3 is considered
for the rock in this analysis.
The other material properties are considered in accordance with the general assumptions made
by the benchmark organizing committee. Although in practice the dynamic modulus of
elasticity is usually considered 25 percent greater than static modulus of elasticity, in this
analysis the supplied static values are respected.
Static analysis
The application of the dead load should consider the manner in which the dam was
constructed. Arch dams are often constructed as independent cantilever blocks separated by
vertical joints. Since these joints are not capable of transferring dead load horizontally until
they are grouted, dead loads should be applied to individual cantilever to simulate this
condition. This may be accomplished by performing dead load analysis in two steps. First,
dead loads are applied to even cantilevers (Set-1) and the stresses are extracted. In second
analysis, the dead loads are applied to odd cantilevers (Set-2) separately and the stresses are
extracted. The addition of the stresses obtained from the two steps is considered as the initial
stress state of the dam and the displacement of the dam due to the dead load is not considered.
Water loads due to the hydrostatic pressures of the normal water level are external forces
acting on the u/s face of the dam. The hydrostatic pressures are applied to the monolithic arch
structure after the construction joints are grouted.
Dynamic modelling
The methods used in this analysis are based on the modal superposition method. The modal
analyses performed for the empty dam show that 90 percent of the mass of the dam is excited
by a frequency range between 1.93 and 10 Hz. The maximum of the seismic is also in the
same range (1-10 Hz).A significant amplification can therefore be expected.
The different approaches used to take into account the fluid-structure interaction are:
a) Generalized Westergaard Added-mass
The added-mass representation of dam-water interaction during earthquake ground shaking
was first introduced by Westergaard [1]. In his analysis of a rigid 2D gravity dam with a
vertical upstream face, Westergaard showed that the hydrodynamic pressures exerted on the
face of the dam due to the earthquake ground motion is equivalent to the inertia forces of a
body of water attached to the dam and moving back and forth with the dam while the rest of
reservoir water remains inactive.
A general form of the Westergaard added-mass concept which accounts for the 3D geometry
[2] can be applied to the earthquake analysis of arch dams.
The general formulation is based on the same parabolic pressure distribution with depth used
by Westergaard, except that it makes use of the fact that the normal hydrodynamic pressure


at any point on the curved surface of the dam is proportional to the total normal
acceleration,

(1)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
179

(2)

Density of water;
Westergaard pressure coefficient;
H: Water depth;
Z: Level of the point on the curved surface of the dam.
The normal pressure

at each point is then converted to an equivalent normal hydrodynamic


force by multiplying by the tributary area associated with that point.
b) Incompressible Finite-Element Added Mass Potential Approach
The added-mass representation of the impounded water can be obtained more accurately by a
finite-element solution of the pressure wave equation, which fully accounts for the complex
geometry of the dam and the reservoir. The impounded water represented by the wave
equation is discretized using a finite element mesh of incompressible liquid elements. The
solution is obtained by numerical procedures with the following boundary conditions:
(3)
- The hydrodynamic pressures at the water free surface are assumed to be zero, that is,
the effects of surface waves are neglected.
- The reservoir bottom and sides, as well as a vertical plane at the upstream end of the
reservoir model, are assumed to be rigid. For rigid boundaries the normal pressure
gradients or the total normal accelerations are zero.
- The normal pressure gradients at the dam-water interface are proportional to the total
normal accelerations of the fluid.
The computed pressures for the nodal points on the upstream face of the dam are then
converted into equivalent nodal forces, from which an added-mass matrix representing the
inertial effects of the incompressible water is obtained.
c) Compressible Water with Absorptive Reservoir Boundary- Substructuring
Method
The substructuring method consists of dividing the complete system into three substructures:
the structure, the water, and the foundation, each of which can be partially analysed
independently of the others. The structure is represented by a 3D finite element, which
permits modelling of a general geometry and linear elastic material properties. The water
domain and the foundation region are represented by boundary elements.
The added mass representation of the impounded water described above ignores the effect of
water compressibility and reservoir boundary absorption. However, the water compressibility
and reservoir boundary absorption can significantly affect the hydrodynamic pressure and
hence response of arch dams to earthquakes [3].
Interaction of the dam with the foundation rock leads to an increase in vibration periods,
primarily due to the flexibility of the foundation rock. Dam-foundation interaction also
decreases the dam response if damping arising from material damping in the foundation rock
and radiation damping associated with wave propagation away from the dam are considered in
the analysis.
Procedures for earthquake response analysis of arch dams including dam-water interaction,
water compressibility, reservoir boundary effects and dam-foundation interaction are
developed by EDF [4]. In this procedure, the radiative damping and the hysteretic damping of
the foundation are also considered. To represent the infinity domain, Greens functions are
used in the fluid domain and in the foundation domain.
In the fluid domain, the Helmholtz equation is discretized using boundary elements.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
180

(4)
The solution is obtained by numerical procedures with the following boundary conditions:
- The hydrodynamic pressures at the water free surface are assumed to be zero, that is,
the effects of surface waves are neglected.
- The normal pressure gradients at the dam-water interface are proportional to the total
normal accelerations of the fluid.
- The normal pressure gradients at the foundation-water interface (reservoir boundary)
are proportional to the total normal accelerations of the fluid.
- The hydrodynamic pressure wave impinging on the reservoir boundary is partly
reflected into the water, and partly refracted (absorbed) into the boundary materials.
The partial absorption at the reservoir boundary is approximately represented by a
reflection coefficient known as , which is the ratio of reflected to incident wave
amplitude.
In the foundation domain, the Navier equation is also discretized using boundary elements. To
take into account the infinity domain, Green functions are used.

(5)
The solution is obtained by numerical procedures with the following assumptions:
- The rock foundation is assumed homogeneous, isotropic and linear elastic.
- A value of 5% is considered for the modal damping.
Results and comparison of the different approaches
Modal Analysis
The interaction of an arch dam with the impounded water leads to a decrease in the dam
vibration frequencies. This is because the dam cannot move without displacing the water in
contact with it. The fact that water moves with the dam increases the total mass that is in
motion. This added mass decreases the natural frequencies of the dam, which in turn affects
the response spectrum ordinate and hence the effective earthquake inertia forces. The
flexibility of the foundation rock also decreases vibration frequencies of the dam.
Table 1 gives the fundamental modes obtained by the 3 approaches investigated in this
analysis. The results show that the Westergaard method gives the largest added-mass value, as
evidenced by its decreasing the fundamental frequency the most. However, this does not
automatically means that Westergaard approach gives the largest stresses, because the
response of the dam also depends on the characteristics of the earthquake ground motion.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
181
Table 1: fundamental period of the dam

Empty
reservoir
Full reservoir

Westergaard Incompressible
fluid elements
Compressible fluid element
with 50% wave absorption
Fundamental frequency 1.93 Hz 1.29 Hz 1.57 Hz 1.49 Hz
Stresses and displacement
Under static loads, the maximum radial displacement reaches 8.32 cm at the top of the central
cantilever. This displacement is only due to hydrostatic pressure acting on the u/s face of the
dam.
The relatively symmetrical deformation of the arch dam leads to the compression of the lower
portion of the d/s face of the dam and to tractions in the abutments on the u/s face of the dam.
The maximum static principal stress (compression) reaches 9.32 MPa at lower portion in the
abutments on the d/s face of the dam. The minimum static principal stress (tensile) also
reaches 4.40 MPa at the lower portion of the u/s of the dam. This tensile stress (4.40 MPa)
could lead to the opening of a joint at the contact between dam and the foundation rock.
However the contraction joint opening/closing is not modelled.
Figures 1 and 2 give the principal stress contours under static loads.

Figure 1: Principal stresses (compression) under static loads

Figure 2: Principal stresses (tensile) under static loads

During the ground motion, the maximum displacement of dynamic vibrations is about 6 cm
with the respect to the initial displaced shape of the dam.
The maximum displacements represented in fig 3 show that the Westergaard method yields
higher radial displacements at the crest of the dam. The substructuring method where we take
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
182
into account the radiative damping and hysteric damping in the foundation as well wave
absorption at the reservoir boundary, yields lower displacements (20% of reduction).


Figure 3: Maximum displacement of the central cantilever at the d/s face of the dam
In general, the Westergaard method yields higher compressive and tensile stresses. Indeed, the
seismic tensile vertical stresses at the u/s face of the main cantilever vary from 2.4 MPa for
the Westergaard method to 0.2 MPa for the substructuring method at a point at of the dam
height (cf fig4).

Figure 4: Vertical stresses at u/s face during the ground motion

0
50
100
150
200
250
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
H
e
i
g
h
t

(
m
)

displacement (m)
min_westergaard min_potentiel_approach min_substructuring
static max_westergaard max_potentiel_approach
max_substructuring
0
50
100
150
200
250
-6.00 -5.00 -4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00
H
e
i
g
h
t
(
m
)

Stresses (MPa)
min_westergaard min_potential_approach min_substructuring
static max_westergaard max_potential_approach
max_substructuring
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
183
Figure 5 gives maximum hoop stresses at the d/s face of the central cantilever during the
earthquake. It shows a tensile stress varying from 1.6 MPa for Sub structuring method to 2.2
MPa for Westergaard and potential approach at the top part of the cantilever; this could lead
to the opening of vertical joints.

Figure 5: Hoop stresses at d/s face during the ground motion

In general, the ground motion increases the maximum principal stress observed at the lower
portion of the abutments on the d/s face of the dam as well the minimum principal stress
observed at the lower portion of u/s face of the dam.
Thus, the maximum compressive stresses during the ground motion varying from 11.26 MPa
for sub structuring method to 12.77 MPa for added mass approach on the d/s face of dam (cf
fig6 right_section). At dam height on the u/s face, the maximum compressive stresses also
varying slightly, from 9.82 MPa for sub- structuring approach to 10.45 MPa (cf fig 6
main_section).
0
50
100
150
200
250
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00
H
e
i
g
h
t
(
m
)

Stresses (MPa)
min_westergaard min_potential_approach min_substructuring
static max_westergaard max_substructuring
max_potentiel_approach
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
184

Figure 6: Maximum principal stresses at u/s face of central cantilever and d/s face of right
section

In figure 7, we notice that the substructuring method increases slightly the tensile stress at the
contact between dam and the foundation (2.84 MPa under static loads and 3.14 MPa during
earthquake). However this amplification is very significant with added-mass methods, thus
with Westergaard added mass, the tensile stress reaches 5.96 MPa during the earthquake and
with the potential approach it is about 5.24 MPa. These tensile stresses could lead to the
opening of the contact between the dam and the foundation.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
185

Figure 7: Minimum principal stresses (tension) at the u/s face of right section
Influence of reservoir boundary absorption
A hydrodynamic pressure wave impinging on the reservoir boundary is partly reflected into
the water, and partly refracted (absorbed) into the boundary materials. If the reservoir
boundary materials are relatively soft, an important fraction of the reservoir water energy can
be absorbed, leading to a major reduction in the dynamic response of the dam. Therefore, the
values of the absorption ratio for the design and safety evaluation of dams subjected to
earthquake loading should be measured or selected conservatively.
The purpose of this section is to show the effect of this absorption on the dynamic response of
the dam by studying 3 cases of absorption (0%, 50% and 100% absorption).
As expected, wave absorption at the reservoir boundary reduces significantly the dynamic
response of the dam. Figure 8 gives the radial spectrum at the crest of the dam and for
fundamental frequency (f1=1.49 Hz), we observe that the correspondent pseudo acceleration
varies from 1.93g (total absorption) to 4.14 g (without absorption).

80
100
120
140
160
180
200
220
240
-2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
H
e
i
g
h
t
(
m
)

Stresses (MPa)
min_westergaard min_potential min_substructuring
static max_westergaard max_potential
max_substructuring
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
186

Figure 8: radial spectrum at the crest of the dam for 3 cases of absorption

Reservoir boundary absorption also decreases compressive and tensile stresses (cf fig 9), as
well as the radial displacement (cf fig 10). In figure 9, we observe that the maximum stress
varies from 8.95 MPa (total absorption) to 11.26 MPa (without absorption) at dam height
on the u/s face of the main section.
Figure 10 gives the radial displacement of the central cantilever for 3 cases of absorption and
we can observe that the crest displacement varies from 11.4 cm (total absorption) to 15.9 cm
(without absorption).

Figure 9: Hoop stress of the central cantilever for 3 cases of absorption
0.00
0.01
0.10
1.00
10.00
0.10 1.00 10.00 100.00
P
s
e
u
d
o
-
a
c
c
e
l
e
r
a
t
i
o
n

e
n

(
g
)

frequence in (Hz)
Without_absorption 50%- absorption total_absoption
0
50
100
150
200
250
-12.00 -10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00
H
i
g
h
t

(
m
)

Stresses(MPa)
min_without_abs static max_without_abs
min_50%_abs max_50%_abs min_total_abs
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
187

Figure 10: Radial displacement of central cantilever for 3 cases of absorption
Conclusion
The different approaches to hydrodynamic effect modelling investigated in the present study
lead to similar behaviour regarding the structural response of the example arch dam. In
general, the Westergaard approach yields higher compressive and tensile stresses, as well as
higher radial displacements. The substructuring method where wave absorption is taken into
account at the reservoir boundary decreases significantly the dynamic response of the dam due
to the increased damping of the coupled system.
Acknowledgements
The authors wish to thank Electricit de France (EDF-CIH) who financed this study.
Particular thanks to Frederic LAUGIER, head of the Civil Engineering Structures
department for his assistance in preparing the study, and to EDF-R&D for their collaboration.
References
[1] Westergaard, H.M (1933), Water pressure on dam during earthquakes, transactions
American Society of Civil engineers, Vol 98.
[2] Kuo, 1982 Fluid structure interactions: Added Mass Computations for incompressible
fluid. Report No. UCB/EERC-82/09, University of California Earthquake Engineering
Research Center, Berkeley.
[3] Fok and Choppra 1985 Earthquake Analysis and Response of Concrete Dams Report
No UCB/EERC-85/07, University of California Earthquake Engineering Research
Center, Berkeley
[4] www.code_aster.org

0
50
100
150
200
250
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
H
i
g
h
t

(
m
)

displacement (m)
min_without_abs static max_without_abs min_50%abs
max_50%abs min_total_abs max_total_abs

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
189



THEME B



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
191



Theme B
Long Term Behaviour of a Rockfill Dam



Formulators:
Camilo Marulanda, Ph.D., & Joan Manuel Larrahondo, Ph.D.
INGETEC
Cra 6 No. 30A-30 Bogot, Colombia
marulanda@ingetec.com.co & larrahondo@ingetec.com.co


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
192
Introduction
Overview: Long Term Rockfill Behhaviour
It is widely accepted that rockfill in dam shells exhibit time-dependent, creep-like behaviour
([1]-[3]). Early work by Terzaghi (1960) [2] proposed that such behaviour was due to
breakage of rock particles at highly-stressed contacts. Once broken, the rockfill particles
would rearrange to attain more stable configurations, thus inducing displacements.
Interestingly, most of the experimental evidence available to date still suggests that the main
cause of the time-dependent rockfill behaviour is particle breakage and crushing, phenomena
enhanced by the presence of water [2]. Furthermore, a number of constitutive models have
been proposed to describe such time dependence, including logarithmic, hyperbolic, and
visco-elastic relationships between long-term strain and time [1].
Earth-core rockfill dams (ECRD) typically show dramatic deformations upon reservoir first
filling [1]. Time-dependent deformation depends on water content of rock particles, such that
yield is highest in saturated rock particles, while tends to fade for very dry materials [2].
Since the work by McDowell and Bolton (1998) [2], successive deformation stages are
recognized in rockfill-type granular materials subjected to compression stresses as the
confining stress is increased. At low stresses, deformation is due to particle rearrangement.
From a micromechanical point of view, rockfill deforms mainly because of interparticle
sliding and rotation [1]. As the confining stress increases, the granular skeleton becomes
gradually blocked, and particle breakage and crushing are triggered. Such mechanism is called
clastic yielding. Finally, for very high stresses the strain vs. log stress plots have upward-
directed concavity. This is attributed to the comminution limit of small particles.
The time-dependent deformation of rockfill can thus be explained by progressive breakage
and crushing of stressed particles [1]. Particle breakage takes place as subcritical propagation
of preexisting microcracks within loaded rock particles. The velocity of crack propagation is a
function of the local stress intensity factor and the prevailing relative humidity. Particle
breakage combines with complex localised crushing at interparticle contacts to boost rockfill
deformation [1].
Not a large number of databases are available that comprise 25 years or more of
instrumentation data from a rockfill dam. The problem proposed in this document consists in
reproducing the development of displacements and stresses of a real-case dam during
construction, reservoir impoundment, and operation.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
193
Benchmark Example: Alberto Lleras Dam (Guavio Hydroelectric Project), Colombia

Dam Description
The case history that is being proposed as benchmark example for the Theme B of the 12th
Workshop on Numerical Analysis of Dams is that of the Alberto Lleras Dam, part of the
Guavio Hydroelectric Project in Colombia.
The Guavio Hydroelectric Project is located in the Cundinamarca region of central Colombia,
some 120 km northeast from Bogot. The project spans between the towns of Ubal, where
the dam is located, and Mambita, where the underground powerhouse is located [4]. The
project harnesses the hydropower potential of the lower Guavio River (1100 m nominal head)
and comprises the following components designed by INGETEC ([4]-[6]):

- The Alberto Lleras Dam: a 245-m high, clay-core rockfill dam, the highest of its kind
in South America [3]. The total dam volume is about 17 x 106 m3 which impounds a
14-km long, 950 x 106 m3 reservoir.
- A 34-m high, 4500 m3/s capacity, ogee-type spillway controlled by radial gates and
two 600-m long tunnels
- A 13.3-km long, 8 m diameter power tunnel and a 505-m high pressure shaft
- A 234-m long, 17-m wide, 35-m high underground powerhose located 560 m below
ground. The total installed generating capacity of the Guavio project is 1600 MW,
currently delivered by eight Pelton turbines.
- A 1160-m long diversion tunnel as well as two other tunnels, 2330 m and 2190 m
long, respectively, that divert the Batatas and Chivor Rivers into the Guavio reservoir
- A 5.2-km long, 8-m diameter tailrace tunnel to return the Guavio river water back to
its main course

The Guavio project was built between 1979 and 1995 [2] and first filling took place in
February 1992. The average flow of the river at the dam site is 62 m/s. Figure 1 presents an
aerial view of the Guavio Project and the Alberto Lleras Dam.
Available instrumentation data span from 1987 through 2013. Specifically, measurements
include data from:

- 78 level survey points on the dam crest and faces
- 8 Vertical Movement Recorders (RMV)
- 24 Pneumatic Settlement Sensors (SNA)
- 40 pressure cells (CP)
- 26 pneumatic piezometers (PZ)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
194

Figure 1: General view of the Guavio Hydroelectric Project and the Alberto Lleras Dam [4]

The objective for the participants of Theme B is to reproduce, via a numerical model, the
displacements observed on the maximum cross section of the Alberto Lleras Dam.
The minimum requirements that shall be incorporated in their numerical model are a three-
dimensional model using the provided simplified topography and the use of the known dam
zoning and the construction sequence.

Geology Overview
The greater Guavio project area comprises Palaeozoic rocks overlain by Mesozoic rocks [5].
The dam site is underlain by Palaeozoic quartzite, argillite, and limestone. At the dam site, the
river features a 600-m deep, narrow canyon. The upper half of the canyon at the dam site (left
abutment) exhibits Cretaceous rocks.
The Palaeozoic formations strike diagonally with respect to the river canyon and dip
upstream. Most of these rocks below 30 m of depth are fresh, hard, and depict low
permeability [5]. The limestone shows frequent karstic features (caverns) which were tackled
via an extensive grouting programme.
No geological faults exist at the dam foundation area [5]. The steep slopes of the canyon
rendered some relief joints parallel to the river flow. No particular treatment was necessary for
these joints.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
195
Input Data
All detailed input data information is provided in this manuscripts attachments. A description
of each database follows.

Dam Geometry
Attachment 1 contains all drawings related to dam location geometry. Specifically, the
following data files are provided:

- Maximum longitudinal section and description of dam zones (as-built; Oct. 1989;
DWG and PDF files)
- Plan view and topography (as built; 50-m elevation contours; Nov. 1989; DWG and
PDF files)
- Dam zoning description table (XLS file), including placed volumes and as-built
compaction specifications
Construction Sequence
Construction of the dam started on June 1983 with foundation excavation work [7]. Fill
placement operations initiated on September 1984 and ran through August 1989.
In general, during the six rainy months of the year (May-October) rockfill was placed and
compacted in the outer portions of the shell. During the remaining six drier months
(November-April), fill placement was concentrated in the central portion of the dam, namely
the core and confining rockfill [7].
Attachment 2 contains all drawings and data related to construction sequence. The following
data files are provided:

- Construction sequence plan (maximum longitudinal section; as-built; Nov. 1989)
- Fill placement sequence summary table (XLS file)
Material Properties
The outer portions of the shell rockfill comprised weathered fragments of quartzite, argillite,
and limestone [7]. On the other hand, the inner part of the dam below the core consisted in
sound, fresh, and clean rockfill.
The dam core consists mainly in shale fragments in a silty clay matrix [7] obtained from a
nearby quarry named San Pedro. The core material features medium plasticity and fines
content >30%.
Attachment 3 contains all drawings and data related to material characterization. Specifically,
the following data files are provided:

- DWG, PDF, and XLS files containing rockfill properties varying with dam depth
- DWG, PDF, and XLS files containing earth-core properties varying with dam depth
- One XLS file containing as-built grain size distribution bands for each dam zone
- One XLS file containing compressibility curves (pre-construction) for the earth core
and shell rockfill materials

For illustration purposes, material parameter data that were employed during the geotechnical
design of the dam are provided next:
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
196
- Earth core and shell rockfill compressibility: an experimental programme was
undertaken prior to construction to obtain design compressibility parameters from both
the earth core material and the rockfill.
30-cm thick samples of dam-core material were subjected to one-dimensional
consolidation tests in a 30-cm diameter oedometer. The earth-core materials featured
maximum grain size of 6.5 cm and natural water content during testing. Samples were
reconstituted via compaction of five layers using a 30-lb hammer falling freely from a
height of 24 in. Between 160 and 200 blows per layer were applied.
The rockfill material compressibility was also studied using the same large oedometer
as that of the core materials. However, rockfill samples were tested dry and using the
same grain size distribution as that of the construction specification (with maximum
grain size of 6.5 cm). Samples were reconstituted by compaction of four layers with a
vibrating table during 1.5 min each layer. During the loading process, the samples
were intentionally inundated at the 16 kg/cm2 loading step.
The following table presents the experimental programme initial conditions and results
in terms of compressibility coefficient. Attachment 3 also contains the compressibility
curves obtained from these tests.

- Plasticity of the earth core material (measured values during the dam geotechnical
design phase)
Liquid limit (%) 39
Plastic limit (%) 26
Plasticity index (%) 13
Instrumentation data
During construction and particularly after reaching 2/3 of the dam height, large displacements
took place within the dam core [7]. As a result, most of the cable of the pneumatic and electric
instruments was destroyed. Overall, about 50% of the original instruments installed were lost
due to construction operations.
Available instrumentation data for this workshop span from 1987 through 2013. Attachment 4
contains all drawings and data related to instrumentation data. Specifically, the following files
are provided:

- Five folders with XLS data files
- Two layouts of installed instrumentation (as-builts; plan view and maximum
longitudinal section; DWG and PDF; July 1990)

A brief description of available instrumentation data follows.
Results
Water content of the < #
4 sieve fraction (%)
Dry density
(Ton/m
3
)
a
v

(10
-3
cm
2
/kg)
Max
Ave
Min
28
24
21
1.84
1.77
1.71
4.7
3.6
2.6
Max
Ave
Min
-
-
-
2.25
2.17
2.03
2.1
1.6
0.8
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
197
Movements
- Surface movements: Data from a total of 63 level survey instruments, labelled 13
through 75 are provided. Survey elevation data span from 1989 through 2013.

- Vertical Movement Recorders (RMV) or crossarm gages: nine folders are provided
containing monthly RMV data, spanning from 1989 through 1995. Additionally, for
the 1996-2008 period, a XLS file is provided with data collected in eight of the nine
available RMVs.

RMV instruments or crossarm gages consist of a series of telescopic pipe sections with
alternate sections anchored to the embankment fill by horizontal channel crossarms
located at certain depth intervals, typically 3 m ([8] See Figure 2). The crossarms
ensure that the coupled pipes move together as compression of the fill progresses.
Measuring points are located at the lower end of each inner pipe; depths to such
measuring points are usually sounded by a probe with spring-loaded sensing pawls,
lowered on a steel tape. To produce a measurement, the probe is lowered just below a
given inner pipe and then raised until the paws latch against the lower end.

- Pneumatic Settlement Sensors (SNA): two folders; instruments 28-41 with monthly
data from 1987 to 1996; instruments 49-58 with monthly data from 1988 to 1995.


Figure 2: RMV (crossarm gage) schematic [8]. Left: telescopic pipe array. Right:
measurement probe
Pressure Cells (CP)
Three folders are provided. All total pressure cell data is reported in kg/cm
2
. Monthly data are
provided as follows:

- CP 1-24: data from 1986 to 1998; CP 1, 3, 5, 6, and 7 are at 1405 masl elevation; CP 8
through 24 are at 1463 masl elevation
- CP 28-37: 1987 to 1998 at 1523 masl elevation

- CP 41-46: data from 1988 to 1997 at 1589 masl elevation
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
198
Pneumatic Piezometers (PZ)
- One XLS file is provided that contains data collected from the dam foundation
piezometers (PZ 82, 83, and 84). Data were collected monthly spanning from 1988
through 2004.
- 15 XLS data files are provided with data collected from the dam core piezometers (PZ
45 through 49) at 1523 masl elevation. Data were collected monthly spanning from
1987 through 2001.
Requested Results
All workshop participants are kindly requested to provide a paper, 15 pages maximum, in
which all assumptions are clearly stated, particularly regarding initial and boundary
conditions.
To facilitate results rating, the participants must also submit their key results under a
prescribed format whereby an Excel file under the name ThB_ResXXXX.xls shall be created
(XXXX represents a name or acronym of the participants organisation).
Fill Displacements
Displacement contours for three longitudinal sections and one cross section along the dam
crest shall be presented. Results at the end of construction, at the condition prior to
impoundment, and after reservoir impoundment shall be included.
Stresses within the Dam
Calculated principal and vertical stresses obtained from the analysis on five locations within
the fill are requested. The results shall be presented both at the end of construction and after
partial and full impoundment.
Creep and Arching Effects
A discussion supported with simulation results is requested regarding long-term displacement
behaviour including possible arching effects within the dam body.
Groundwater Flow
The participants are requested to submit the resulting groundwater flow net under 1620 masl
reservoir level water level conditions.
Acknowledgements
The formulators are grateful to EMGESA Colombia for allowing the use of the
instrumentation data for this workshop
References
[1] Oldecop, L. A. and Alonso, E. E. (2007). Theoretical Investigation of the Time-
Dependent Behaviour of Rockfill. Gotechnique, V. 57, No. 3, pp. 289301
[2] Oldecop, L. A. and Alonso, E. E. (2001). A Model for Rockill Compressibility.
Gotechnique, V. 51, No. 2, pp. 127-139
[3] Justo, J. L., Durand, P. y Justo, E. (2003). Un Modelo Tridimensional para el Estudio de
la Fluencia en Presas de Materiales Sueltos. Rev. Int. Mt. Num. Clc. Dis. Ing. V. 19,
No. 3, pp. 313-330 (in Spanish)
[4] INGETEC S.A. (2013). Hydropower Projects Guavio
http://www.ingetec.com.co/experiencia-ingles/textos-proyectos-ingles/proyecto-
hidroelectricos-ingles/guavio-ingles.htm (Accessed 2013)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
199
[5] Marulanda, A. and Amaya, F. (1989). The Design and Construction of Colombias
Guavio Dam. Water Power and Dam Construction, December 1989. pp. 41-60
[6] INGETEC S.A. (2013). Hydroelectric Developments.
http://www.ingetec.com.co/brochures-ingles/BROCHURE-HIDROELECTRICAS-
INGLES.pdf (Accessed 2013)
[7] Amaya, F. and Marulanda, A. (1993). Behavior of Guavio Dam during Construction and
First Filling. Proceedings of the International Symposium on High Earth-Rockfill Dams.
ICOLD-CSHEE. Beijing, China.
[8] Dunnicliff, J. (1988). Geotechnical Instrumentation for Monitoring and Performance.
John Wiley & Sons. 577 p.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
201



PAPERS

THEME B


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
203
Solution to Theme B
Long Term Behaviour of a Rockfill Dam
F. Ezbakhe
1
, I. Escuder-Bueno
1

1
Universidad Politcnica de Valencia, Spain
E-mail: iescuder@hma.upv.es
Preface
The authors acknowledge that the solution herein presented is very simplified and should not
be taken as a professional reference by any means.
However, with the purpose of enabling the formulator to estimate the order of magnitude of
the lack of accuracy and/or physical inconsistencies by using methods as simplified and
empirical as those herein presented, the paper may be of certain interest.
Introduction
Theme B of the 12th Benchmark Workshop on Numerical Analysis of Dams (ICOLD)
consists in the evaluation of the long term behavior of a rockfill dam, analyzing the case of the
Alberto Lleras Dam, part of the Guavio Hydroelectric Project in Colombia.
The objective is to reproduce, via a numerical model, the behavior of the dam during the
construction, impoundment and exploitation, providing the evolution of displacements and
stresses within the dam as results.
For the solution presented in this paper, the problem has been solved by FLAC3D (Itasca), a
finite differences program developed by Itasca Consulting Group, Minnesota. This software
uses the finite differences method to analyze the mechanical behavior of a continuous medium
in 3D until it reaches an equilibrium state.

Figure 1: View of the Alberto Lleras Dam
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
204
Statement of the problem
Alberto Lleras Dam is a clay-core rockfill dam which main features are given in the next
table:
Table 1: Main features of the Alberto Lleras Dam
Height 245m
Width 950m
Dam volume 7hm3
Reservoir volume 950hm3

Construction of the dam lasted five years, from September 1984 to August 1989. In general,
during the six rainy months of the year (May-October) rockfill was placed and compacted in
the outer portions of the shell. During the remaining six drier months (November-April), fill
placement was concentrated in the central portion of the dam. The impoundment started on
February 1992 and run through September 1992.

The dam consists in three rockfill zones, with similar mechanical parameters but different
granulometric composition. It includes also two transition zones and a clay core, featuring a
medium plasticity clay matrix.

Figure 2: View of the Alberto Lleras Dam

The dam site is underlain by quartzite, argilite and limestone, with no geological faults. The
formations strike diagonally with respect to the river canyon and are fresh, hard and depict
low permeability.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
205
Features of the implemented model
Geometrical model
From a simplification of the real topography and geometry of the problem, a 3D geometrical
model of the dam was implemented, formed by 7286 nodes and 5800 solid elements. Next
figures show the FLAC 3D geometrical model of the dam.



Figure 3: FLAC 3D geometrical model
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
206
Mechanical model
The mechanical model selected for the solution herein presented is linear-elastic. This model
was chosen for its simplicity and wide applicability solving soil and rock mechanics
problems.
In this model, strain increments generate stress increments according to the linear and
reversible law of Hooke:

(1)

where the Einstein summation convention applies, _ij is the Kroenecker delta symbol, and
_2 is a material constant related to the bulk modulus, K, and shear modulus, G, as:

(2)

New stress values are then obtained from the relation:

(3)

Next table provides the values that have been adopted for the mechanical model:
Table 2: Mechanical parameters

Density
(kg/m3)
Bulk modulus (Pa) Shear modulus (Pa)
Core 2010 7.5106 5.4106
Rockfill 2750 8.5107 4.7107
Foundation 2500 41010 31010

The boundary conditions applied are:
- Plane x=-300 (lateral surface). Movements constrained on the x-axis
- Plane x=400 (lateral surface). Movements constrained on the x-axis
- Plane y=-300 (front surface). Movements constrained on the y-axis
- Plane y=1300 (back surface). Movements constrained on the y-axis
- Plane z=-300 (underside). Movements constrained on all axis
Constructive behaviour
The construction process may affect the stress state, particularly if excess pore pressures
develop in the soils and do not dissipate completely during the construction stages. In
addition, staged modeling of the embankment lift construction also provides a better
representation of the stresses in the embankment.
In this case, due to the difficulty involved in modeling the actual construction sequence, it has
been simplified considering that it takes place in uniform layers of 10m high. Thus, the
numerical process is to activate each layer, assigning materials and properties, and calculating
the model to reach equilibrium.
Post-constructive behaviour
In Geotechnical Engineering, time-dependent settlement is associated to the process of
consolidation. This behavior would be therefore determined by the rate at which water is able
to flow through the ground pores under a certain hydraulic gradient.
Boughton (1970) found that this settlement also occurred during the construction process. For
calculation purposes, Boughton got a good fit for the form of post-construction deformations
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
207
by applying the entire weight of the structure at once. Then, adopting a scale factor, he got a
good fit for the magnitude of the deformations.
In this case, the scale factor used is 1.05 for dry behavior (from the completion of construction
to the start of filling) and 1.11 for the behavior of the wet rockfill (with filled reservoir).
Results
Settlements
Settlements of the dam body are obtained at three different times:
- August 1989 completed the construction of the dam
- January 1992 immediately before the first filling
- September 1992 1630 masl reservoir

Settlements are also obtained in nine point of the dam:
Table 3: Settlements in the dam

aug-89 jan-92 sep-92

Settlement
(m)
Elevation
(masl)
Settlement
(m)
Elevation
(masl)
Settlement
(m)
Elevation
(masl)
M-30 0.00 1641.081 0.810 1640.271 0.971 1640.110
M-36 0.00 1641.020 0.875 1640.145 1.043 1639.977
M-48 0.00 1641.751 0.363 1641.388 0.470 1641.281
M-53 0.00 1640.420 0.290 1640.130 0.368 1640.052
M-64 1.54 1604.027 1.974 1603.594 2.070 1603.498
M-66 1.60 1570.985 1.827 1570.758 2.032 1570.553
M-72 1.46 1515.672 1.617 1515.513 1.715 1515.415
SNA-30 3.60 1517.435 4.037 1516.995 4.593 1516.439
SNA-38 3.42 1518.242 3.872 1517.790 3.991 1517.671


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
208

Figure 4: Settlements in August 1989


Figure 5: Settlements in January 1992


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
209


Figure 6: Settlements in September 1992
Stresses
Next figure represents the stresses of the dam after dam construction:


Figure 7: Principal stresses in August 1989
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
210
Principal stresses (s1) and vertical stresses (sv) are also obteined in five different points:
Table 4: Stresses in the dam
@ EOC Reservoir at EL 1575 Reservoir at EL 1630
ID s
1
[Kpa] s
v
[Kpa] s
1
[Kpa] s
v
[Kpa] s
1
[Kpa] s
v
[Kpa]
CP-11 1594.3 1556.5 1619.1 1599.9 1752.3 1725.6
CP-6 1345.2 1281.2 3015.2 1964.7 3237.5 2482.1
CP-19 1909.4 1554.0 2240.3 1914.9 2806.5 2019.0
CP-22 2871.0 1459.6 2884.9 1947.6 2916.8 1992.0
CP-3 2769.7 2653.6 3438.1 2949.6 3475.0 2969.2
Conclusion
In order to study the long dams long term behavior, a linear elastic analysis has been realized.
This means that a linear analysis is performed for each time step and when the equilibrium is
reached the calculus moves to next instant of time.
The implemented model considers the dam body as a structure without interfaces and permits
the evolution of the displacements and stresses over time.
The simulation of this evolution can resembles to some extent to the data collected by
auscultation, with a margin of error due to the different simplifications made in the model.
References
[1] Escuder, I. Estudio del comportamiento tenso-deformacional de pedraplenes inundables
mediante simulaciones numricas formuladas en diferencias finitas y calibradas con
lecturas de instrumentacin. Doctoral Thesis
[2] Itasca Consulting Group, Inc., Fast Lagrangian Analysis of Continua in 3 Dimensions
version 3.1: Theory and Background, Optional features, Minneapolis, Minnesota, 2006
[3] ICOLD, Guidelines for use of numerical models in dam engineering, Bulletin 155



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
211



THEME C


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
213
Theme C
Computational Challenges in Consequence
Estimation for Risk Assessment: Numerical Modelling, Uncertainty
Quantification, and Communication of Results
Part 1 Hydraulic Modelling and Simulation

Sponsoring Organizations:
U.S. Army Corps of Engineers (USA)
U.S. Department of Homeland Security (USA)
Formulators:
Yazmin Seda-Sanabria (Formulation Team Co-Chair)
National Program Manager, Critical Infrastructure Protection & Resilience Program, U.S.
Army Corps of Engineers, Headquarters, 441 G Street NW (ATTN: CECW-HS), Washington,
DC 20314 (USA), Email: yazmin.seda-sanabria@usace.army.mil
Enrique E. Matheu (Formulation Team Co-Chair)
Chief, Critical Lifelines Branch, Sector Outreach and Programs Division, Office of
Infrastructure Protection, U.S. Department of Homeland Security, 245 Murray Lane
Arlington, VA 20598-0608 (USA), Email: enrique.matheu@hq.dhs.gov
Timothy N. McPherson (Formulation Team Technical Lead)
R&D Group Leader, Energy and Infrastructure Analysis, Los Alamos National
Laboratory, Los Alamos, New Mexico 87544 (USA), Email: tmac@lanl.gov
Mustafa Altinakar
Director and Research Professor, National Center for Computational Hydroscience and
Engineering, The University of Mississippi, Brevard Hall Room 327, P.O. Box 1848,
University, MS 38677-1848 (USA), Email: altinakar@ncche.olemiss.edu
Mark Jourdan
Research Hydraulic Engineer, Coastal and Hydraulics Laboratory, U.S. Army Engineer
Research and Development, 3909 Halls Ferry Road, Vicksburg, Mississippi 39180 (USA),
Email: mark.r.jourdan@usace.army.mil
Michael K. Sharp
Technical Director, Geotechnical and Structures Laboratory, U.S. Army Engineer Research
and Development, 3909 Halls Ferry Road, Vicksburg, Mississippi 39180 (USA), Email:
michael.k.sharp@usace.army.mil

For technical information, please contact Dr. Tim McPherson at tmac@lanl.gov

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
214
Introduction
In the last decade, computational capacity has grown dramatically such that multiprocessor
computing techniques are now widely available. This increase in resource availability has
allowed the development of a vast array of new models for flooding and consequence
assessment. Many of these models are computing flood wave propagation at extremely high
temporal and spatial resolutions. When these models are coupled to equally complex models
of population mobility, infrastructure impact and economic consequence, simulation
frameworks are created that can support a paradigm shift from standard approaches to dam
risk analysis.
Although computational advances have increased the availability and applicability of novel
tools, there is a deficiency in benchmarks on the use of those tools in risk assessments. An
obvious application of increased availability and efficiency of computational resources is to
conduct probabilistic risk assessments using Monte Carlo techniques, but the application of
such approaches entails a wide range of assumptions and technical decisions regarding the
management of uncertainty such as variable uncertainty, parameter uncertainty, uncertainty in
probabilistic sub-model, measurement error, computational errors, and numerical
approximation to name a few.
Universities, engineering companies and regulatory bodies are invited to contribute to the
benchmark and take part in the discussion of results gained.
This document is part 1 in a 2-part series for Theme C. Part 1 pertains to the hydraulic
modelling and simulation of the dam breach and subsequent flood wave and provides details
regarding the available data, dam geometry and failure, and expected modelling and
simulation solution requirements. Part 2 primarily focuses on consequence estimation using
the modelling and simulation results from Part 1.
Benchmark Focus
The numerical problem proposed for the workshop consists of estimating the consequences of
failure of a dam near populated areas with complex demographics, infrastructure and
economic activity. The dam in question will be near the city of Hydropolis, a virtual testbed
for flood risk analysis to be built in preparation for the benchmark study.
Theme C participants are free to select the type and sophistication of the simulation engines
used to solve the problem, including 1-d, 2-d and 3-d flood simulation tools, Population at
Risk (PAR) and Loss of Life (LOL) estimation techniques, and infrastructure and
consequence assessment models.
Flood Modelling and Simulation
The following sections are intended to provide information regarding the data provided for the
dam failure modelling and simulation benchmark. Specifically this information includes the
topographic data and dam geometric and construction information. It is not
the intent of this benchmark to set requirements as to which modelling and simulation
environment should be used. Therefore, the descriptions and data provided are intended to be
useful to a wide range of modelling and simulation environments at many levels of fidelity.
Dam Information
A hypothetical embankment dam was constructed in a mountainous region. The high- hazard
dam sits directly above a lightly populated area and 3.5 kilometres away from an urban
environment. The front and rear views of this dam are shown in Figure 2-1 and 2-2,
respectively.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
215
The primary function of the dam is flood control for heavy snowmelt and strong monsoonal
weather patterns. In addition, the reservoir provides some water supply and recreational
activities to nearby communities. The following sections provide more detail regarding the
geometry and the construction of the hypothetical dam.


Figure 1: Front view of dam and surrounding topography


Figure 2: Rear view of dam, reservoir, and surrounding topography

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
216
Dam Geometry
The hypothetical embankment dam is considered high head (61 m) with moderate storage (38
million cubic meters). An overview of the dam geometry is provided in Table 5 and a cross-
sectional profile is shown in Figure 3.
Table 5: Dam geometric parameters
Parameter Description Value
Dam Location (x, y) 4499.66, 6681.57
Crest Length (m) 360
Crest Width (m) 24
Crest Elevation (m) 272
River Bed Elevation (m) 211
Upstream Embankment Slope (?H:1V) 3
Downstream Embankment Slope (?H:1V) 3


Figure 3: Cross-sectional view of the hypothetical dam

The storage capacity at crest elevation is more than 38 million cubic meters. The stage-
volume curve for the reservoir is shown in Table 6.
Table 6: Reservoir stage-volume curve

Elevation (m)
Surface
Area (m)

Volume (m)

Elevation (m)
Surface
Area (m)

Volume (m)
211 0 0 243 605,559 6,214,463
213 898 266 245 682,169 7,500,447
215 7,812 7,432 247 762,461 8,944,612
217 25,502 39,871 249 848,052 10,553,044
219 43,821 109,250 251 916,938 12,322,332
221 62,409 216,078 253 978,640 14,218,482
223 85,038 364,294 255 1,035,581 16,233,223
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
217
225 117,544 565,695 257 1,094,197 18,365,080
227 153,732 836,860 259 1,155,316 20,614,116
229 190,728 1,181,477 261 1,216,975 22,987,820
231 222,068 1,594,314 263 1,278,543 25,483,100
233 252,239 2,067,481 265 1,338,586 28,100,382
235 294,797 2,613,155 267 1,398,359 30,837,356
237 351,808 3,254,198 269 1,467,826 33,701,532
239 461,611 4,084,756 271 1,543,126 36,712,416
241 531,463 5,077,507 272 1,584,052 38,276,344

Dam Construction Information
Homogenous Dam
The dam is a rolled earth fill structure composed of predominantly sandy clays and clayey
sands. Compaction was achieved using 4 passes of a 50-ton pneumatic-tired roller on 0.3-
meter loose lifts.
Strength properties for the dam were developed from results from undrained triaxial strength
tests. The drained strength parameters were based on 5% axial strain and were selected to
represent initial confining stresses up to 478 kPa. The undrained strengths were interpreted
from an approximate evaluation of S
u
/o'
mc
ratios estimated from the reported undrained tests.
The undrained strength S
u
was taken as one-half the maximum deviator stress for axial strains
up to 10%. The estimated strengths are summarized in Table 7.
Table 7: Selected parameters for dam
Parameter Description Value
Strength Parameters:
c' Effective (drained) cohesion in kPa 19.15 kPa
|' Effective (drained) friction angle 14

S
u
Undrained strength in kPa 43.09+
0.175o'
mc

Stiffness Parameters:
V
s1
Shear wave velocity at o'
v
= 1 atm 152.4 m/s
G
max,1
Maximum shear modulus at o'
v
= 1 atm 46443 kPa
Others:

sat
Saturated unit weight
2002 kg/m
k Permeability 1.910
-6
cm/s

Foundation
Foundation stiffness parameters were based on shear wave velocity values without reduction
for strain level or changes in effective stress. A simplified stiffness distribution was used to
reflect the gradual increase in stiffness with depth. The selected properties are summarized in
Table 8.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
218
Table 8: Selected parameters for foundation
Parameter Description Value
Base of dam to
depth of
3.6 m
Foundation
below
3.6 m
Stiffness Parameters:
V
s Shear wave velocity 167 m/s 185 m/s
G Shear modulus
6.7510
5
kPa 8.33 10
6
kPa
v
Poissons ratio 0.25 0.25
Others:

Unit weight 2,242kg/m 2,402 kg/m
K Permeability 9.510
-7
cm/s 9.510
-7
cm/s

Dam Failure Guidance
The dam failure for this benchmark takes place when the pool elevation is at crest elevation.
The mode of failure will be assumed as an overtopping failure. While it is open for
participants to conduct detailed (physically-based) modelling of the breaching process as part
of the benchmark, several breach parameter estimation approaches available in the literature
are summarized in the appendix of this document.
Data Provided
In addition to the information provided in this document, gridded data representing the
topography and the land use classification are provided. All gridded datasets conform to the
same domain and cellsize. The resolution of data provided is considered a base dataset and
may be altered if required by the modelling and simulation environment used by the
participant.
Digital Elevation Data
Two digital elevation models (DEM) are provided to benchmark participants:
- DEM representing pre-dam construction.
- DEM representing post-dam construction.
Participants in the benchmark may decide which DEM is more appropriate for use based on
individual requirements for modelling and simulation.

Land Use/Cover
Benchmark participants are provided a gridded dataset representing the hypothetical land
use/cover for the simulation region. These data follow the classification guidance and values
provided in the National Land Cover Dataset (NLCD). For completeness, a description of
land use classifications is provided in Table 9.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
219
Table 9: Land use/cover classifications and descriptions
NLCD Class Description
11 Open Water
12 Perennial Ice/Snow
21 Developed-Open Space
22 Developed- Low Intensity
23 Developed- Med. Intensity
24 Developed- High Intensity
31 Barren Land
41 Deciduous Forest
42 Evergreen Forest
43 Mixed Forest
52 Shrub/Scrub
71 Grassland/Herbaceous
81 Pasture/Hay
82 Cultivated Cropland

Correlation of these classifications to surface roughness for modelling and simulation should
be reported by the participant.

Flood Modelling and Simulation Reporting Requirements
For consistency between benchmark participants, each participant is requested to generate
information described in Table 10 from the results of the flood simulation. These data focus
on the hydraulics of the simulation.
Table 10: Reporting requirements for flood modelling and simulation
Required Data Data Description
Breach Discharge This hydrograph should show the discharge
rate from pre-failure to empty reservoir, in
units of m3/
Cross-Section Discharge



These shall consist of a complete hydrograph
for unsteady simulation environments and
peak discharges for steady-state simulation
environments
Peak Flood Depths

Gridded dataset representing the peak flood
depth, units of meters
Flood Wave Arrival Time

Gridded dataset with flood wave arrival time
in 5-minute intervals
Peak Unit Flow Rate Gridded dataset with a value representing the
peak unit flow rate in units of m/s
Flooded Area Summation of the total flooded area, units of
m. In addition, participants will provide
flooded area categorized by range of flood
depths at .5 meter intervals

The locations in which participants should provide hydrographs are shown in Table 11

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
220
Table 11: Cross-section locations
Cross Section ID X Location (m) Y Location (m)
1 4737.18 6755.39
2 5971.79 7053.10
3 7486.60 7582.85
4 9100.84 7773.91
5 10716.96 7397.67

Solution Metadata
As described above, Participants are expected to provide gridded data for some of the
reporting requirements. These gridded data files must conform to the metadata of the original
data provided in the benchmark. These are summarized in Table 12.
Table 12: Gridded data metadata requirements
Metadata Parameter Value
Left Extent 0
Bottom Extent 0
Right Extent 25831.81905
Top Extent 9930.941439
Cellsize 9.4760892
Columns 2726
Rows 1048
Acronyms and Abbreviations
DEM Digital Elevation Model
NLCD National Land Cover Dataset
kPa Kilopascal

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
221
Theme C
Computational Challenges in Consequence
Estimation for Risk Assessment: Numerical Modelling, Uncertainty
Quantification, and Communication of Results
Part 2 Consequence Estimation

Sponsoring Organizations:
U.S. Army Corps of Engineers (USA)
U.S. Department of Homeland Security (USA)
Formulators:
Yazmin Seda-Sanabria (Formulation Team Co-Chair)
National Program Manager, Critical Infrastructure Protection & Resilience Program, U.S.
Army Corps of Engineers, Headquarters, 441 G Street NW (ATTN: CECW-HS), Washington,
DC 20314 (USA), Email: yazmin.seda-sanabria@usace.army.mil
Enrique E. Matheu (Formulation Team Co-Chair)
Chief, Critical Lifelines Branch, Sector Outreach and Programs Division, Office of
Infrastructure Protection, U.S. Department of Homeland Security, 245 Murray Lane
Arlington, VA 20598-0608 (USA), Email: enrique.matheu@hq.dhs.gov
Timothy N. McPherson (Formulation Team Technical Lead)
R&D Group Leader, Energy and Infrastructure Analysis, Los Alamos National
Laboratory, Los Alamos, New Mexico 87544 (USA), Email: tmac@lanl.gov
Mustafa Altinakar
Director and Research Professor, National Center for Computational Hydroscience and
Engineering, The University of Mississippi, Brevard Hall Room 327, P.O. Box 1848,
University, MS 38677-1848 (USA), Email: altinakar@ncche.olemiss.edu
Mark Jourdan
Research Hydraulic Engineer, Coastal and Hydraulics Laboratory, U.S. Army Engineer
Research and Development, 3909 Halls Ferry Road, Vicksburg, Mississippi 39180 (USA),
Email: mark.r.jourdan@usace.army.mil
Michael K. Sharp
Technical Director, Geotechnical and Structures Laboratory, U.S. Army Engineer Research
and Development, 3909 Halls Ferry Road, Vicksburg, Mississippi 39180 (USA), Email:
michael.k.sharp@usace.army.mil

For technical information, please contact Dr. Tim McPherson at tmac@lanl.gov

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
222
Introduction
In the last decade, computational capacity has grown dramatically such that multiprocessor
computing techniques are now widely available. This increase in resource availability has
allowed the development of a vast array of new models for flooding and consequence
assessment. Many of these models are computing flood wave propagation at extremely high
temporal and spatial resolutions. When these models are coupled to equally complex models
of population mobility, infrastructure impact and economic consequence, simulation
frameworks are created that can support a paradigm shift from standard approaches to dam
risk analysis.
Although computational advances have increased the availability and applicability of novel
tools, there is a deficiency in benchmarks on the use of those tools in risk assessments. An
obvious application of increased availability and efficiency of computational resources is to
conduct probabilistic risk assessments using Monte Carlo techniques, but the application of
such approaches entails a wide range of assumptions and technical decisions regarding the
management of uncertainty such as variable uncertainty, parameter uncertainty, uncertainty in
probabilistic sub-model, measurement error, computational errors, and numerical
approximation to name a few.
Universities, engineering companies and regulatory bodies are invited to contribute to the
benchmark and take part in the discussion of results gained.
This document is part 2 in a 2-part series for Theme C. Part 1 pertains to the hydraulic
modelling and simulation of the dam breach and subsequent flood wave and provides details
regarding the available data, dam geometry and failure, and expected modelling and
simulation solution requirements. Part 2 focuses on the consequences of the flood using the
modelling and simulation results from Part 1.
Benchmark Focus
The numerical problem proposed for the workshop consists of estimating the consequences of
failure of a dam near populated areas with complex demographics, infrastructure and
economic activity. The dam in question will be near the city of Hydropolis, a virtual testbed
for flood risk analysis to be built in preparation for the benchmark study.
Theme C participants are free to select the type and sophistication of the simulation engines
used to solve the problem, including 1-d, 2-d and 3-d flood simulation tools, Population at
Risk (PAR) and Loss of Life (LOL) estimation techniques, and infrastructure and
consequence assessment models.
Consequence Estimation
A full consequence assessment for dam failure generally includes four main categories- public
health and safety, economic impact, psychological impact, and governance/mission impact. At
a minimum, the consequence assessment should focus on
impacts related to human consequences and direct economic impact. This document outlines
the requirements for consequence assessment within Theme C of the benchmark, addressing
human consequences (e.g., population at risk and loss of life) and direct economic impact.
The intent is not to limit the approaches or techniques used by participants but to guide the
assessment into categories for comparison. Additional consequence assessments are also
encouraged, but will not be used in the assessment of uncertainty.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
223
Dam Failure Assumptions
Over a series of fall rainfall events, the reservoir slowly approaches crest elevation. Due to the
frequency of fall rainfall events, there are few people recreating at or near the facility. At
11:00 pm on Saturday evening, the dam fails, releasing a torrent of water through the canyon
and towards Hydropolis. The failure was sudden and unexpected, and for those in the canyon
witnessing the failure, there are no mobile phone services. The reservoir and dam does not
have a completed emergency action plan, thus the potential impact related to failure is
unknown to residents in Hydropolis.
Data Provided
Participants are requested to evaluate the downstream consequences as it relates to population
at risk, loss of life and direct economic impact. Participants in Benchmark Theme C are
provided additional data to assist in consequence assessment. This data includes Hydropolis
census information in shapefile format and parcel data in shapefile format.

Census Data
Participants are provided with a shapefile representative of census data for the city of
Hydropolis. This data includes information regarding the population and the economic
activity. Table 5 and Table 2 provide attribution information for these data.
Table 1: Population data field names and descriptions
Field Names Description
Total Total Male and Female
Male Total Male
Munder5 Male: Under 5 years
M5to9 Male: 5 to 9 years
M10to14 Male: 10 to 14 years
M15to17 Male: 15 to 17 years
M18to19 Male: 18 and 19 years
M20 Male: 20 years
M21 Male: 21 years
M22to24 Male: 22 to 24 years
M25to29 Male: 25 to 29 years
M30to34 Male: 30 to 34 years
M35to39 Male: 35 to 39 years
M40to44 Male: 40 to 44 years
M45to49 Male: 45 to 49 years
M50to54 Male: 50 to 54 years
M55to59 Male: 55 to 59 years
M60to61 Male: 60 and 61 years
M62to64 Male: 62 to 64 years
M65to66 Male: 65 and 66 years
M67to69 Male: 67 to 69 years
M70to74 Male: 70 to 74 years
M75to79 Male: 75 to 79 years
M80to84 Male: 80 to 84 years
M85over Male: 85 years and over
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
224
Female Total Female
Funder5 Female: Under 5 years
F5to9 Female: 5 to 9 years
F10to14 Female: 10 to 14 years
F15to17 Female: 15 to 17 years
F18to19 Female: 18 and 19 years
F20 Female: 20 years
F21 Female: 21 years
F22to24 Female: 22 to 24 years
F25to29 Female: 25 to 29 years
F30to34 Female: 30 to 34 years
F35to39 Female: 35 to 39 years
F40to44 Female: 40 to 44 years
F45to49 Female: 45 to 49 years
F50to54 Female: 50 to 54 years
F55to59 Female: 55 to 59 years
F60to61 Female: 60 and 61 years
F62to64 Female: 62 to 64 years
F65to66 Female: 65 and 66 years
F67to69 Female: 67 to 69 years
F70to74 Female: 70 to 74 years
F75to79 Female: 75 to 79 years
F80to84 Female: 80 to 84 years
F85over Female: 85 years and over
Table 2: Economic table field names and descriptions
Field Names Description
jobs11 Agriculture, Forestry, Fishing and Hunting
jobs21 Mining, Quarrying, and Oil and Gas Extraction
jobs22 Utilities
jobs23 Construction
jobs3133 Manufacturing
jobs42 Wholesale Trade
jobs4445 Retail Trade
jobs4849 Transportation and Warehousing
jobs51 Information
jobs52 Finance and Insurance
jobs53 Real Estate and Rental and Leasing
jobs54 Professional, Scientific, and Technical Services
jobs55 Management of Companies and Enterprises
jobs56
Administrative and Support and Waste Management and Remediation
Services
jobs61 Educational Services
jobs62 Health Care and Social Assistance
jobs71 Arts, Entertainment, and Recreation
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
225
jobs72 Accommodation and Food Services
jobs81 Other Services (except Public Administration)
jobs92 Public Administration

Parcel Data
Benchmark participants are provided parcel data in shapefile format. This includes
information regarding the zoning (e.g., residential, commercial, etc.) and the structure (e.g.,
stories, basement, quality). Table 3 provides a description of the zone classification provided
in this data.
Table 3: Parcel descriptions for zone classifications
Code Description
A-1 Rural Agricultural1-Acre Minimum Zone
A-2 Rural Agricultural2-Acre Minimum Zone
AP Airport Protection Overlay Zone
C-1 Neighborhood Commercial Zone
C-2 Community Commercial Zone
C-3 Heavy Commercial Zone
IP Industrial Park Zone
M-1 Light Manufacturing Zone
O-1 Office & Institutional Zone
P Parking Zone
P-R Reserve Parking Zone
R-1 Residential Zone: Houses
R-1 Single Family Residential Zone
R-2 Residential Zone: Houses, Townhomes & Medium Density Apartments
R-3 Residential Zone: Houses, Townhomes & High Density Apartments
RA-1 Residential and Agricultural Zone, Semi-Urban Area
RA-2 Residential and Agricultural Zone
R-D Residential and Related Uses Zone, Developing Area
R-LT Residential Zone: Houses & Limited Townhomes
RO-1 Rural and Open Zone
RO-20 Rural and Open Agricultural Zone
R-T Residential Zone: Houses & Townhomes
SD-LC-1 Limited Neighborhood Commercial Zone
SD-RO Residential/Office Zone
SU-1 Special Use Zone
SU-2 Special Neighborhood Zone, Redeveloping Area
SU-3 Special Center Zone
UCO Urban Conservation Overlay Zone
WO Wall Overlay Zone

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
226
Consequence Assessment Reporting Requirements
All participants should summarize methods, techniques, assumptions, and key results in a
paper not to exceed 15 pages. In addition, each participant is requested to generate the
information described in Table 4 using results of the consequence assessment. These data
focus on the consequences of failure.
Table 4: Reporting requirements for consequence assessment
Required Data Data Description
Population at
Risk
Participants should use the provided consequence results spreadsheet to
populate the requested information. This information includes age
demographic breakdowns by time to flooding and depth of flooding. A
gridded dataset indicating the spatial variability of population at risk
(regardless of age) is also requested.
Loss of Life
Participants should use the provided consequence results spreadsheet to
populate the requested information. The loss of life estimation includes
breakdowns by time from breach failure. A gridded dataset indicating the
spatial variability of loss of life is requested.
Flood Severity
Grid
Gridded dataset representing the peak flood severity, classified as low,
medium and high (e.g., 1, 2, 3) severity. Participants are expected to
describe the assumptions used in categorizing flood severity in the
submitted paper.
Direct Economic
Impact
Participants should use the provided consequence results spreadsheet to
populate the requested information. This information includes break
downs by time from dam failure. A gridded dataset indicating the spatial
variability of direct economic impact is also requested.

Consequence Metadata
As described above, participants are expected to provide gridded data for some of the
reporting requirements. These gridded data files must conform to the metadata of the original
data provided in the benchmark. These are summarized in Table 5.
Table 5: Gridded data metadata requirements
Metadata Parameter Value
Left Extent 0
Bottom Extent 0
Right Extent 25831.81905
Top Extent 9930.941439
Cellsize 9.4760892
Columns 2726
Rows 1048


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
227



RESULT COMPARISON OF THE
PARTICIPANTS

THEME C



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
229
Results Comparison

Theme C
Computational Challenges in Consequence
Estimation for Risk Assessment
Introduction
Theme C of the 12
th
International Benchmark Workshop on the Numerical Analysis of Large
Dams involved 8 participants who submitted at least partial solutions of the Hydropolis Dam
Break Case Study. This synthesis is not a complete representation of each solution. In many
cases, participants conducted their own sensitivity/uncertainty analyses. For brevity, this
synthesis document considered only one set of solutions from each participant. Full
description of solutions can be found in the papers authored by each participant team.
This synthesis of results is based on statistical methods to facilitate comparison to the greatest
extent possible. The Kappa statistic is used to quantify the similarities among the participants
study results. Kappa corresponds to a numerical rating of the degree of agreement between
two raters, which in this case are the outputs (e.g., gridded data peak flood depths) of two
participant models. Kappa is calculated in Equation 1.

(1)

where P(a) is the relative observed agreement between the two models and P(e) is the
probability that the agreement is due to chance. Kappa ranges from -1 to 1. If the two models
are in perfect agreement then Kappa equals 1, whereas if there is no agreement among the
models except what would be expected by chance. A Kappa value less than 0 indicates more
disagreement than what one can expect by chance. A commonly cited scale for interpreting
Kappa values is given by Landis and Koch
1
and reported in the following Table 1. The Kappa
statistic quantifies the aggregate agreement between two different spatial grids, but does not
provide any spatial information about where the two models differ more or less.
Table 1: Kappa value interpretation as proposed by Landis
Kappa Agreement Interpretation Agreement Index
(used in Results Tables)
< 0 Less than chance agreement Chance
0.01-0.20 Slight agreement Slight
0.21-0.40 Fair agreement Fair
0.41-0.60 Moderate agreement Moderate
0.61-0.80 Substantial agreement Substantial
0.81-0.99 Almost perfect agreement Almost perfect


1
Landis J.R., and Koch G.G., The measurement of observer agreement for categorical data. 1977 Biometrics
33, 159-174
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
230
Overview of Participation
The benchmark case includes a diverse set of analyses and several participants only submitted
results against a subset of the complete problem. Table 2 lists each of the lead authors, an ID
assigned to the participant team for solution comparison, and an indication of submission
completion for topics of comparison.
Table 2: Summary of Participant Submissions
Authors ID Hydraulic
Solution
Population
at Risk
Solution
Loss of
Life
Solution
Economic
Solution
Davison et al. HRW Yes Yes Yes Yes
Bent et al. LANL Yes Yes Yes Yes
Mancusi et al. MANC Yes Yes Yes Yes
Williams & Buchanan MMC Yes Yes Yes Yes
Thames & Kalyanapu TK Yes Yes Yes Yes
Altinakar et al. MUST Yes Yes Yes No
Saberi et al. SAB Yes No No No
McVan et al. MCV Yes Yes No No
Discussion
Breach formulation
Participants used a range of models, including physics-based breach models using dam
material information and regression equations based on previous dam failures. Figure 1 shows
the breach hydrographs produced by each team. It is clear that choice of model, method, and
parameters can significantly affect the timing and the magnitude of the peak discharge. In
general, models using regression equations had a much earlier peak than physics-based
models. The SAB and HRW hydrographs were shifted in time to better illustrate the data for
comparison. The peak discharge ranges in time from under 0.5 hours to more than 18 hours
and the magnitude of peak discharge ranges from near 10,000 cubic meters per second (cms)
to more than 40,000 cms. SAB, MMC, LANL, MANK, and TK all have peak discharges
above 20,000 cms, while MCV, MUST, and HRW have peaks below 20,000 cms.


Figure 1: discharge hydrographs
~4 hr shift for
comparison
~16 hr shift for
comparison
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
231
Hydrodynamics
Each participant used different models, which solve the same equations using different
numerical schemes. The most notable difference between models appeared to be the
difference in mesh or grid type used. Models using structured grids tended to have more
similar results relative to peak depths than models using irregular meshes. This may be a
result of interpolation between the two types of meshes for comparison. A wide variety of
techniques were used by the participant teams to produce the necessary flood output data for
consequence analyses. Each participant submission details the methods used. These
differences can be a strong explanatory variable in the comparison of consequence results. For
example, dam failure hydrodynamic simulation results are dependent on the input datasets
(e.g., breach discharge, topography, roughness) and modeling\method approach. Participants
used different approaches to estimate roughness coefficients. Each participant was required to
interpret the roughness coefficient associated with an NLCD land use type. Because there
isnt a single accepted dataset, there were a range of coefficients used by participants.

Cross Sections
The participants were provided a single point location for each section, and the participants
estimated the line upon which flow should be summed. Figure 2 shows the hydrographs
submitted by each of the participants for all cross sections. SAB and HRW have the same
time shifts as previously noted.


Figure 2: Hydrographs for all cross sections submitted by benchmark participants
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
232
Peak Flood Depths
The peak flood depth was defined by all participants as the peak depth at every cell regardless
of the time at which it occurred. The maximum flooded area and flooded area categorized in
half-meter increments of peak depths are summarized in Figure 3 and Figure 4. The total
flooded area ranges from near 47 km to just under 30 km.
The Kappa statistic was applied to the gridded peak flow depths. A Kappa value was
calculated for each unique combination of two models. The results of the peak flood depth
Kappa analysis are summarized in Table 3. There is generally strong agreement between the
models. The chance and moderate differences between MCV and SAB and other models is
likely due to interpolation of the irregular mesh output from those submissions to a new mesh
for comparison with other models.


Figure 3: Summary of total flooded area


Figure 4: Flood area summarized by half-meter increments of peak depth


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
233
Table 3: Kappa statistic (model similarity) for peak flood depth
HRW LANL MANC MMC TK MUST MCV SAB
HRW Substantial Substantial Substantial Substantial Substantial Moderate Chance
LANL Substantial Substantial Substantial Substantial Substantial Moderate Chance
MANC Substantial Substantial Substantial Substantial Substantial Moderate Chance
MMC Substantial Substantial Substantial Substantial Substantial Moderate Chance
TK
Substantial Substantial Substantial Substantial Substantial Moderate Chance
MUST
Substantial Substantial Substantial Substantial Substantial Moderate Chance
MCV
Moderate Moderate Moderate Moderate Moderate Moderate Chance
SAB
Chance Chance Chance Chance Chance Chance Chance

Flood Wave Arrival Time
There were differences among participants in defining thresholds for determining when the
flood wave arrived. For example, LANL and MUST defined the flood wave arrival as the time
at which a grid cell becomes wet without implying a threshold. MMC defined the flood wave
arrival as the time in which the depth of water in a grid cell reaches 0.6 meters. Agreement
between participants was assessed using the Kappa Statistic and are shown in Table 4. The
results indicate that the level of agreement between any two models varies more for the flood
wave arrival time than for the peak flood depths. It is noted again that HRW used a breach
hydrograph with a much longer lag time to peak discharge than other participants.
Table 4: Kappa statistic (model agreement) for flood wave arrival time

HRW LANL MANC MMC TK MUST MCV SAB
HRW
Chance Chance Chance Chance Chance Chance Chance
LANL Chance Slight Substantial Slight Fair Substantial Moderate
MANC Chance Slight Slight Chance Fair Slight Fair
MMC Chance Substantial Slight Slight Moderate Substantial Substantial
TK Chance Slight Chance Slight Slight Slight Slight
MUST Chance Fair Fair Moderate Slight Fair Substantial
MCV Chance Substantial Slight Substantial Slight Fair Moderate
SAB Chance Moderate Fair Substantial Slight Substantial Moderate


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
234
Peak Unit Flow Rate
Table 5 shows the level of agreement between participants for Peak Unit Flow Rate as
determined by the Kappa analysis. The results show less agreement between two models for
the peak unit flow rate than for peak flood depth and flood wave arrival.
Table 5: Kappa statistic (model agreement) for peak unit flow rate
HRW LANL MANC MMC TK MUST MCV SAB
HRW Fair Moderate Slight Slight Substantial Fair Fair
LANL Fair Fair Substantial Chance Fair Slight Slight
MANC Moderate Fair Fair Slight Moderate Slight Slight
MMC Slight Substantial Fair Chance Fair Slight Slight
TK Slight Chance Slight Chance Slight Slight Chance
MUST Substantial Fair Moderate Fair Slight Fair Fair
MCV Fair Slight Slight Slight Slight Fair Slight
SAB Fair Slight Slight Slight Chance Fair Slight
Consequences
Population
The spatial distribution of population within the flooded area varied among the participants.
MCV, MANC, and MUST uniformly distributed the population available in the census blocks
to properly sum the affected population within partially flooded census blocks. HRW, TK,
and MMC redistributed the population to the parcel data provided. MMC additionally
accounted for residential population and workforce population within the parcels. Finally,
LANL used imperviousness defined by developed areas in the NLCD to distribute the
population from each census block.

Population at Risk
Population results include total PAR and PAR by age (e.g., under 14 and over 65 years).
These are summarized in Figure 5 and Figure 6.


Figure 5: Summary of total, under 14-yr, and over 65-yr PAR

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
235
Despite the wide range of peak discharges and subsequent flooded areas, the total sum of PAR
is fairly consistent among participants. In addition, the participants consistently indicate that
the majority of the PAR is located in flood water less than 2 meters deep. While there is
consistency in total population by depth, PAR is less consistent with respect to flood arrival
time. This indicates that PAR is perhaps sensitive to breach formulation. For example, HRW
had nearly a 16-hour lag time for peak discharge. Therefore, the PAR in their solution is not
generally at risk until more than 3 hours. This information is likely to have an impact on loss
of life models that take into account warning time and evacuation plans.


Figure 6: PAR by peak depth

Flood Severity
Figure 1-7 shows the standard deviation among the different flood severity calculations.


Figure 7: Flood severity standard deviation

There are significant differences on how the flood severity is defined across the participants.
The methods used by TK and MMC were not described. HRW defined flood severity based
on effects to population who are exposed (i.e., not within buildings) to the flood waters.
LANL, MUST, and MANC used the U.S. Department of Homeland Securitys Dam Sector
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
236
consequence estimation guidelines.
2
These classifications are based more on structural
stability and population within buildings than people directly exposed to the flood. There is
complete agreement between the models on the severity assigned within the canyon (standard
deviation of 0). Differences in severity classification are noted within the floodplain area,
which the area where population in this case study is located.

Loss of Life
The total loss of life is shown in Figure 8. Although the method in which the loss of life was
estimated varied between participants, the majority of estimates for loss of life was
approximately 2,000 people. HRWs loss of life estimate was the same order of magnitude as
the other teams, but nearly twice as great. An explanatory factor in this difference could be the
different technique used to calculate flood severity.


Figure 8: Total loss of life

Economics
There were differences in definition of direct economic impact. For example, LANL defined
the direct economic impact as lost jobs, wages, and business outputs. LANL was the only
participant to use business gross domestic product (GDP) per employee, but they did not
include structural damage, which they defined as an insurable loss and not a direct economic
impact. As such, LANL economic results are reported on a $/day basis, but did not estimate
the duration of the impact. MAN, MMC, TK, and HRW all evaluated structural damages
using the provided parcel information. The key challenge using the parcel data was assigning
values to the occupancy classes. Because this information was not provided by the
formulation team, each participant had to assume this information. The data used by
participants ranged from available data in models, economic subject matter experts, and
values reported in the literature. Figure 9 shows the total economic impact. Because of the
difference in methods (e.g., GDP vs. insurable losses), and the assumption regarding asset
value, the economic impact values range significantly. LANL reported $665,000/day, but the
duration of the event is unknown. At the high end, MANC reported $2.6 Billion based on
insurable losses.

2
Department of Homeland Security, 2011. Estimating loss of life for dam failure scenarios.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
237

Figure 9: Summary economic impact submitted by the participants
Conclusion
In general, the results from each participant team are similar in terms of hydrodynamics.
However, flood wave arrival times were different between teams. This is attributable to
differences in the calculation of the breach hydrograph and differences in regression and
physics-based formulations in addition to definition of thresholds used to quantify the arrival
of the flood. The largest differences in peak flood depths were likely due to the requirement
for the teams using irregular meshes to report output in a structured grid format.
PAR estimates were also similar across teams. Differences are primarily due to different
methods used by participants to spatially distribute population. The methods included uniform
distribution of the population within the census block, uniform distribution of the population
among parcels within the census blocks, and uniform distribution of the population among
develop land within the census block. The most significant differences in the population
estimates appeared to be the time after failure at which population is at risk. This has
significance relative to flood wave warning time and potential evacuations.

Finally, there were some significant differences in the economic impact. First, there were
differences in interpretation of direct economic impact. This was either interpreted as business
losses based on GDP or structural damage/insurable losses using parcel data. In addition,
within the insurable loss estimation, there were significant differences in assessing the value
of structures within parcels.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
239



PAPERS

THEME C



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
241
A Benchmark study on dam breach and consequence
estimation using EMBREA and Life Safety Model
M. Davison
1
, M. Hassan
1
, O. Gimeno
1
, M. van Damme
1
and C. Goff
1

1
HR Wallingford, Howbery Park, Wallingford, Oxfordshire, OX10 8BA, United Kingdom
E-mail: m.davison@hrwallingford.com
Abstract
This paper presents the modelled consequences of a hypothetical dam breach as laid out in
Theme C of the 12
th
international Benchmark Workshop on Numerical Analysis of Dams. The
EMBREA model was used to model the failure of the dam due to headcut erosion and to
derive a breach hydrograph which was then used in InfoWorks ICM to model the 2D flood
spreading. The results of the flood model have been used to calculate how severe the flood
would be in terms of the total population at risk, loss of life using the Life Safety Model and
economic impact. The paper shows that the consequences of the hypothetical flood will be
severe in terms of casualties and economic damage.
Introduction
This paper presents the modelling of a potential breach of a hypothetical dam and the
estimation of the consequences in the populated areas and economic activities downstream,
within Theme C of the 12
th
International Benchmark Workshop on Numerical Analysis of
Dams [1]. The paper is structured in three sections. Section 1 describes the dam breach model,
which provides indicative predictions of a number of breach scenarios; Section 2 focuses on
the flood wave propagation and Section 3 presents the human consequences and direct
economic impact of the dam failure.
Dam Breach Modelling
Choice of Breach Model
During the last 15 years, HR Wallingford has undertaken the development of a new breach
model, called EMBREA (previously known as HR BREACH). The IMPACT project [2] in
2005, Peeters et al. [3] in 2011, and Morris et al [4] in 2012 have demonstrated that EMBREA
performs the best among the available breach models. On that basis, EMBREA was used to
undertake the modeling for this benchmark work.
Selection of Failure Modes
EMBREA can simulate breach failure by overtopping and piping. For this study overtopping
is considered to be the main dam failure mode, but piping was also investigated in one of the
scenarios.

Overtopping failure results in breach formation either due to headcut erosion or surface
erosion. With headcut formation the erosion of material forms steps in the downstream face of
the embankment which progressively grow in size and cut through the embankment. When
the retreating headcuts reach the upstream side of the crest, the hydraulic control rapidly
reduces in height and rapid failure ensues. With surface erosion, material is eroded from the
embankment face and the crest area more smoothly leading to a reduced cross sectional
embankment profile.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
242
The dominant process depends mainly on the material used in the dam construction. The
hypothetical dam in this instance is a rolled earth fill structure composed of predominantly
sandy clays and clayey sands with a relatively low permeability (1.9x10
-6
cm/s). Based on the
type of soil and the permeability provided, and using Shevnin et al. [5], its clay content is
estimated to be more than 25%. In addition to the clay content, the matric suction in
unsaturated soils adds to the cohesive properties of soils. The low permeability of the
embankment soil indicates the high likelihood of the presence of an added cohesion due to
matric suction throughout the erosion process. The overall cohesive properties make headcut
erosion the most likely erosion mode during overtopping failure.

The headcut erosion process is modeled in EMBREA using an erodibility coefficient (K
d
),
and a headcut migration coefficient (C). These parameters are usually measured on site or in a
soil laboratory, but as the dam on this study is hypothetical, they were estimated based upon
the soil clay content and compaction effort. To assess their uncertainty, the Monte Carlo
analysis feature in EMBREA was used as described in below.

A number of modeling runs were undertaken to establish the potential worst case scenarios
for the hypothetical dam failure. Within these runs, consideration was given to the uncertainty
and sensitivity of results to various aspects such as material erodibility, failure mode and
erosion processes, as follows:

- Runs 1, 2 and 3. In these runs EMBREA was used to perform a Monte Carlo
analysis to the overtopping failure with headcut (1000 runs with a triangular
probability distribution for K
d
and corresponding C values as given in Table 4). The
output of the Monte Carlo run analysis was a frequency distribution of the peak
breach outflow. The median value of this distribution was taken as the base run (Run
1). The corresponding values of the 75 % (Run 2) and 25 % (Run 3) exceedance
probabilities were taken as the lower and upper limits respectively.
- Run 4. In this run, the erosion process was changed to overtopping through
surface erosion rather than headcut to assess the impact of a different erosion
process on the breach hydrograph. Other inputs were identical to Run 1.
- Run 5: In this run, a good grass protection is assumed to be present on the
downstream face of the embankment to assess its impact on delaying the breach
initiation and hence the wave arrival time at downstream locations. Other inputs
were identical to Run 1.
- Run 6: In this run, the failure mode was changed to piping rather than overtopping
to assess the impact of a different failure mode on the breach hydrograph. It should
be noted that surface erosion is used in the model once the top of the pipe becomes
unstable and fails (i.e. failure mode changes from piping to overtopping). Other
inputs were identical to Run 1.
Model set up
This section provides a description of the model set up, including modeling boundary
conditions, initial conditions, dam geometry and soil properties. The selection of such data
was mainly based upon the data provided to participants in [1]. Due to the size of the
reservoir, any inflow into the reservoir during the breach event has a negligible effect on the
reservoir water levels, and has therefore not been included. Based on investigation of the
immediate reach downstream of the dam, a low tailwater level was assumed meaning that the
breach flow was not drowned. The dam geometry data is given in Table 1. Table 2 shows
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
243
the initial conditions used in each series of runs. The dam failure for this benchmark takes
place when the pool elevation is at crest elevation. Table 3 shows soil properties used in each
series of runs and Table 4 shows K
d
and C values for each run. The K
d
values and C values
were obtained from the qualitative relationships for different soil types provided in the
EMBREA user manual version 1.3.

Table 1: Dam
geometry


Crest level
Foundation
level

Crest Width

Crest length
Upstream
slope
Downstream
slope
272 mAD 211 mAD 24m 360m 1:3 1:3

Table 2: Initial
conditions

Run
No.
Pipe diameter
(m)
Pipe Level
(mAD)
Initial water
level (mAD)
Initial breach
depth (m)
Initial breach
width (m)
1-5 NA NA 272 0.25 0.5
6 0.03 212 272 NA NA

Table 3: Soil
properties

Porosity 0.40 Friction Angle (degrees) 14
Unit Wt (KN/m
3
) 19.64 Cohesion (KN/m
2
) 19.15
Mannings n 0.025 Plasticity Index 10

Table 4: K
d
and C
values

Run No. Erodibility coefficient, K
d
(cm
3
/N-s) Headcut migration coefficient, C (s
-2/3
)
Monte
Carlo
Triangular probability distribution
with a lower value of 1, mid value
of 5 and an upper value of 10
Values calculated from K
d
values
based on EMBREA user manual
version 1.3
1, 5 and 6 5.5 0.0032
2 4.3 0.0026
3 6.8 0.0040
4 5.5 NA
Modelling results, observations and conclusions
Figure 1 shows a comparison of the breach outflow results for Runs 1 to 6. Runs 1, 2 and 3
hydrographs show that K
d
and C values have a significant effect on the breach peak outflow
and initiation time in the case of overtopping failure due to headcut formation. Increasing
them by a factor of 1.3 (Run 1) and 1.6 (Run 3) has increased the breach peak outflow by
about 40 and 100% compared to run 2 respectively. Run 4 shows that changing the type of
erosion process results in a significant reduction in the peak breach outflow and initiation
time. The peak outflow was reduced by about 78% compared to Run 1 with less than one hour
of initiation time. Run 5 shows that for the case considered, a grass protection increases the
initiation time of the breach with a negligible change in the peak breach outflow compared to
Run 1. Run 6 shows that changing of the failure mode on breach formation to piping
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
244
significantly reduces the peak breach outflow and initiation time. The peak outflow was
reduced by about 30 % compared to Run 1, with the shortest initiation time of all the runs.
Figure 1: Breach outflow hydrograph for Runs 1 to 6
Flood Modelling
Data provided
The data provided for the flood model to participants was:
- digital elevation model (DEM) representing pre-dam and post-dam construction
- gridded dataset of land use/cover
The horizontal resolution of both datasets is about 9.5m. The land use/cover data follows the
classification guidance and values provided in the National Land Cover Dataset (NLCD),
2006.
Roughness estimation
Surface roughness values for the flood simulation were estimated for each land use/cover
class using the Conveyance and Afflux Estimation System (CES/AES) software [6]. The tool
provides a database of roughness values extracted from various sources in the literature and
given as mid, upper and lower estimates covering the range of roughness values expected
within each system. This provides some measure of the uncertainty associated with the
estimation of roughness by deriving an upper and a lower roughness credible scenario. The
values assigned to each land use/cover class are shown in Table 5, as well as the total areas
covered in the test bed region. For the developed areas, the low and high roughness values
correspond to the lowest and highest values derived from the possible share between
pervious/impervious areas given in the descriptions.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
245
Table 5: Land use/cover estimated roughness


Class

ID

Description

Area
Winter Summer
Mid Low High Mid Low High
Water 11 Open Water 16.9% 0.010 0.010 0.010 0.010 0.010 0.010





Developed

21
Developed,
Open Space

5.2%

0.029

0.025

0.034

0.029

0.025

0.034

22
Developed,
Low Intensity

10.7%

0.030

0.026

0.035

0.030

0.026

0.035

23
Developed,
Medium Intensity

5.1%

0.031

0.026

0.036

0.031

0.026

0.036

24
Developed
High Intensity

0.5%

0.032

0.025

0.036

0.032

0.025

0.036

Barren

31
Barren Land
(Rock/Sand/Clay)

0.2%

0.022

0.017

0.028

0.022

0.017

0.028


Forest
41 Deciduous Forest 28.4% 0.073 0.049 0.112 0.102 0.072 0.162
42 Evergreen Forest 7.9% 0.251 0.151 0.341 0.251 0.151 0.341
43 Mixed Forest 0.1% 0.142 0.091 0.205 0.142 0.091 0.205
52 Shrub/Scrub 20.9% 0.073 0.048 0.112 0.102 0.072 0.162

Herbaceous

71
Grassland/
Herbaceous

3.5%

0.046

0.027

0.083

0.046

0.027

0.083
Planted/
Cultivated

81

Pasture/Hay

0.4%

0.046

0.027

0.083

0.046

0.027

0.083
Wetlands 90 Woody Wetlands 0.2% 0.054 0.039 0.064 0.054 0.039 0.064
Flood model
The 2D Infoworks ICM software was used to simulate spreading of the flood given by the
breach hydrograph into the valley downstream of the dam. Three roughness scenarios were
considered that correspond with the mid, low and high roughness values for each land use
class. For each roughness scenario, the model was run for each of the six dam breach
hydrographs from the breach model. The hydrographs were used as an inflow boundary
condition for the flood model. This was represented using the breach width and depth
corresponding to the maximum breach outflow in combination with the post-dam construction
digital elevation model.
Results
The results of the flood model presented in this section correspond to the overtopping failure
with headcut failure mechanism only, as this is the most likely failure mode for the dam. The
results for other failure modes can be provided upon request.

Figure 3 shows the attenuation of the breach hydrograph of the median scenario (Run 1)
between the five cross sections on Figure 2. As shown, the peak flow is reduced by 10% when
the flood arrives at cross section 5, which takes about 14 min from the moment of the peak
breach discharge. In the scenario with 75% exceedance probability (Run 2), the peak flow
attenuation is of a similar amount (9%), but the wave takes 10 min to reach cross section 5.
Finally, for the scenario with 25% exceedance probability (Run 3), the peak flow hardly
decreases (see Table 6).

The flood extents for the three scenarios are very similar, with the flood spreading to the
valley on the north between sections 3 and 4 for the median and 75% scenarios (see Figure 2).
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
246
The peak flood depths for the 25%, the Median, and 75% exceedance probability scenarios
are given in Table 7. Flood depths are mostly below 2m for all three scenarios, and below 4m
for more than 90% of the flooded area. The spatial variability of the peak flood depths is
shown in Figure 4 at 0.5m intervals for the median scenario. The higher flood depths in the
city occur between cross sections 3 and 5 and also along the south end of the flood extent
downstream section 5. That is also the main flowpath through the plain, with unit flow rates
mainly between 4 and 10 m
2
/s (Figure 5), increasing to 30m
2
/s near section 5. The area of the
city with higher unit flow rates up to 80m
2
/s- is between cross sections 4 and 5.

The arrival time of the first inundation relative to the time at which a breach flow initiates is
shown in Figure 6. The flood reaches the populated area after 1.5h to 2h, but the majority of
the city starts flooding after 2h. That is the arrival time of the small flow at the beginning of
the dam breach initiation. However, the peak flow travel time in Figure 7 shows that the flood
wave peak travels from the dam to the city in less than 15min, reaching the further edge in
less than 90min, and hence the peak flow travel time is the travel time to consider.





















Figure 2: Cross-section locations and flood extent


Figure 3: Cross section discharges median run
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
247



Scenario
Table 6: Cross section discharges
Breach Cross Section 5
Peak flow (m
3
/s) Time Peak flow (m
3
/s) Time
Median 13,956 18:51 12,540 (-10.1%) 19:05 (14min)
75% Exc. Prob. 19,257 15:05 17,537 (-8.9%) 15:15 (10min)
25% Exc. Prob. 9,908 23:51 9,877 (-0.3%) 24:05 (14min)



























Figure 4: Peak flood depths (m) for the median scenario

Table 7: Flooded area (m
2
)


Peak Flood Depth

25% exceedance

Flooded area (m)
75% exceedance

Range (m)
probability
Median

probability
0 - 0.5 7,684,136 3,842,921 3,776,472
0.5 - 1 9,977,083 8,084,897 8,525,347
1 - 1.5 5,110,216 6,912,247 7,732,446
1.5 - 2 3,012,575 3,611,785 4,025,567
2 - 2.5 2,115,959 2,523,993 2,710,321
2.5 - 3 1,231,376 2,305,789 1,967,706
3 - 3.5 515,700 1,875,305 1,380,169
3.5 - 4 355,773 1,185,490 696,280
4 - 4.5 186,327 683,978 424,916
4.5 - 5 136,850 463,618 386,304
5 - 5.5 143,135 354,426 256,907
5.5 - 6 95,004 222,336 162,801
6 - 6.5 61,959 151,037 150,050
6.5 - 7 57,559 131,372 140,801
7 - 7.5 61,421 113,772 82,433
7.5 - 8 58,368 86,025 71,119
>8 522,524 900,657 973,392
TOTAL 31,325,966 33,449,648 33,463,027
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
248



















Figure 5: Peak unit flow rate in m
2
/s for the median scenario




















Figure 6: Inundation arrival time from initiation time for the median scenario





















Figure 7: Peak flow travel time
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
249
Consequence Estimation
Flood severity
Flood Severity has been classified according to the Defra methodology presented in technical
reports [7, 8]. The method describes the flood hazard as a function of velocity, depth and the
presence of debris as

(1)
Where, HR is the (flood) Hazard Rating, d is the Depth of flooding (m), v is the velocity of
the floodwater in (m/s), and DF is the debris factor (assumed to be 0.5 for depths <0.25m, and
1.0 for depths >0.25m).

Four flood hazard classes were established based on the thresholds of flood hazard rates (HR)
that cause people to lose stability. Thresholds are derived from the experiments on people
with different height multiplied by mass values. For this study, a 3 class categorization (low,
medium and high) was required. Therefore the thresholds have been redefined as in Table 8.
Figure 8 shows that the flood severity is mostly high in the flooded area.

Table 8: Flood Severity classification


ID

Class
Flood Hazard
Rate (HR)

Description
1 Low < 0.75 Caution. Flood zone with shallow flowing water or deep
standing water
2 Medium 0.75 - 1.25 Dangerous for some (i.e. children, the elderly and the
infirm). Danger: Flood zone with deep or fast flowing
water
3 Significant
to high
>1.25 Dangerous for most people to all people.
Danger to extreme danger: flood zone with deep fast
flowing water





















Figure 8: Flood severity
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
250
People at risk
Given the reporting requirements of the people at risk by time intervals and flood depths, the
calculation of the number of people at risk was based on the flood extent from the hydraulic
model and the population from the census data and the buildings. The average number of
people in a building was calculated by dividing the total population by the number of
buildings for each census zone. To calculate the number of people at risk from flooding the
number of buildings in the maximum flood extent was determined from GIS and the
population summed. Table 9 shows the total number of people at risk per age group and flood
depth. Most of the people are at areas with flood depth below 1.5m.

Table 9: Number of people at risk per age group and flood
depth

Peak Flood Depth
Range (m)
Total Population At
Risk
12-yr and Under
Population at Risk
65-yr and Over
Population at Risk

0 - 0.5 4,608 693 841
0.5 - 1 10,189 1,528 1,878
1 - 1.5 5,336 763 1,139
1.5 - 2 1,209 146 326
2 - 2.5 710 113 163
2.5 - 3 621 118 111
3 - 3.5 497 106 72
3.5 - 4 369 93 28
4 - 4.5 283 71 21
4.5 - 5 252 63 21
5 - 5.5 211 55 17
5.5 - 6 182 47 13
6 - 6.5 123 33 7
6.5 - 7 91 23 8
7 - 7.5 52 13 4
7.5 - 8 43 11 3
>8 128 37 6
Total 24,904 3,913 4,658

Loss of life
As with the number of people at risk from flooding, the calculation of loss of life uses the
same population distribution. There are a number of methods to calculate loss of life. Brown
and Graham [9] is a very simple method where the population at risk is multiplied by a factor
depending on the length of time between the breach and flood warning. DeKay and
McClelland [10] use a simple method that uses different equations for high and low force
flooding. Graham [11] expands on the simple methods of Brown and Graham [9] and DeKay
and McClelland [10] using the concept of flood severity, flood warning and the understanding
of the exposed population to vary the fatality rate. The UK Flood Risk to People methodology
[7] calculates the loss of life based on the hazard rating, the type of people exposed and the
sensitivity of the area. This is similar to the method of Graham [11] used in the US, although
it is more advanced because it uses the magnitude of the flood expressed by the spatially
variable hazard rating to define fatality rate rather than a global fatality rate. The sensitivity of
the area is defined by the time of travel of the flood wave, type of housing, and the presence
of flood warning and evacuation plans. This calculation is most appropriate for large areas for
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
251
example census zones. Finally, the Life Safety Model [12] is an agent based model that
determines the fate of each person that is exposed to the flood in a dynamic way through time
and space. The model is designed for improving evacuation plans. The loss of life function in
the model is based on the water depth and velocity. This function can be used in a static way
to calculate loss of life from the maximum DV (Depth x Velocity).

After consideration of all the above methodologies, the approach taken was to primarily use
the loss of life function from the Life Safety Model (LSM) to provide an estimate of loss of
life based on people distributed in buildings. This has been chosen ahead of the UK Risk to
People (UK R2P) method because of the requirement to produce gridded output at a high
resolution.

The Theme C2 information [13] states that the dam fails at 11pm on Saturday night with no
witnesses that are able to raise an alarm, and that there is no evacuation plan in place. It has
therefore been assumed that the total population are in their homes and receive no warning for
the loss of life calculations. The loss of life estimated for this pilot from the Life Safety Model
approach is 4,966. That is based on conservative assumptions that a) people are on the
ground, and b) take no measures to avoid the flood. The spatial variability of fatalities in
Figure 9 show that the area between cross sections 3 and 5 is the most affected. All fatalities
are after 3h of the breach initiation.

An alternative scenario where people shelter in buildings has been performed with the Life
Safety Model where loss of life occurs if the building is submerged or collapsed. Assuming
that people shelter in buildings the number of fatalities is estimated to be 3,075. A more
realistic estimate can be produced assuming that people take refuge in multi-storey buildings
or respond to a warning and evacuate the area, for which the dynamic Life Safety Model
would be appropriate. A comparison of the methods is shown in Table 10.

The UK R2P method was used as a secondary check of LLOL. It is an area wide calculation
which uses a simple vulnerability measure multiplied by the Hazard Rating to define the
number of people at risk. The number of injuries is then estimated as 2 times the number of
old and sick people as a proportion of the number at risk. The fatality rate is 2 times the
Hazard Rating, and the number of fatalities is the number of injured multiplied by the fatality
rate. This produces a significantly lower estimate of loss of life than the LSM because it
assumes that young and healthy people can escape the flood. In this case the lack of flood
warning means that the whole population should be considered at risk. The calculation has
been modified so that the whole population is exposed producing an estimate of loss of life of
4,134.

Table 10: Loss of life


Estimation
method

Life Safety
Model
approach 1
Life Safety
Model approach
2 (people shelter
in buildings)

UK Risk to
people
methodology
Modified
UK Risk to
people
methodology
Loss of life 4,966 3,075 1,150 4,134
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
252

















Figure 9: Loss of life from Life Safety Model, cell size ~9.5m x 9.5m
Economic damage
Depth damage curves have been taken from the MCM [14] 2010 update for residential and
non-residential properties that are similar to the description given in the building data
provided. These curves were developed for UK properties. The damages have been converted
to US$ using an exchange rate of 1.51, but no other adjustments have been made to account
for differences in US and UK properties. Alternatively there is a US method that relates the
damage as a percentage of the building value for given water depths [15], but this method
could not be used because building values were not provided in the building data.

The total economic damage for the base dam breach has been estimated as US$1,237,582,738.
The spatial variability of the economic impact is shown in Figure 10, the commercial
buildings having the higher damage per grid cell (~89.8m
2
).





















Figure 10: Direct economic impact
Observations
It would have been preferable to distribute the number of people across the residential
buildings based on knowledge of household groupings. However, the census data provided
was not detailed enough. In addition the building data provided contained the overall footprint
rather than the number of individual properties (in the case of flats). This meant that some
census zones had large population with little or no residential buildings. For example census
zone 30 has 3 large properties classified as commercial but a population of 565. To reduce
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
253
uncertainty in the distribution of people within a census area, more detail is required on the
demographics and number and type of buildings. The LSM can also model the behaviour of
people in a flood situation as they seek to avoid the flood, thereby reducing the predicted loss
of life.
Conclusions
The paper presents the consequences of the breach of the hypothetical dam as described in
Theme C of the 12
th
International Benchmark Workshop on Numerical Analysis of Dams.

The dam breach analysis undertaken using EMBREA highlights the importance of choosing
the correct erosion method and the impact of choosing the correct erodibility of the soil.
Based on the data provided, it was estimated that the embankment fails according to headcut
erosion leading to a steep hydrograph with a high peak discharge, the consequences of which
are more severe than when an embankment fails due to surface erosion or piping.

The effects of the different roughness scenarios considered on the flood model was negligible
compared to the impact of the different hydrographs used as input, emphasizing again the
need for a good dam breach model. The flood model results provide useful information for
warning and evacuation planning. The model indicates that the first flow from the dam breach
reaches the populated area 1.5h to 2h after the breach initiates, but the peak flow travels from
the dam to the city in less than 15min, reaching the further edge in less than 90min.

The total direct economic impact is estimated at more than US$1.24b. The loss of life
calculations performed with the static Life Safety Model show a loss of life of 4,966, which
represents nearly the 20% of the total people at risk.
References
[1] 12
th
International Benchmark workshop on numerical analysis of dams, "THEME C:
Computational Challenges in Consequence Estimation for Risk Assessment:
Numerical modelling, Uncertainty Quantification, and Communication of Results.
Part 1- Hydraulic Modelling and Simulation," ed: International Commission on Large
Dams (ICOLD), 2013.
[2] M. W. Morris, M. A. A. M. Hassan, and K. A. Vaskinn, "Conclusions and
recommendations from the IMPACT Project: WP2 Breach Formation," 2004.
[3] P. Peeters, T. Van Hoestenberghe, L. Vincke, and P. Visser, "SWOT analysis of
breach models for common dike failure mechanisms," presented at the 34th IAHR
World Congress - Balance and Uncertainty, 33rd Hydrology & Water Resources
Symposium, 10th Hydraulics Conference, Brisbane Australia, 2011.
[4] M. W. Morris, M.A.A.M. Hassan, T.L. Wahl , R.D. Tejral, G.J. Hanson, and D.M.
[5] Temple, "Evaluation and development of physically-based embankment breach models,"
presented at the 2nd European Conference on Flood Risk Management, Rotterdam, The
Netherlands, 2012.
[6] V. Shevnin, O. Delgado-Rodrguez, A. Mousatov, and A. Ryjov, "Estimation of
hydraulic conductivity on clay content in soil determined from resistivity data,"
Geofisica Internacional, vol. 45, 2006.
[7] Defra and Environment Agency, "Reducing uncertainty in river flood conveyance,
phase 2, Conveyance manual," Environment Agency, report2004.
[8] Defra and Environment Agency, "R&D outputs: Flood Risk to People Phase 2.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
254
[9] Guidance Document," 2006.
[10] Defra and Environment Agency, "Framework and guidance for Assessing and
Managing Flood Risk Development, Flood Risk Assessment Guidance for New
Development," Technical report2005.
[11] C. A. Brown and W. J. Graham, "Assessing the Threat to Life from Dam Failure,"
[12] Water Resources Bulletin, vol. 24, pp. 1303-1309, 1988.
[13] M. L. DeKay and G. H. McClelland, "Predicting Loss of Life in Cases of Dam Failure
and Flash Flood," Risk Analysis, vol. 13, pp. 193-205, 1993.
[14] W. J. Graham, "A Procedure for Estimating Loss of Life Caused by Dam Failure," ed.
[15] Denver, Colorado: U.S. Department of Interior, Bureau of Reclamation, Dam Safety
[16] Office, 1999.
[17] BC Hydro, "Dam Safety, BC Hydro Life Safety Model: Formal model description,"
ed: BC Hydro,, 2004.
[18] I. B. w. o. n. a. o. dams, "THEME C: Computational Challenges in Consequence
Estimation for Risk Assessment: Numerical modelling, Uncertainty Quantification,
and Communication of Results. Part 2- Consequence Estimation," International
Commission on Large Dams (ICOLD) ed, 2013.
[19] E. Penning-Rowsell, C. Johnson, S. Tunstall, S. Tapsell, J. Morris, J. Chatterton, et al.,
"The benefits of Flood and Coastal Risk Management: A manual of assessment
techniques," ed. Middlesex: Middlesex University Press., 2005.
[20] C. Scawthorn, P. Flores, N. Blais, H. Seligson, E. Tate, S. Chang, et al., "HAZUS-MH
flood loss estimation methodology. II. Damage and loss assessment," Natural Hazards
Review, vol. 7, pp. 72-81, 2006.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
255
Uncertainty in Two-Dimensional Dam-Break Flood
Modeling and Consequence Analysis
M.S. Altinakar
1
, M.Z. McGrath
1
, V.P. Ramalingam
1
, D. Shen
1
, Y. Seda-Sanabria
2
and
E.E. Matheu
3

1
National Center for Computational Hydroscience and Engineering, The University of
Mississippi, Brevard Hall Room 327, P.O. Box 1848, University, MS 38677-1848, USA
2
Critical Infrastructure Protection and Resilience Program, Office of Homeland Security, U.S.
Army Corps of Engineers, Headquarters, Washington, DC 20314, USA
3
Critical Lifelines Branch, Sector Outreach and Programs Division, Office of Infrastructure
Protection, National Protection and Programs Directorate, Washington, DC 20528, USA
E-mail: altinakar@ncche.olemiss.edu
Abstract
This paper presents the preliminary results of an extensive study to evaluate and quantify
uncertainty in two-dimensional numerical dam-break flood modeling and consequence
analysis based on a benchmark test case. The benchmark problem consists of a hypothetical
61m-high embankment dam in a mountainous region with lightly populated urban areas
located downstream. Although the benchmark problem was intended mainly for estimating
the uncertainty in the evaluation of the consequences of failure of a dam near populated areas
with complex demographics, infrastructure and economic activity, the present study also
investigated the uncertainty in two-dimensional numerical modeling based on three control
variables. DSS-WISE numerical model was used to calculate dam-break flood and potential
loss-of-life for a total of 120 cases, which represented 40 random pairs of breach width and
breach formation time, each computed with three different sets of Mannings coefficients
defined based on land use/cover type. Computed results include raster maps of maximum
flood depth, maximum flood discharge per unit width and flood arrival time as well as
hydrographs at 7 cross sections. Analysis of population at risk (PAR) impacted by the flood
and loss-of-life were also computed for all 120 simulations. The paper presents some
preliminary results based on the statistical analysis of results.
Introduction
Test case consists of synthetic data specially designed for the benchmark study. A
hypothetical 61m-high embankment dam located in a mountainous area impounds a reservoir
having a normal storage of 38 million cubic meters. A lightly populated urban area, called
Hydropolis, is located about 3.5 km downstream from the dam. The benchmark test case data
for hydrodynamic simulation consists of (1) digital elevation model (DEM) of the area of
interest with and without the dam; (2) dam geometric characteristics; (3) reservoir stage-
volume and stage surface area curves; (4) dam and foundation material; (5) dam failure
scenario and conditions ate the time of failure; and (6) gridded data of land use/cover.

In addition the following data is provided for consequence estimation: (1) shapefile of the
census data for Hydropolis with breakdown into various age and gender classes; (2) shapefile
for twenty economic activity classes; (3) shapefile for parcel data with 28 zone classes
(residential, commercial, etc.). The complete dataset and supporting documents have been
published on the conference website under Theme-C and will not be repeated here. It is
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
256
assumed that the dam fails without warning when the water surface in the reservoir is at the
crest elevation and the population downstream cannot be warned.
Two-Dimensional Dam-Break Flood Simulation Model
Two dimensional simulations of dam-break flood were performed using the DSS-WISE
software developed by the National Center for Computational Hydroscience and Engineering
(NCCHE), The University of Mississippi. DSS-WISE is an integrated platform, which
couples a state-of-the-art two dimensional numerical model with GIS-based pre- and post-
processors, that are developed as an extension to ArcGIS software developed and
commercialized by ESRI. Its numerical solver CCHE2D-FLOOD solves the conservative
form of the two-dimensional shallow water equations, given as

[]

[]

(1)

in which, is the vector of conserved variables, and are flux vectors in and
directions, and . These are defined as

[

] [

] [

]
[

)
]

(2)

where, is the flow depth, and components of unit discharge in and directions,


the bed elevation,

the net source/sink mass flux per cell area per unit time, and the
gravitational acceleration. The slopes of energy grade line are given by Mannings equation


(3)

with as the Mannings coefficient of roughness. Note that and correspond to the
components of the unit discharge in and directions, respectively, i.e.,

and

. Finite
volume discretization of Equation (1) over a regular Cartesian mesh provides an explicit
equation for advancing the values of the conserved variables in time:

) (

) (

) (

4

In Equation (4), and define the cell size in and directions, and the time step size.
The intercell fluxes are computed using the first order Harten-Lax-van Leer-Contact (HLLC)
approximate Riemann solver by Toro et al. (1994), which was implemented following the
methodology described by Kim et al. (2007). The details can be found in Altinakar et al.
(2013a), Altinakar (2012) and Altinakar and McGrath (2012). The DSS-WISE software and
its solver CCHE2D-FLOOD have been fully verified using analytical solutions and validated
using laboratory and field data. Blind tests have also been undertaken in collaboration with
U.S. Army Corps of Engineers (Altinakar et al., 2012).
Data for Benchmark Test Case
Figure 1 shows the digital elevation model without dam. Open water surface to the East of the
area of interest was converted into NoData. The numerical model treats NoData cells as an
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
257
internal free outflow boundary. This prevents flood propagating over the open water surface,
which would otherwise be perceived as bed topography. The dam is represented in the DEM
as an idealized structure representing a wall of constant thickness. The embankment dam has
a crest width of 24m and the upstream and downstream slopes of 3H:1V. Considering that the
dam has a height of 61m, the width of the idealized wall was chosen as 207m, which
corresponds to the average width of the embankment dam. The overtopping failure was
assumed to take place when the reservoir reaches crest elevation. Thus, the initial condition
models the reservoir as a stagnant water body with surface elevation equal to crest elevation.


Figure 1: DEM without the dam (pre-construction) with open water area removed.
Selection of Test Cases
Three parameters were chosen to investigate the uncertainty of unsteady hydrodynamic
simulations: (1) bottom breach width,

; (2) breach formation time,

; and (3) the set of


Mannings coefficient of roughness, .
Prediction of Breach Parameters
Empirical equations have been developed to estimate breach parameters based on the analysis
of the data from historic dam failures (Wahl, 2004). The present study uses the empirical
equations given by Froehlich (1995) to compute the breach width and breach formation time

(5)

(6)

where

is the average breach width (m), i.e. the breach width at half of the breach height

(m),

is a constant equal to 1.4 for overtopping failures, and

is the reservoir volume


at the time of failure (m3). Equations (5) and (6) were obtained based on data from past
failures of zoned earthen, earthen with a clay core, and rockfill embankment dams. The dam
heights ranged from 3.66m to 92.6m (with 90%<30m and 76%<15m), the volume of water in
the reservoir ranged from 0.0130106 m3 to 660.0106 m3 (with 87%<25.0106 m3 and
76%<15.0106 m3) and the lake surface areas ranged from 0.045km2 to 2,165km2. The data
for the benchmark test case (61m, 38106 m3, and 1.584 km2), which is well within the range
of cases used to establish Equations (5) and (6). Thus, the average breach width and breach
formation time are calculated as

and

, respectively. Considering the


Open water area removed by
changing elevations to No Data
Elevation (m)
1823.5
67.2
Breach Cross Section
Cross Section 6
Idealized Dam
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
258
side slopes of 1.4H:1V [6], the bottom with and top width of the trapezoidal breach cross
section are

and

, respectively.
Uncertainty of prediction of breach parameters for embankment dams was investigated by
Wahl (2004) [12]. It was shown that the prediction errors using empirical equations, including
Froehlich (1995) [6], exhibit a lognormal probability distribution (base 10) defined by a mean
value, , and a standard deviation, . Table 1 summarizes the uncertainty estimates of the
breach parameters predicted using Froehlich (1995). Using the values given in Table 1, the
95% confidence interval for the breach bottom width extends from 25m to 146m whereas the
95% confidence interval for breach formation time extends from 0.3hr to 4.9hr.

Table 1: Uncertainty Estimates for Froehlich (1995) Equations (taken from Wahl, 2004).
Parameter Equation
Number of Cases
(log
cycles)

(log
cycles)
95%
Confidence
interval*
Before outlier
exclusion
After outlier
exclusion
Breach width Eq. (5) 77 75 +0.01 0.195 0.40 2.4
Breach formation time Eq. (6) 34 33 -0.22 0.32 0.38 7.3
*95% confidence interval is given around a hypothetical predicted value of 1.0

Mannings Coefficients
Mannings coefficients were assigned based on the given land use/cover data with 16 size
classes that follow the classification guidance and values provided in the U.S. National Land
Cover Dataset (NLCD). The land use/cover classes for the area of interests are shown in
Figure 2. Table 2 shows the three sets of Mannings coefficient used in the present study. It
should be noted that the area of interest does not have classes 12 and 95. Figure 3 provides a
visual comparison of the three sets of Mannings coefficients for each land use/cover class.


Figure 2: Map of land use/cover classifications.
Latin Hypercube Sampling for Breach Parameters
Latin Hypercube Sampling (LHS) method was used to select 40 random pairs of bottom
breach width,

and breach formation time,

. LHS is a stratified-random method which


provides an efficient way of sampling variables from their distributions (Iman et al. 1981).
The cumulative distribution (normal distribution of logarithms) of both variables were divided
into 40 intervals of equal probability. For each variable, a value is randomly selected from
every interval. The selected values are paired randomly to achieve an optimal filling of the
parameter-uncertainty space defined by the 4040 matrix of

and

. Design of the LHS


was accomplished using the lhsdesign function in Matlab software using an option that
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
259
maximizes the minimum distance between the pairings. Figure 4a shows the LHS design used
in the present study. Since

and

have lognormal distributions (base 10), the probability


values,

, used in LHS were converted into real values using the following expressions:
Table 2: Land Use/Cover Classes and Three Sets of Corresponding Mannings Coefficients.
NLCD
Class
Description
(

)
DSS-
WISE
(1)

FEMA
HAZUS
(2)

NOAA
C-CAP
(3))

Min.
Value
Average
Value
Max.
Value
11 Open Water 0.0330 0.0010 0.0250 0.0010 0.0197 0.0330
12 Perennial Ice/Snow 0.0100 0.0100 0.0100 0.0100 0.0100 0.0100
21 Developed-Open Space 0.0404 0.0200 0.0350 0.0200 0.0318 0.0404
22 Developed- Low Intensity 0.0678 0.0500 0.1200 0.0500 0.0793 0.1200
23 Developed- Med. Intensity 0.0678 0.1000 0.1200 0.0678 0.0959 0.1200
24 Developed- High Intensity 0.0404 0.1500 0.1200 0.0404 0.1035 0.1500
31 Barren Land 0.0113 0.0900 0.0300 0.0113 0.0438 0.0900
41 Deciduous Forest 0.1000 0.1000 0.1600 0.1000 0.1200 0.1600
42 Evergreen Forest 0.1000 0.1100 0.1800 0.1000 0.1300 0.1800
43 Mixed Forest 0.1200 0.1000 0.1700 0.1000 0.1300 0.1700
52 Shrub/Scrub 0.0400 0.0500 0.0800 0.0400 0.0567 0.0800
71 Grassland/Herbaceous 0.0400 0.0340 0.0350 0.0340 0.0363 0.0400
81 Pasture/Hay 0.0350 0.0330 0.0500 0.0330 0.0393 0.0500
82 Cultivated Cropland 0.0700 0.0370 0.1000 0.0370 0.0690 0.1000
90 Woody Wetlands 0.1500 0.1000 0.1500 0.1000 0.1333 0.1500
95 Herbaceous Wetlands 0.1825 0.0450 0.0500 0.0450 0.0925 0.1825
(1)
Altinakar et al. (2013b);
(2)
Luettich and Westerink (2009);
(3)
ARCADIS (2011)


Figure 3: Comparison of three sets of Mannings coefficients assigned to NLCD classes.

]
;

]
(7)

In the above expression,

stands for the inverse of the cumulative


distribution function of standard normal distribution, with mean equal to zero and standard
deviation equal to 1, at the probability value of

. For each variable, i.e.

or

, its
corresponding and values were used. The selected 40 pairs are provided in Table 3. In
Figure 4a, the numbers for the data points correspond to combination numbers in Table 3.
Figure 4b and Figure 4c show that the randomly selected of

and

values display a
lognormal probability distribution, as expected.
Assumptions Regarding the Progression of the Breach Geometry
0.0000
0.0200
0.0400
0.0600
0.0800
0.1000
0.1200
0.1400
0.1600
0.1800
0.2000
11 12 21 22 23 24 31 41 42 43 52 71 81 82 90 95
M
a
n
n
i
n
'
s

C
o
e
f
f
i
c
i
e
n
t
,

n

(
m
-
1
/
3
s
)
NLCD Class
DSS-WISE FEMA HAZUS NOAA C-CAP
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
260
Froehlich (1995) expression, used for estimating breach characteristics, assumes that the final
breach cross section for overtopping failures is a trapezoidal with side slopes of 1.4H:1V.
Figure 5a shows the mean, minimum and maximum breach sizes selected by LHS. The
elevations of the cells under the footprint of the idealized dam were initially raised to the level
of the crest. As soon as the simulation starts, the breach occurs by gradually bringing down
these cells to the original valley elevation during the breach formation time,

, as shown in
Figure 5b. The cell elevations are not permitted to go below the original bed elevation.


Figure 4: (a) LHS design obtained by maximizing minimum distance between data pairs; (b)
Histogram of randomly selected breach widths and the fitted lognormal distribution; (c)
Histogram of randomly selected breach formation times and the fitted lognormal distribution.
Table 3: List of 40 pairs of breach width and breach formation time selected using LHS.
Pair
No


(m)


(hr)
Pair
No


(m)


(hr)
Pair
No


(m)


(hr)
Pair
No


(m)


(hr)
1 41.9 2.2 11 38.3 1.4 21 70.6 0.4 31 88.4 0.3
2 51.8 1.0 12 130.3 2.1 22 46.2 2.8 32 53.2 1.4
3 103.5 0.6 13 116.9 1.6 23 60.7 0.9 33 49.6 0.7
4 55.1 5.2 14 68.8 0.6 24 88.0 1.0 34 78.4 2.6
5 57.0 2.0 15 58.6 1.7 25 113.1 1.2 35 92.9 0.5
6 59.0 0.6 16 83.6 1.3 26 98.5 3.0 36 63.6 0.8
7 30.6 1.8 17 80.6 0.5 27 44.7 1.1 37 31.8 1.1
8 20.0 0.4 18 37.2 0.2 28 27.4 0.7 38 73.8 0.9
9 74.4 3.4 19 66.7 2.5 29 65.7 1.5 39 47.5 4.9
10 40.1 0.8 20 177.3 1.7 30 48.9 1.3 40 36.0 0.7


Figure 5: (a) Breach geometry for mean, minimum and maximum breach bottom widths; and
(b) Linear progression of the breach geometry assumed in the numerical model.

0 50 100 150 200 250
0
2
4
6
8
10
12
14
Breach Bottom Width (m)
N
u
m
b
e
r

o
f

O
b
s
e
r
v
a
t
io
n
s
Histogram of Data and Lognormal PDF
0 2 4 6 8 10 12
0
2
4
6
8
10
12
14
16
Breach Formation Time (hr)
N
u
m
b
e
r

o
f

O
b
s
e
r
v
a
t
i o
n
s
Histogram of Data and Lognormal PDF
(a) (b) (c)
Histogram of Generated W
b
Data and Lognormal PDF Histogram of Generated W
b
Data and Lognormal PDF
N
u
m
b
e
r

o
f

O
b
s
e
r
v
a
t
i
o
n
s
Breach Bottom Width, W
b
, (m) Breach Formation Time, t
f
, (hr)
N
u
m
b
e
r

o
f

O
b
s
e
r
v
a
t
i
o
n
s
200
220
240
260
280
300
100 150 200 250 300 350 400 450 500
E
l
e
v
a
t
i
o
n

(
m
)
Chainage (m)
Cross section Dam Crest Mean Breach
Minimum Breach Maximum Breach
200
210
220
230
240
250
260
270
280
290
300
100 150 200 250 300 350 400 450 500
E
l
e
v
a
t
i
o
n

(
m
)
Chainage (m)
t = t
f
/ 4
t = 2t
f
/ 4
t = 3t
f
/ 4
t = t
f
t = 0
(a) (b)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
261
Results of Hydrodynamic Simulations
A total of 120 numerical simulations were performed by combining 40 randomly chosen pairs
of breach width and breach formation time with three sets of Mannings coefficients. Each
simulation provided three raster files (maximum flow depth,

, flood arrival time,

,
and maximum specific discharge,

), and discharge hydrographs at seven cross sections


(dam crest, five pre-defined cross sections along the main downstream flow, and one cross
section to capture the flows towards north) recorded with a time interval of two minutes.


Figure 6: Statistics of maximum flood depth,

(m), based on 120 simulations.


Maps of Maximum Depth, Flood Arrival Time and Maximum Discharge per Unit Width
The set of 120 raster files were used to calculate maps of ensemble average, standard
deviation, and minimum and maximum values of the maximum flood depth (Figure 6), the
flood arrival time (Figure 7), which is defined as the time at which the dry computational cell
becomes wet, regardless of the depth, and the maximum discharge (Figure 8). In these maps,
cells with a null value are not displayed. As expected, the highest flow depths and unit
discharges occur in the narrow valley immediately downstream of the dam. Comparison of
minimum and maximum value maps of these three variables indicate that large differences
exist between individual runs. The largest standard deviations of maximum flood depth and
maximum unit discharge occur in the narrow valley downstream of the dam whereas the
largest standard deviation values for the arrival time are observed farthest from the dam. The
area on the northern part is inundated only for certain simulations. The longest time arrival
and largest standard deviations of arrival time are observed in this area (Figure 7).

An understanding of the sensitivity of results to different parameters can be gained by
analyzing and comparing results files of individual simulations. Figure 9 shows a comparison
of the flood maps of selected pairs of breach width and breach formation time with all three
sets of Mannings coefficients. The upper right hand corner shows a plot of all 40 random
pairs selected with LHS in terms of real values of

and

. This should be compared with


the plot in probability space shown in Figure 4a. As shown, there are no simulations in the
upper right corner of the plot corresponding to large breach widths with long formation times.

The inspection of Figure 9 shows that the area on the north part of the DEM is inundated only
for certain combinations of parameters. In

space, the simulations with inundation of


0.25
0.50
0.75
1.0
2.0
3.0
4.0
5.0
8.0
12.0
17.0
22.0
27.0
32.0
37.0
42.0
47.0
52.0
57.0
62.0
meter
Average H
max
(m) St. Dev. of H
max
(m)
Minimum Value of H
max
(m) Maximum Value of H
max
(m)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
262
north part correspond to small breach formation times and large breach widths. In addition,
higher Mannings values favor inundation of the north part of the DEM. The main land
use/cover types in the narrow valley are pasture/hay, woody wetlands, and evergreen forest.
The n-values from NOAA C-CAP for these land use/cover types are the highest resulting in
higher flow depths and, thereby spilling of the flood into northern region of the DEM for pair
numbers 3, 6. 8, 18, and 31.


Figure 7: Statistics of flood arrival time,

(hr), based on 120 simulations.



Figure 8: Statistics of maximum discharge per unit width,

(m
2
/s), based on 120
simulations.
Hydrographs
Discharge hydrographs were recorded at seven cross sections. These include the breach cross
section and five cross sections specified in the benchmark test case formulation. An additional
cross section was also introduced in order to measure the flood discharge propagating to the
north of the DEM under certain combinations of control parameters (see Figure 9). Due to the
Average t
arr
(hr) St. Dev. of t
arr
(hr)
Minimum Value of t
arr
(hr) Maximum Value of t
arr
(hr)
0.25
0.50
0.75
1
2
3
4
5
6
7
8
9
10
11
12
16
20
24
30
34
hours
Average q
max
(m) St. Dev. of q
max
(m)
Minimum Value of q
max
(m) Maximum Value of q
max
(m)
m
2
/s
3
4
4.65
5
6
8
10
14.86
25
50
100
150
200
250
300
350
400
450
500
562
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
263
lack of space, only limited number of results will are presented in this paper. Additional
details will be provided in subsequent publications.

Figure 10 shows the discharge hydrographs at the breach cross section (left plot) and the
relationship between peak discharge and the control parameters

(middle plot) and

(right
plot). Runs with different n-values are distinguished with different symbols/colors. Since at
the breach cross section the influence of the Mannings coefficient is not fully felt, for a given

and

pair, the hydrographs obtained with different n-values coincide fairly well. The
peak discharge does not seem to be highly correlated with the breach width (middle plot) but
rather with the breach formation time (right plot). The peak discharge is inversely
proportional with the breach formation time and all the data collapses onto a single curve.

A similar plot is shown for cross section 5 in Figure 11. The peak discharges of hydrographs
are attenuated. For a given

and

pair, the hydrographs computed with different sets of n-


values have different arrival times. The peak discharges are not correlated with the breach
width (middle plot) but they still show a relatively good correlation with the breach formation
time (right plot) despite the fact that the differences in the peak discharge for simulations with
different n-values are larger for

and

pairs.


Figure 9: Comparison of flood depth maps for selected simulations.
DSS-WISE n-values
FEMA HAZUS n-values NOAA C-CAP n-values
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
264

Figure 10: Discharge hydrographs and the peak discharge at the breach cross section (DSS-
WISE: black squares, FEMA HAZUS: red plus signs, and NOAA C-CAP: blue crosses).

Figure 11: Discharge hydrographs and the peak discharge at cross section 5 (DSS-WISE:
black squares, FEMA HAZUS: red plus signs, and NOAA C-CAP: blue crosses).
Consequence Analysis
In the present study the consequence analyses were restricted to the analysis of population at
risk (by age group) to be affected by flood based on flood arrival time and peak discharges per
unit width. Direct or indirect economic analyses were not performed.
Population by Census Block and PAR Distribution
Census block polygons for Hydropolis, colored by total population, are shown in Figure 12.
Other polygon attributes include population by age and gender group and jobs in 20 different
classes. The loss of life analysis in the present study was based only on total population. The
map of population at risk per computational cell shown in Figure 13 was computed by
assuming a uniform distribution of population in each census block. As it can be seen, there
are two major urban areas (see also Figure 2). The urban area on the west is located in the
narrow valley downstream of the dam and directly on the path of the flow. The urban area on
the north east corner of the map is farther from the dam and located on a relatively flat area.

20 40 60 80 100 120 140 160 180
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
4 Breach Cross Section
Breach Width(m)
M
a
x
.

c
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
0 1 2 3 4 5 6
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
4 Breach Cross Section
Breach Formation Time (s)
M
a
x
.

c
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10
4
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
4 Breach Cross Section
time (s)
C
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
20 40 60 80 100 120 140 160 180
0
0.5
1
1.5
2
2.5
3
3.5
x 10
4 Cross Section 5
Breach Width(m)
M
a
x
.

c
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
0 1 2 3 4 5 6
0
0.5
1
1.5
2
2.5
3
3.5
x 10
4 Cross Section 5
Breach Formation Time (s)
M
a
x
.

c
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 10
4
0
0.5
1
1.5
2
2.5
3
3.5
x 10
4 Cross Section 5
time (s)
C
r
o
s
s

s
e
c
t
i
o
n

d
i
s
c
h
a
r
g
e

(
m
3
/
s
)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
265

Figure 12: Total population by census block (only census blocks with population are shown).
PAR Analysis
The number of PAR by age group for each one of the 120 simulations were counted for 7
flood arrival-time intervals and for 17 depth intervals and recorded in tables.
shows the average value, standard deviation, minimum and maximum values of PAR numbers
by arrival time interval. Some of the information listed in
is summarized in Figure 14. The plot on the right hand side of Figure 14 shows that the range
between cumulative values of minimum and maximum PAR can be significant. Final
minimum and maximum cumulative PAR values correspond to 4.4% of 268% of the final
average cumulative PAR.

Figure 13: PAR distribution per computational cell (10m10m).


1 - 42
43 - 81
82 - 152
153 - 278
279 - 481
482 - 943
944 - 1987
1988 - 3338
Total Population
82
1
PAR / 100m
2
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
266
Table 4: Statistics of PAR numbers by age group and flood arrival time interval.
Age Value 0-15min 15-30min 30-60min 60-90min 90-120min 120-180min >180min
T
o
t
a
l

P
A
R

Average 0.186 12.782 38.370 75.021 65.333 54.556 14.201
St. Dev. 0.548 14.597 28.768 47.800 25.632 39.032 26.624
Min. 0.043 0.019 8.566 1.110 0.532 1.210 0.087
Max. 5.110 52.346 144.341 148.087 95.128 119.794 132.010
P
A
R


1
4

Average 0.035 3.296 7.584 10.381 9.437 7.987 2.037
St. Dev. 0.147 3.802 3.101 6.789 4.100 6.049 4.041
Min. 0.006 0.002 2.014 0.286 0.140 0.157 0.010
Max. 1.286 13.409 19.903 21.096 14.424 18.265 20.306
1
4
<
P
A
R


<

6
5

Average 0.129 8.623 25.142 48.743 42.874 35.977 9.376
St. Dev. - - - - - - -
Min. 0.031 0.014 5.926 0.746 0.346 0.853 0.061
Max. 3.507 35.341 93.943 96.662 59.286 77.542 87.380
P
A
R


6
5

Average 0.021 0.862 5.644 15.898 13.022 10.592 2.789
St. Dev. 0.030 0.986 7.590 10.011 4.963 7.275 4.983
Min. 0.007 0.004 0.625 0.078 0.046 0.200 0.016
Max. 0.317 3.597 30.495 30.329 21.418 23.987 24.324


Figure 14: (Left) Average PAR by age group and by arrival time; (Right) Cumulative values
of average, minimum and maximum total PAR impacted by the flood as a function of time.

and Table 6 show the average value, standard deviation, minimum and maximum values of
PAR numbers by flood depth interval in two separate tables. Some of the PAR information
listed in Table 5 and Table 6 is summarized in Figure 14. The plot on the right hand side of
Figure 14 shows that the range between cumulative values of minimum and maximum PAR
can be significant. This figure shows that majority of PAR impacted by the flood (93.5%) will
be subjected to a maximum flood depth of 3.5m. The range between minimum and maximum
PAR values is significant. The largest differences are observed for the depth interval 0.0-
0.5m. Minimum (98.6% PAR in less than 3.5m depth) and maximum (87.8% PAR in less
than 3.5m depth) PAR both show the same tendency as the average PAR. The average,
minimum and maximum values of the inundated area listed in Table 5 and Table 6 are plotted
in Figure 16 as a function of flow depth. As it can be seen most of the inundated area (95.1%)
will be subjected to water depths less than or equal to 5m. The largest differences between
minimum and maximum values occur for the depth interval 0.0-0.5m.


0
100
200
300
400
500
600
700
800
0-15min 15-30min 30-60min 60-90min 90-120min 120-180min >180min
N
u
m
b
e
r

o
f

P
A
R
Cumulative Total PAR as a Function of Time
Cumul. Total PAR Min Cumul. Total PAR Max Cumul. Total PAR
0
10
20
30
40
50
60
70
80
0-15min 15-30min 30-60min 60-90min 90-120min 120-180min >180min
N
u
m
b
e
r

o
f

P
A
R
Average PAR
PAR 14 14 < PAR < 65 PAR 65 Total PAR
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
267
Table 5: Average and standard values of PAR numbers by age group and depth interval.
Depth (m)
Total PAR PAR 14 14 < PAR < 65 PAR 65 Inundated Area (m2)
Min Max Min Max Min Max Min Max Min Max
0.0m-0.5m 55.53 19.25 8.21 2.91 37.03 - 10.29 3.48 7,333,416 3,206,673
0.5m-1.0m 73.38 15.41 11.10 2.16 48.74 - 13.54 3.29 9,018,780 1,693,804
1.0m-1.5m 67.36 23.75 9.97 3.68 44.20 - 13.19 4.05 8,709,363 1,270,547
1.5m-2.0m 26.36 12.31 3.63 1.83 16.67 - 6.06 2.33 5,856,884 1,471,850
2.0m-2.5m 10.30 4.18 1.37 0.43 6.23 - 2.70 1.27 3,358,918 1,008,449
2.5m-3.0m 4.65 0.87 0.81 0.15 2.83 - 1.01 0.41 2,018,934 709,981
3.0m-3.5m 3.23 0.47 0.67 0.19 2.02 - 0.54 0.24 1,453,657 533,611
3.5m-4.0m 2.57 0.67 0.62 0.20 1.67 - 0.28 0.14 1,337,561 659,157
4.0m-4.5m 2.24 0.83 0.58 0.21 1.50 - 0.16 0.06 674,860 362,878
4.5m-5.0m 2.18 0.85 0.56 0.21 1.46 - 0.16 0.07 325,181 106,113
5.0m-5.5m 2.02 0.90 0.52 0.23 1.35 - 0.15 0.07 260,150 95,178
5.5m-6.0m 1.81 0.95 0.46 0.24 1.22 - 0.13 0.08 218,419 85,614
6.0m-6.5m 1.63 1.08 0.41 0.27 1.10 - 0.12 0.09 185,084 79,717
6.5m-7.0m 1.41 1.09 0.36 0.27 0.95 - 0.10 0.09 161,414 73,249
7.0m-7.5m 1.17 1.06 0.30 0.27 0.79 - 0.08 0.08 137,548 63,610
7.5m-8.0m 1.03 1.13 0.26 0.28 0.70 - 0.08 0.09 119,563 57,089
>8.0m 3.55 4.92 0.92 1.21 2.40 - 0.23 0.36 2,226,929 550,383
Table 6: Minimum and maximum values of PAR numbers by age group and depth interval.
Depth (m)
Total PAR PAR 14 14 < PAR < 65 PAR 65 Inundated Area (m2)
Min Max Min Max Min Max Min Max Min Max
0.0m-0.5m 38.76 134.98 5.33 20.13 25.84 89.84 7.59 25.01 3,502,773 17,674,779
0.5m-1.0m 49.03 102.42 7.00 15.17 33.25 67.26 8.78 19.99 6,067,713 11,718,682
1.0m-1.5m 12.73 92.40 2.05 14.09 7.80 61.01 2.88 17.30 4,253,739 10,358,269
1.5m-2.0m 7.50 45.04 1.32 6.95 4.98 29.01 1.20 9.09 2,131,314 7,690,691
2.0m-2.5m 3.93 19.90 0.80 2.95 2.82 12.39 0.30 4.56 740,011 5,280,020
2.5m-3.0m 2.59 7.23 0.54 1.09 1.84 4.38 0.21 1.76 422,222 3,035,832
3.0m-3.5m 1.25 4.61 0.30 1.15 0.85 2.59 0.10 0.87 447,365 2,560,271
3.5m-4.0m 0.67 3.63 0.16 0.93 0.46 2.08 0.05 0.62 171,062 2,481,161
4.0m-4.5m 0.29 4.16 0.07 1.02 0.20 2.78 0.02 0.36 95,543 1,333,115
4.5m-5.0m 0.28 4.03 0.08 1.00 0.19 2.70 0.01 0.33 78,662 521,537
5.0m-5.5m 0.23 3.65 0.07 0.93 0.16 2.42 0.01 0.31 73,453 442,965
5.5m-6.0m 0.17 3.69 0.06 0.91 0.11 2.45 0.00 0.33 74,621 418,002
6.0m-6.5m 0.02 4.07 0.01 1.01 0.01 2.71 0.00 0.35 66,539 330,630
6.5m-7.0m 0.00 4.05 0.00 1.00 0.00 2.70 0.00 0.35 59,984 298,393
7.0m-7.5m 0.00 4.10 0.00 1.02 0.00 2.72 0.00 0.35 44,808 276,752
7.5m-8.0m 0.00 4.26 0.00 1.05 0.00 2.84 0.00 0.36 39,151 277,381
>8.0m 0.02 20.47 0.00 5.10 0.01 13.92 0.00 1.44 1,409,981 3,827,117

Loss of Life Analysis
The loss of life is computed using the U.S. Bureau of Reclamation DSO-99-06 Procedure
(Graham, 1999). In this method, the fatality rate of PAR in a given area, which in the case of a
two dimensional numerical simulation can be taken as the computational cell, is computed
from Table 6 based on three parameters: flood severity, warning time, and flood severity
understanding. Warning time can be computed as the time difference between the time the
warning issued and the arrival time of the flood. Flood severity understanding is a parameters
that must be appreciated based on the characteristics of the population in the inundation area.
Since in the benchmark test case it is assumed that the population cannot be warned, the flood
severity understanding is no longer applicable, and Table 6 reduces to the three highlighted
lines. There are no clear guidelines for estimating flood severity. In the present study, the
flood severity was computed based on maximum specific discharge,

, which is equal to
the product of depth and flow speed. The limits of high, medium, and low flood severity
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
268
shown in Table 7 are taken from guidelines published by the U.S. Department of Homeland
Security (DHS, 2011).


Figure 15: (Left) Average PAR by age group and by flood depth; (Right) Average, minimum
and maximum total PAR impacted by the flood as a function of depth.


Figure 16: Average, minimum and maximum values of inundated area as a function of depth.

Based on the flood severity criteria given in Table 8 and the fatality rates in Table 7, the loss-
of-life (LOL) maps for each simulation was established using the corresponding

raster.
The average value, standard deviation, and minimum and maximum values of loss of life for
each computational cell were then computed from 120 loss-of-life maps. Figure 17 shows the
average value, standard deviation, minimum and maximum values of estimated loss of life
(LOL) based on 120 simulations. As expected, the largest loss of life numbers are observed in
the western urban area located in the narrow valley. This area is closer to the dam and directly
on the path of the dam-break flood. The

values are much larger. The urban area on the


north eastern corner of the map has relatively small values of LOL due to lower values of

. It is important to note that the areas with no loss of life and with zero standard
deviation are not displayed. Looking at the standard deviation map, one can observe that there
are areas with zero standard deviation. These areas can be regarded as areas where the LOL
has highest likelihood.

0
10
20
30
40
50
60
70
80
N
u
m
b
e
r

o
f

P
A
R
Average PAR
PAR 14 14 < PAR < 65 PAR 65 Mean Total PAR
0
20
40
60
80
100
120
140
160
N
u
m
b
e
r

o
f

P
A
R
Minimum, Avarage and Maximum Total PAR
Mean Total PAR Minimum Total PAR Maximum Total PAR
0
2,000,000
4,000,000
6,000,000
8,000,000
10,000,000
12,000,000
14,000,000
16,000,000
18,000,000
20,000,000
I
n
u
n
d
a
t
e
d

A
r
e
a

(
m
2
)
Minimum, Avarage and Maximum Values of Inundated Area
Avg. Inundation Area Min. Inundation Area Max. Inundation Area
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
269
Table 7: Fatality rates for loss-of-life estimation from dam failure (Graham, 1999).
Flood Severity
Warning Time
(Minutes)
Flood Severity
Understanding
Fatality Rate (Fraction of PAR projected to die)
Suggested Value Suggested Range
HIGH
Not applicable 0.75 0.30 1.00
15 to 60
Vague Apply the values shown above to the PAR values
who remain in the dam failure floodplain after
warnings are issued. No guidance is provided on
how many people will remain in the flood plain.
Precise
More than 60
Vague
Precise
MEDIUM
Not applicable 0.15 0.03 0.35
15 to 60
Vague 0.04 0.01 0.08
Precise 0.02 0.005 0.04
More than 60
Vague 0.03 0.005 0.06
Precise 0.01 0.002 0.02
LOW
Not applicable 0.01 0.0 0.02
15 to 60
Vague 0.007 0.0 0.015
Precise 0.002 0.0 0.004
More than 60
Vague 0.0003 0.0 0.0006
Precise 0.0002 0.0 0.0004
Table 8: Flood severity criteria used for loss of life analysis.
Flood severity Criteria in SI Units Criteria in US Customary Units
LOW


MEDIUM


HIGH



Figure 17: Average value, standard deviation, minimum and maximum values of estimated
loss of life (LOL) based on 120 simulations.
Conclusion
The uncertainty in 2D numerical modeling of dam-break flood and the resulting loss of life
were studied using 120 simulations representing 40 random pairs of breach width and breach
formation time calculated with three different sets of Mannings coefficients. The results
show that large variations in the extent of the inundated area, water depths, and loss-of-life
occur based on the particular combination of control parameters. The paper presents the
average value and standard deviation of selected computed results as well as their upper and
lower bounds. Peak discharge at the breach cross section and the cross sections downstream
seem to be highly correlated with the breach formation time rather than the breach width.
Average LOL St. Dev. of LOL
Minimum Value of LOL Maximum Value of LOL
0.00 - 0.02
0.02 - 0.04
0.04 - 0.06
0.06 - 0.08
0.08 - 0.10
0.10 - 0.12
0.12 - 0.14
0.14 - 0.16
0.16 - 0.18
0.18 - 0.20
0.20 - 0.22
0.22 - 0.24
0.24 - 0.26
0.26 - 0.28
0.28 - 0.30
0.30 - 0.32
0.32 - 0.34
0.34 - 0.36
0.36 - 0.38
# of Deaths
1

-
4
2
4
3

-
8
1
8
2

-
1
5
2
1
5
3

-
2
7
8
2
7
9

-
4
8
1
4
8
2

-
9
4
3
9
4
4

-
1
9
8
7
1
9
8
8

-
3
3
3
8
P
A
R
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
270
Only part of the results could be presented. Additional findings will be presented in
subsequent publications.
Acknowledgements
The research for the present benchmark study and the paper was sponsored by the U.S. Army
Corps of Engineers, Headquarters Office of Homeland Security. The development of the
DSS-WISE software was funded by the Department of Homeland Security-sponsored
Southeast Region Research Initiative (SERRI) managed by the Department of Energys Oak
Ridge National Laboratory.
References
[1] Altinakar, M.S., McGrath, M.Z., Yucel, O., and Ger, M. (2013a). Simulation of Super-
critical Flow in an Open Channel Bend Using Cut-Cell Boundary Method. 6
th
Int.
Perspective on Water Res. and Env. (IPWE 2013), January 7-9, 2013, Izmir Turkey.
[2] Altinakar, M.S., McGrath, M.Z., Ramalingam, V.P., Strati, O. (2013b). Dam Break Flood
Simulation Doing It Faster and Simpler: DSS-WISE, DSS-WISE Lite, and DSAT.
1-day Short Course, ASDSO West Regional Conf., May 8, 2013, Seattle, Washington.
[3] Altinakar, M.S. (2012). Advances in Numerical Modeling and Simulation of Floods,
Inundation Mapping and Consequence Analysis. Keynote Lecture presented at the 10th
International Congress on Advances in Civil Engineering (ACE 2012), October 17-19,
2012, Middle East Technical University (METU), Ankara, Turkey.
[4] Altinakar, M.S., McGrath, M.Z., Matheu, E.E., Ramalingam, V.P., Seda-Sanabria, Y.
Breitkreutz, W., Oktay, S., Zou, J.Z., Yezierski, M. (2012): Validation of Automated
Dam-Break Flood Simulation and Assessment of Computational Performance, Proc. Dam
Safety 2012, September 16-20, Denver, Colorado.
[5] Altinakar, M.S. and McGrath, M.Z. (2012). Parallelized Two-Dimensional Dam-Break
Flood Analysis with Dynamic Data Structures. ASCE-EWRI, 2012 World Environmental
& Water Resources Congress, May 20-24, 2012, Albuquerque, New Mexico.
[6] ARCADIS (2011). ADCIRC Based Storm Surge Analysis of Sea Level Rise in the
Galveston Bay and Jefferson County Area in Texas. Report prepared for The Nature
Conservancy, Nov 28, 2011.
[7] DHS (2011). Estimating Loss of Life for Dam Failure Scenarios. U.S. Department of
Homeland Security, September 2011. Washington DC.
[8] Froehlich, D.C. (1995a). Embankment Dam Breach Parameters Revisited. Water Res.
Engineering, Proc. 1995 ASCE Conf. on Water Res. Eng., New York, pp. 887-891.
[9] Graham, Wayne J. (1999). A Procedure for Estimating Loss of Life Caused by Dam
Failure. DSO-99-06, U.S. Bureau of Reclamation, Denver, Colorado, September 1999.
[10] Iman, R.L.; Helton, J.C.; and Campbell, J.E. (1981). An Approach to Sensitivity Analysis
of Computer Models, Part 1. Introduction, Input Variable Selection and Preliminary
Variable Assessment. Journal of Quality Technology, Vol. 13, No. 3, pp. 174183.
[11] Kim, D.-H., Cho, Y.-S., and Yi, Y.-K. (2007). Propagation and Run-up of Nearshore
Tsunamis with HLLC Approximate Riemann Solver. Ocean Engineering, Volume 34,
Issues 8-9, June 2007, pp. 1164-1173.
[12] Luettich, R. and Westerink, J. Storm Surge Inundation Modeling: State of Science.
Hurricane Surge Workshop, February 11, 2009, St. Petesburg, Florida
[13] Toro, E. F., Spruce, M., Speares, W. (1994). Restoration of the Contact Surface in the
HLL-Riemann Solver. Shock Waves, Vol. 4, No. 1, pp. 25-34.
[14] Wahl, T.L. (2004). Uncertainty of Predictions of Embankment Dam Breach Parameters.
Journal of Hydraulic Engineering, Vol. 130, No. 5, pp. 389-397.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
271
Risk assessment for hypothetical dam break
A method for the rapid and consistent evaluation
L.Mancusi
1
, L.Giosa
2
, A.Cantisani
2
, A.Sole
2
and R.Albano
2

1
Sustainable Development and Energy Resources Department, Research on Energy Systems
(RSE) Spa, via R. Rubattino 54, 20134 Milano, Italy
2
School of Engineering, University of Basilicata, viale dellAteneo Lucano 10, 85100
Potenza, Italy
E-mail: leonardo.mancusi@rse-web.it
Abstract
This paper details the technical contribution to the theme of flood risk analysis as
consequence of a dam failure. According to the numerical problem proposed for the
workshop, the analysis consists of the evaluation of the dam break and its consequences. The
simulation includes two scenarios of dam breach: the scenario 1 that represents the case of an
easy erodible dam and the scenario 2 the case of an erosion resistant dam. For each scenario, a
dam failure discharge hydrograph was calculated and the subsequent flood wave and
consequences have been evaluated.
The methodology adopted involves, for a first part, the use of standard models for the
hydraulic modelling of the dam breach and flood wave propagation. For the second part, a set
of GIS scripts was written, tested and developed using the python scripting language to obtain
a rapid appraisal of consequences for the population and to assess the direct economic
damages for residential, commercial, and industrial buildings.
Since the latter elaboration depends greatly on the type and the level of detail of available
data, in this study we have been used data as generic as possible and GIS scripts that allow,
for a great variety of cases, a rapid initial assessment.
Introduction
Modern society considers it essential to increase the safety of the infrastructure. Risk analysis
is a helpful tool for the evaluation and management of risks which can affect people,
environment and human development. The purpose of this paper was to demonstrate the
application of an example of quantitative risk assessment technique that consists of estimating
the consequences of failure of a dam near populated areas with complex demographics,
infrastructure and economic activity.
The first chapter concerns the hydraulic modelling and simulation of the dam breach, the
second the subsequent flood wave propagation and the last focuses on consequence
estimation.
Dam failure
This section includes the description of modelling of the breaching process and the
subsequent discharge hydrograph for the hypothetical overtopping of the dam.
Two methods have been adopted and the results are named respectively scenario 1 and 2.
Scenario 1
In this case were used, a statistical method available in the literature for the dam failure peak-
discharge estimation of and a physically based mathematical model for calculating the total
discharge hydrograph.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
272
P. Molinaro [1] utilised 31 data sets (predominantly earthfill and some rockfill) extracted
from the report of J. E. Costa [2] to develop a relationship from the peak-discharge, the
height of dam and the reservoir volume at time of failure.
221 . 0
3
2
5
max
116 . 0
|
.
|

\
|
=
H
V
H g
Q

(1)

Where: Qmax = peak-discharge (m
3
/s)
g = gravity of Earth that has an approximate value of 9.81 m/s
2

H = height of the dam (m)
V = volume of water at breach time (m
3
)
Applying the formula (1) for this case (H=61 m and V=30.3*10
6
m
3
) we obtain:
Q
max
=32800 m
3
/s

The latter value was used to calibrate parameters of the mathematical model.
The mathematical model was developed by Molinaro [3] and simulates the breach
development process through an earthen dam due to overtopping.
The model is developed by coupling the conservation of mass of the reservoir inflow,
spillway outflow, and breach outflow with the sediment transport capacity of the quasi-steady
uniform flow along an erosion-formed breach channel. The rate at which the breach is eroded
is evaluated using the Engelund and Hansen [4] sediment transport relation.
The dam is modeled as an isosceles triangle formed by a noncohesive material of uniform
diameter D. The storage characteristics of the reservoir are described by specifying a table of
volume vs. water elevation.
The overtopping failure simulation starts by assigning a small initial breach whose bottom
elevation must be below the reservoir water level. The first stages of erosion are along the
downstream face of the dam while the breach bottom erodes vertically downward. An erosion
triangular channel is gradually cut into the downstream face of the dam. The sides of the
breach channel has a constant angle () with the vertical which is a function of the internal
friction () of dams material. The flow into the channel is determined by the broad-crested
triangular weir relationship.
The breach bottom is allowed to progress downward until it reaches the bottom elevation of
the dam, subsequently the channel becomes trapezoidal with the sides that maintain the same
slope of the previous triangle.
The following figure Figure 1 shows the sequence of the simulation of the breach formation.


Figure 1: Breach formation sequence
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
273
The most important parameter for model calibration is the characteristic diameter D. In this
case, taking into account the result of the equation 1, we have adopted the value D = 0.01 m.
The discharge hydrograph obtained is shown in the Figure 3.
This hydrograph was used for the first propagation of a flood wave referred to the next
chapter. It is characterized by Q
max
=28936 m
3
/s and a breach formation time of 2040 sec
(0.57 hrs).
Scenario 2
In addition to the above, a second method was applied for the evaluation of failure discharge
hydrograph. Even in this case a regression equations was used for the estimation of dam
breach parameters, and then a mathematical model was applied.
Table 1 summarizes the resulting breach parameters (Wb: bottom width of the breach and
tf: breach formation time) computed by several approaches available in the literature.
Table 1: Breach Parameters
Method Wb (m) tf (hrs)
MacDonald and Langridge - Monopolis (1984) 167 1.99
Floehlich (1995a) 147 0.66
Floehlich (2008) 110 0.57
Von Thun and Gillette (1990) 207 1.47

In the last column you can see that the breach formation time ranges from 0.57 to 2 hours. As
in scenario 1 the resulting time corresponds with the minimum, then for this scenario the case
of the maximum value of tf=2 hours was investigated.
For performing the dam breach outflow hydrograph computation, HEC-RAS model was
adopted. The implementation of these breach parameters in the HEC-RAS modelling system
is depicted on Figure 2. The resulting discharge hydrograph compared to the previous
scenario is depicted in Figure 3. As can be seen in the latter figure, the second hydrograph,
having the same volume, is characterized by a peak value cut in half compared to the first, but
a duration in time approximately double.
For further analysis, we can consider the two hydrographs: the first as representing of a easy
erodible dam and the second an erosion resistant dam.


Figure 2: HEC-RAS Dam Breach Model of scenario 2

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
274

Figure 3: Discharge hydrograph of scenario 2 compared with scenario 1
Flood simulation
Hydraulic modelling has carried out using the MIKE 21 software by the DHI Water
Environment Health to simulate flood wave propagation in the river and to describe the
inundation on the floodplain. This software solves the shallow water equations by means of a
finite difference scheme. For each scenario, the simulation has been constructed using, as
upstream boundary condition, the corresponding discharge hydrograph for the hypothetical
overtopping of the dam.
According to Bunya et al. [5], the Manning n coefficient is spatially assigned associating the
value of n with the land cover definition of 2001 from the USGS National Land Cover Data
(NLCD) (Table 2), These values are selected or interpolated from standard hydraulic
literature.
Table 2: Manning n value for 2001 NLCD classification
Lu
Code
Description n Mann.
(s/m
1/3
)
Lu
Code
Description n Mann.
(s/m
1/3
)
11 Open Water 0.020 42 Evergreen Forest 0.180
12 Perennial Ice/Snow 0.022 43 Mixed Forest 0.170
21 Developed-Open Space 0.050 52 Shrub/Scrub 0.070
22 Developed-Low Intensity 0.120 71 Grassland/Herbaceous 0.035
23 Developed-Med Intensity 0.120 81 Pasture/Hay 0.033
24 Developed-Hight Intensity 0.121 82 Cultivated Cropland 0.040
31 Barren Land 0.040 90 Woody Wetlands 0.140
41 Deciduous Forest 0.160 95 Herbaceous Wetlands 0.035

The results of flood modelling consist of values, for each grid cell in the study area, depth (m)
and the two components of the vector unit flow rate (m
2
/s) for 15 minute intervals and the
envelope of their maximum. Using a GIS scripts, hydrographs flow at different cross sections
were extracted. Some of these are shown in Figure 4.

Dam Failure discharge hydrographs for scenario 1 and 2
0
5000
10000
15000
20000
25000
30000
35000
0 20 40 60 80 100 120
Time (min)
Q
(
m
3
/
s
)
Discharge for scenario 1
Discharge for scenario 2
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
275

Figure 4: Discharge hydrographs of the two scenarios at different cross sections

Comparing the hydrographs of two scenarios in different cross sections we may note that:
while having at the initial cross-section a large difference in peak flow rate, however, during
the process of propagation downstream, an attenuation of the difference occurs.
Impacts assessment
The potential risks associated with the failure or disruption of dams could be considerable and
potentially result in significant destruction, including loss of life, massive property damage,
and severe long-term consequences. The following sections contain the analysis of two
categories: public safety and direct economic impact.
Population at Risk and Loss of Life estimation
The analyzes described in this section have been carried out by adopting the published
guidelines of the report [6] that provides guidelines and recommendations for estimating loss
of life resulting from dam failure or disruption.
The results of flood modelling and the data from the population census are used. Geographic
analyzes were carried out using Map Algebra techniques implemented in a set of scripts
written, tested and developed using the python scripting language and the Open Sources
GDAL libraries and NumPy Python module. To combine multiple maps in Map Algebra all
data have been converted into grid format.
The outputs of the hydrodynamic model have been processed to derive the information
required for the analysis. Using a GIS scripts, a Flood Wave Arrival Time grid was
obtained, in addition the two components of the vector unit flow rate are combined to obtain
the maximum Peak Unit Flow Rate values (m
2
/s). These values, called parameter DV, are
representative of the general level of destructiveness that would be caused by the flooding.
The DV values are then categorized, as suggested in the Figure 5 extracted from guidelines,
into ranges of values which define low, medium, and high severity zones.

Comparison of hydrographs of the two scenarios
at different cross sections
0
5000
10000
15000
20000
25000
30000
35000
0 20 40 60 80 100 120 140 160
Time (min)
Q

(
m
3
/
s
)
Scenario1 - Cross-sec.1
Scenario1 - Cross-sec.5
Scenario2 - Cross-sec.1
Scenario2 - Cross-sec.5
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
276

Figure 5: Flood severity rating criteria reported in the guidelines

The vector polygons of the population census block were converted into grid format: the
hypothesis assumed for the different values of the fields is that their distribution within the
polygon is homogeneous.
By overlaying grid maps of flood with the grid of the population is achieved as a result the
map of Population at Risk (PAR).
The estimate of loss of life is finally obtained by multiplying the PAR with the Fatality Rate
(Fraction of people at risk projected to die).
The latter was obtained by using the values of the Table 3 (from tab. 4 of guidelines) as a
function of warning time and flood severity.
Table 4 shows the results for the two scenarios for the hypothesis of event occurred at night
(understanding=vague). In the case of the first scenario, there is a population at risk greater
than 11 percent, while the estimate of the largest loss of life is almost 30 percent. The last
result is mainly caused by the shorter warning time of the first scenario.
It should be noted that in each case that the differences in terms of the consequences are less
than the differences of the peak discharge of the two scenarios at the breach of the dam, . This
result is due to the fact that the volume released from the dam is still the same.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
277
Table 3: Recommended Fatality Rates for Estimating Loss
of Life as reported in the guidelines
Flood
Severity
Warning Time
(min)
Understanding
Fatality
Rate
HIGHT Not applicable Not applicable 0.75
MEDIUM
No warning Not applicable 0.15
15 to 60
Vague 0.04
Precise 0.02
More then 60
Vague 0.03
Precise 0.01
LOW
No warning Not applicable 0.01
15 to 60
Vague 0.007
Precise 0.002
More then 60
Vague 0.0003
Precise 0.0002
Table 4: Population at Risk and Loss of Life estimation
Time
Interval
(min)
Total Population At Risk
14-yr and Under
Population at Risk

65-yr and Over
Population at
Risk
Loss of Life
Scen. 1 Scen. 2 Scen. 1 Scen. 2 Scen. 1 Scen. 2 Scen. 1 Scen. 2
0-15 0 3 0 0 0 0 0 2
15-30 2 4 0 0 0 1 2 2
30-60 4 529 2 794 1 169 732 308 196 1 783 1 262
60-90 5 111 641 721 110 1 057 105 2 0
90-120 10 411 8 529 1 491 1 084 2 163 1 996 0 0
120-180 8 935 12 346 1 350 1 899 1 662 2 290 0 0
>180 688 1 911 85 265 121 360 0 0
Total 29 676 26 228 4 816 4 090 5 311 4 948 1 787 1 266

Peak
Flood
Depth
Range
(m)
Flooded Area (m
2
)
Total Population
At Risk
14-yr and Under
Population at
Risk
65-yr and Over
Population at
Risk
Scen. 1 Scen. 2 Scen. 1 Scen. 2 Scen. 1 Scen. 2 Scen. 1 Scen. 2
0.0-0.5 8 308 132 4 088 784 5 854 4 825 915 706 1 017 937
0.5-1.0 6 025 869 7 190 976 5 683 6 770 895 1 021 987 1 237
1.0-1.5 7 854 660 7 858 701 8 937 7 769 1 373 1 164 1 664 1 453
1.5-2.0 5 612 806 5 141 466 3 768 2 879 543 391 781 676
2.0-2.5 3 711 550 2 970 641 1 562 1 038 197 118 404 304
2.5-3.0 2 547 251 1 901 885 510 380 72 60 137 98
3.0-3.5 1 746 448 1 453 532 274 292 43 49 68 70
3.5-4.0 1 279 417 711 995 209 220 37 53 49 28
4.0-4.5 546 680 392 859 125 228 35 62 8 17
4.5-5.0 391 332 310 246 153 237 43 65 11 17
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
278
5.0-5.5 380 736 209 944 211 216 57 54 15 13
5.5-6.0 242 630 250 352 190 271 51 67 15 16
6.0-6.5 201 323 294 532 194 346 53 90 14 25
6.5-7.0 192 882 221 527 162 337 44 83 9 30
7.0-7.5 176 180 128 947 176 154 44 37 11 11
7.5-8.0 186 417 101 919 254 84 61 18 16 7
>8 1 880 693 850 640 1 414 182 353 52 104 9
Total 41 285 006 34 078 946 29 676 26 228 4 816 4 090 5 311 4 948

Direct Economic Impact estimation
Methods and values of the parameters used in this section are drawn mostly from the report
[7]. They concern the assessment of the direct economic damages for residential, commercial,
and industrial buildings. The input data consist of map of land use and parcel zone map of the
study area. As in the previous paragraph, for the analysis, all the data are preliminarily
converted into grid format.
The following assessments do not take into account Agricultural, Roads, Infrastructure and
Vehicles damages. The assessment however allows the estimation of the damage to buildings
and their contents, and when applied to different scenarios allows an effective comparison of
the impact.
The extent of damage to the buildings and its contents is estimated from the depth of flooding
by the application of a depth-damage curve associated with each occupancy type.
Depth damage curves demonstrate the relationship between the depth of the flood relative to
the first finished floor level of buildings and the damage caused to the structures and contents.
Damages are typically expressed as a percentage of depreciated building replacement value.
Adopting a non-traditional approach, the adopted method models directly the content damage
as a percentage of structure value rather than using a content-to structure value ratio.
Not having a map of buildings, the area covered by the buildings has been derived from the
land use map according to the hypothesis of Building Coverage shown in the following table.
Table 5: Relationship between Land Use and Building Coverage
Lu
Code
Description Building
Cover. %
Lu
Code
Description Building
Cover. %
11 Open Water 0% 42 Evergreen Forest 0%
12 Perennial Ice/Snow 0% 43 Mixed Forest 0%
21 Developed-Open Space
10% 52 Shrub/Scrub 0%
22 Developed-Low Intensity
20% 71 Grassland/Herbaceous 0%
23 Developed-Med Intensity
35% 81 Pasture/Hay 0%
24 Developed-Hight Intensity
50% 82 Cultivated Cropland 0%
31 Barren Land 0% 90 Woody Wetlands 0%
41 Deciduous Forest 0% 95 Herbaceous Wetlands 0%

To calculate damages, each structure must be assigned to a structure occupancy type. For each
structure occupancy type an estimated replacement value and a structure depth-damage and a
content depth-damage relationship must be defined.
In our case, replacement values were extracted from the "Table C-3 Estimated Replacement
Value", depth-damage relationship from Table C-1 Depth Damage Curves, Defining
Damages as a Percentage of Depreciated Building Value for Depth of Flooding Above Floor
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
279
Height and the height of the floor of buildings from the ground level was taken from "Table
C-2 Foundation Heights" of report [7]. The following table contains the list of occupancy
type categories adopted.
Table 6: Occupancy types
Occup.
Type
Description Unit Cost
($US/sqm)
Origin data of Depth Damage Curve
RES1 Residential One Story,
No Basement
1711 USACE Generic Depth Damage
Curves for residential buildings
RES2 Residential Two or more
Stories, No Basement
3336 "
COM Commercial buildings 1528 USACE depth damage, as used in
Ford (2005) [8]
IND Industrial buildings 1528 "
PUB Public buildings 1711 "
FAR Homestreads 1711 "

The Figure 6 below shows the graph of the depth-damage curves.


Figure 6: Depth damage curves
To assign at each parcel the occupancy type we chosen the values according to the Table 7.
Table 7: Reclassify table: from parcel ZONINGCATE to occupancy type
ZONINGCATE Stories Occupancy Type
COMMERCIAL any COM
INDUSTRIAL / WHOLESALE / MANUFACTURING any IND
INSTITUTIONAL / GOVERNMENT any PUB
OFFICE 1 RES1
OFFICE 2 RES2
OFFICE 3 RES2
OPEN SPACE / RECREATION / AGRICULTURAL any FAR
RESIDENTIAL 1 RES1
RESIDENTIAL 2 RES2
RESIDENTIAL 3 RES2
Structural Depth-Damage Curves as a Percentage of Depreciated
Building Value for Depth of Flooding Above on the Ground Level
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
-1 0 1 2 3 4 5
water depth (m)
D
a
m
a
g
e

(
%
)
RES1_S
RES2_S
COM_S
IND_S
PUB_S
FAR_S
Contents Depth-Damage Curves as a Percentage of Depreciated
Building Value for Depth of Flooding Above on the Ground Level
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
-1 0 1 2 3 4 5
water depth (m)
D
a
m
a
g
e

(
%
)
RES1_C
RES2_C
COM_C
IND_C
PUB_C
FAR_C
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
280
RESIDENTIAL / AGRICULTURAL any FAR
UTILITIES / TRANSPORTATION any RES1

The results of applying the method for the two scenarios are listed in the following table.
Table 8: Direct Economic Impact
Time Interval
(min)
Direct Economic Impact ($US)
Scenario 1 Scenario 2
0-15 0 0
15-30 0 0
30-60 499 417 507 333 141 773
60-90 671 323 845 151 037 801
90-120 839 403 518 854 746 410
120-180 577 769 953 822 621 093
>180 12 807 947 64 207 745
Total 2 600 722 770 2 225 754 822
The results show that the total damage, in the case of the first scenario, are greater than 14%
and that difference occurs in the first 120 minutes. Also in terms of economic loss the
difference between the two scenarios are less than the differences of the peak discharge in the
breach of the dam.
Conclusion
In this paper we present the results of the analysis of a possible dam failure. The development
of a dam break is a complex process involving numerous uncertainties: the methodology
adopted in this work is a medium-scale approach type and can be used for the rapid and
consistent evaluation of consequences for the population and to assess the direct economic
damages for residential, commercial, and industrial buildings. Rapidity is allowed by using
aggregate data: maps of land-use, population census and parcel zone. Consistency is required
to ensure comparability between evaluations. For that reason the method can be used to
prioritize corrective actions to achieve the greatest and quickest possible risk reduction or for
identification of the most effective and better-justified measures of risk mitigation.
The comparison carried out for the two scenarios is an example of use of the methodology to
estimate the sensitivity of results with respect to an uncertain parameter which is the breach
formation time.
Acknowledgements
This work was carried out also thanks to the Research Fund for the Italian Electrical System
under the Contract Agreement between RSE S.p.A.(Research for Energetic System) and the
Ministry of Economic Development - General Directorate for Nuclear Energy, Renewable
Energy and Energy Efficiency, stipulated on July 29, 2009 in compliance with the Decree of
March 19, 2009.
References
[1] Molinaro P. (1990). Metodi statistici per la stima della portata di picco defluente in una
valle per la rottura di una diga in materiale sciolto. ENEL DSR-CRIS Milano.
[2] Costa J. E. (1995). Floods from dam failures. U.S. Geological Survey, Open-File Report
85-560 Denver Colorado.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
281
[3] Molinaro P. (1986). Modello matematico della breccia che si sviluppa in uno sbarramento
in materiale sciolto per tracimazione. ENEL DSR-CRIS Milano.
[4] Engelund F., and E. Hansen (1972). A Monograph on Sediment Transport in Alluvial
Streams, Teknisk Folag, Copenhagen.
[5] S. Bunya, J. C. Dietrich, J. J. Westerink, B. A. Ebersole, J. M. Smith, J. H. Atkinson, R.
Jensen, D. T. Resio, R. A. Luettich, C. Dawson, V. J. Cardone, A. T. Cox, M. D. Powell,
H. J. Westerink, H. J. Roberts. (2010). A High-Resolution Coupled Riverine Flow, Tide,
Wind, Wind Wave, and Storm Surge Model for Southern Louisiana and Mississippi. Part
I: Model Development and Validation. Monthly Weather Review 138:2, 345-377.
[6] Dams Sector. Estimating Loss of Life for Dam Failure Scenarios. U.S. Department of
Homeland Security, Washington, D.C. , September 2011.
[7] Department of Water Resources Division of Flood Management (2008). Flood Rapid
Assessment Model (F-RAM) Development 2008, State of California, The Resources
Agency
[8] David Ford Consulting Engineers (2005). Urban flood scenario: Sacramento area levee
breach scenario, report prepared for California Department of Water Resources,
December 5
th
2005.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
283
2-D hydraulic modelling of a dam break scenario
O. Saberi
1
, C. Dorfmann
2
and G.Zenz
3

1, 2
PhD student, Institute of Hydraulic Engineering and Water Resources Management,
Stremayrgasse 10/2, A-8010 Graz, AUSTRIA
3
Univ. Prof. Institute of Hydraulic Engineering and Water Resources Management,
Stremayrgasse 10/2, A-8010 Graz, AUSTRIA
E-Mail: omid.saberi@student.tugraz.at
Abstract
Dam failure is a catastrophic event, and study on the structure of dam is important in the field
of water resource engineering. The reason is the risk to life and property below the dam
structure. The dam failure occur due to many reasons, some of these reasons are related to
structural problems while others are related to the hydraulic conditions. For embankment
dams, main reasons are overtopping and piping through the dam body or foundation.
However there are some situations that make dam failure happen suddenly, like earthquakes,
landslides or wars.
This study provides a two-step numerical simulation of dam failure analysis for an
embankment dam which its failure happened in overtopping conditions. First step is about
simulating breach of the dam and calculating breach hydrograph, maximum discharge of
breach, initial time of the breach formation, and time of maximum breach discharge and other
parameters of breach. BREACH GUI software was used for this step. The second part
simulates flood after dam break. TELEMAC2D software was used for simulating the flood.
The results include the travel time (warning time) of the flood wave to various locations in the
downstream valley and the representative valley cross-sections depicting flow depth and unit
flow rates. Furthermore the results of this paper can provide information to build an
inundation map which can help to develop risk management analyzes.
Introduction
A hypothetical embankment dam was constructed in a mountainous region. This high hazard
dam sits directly above a lightly populated area which is 3.5 kilometers away from an urban
environment. The primary function of this dam is flood control for heavy snowmelt and
strong monsoonal weather patterns. In addition, the reservoir provides some water supply and
recreational activities to nearby communities. The dam failure for this dam takes place when
the pool elevation is at crest elevation. The mode of failure is assuming an overtopping failure
[6].
Two primary tasks in the analysis of a potential dam failure are the prediction of the reservoir
outflow hydrograph and the routing of that hydrograph through the downstream valley to
determine the dam failure consequences. When populations are located close to a dam, it is
important to accurately predict the breach outflow hydrograph and its timing relative to events
in the failure process that could trigger the start of evacuation efforts.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
284
Breach Parameter Estimation
Empirical Method
The empirical method are used to predict breach parameters estimation and breach peak
discharge and breach hydrograph estimation including geometry of breach, time to reach
failure and peak flow discharge, the empirical methods are get from documented failures.
The recommended empirical method for predicting dam breach parameters are
MacDonald&langridge-Monopolis (1984), Froehlich (2008), Froehlich (1995 a) and Von
Thun and Gillite. In this paper we calculate the dam breach parameters with Froehlich (2008)
Formula. For more details we refer to the 12th International Benchmark Workshop [6].
Froehlich (2008) is depending only on the height of breach (

, the reservoir volume


and the breach side-slope, also we have the failure mode factor (

) in the Froehlich (2008)


for distinguish between overtopping and piping failure.
After put mentioned parameters in the Froehlich (2008) formula we have breach hydrograph
in the below shape:
Table 1: Results of the Froehlich (2008)


Froehlich (2008)
39500
0.57 110



Figure 1: Outflow hydrograph from the Froehlich (2008)
BREACH modelling
In this project we use the National Weather Service (NWS) software Brach GUI for
predicting the breach parameters and the resulting outflow hydrograph of an earthen dam [1].
BREACH model was primarily developed by Fread [9] from the National Weather Services.
Since 1988 it has been used broadly. This software use a physically-based mathematical
model using the principles of hydraulics, sediment transport, soil mechanics, the geometric
and material properties of the dam, and the reservoir properties [1]. Furthermore it uses
critical properties which are measurable from dam material descriptions. Therefore it
considers a more robust model. However it is the responsibility of the engineer to determine
the appropriate combination of values as the measure characteristic are normally changing
with in a wide range.
Available parameter of breach hydrograph
The dam used in this study is a homogenous embankment dam which is a rolled earth fill
structure and composed of predominantly sandy clays and clayey sands. It is located in a
0
10000
20000
30000
40000
50000
0 0.2 0.4 0.6
D
i
s
c
h
a
r
g
e

(
m

*
1
0
0
0
/
s
)

Time (hours)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
285
mountain region. The width of crest is 24 m, the length of crest is 360 m and the height of the
dam is 61m. The total water volume of water behind the dam is 38,276,344 m. The distance
from this dam to the population area is 3.5 km. The dam failure will take place when the
water level behind the dam is at crest elevation and the mode of failure will be over topping
failure. For more details we refer to the 12th International Benchmark Workshop [6]. As base
inflow we assumed a discharge equal to 100 m/s. In Table 1 the main physical parameters of
the dam are listed.
Table 2: Strength parameters of soil [6]
Effective cohesion (KPa) 19.15
Effective friction angle 14
Undrained strength (KPa) 43.09+0.175


Shear Wave velocity (m/s) 152.4
Maximum shear modulus (KPa) 46443
Saturated unit weight (kg/m3)

2002
Permeability (cm/s) 1.9*10^-6
Calculating input parameter for Breach software
For start a simulation with the BREACH software we need to calculate some parameters as
input data. The most important input parameters to be provided in the BREACH software are
the mean grain size D50 (mm), the initial elevation of water surface, the elevation of the top
of the dam, the elevation of the bottom of the dam, the average plasticity index (PI), the
critical shear stress coefficient, the slope of the upstream and downstream faces of the dam
(1:3), the base inflow hydrograph, the surface area behind the dam vs. the elevation of the
dam, the porosity specifications of the dam material and the breach parameter. Table 1 and the
equations 1-4 according to [] can help with defining these input parameters:

(1)
where V is the shear wave velocity (m/s), is the density of material (kg/m3) and G is the
shear modulus (KPa)

(2)
where

is the specific gravity, is the unit weight of soil (N/m) and

is the unit weight


of water (N/m)

(3)

where

is the specific gravity,

is the unit weight of water (N/m) and e is the void ratio



(4)

where e is the void ratio (%) and n is the porosity (%).
With the parameters in Table 1 and the equations 1-4 the porosity n is calculated as 49%. D50
can be considered approximately equal to 0.066 mm [4]. The basis for this approximation
comes from soils mechanics fields.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
286
When the diameter of Clay soil is less than 0.002 mm and the diameter of sand soil is between
and 0.05 and 2.00 mm then sandy clay is a soil with 33 percent till 55 percent clay and up to
65 percent sand [5].
The average plasticity index (PI) is calculated according to the USCS plasticity chart and the
relationship equations of Skempton and Henkel (1953) [3], Psterman (1959) [3] and
Bjerum(1960) [3] which is equal to 42%. Furthermore PI will be used in the following
formulation to calculate two more parameters required by BREACH software (CA, CB) [1]:



(5)

where

is the critical shear stress (kPa), PI is the plasticity index (%) and CA, CB are the
critical shear stress parameters. Having PI=42% and

derived from Table 1, CA and CB are


calculated as .
Output of the BREACH software
Having all the above inputs into the software we can run the BREACH model to calculate the
outflow hydrograph given in Figure 1.

Figure 2: Outflow hydrograph from the breach software

Furthermore the software output provides some other parameters which can be used in risk
management studies. These parameters are given in Table 2:

Table 3: Results of the BREACH software
Maximum discharge of breach (m3/s) 41600
Time for peak out flow (hours) 4.62
Time for starting outflow (hours) 4.44
Final depth of breach (m) 61
Top width of breach at peak breach flow (m) 250
Time to reach final breach bottom elevation (hours) 4.97
Bottom width of breach at peak breach flow (m) 15
0
5
10
15
20
25
30
35
40
45
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
D
i
s
c
h
a
r
g
e

(
m

*
1
0
0
0
/
s
)

Time (hours)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
287
Comparison between result of Froehlich (2008) and BREACH modelling
After calculating breach hydrograph from Froehlich (2008) and breach modeling in follow
picture, we find out all the dam breach parameters including the failure time, the peak
discharge outflow and the average of breach width of Froehlich less than BREACH software.

Figure 3: merged discharge of cross sections (m3)/s
2-D hydrodynamic flood with Telemac-2D
The flood from dam break can go to downstream valley and in this part we are going to
determine dam failure consequences and calculating the water surface, the water depths, the
unit flow rates as well as the flow rates in various cross sections. The TELEMAC-2D
modelling software is used for this purpose. TELEMAC-2D is a numerical simulation
software and belongs as module to the open source Telemac-Mascaret suite for the
simulations of hydrodynamic flow, contaminant and sediment transport. It solves the two-
dimensional depth-averaged Saint Venant equations for free surface flow. A detailed
description of the Telemac System is given in Hervouets book (2007)[7]. The main results of
Telemac-2D are the water depths and the depth- averaged velocity components.
The original geometry provided by the ICOLD benchmark specifications was modified by
deleting a priori non flooded mesh areas in order to save computation time. Table 3 shows the
resulting mesh properties:

Table 4: Mesh properties







The upstream boundary condition is located at the dam and the hydraulic boundary condition
is the unsteady flow hydrograph from the BREACH software, see Figure 1. The downstream
boundary condition is a free outflow.

0
5000
10000
15000
20000
25000
30000
35000
40000
45000
4.463 4.963 5.463
D
i
s
c
h
a
r
g
e

(
m

*
1
0
0
0
/
s
)

Time (hours)
BREACH software
Froehlich (2008)
Number of element: 213750
Interior elements: 210652
Edge elements: 3038
Number of nodes: 108433
Edge nodes: 3114
Interior nodes: 105319
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
288
Results overview
The results of TELEMAC-2D are stored in the binary SEARAFIN format which can be
analyzed and visualized by using the free software Fudaa or BlueKenue [8]. In this project
BLUE KENUE software was used.
In the following the results are shown graphically for breach discharge, cross section
discharge, peak flood depth, flood wave arrival time, peak unit flow rate and flooded area.
Furthermore interpretation on each subject is given along with the output figure.
Breach Discharge
This result was calculated in the first part with BREACH GUI software (shown in Figure 1).
Also from the calculated breach hydrograph we find out:
- With the inflow of 100 m/s into the dam reservoir it takes 4.36 hours for the dam
breach initiation.
- The maximum discharge takes place after 648 s after the breach initiation.
- The duration of flood is 2844 s.
Cross-Section Discharge
For the calculation of the cross section discharge the following steps are performed:
- First the unit flow rates are calculated (m/s)
- Second the unit flow rates are integrated along the cross section width
Figure 2 shows the discharges in various cross sections and the reduction of the peak flows
going downstream from cross section 1 to 5.


Figure 4: merged discharge of cross sections (m3)/s

0
5000
10000
15000
20000
25000
30000
35000
40000
45000
00:00:00 00:28:48 00:57:36 01:26:24 01:55:12 02:24:00 02:52:48
D
i
s
c
h
a
r
g
e

(
m
3
/
s
)

time (HR)
Discharge of cross sections (m3/s)
cross section 1
cross section 2
cross section 3
cross section 4
cross section 5
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
289
Peak flood depths
In BLUE KENUE software we can also calculate the maximum flood depths in the project
area. Figure 3 shows the results for the peak flood depths where the blue color indicates the
safe area with maximum water depths less than 0.5 m and the red color indicates dangerous
areas with minimum water depths of 4.5 m.


Figure 5: Peak flood depths (m)
Flood wave arrival time and water depths
In this part we calculated the arrival time of flood. This will help us to find out how long it
takes for the flood to reach each part of downstream area. BlueKenue software visualizes the
water depths separately for each chosen time interval. It means BlueKenue gives us one
gridded dataset for each time intervall. Figures 4, 5 and 6 shows the water depths or flood
wave arrival times for certain time intervals.


Figure 6: Water depths after 30 minutes
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
290

Figure 7: Water depths after 2 hours and 30 minutes

As we can see in Figure 4, 5 and 6 these visualizations are useful to analyze the flood wave
arrival time and the locations of impact.
Peak unit flow rate
The unit flow rate is given in equation 6:

(6)

where q is the unit flow rate (m/s), U and V are the water velocity components in x and y
direction (m/s), and H is water depth (m).
The peak unit flow rate in every mesh node can be extracted from the time varying unit flow
rate and is shown in Figure 7.


Figure 8: Peak unit flow rate (m/s)
Flooded area
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
291
For calculating the flooded area we can define the total wet area by defining iso-lines. This
option categorizes the flooded area for arbitrary water levels.
The total flooded area is given in Table 4 and the flooded areas for different water depths with
interval of 0.5 m is given in Table 5.
Table 5: Total flooded area
Total flooded area (m) 47626580
Table 6: Flooded areas, intervals 0.5 m
Water heights (m) Flooded area (m) *10^3
>8 4179
8 4290
7.5 4805
7 5565
6.5 6827
6 7910
5.5 9131
5 10269
4.5 11569
4 13349
3.5 15432
3 18145
2.5 22200
2 27109
1.5 32959
1 38433
0.5 41043

The results of these tables can be used in the risk management study for calculating the
damage property and loss of life.
Conclusion
The first part of this study describes the BREACH model for predicting the breach
hydrograph of a hypothetical embankment dam. The reason for the failure of dam is assumed
to be overtopping. The following results can be deduced:
- The maximum outflow in this project is calculated as 41558 m/s. The water flow is
transferred into the downstream direction of the dam which can cause catastrophic
consequences.
- The maximum discharge happens near the dam place at cross section number 1.
- One of the most important parameter in the risk management is the alarm time. In this
study we have calculated the break initiation time as well as the formation time.
Furthermore these data can help us to determination the Property damage and loss
lives.
The above mentioned results used in the successive 2-D depth-averaged calculations with
TELEMAC-2D. The following conclusions can be made based on the 2-D simulation
results:
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
292
- The height of water and the maximum height of water which are important parameters
in risk management fields are calculated to define inundation regions. These results
also can be used to define warning zones according to the flooded areas.
- The wave arrival time is analysed for 15 minutes intervals. This time model shows us
how long it takes for the flood to reach each part of the downstream area which is
useful for reducing loss of human lives in risk management studies.
Acknowledgements
I would like to express my deepest appreciation to all those who provided me the possibility
and help me to complete this report.
References
[1] Janice Sylvestre, Developer, 2010, BREACH GUI (BREACHJ) Version 1-0-0 08-01-
2010
[2] TELEMAC modelling system, 2D hydrodynamics, TELEMAC-2D software Version 6.0,
USER MANUAL
[3] A. S. Balasubramaniam, H. Cai, D. Zhu, C. Surarak and E. Y. N. Oh,Griffith School of
Engineering Griffith University, Gold Coast, Australia, Settlements of Embankments in
Soft Soils
[4] AFTA 2005 Conference Proceedings, Placement of Riparian Forest Buffers to Improve
Water Quality
[5] Braja M.Das, 2008. Advance soil mechanic, third edition. This edition published 2008
[6] By Taylor & Francis 270 Madison Ave, New York, NY 10016, USA
[7] 12
th
International Benchmark Workshop on numerical analysis of dam, 2013, Austria,
Graz
[8] Hydrodynamics of Free Surface Flows: Modelling with the Finite Element Method by
Jean-Michel Hervouet(May 29, 2007)
[9] CHC - Canadian Hydraulics Centre, National Research Council, 2010. Blue Kenue,
Reference Manual, August 2010.
[10] The NWS simplified dam-break flood forecasting model by Jonathan N.wetmore and
Danny L.Fread (Reviesed 12/18/91) by Danny L.Fread, Janice M.Lewis, and Stephen
M.Wiele



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
293
Risk Assessment Analysis of a Hypothetical Dam
Breach Using Adaptive Finite Element Methods
Adaptive Hydraulic Model, ADH
D. McVan
1
, J. Ellis
2
, G. Savant
3
and M. Jourdan
1

1Research Hydraulic Engineer, Engineer Research and Development Center, U.S. Army
Corps of Engineers, Vicksburg, MS, 39180. E-mail: Darla.C.McVan@USACE.Army.mil
2Engineering Technician, Bowhead Science and Technology, LLC, and Onsite Contractor,
Engineer Research and Development Center, U.S. Army Corps of Engineers, Vicksburg, MS
39180.
E-mail: Jeffrey.D.Ellis@usace.army.mil
3
Research Water Resources Engineer, Dynamic Solutions LLC and Onsite Contractor,
Engineer Research and Development Center, U.S. Army Corps of Engineers, Vicksburg, MS
39180.
E-mail: gaurav.savant@usace.army.mil
Abstract
A virtual computational test bed was developed by a committee sponsoring a session at the
International Commission on Large Dam (ICOLD) 12th International Benchmark Workshop.
The purpose of this virtual test bed was to serve as a platform for comparison of dam breach
models and methodologies. This paper provides the results of one of the model comparisons.

The ADH model, developed by the U.S. Army Engineer Research and Development Center
(ERDC) was used to simulate a catastrophic dam failure and gradual dam failure. The ADH
Model is described, as well as the domain, setup, and boundary conditions for the specific
simulations.

Results are provided as gridded and tabular data, as well as hydrographs. The gridded results
include peak flood depths, flood wave arrival times, peak unit flow rates and total population
at risk. Tabular results include flooded area and population as risk (including age
demographics). Hydrographs are presented for both the dam failure discharge and at five
down stream cross sections. Loss of life and economic impacts were not calculated.
Introduction
Computational capacity, along with multi-processors computing techniques, has grown
resulting in development of new models for flooding and consequence/risk assessment.
These new models which are computing flood wave propagation at higher spatial resolutions
can be coupled with risk assessment models which allow for more detailed analyses of the
effects of these flows.

This paper represents results of the numerical dam breach done using the virtual testbed
developed for the International Commission on Large Dam (ICOLD) 12th International
Benchmark Workshop. This virtual testbed was developed as a tool for comparison of
different numerical model results and the model capabilities. The testbed includes a
hypothetical dam above a hypothetical city named Hydropolis. The focus of this study was to
estimate the consequences of failure of the dam. Demographics, infrastructure and dam
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
294
characteristics were provided so that comparison of these models could be made which will
aid in the development of simulation frameworks that can support dam risk analysis. Two
model scenarios were modeled; the first model represents a catastrophic event in which the
entire dam is destroyed instantaneously. The second model represents a gradual breach using
Froehlich (1995a) equations.
Methodology
Model Code Description
ADH is a state-of-the-art code developed by the U.S. Army Engineer Research and
Development Center (ERDC) to simulate both saturated and unsaturated groundwater,
overland flow, three-dimensional Navier-Stokes flow, and two and three-dimensional shallow
water equations, including super-critical flows and shock capturing (Berger el.al. 2010). The
ADH code is parallelized and is capable of running on high performance computing systems.
ADH is also an implicit code meaning its time step size is not limited by the element size as
with explicit codes. These features make ADH computationally efficient for large-scale
applications. ADH has been utilized to study varied phenomena such as estuarine circulation
(McAlpin et al. 2009; Tate et al. 2010; Martin et al. 2010, Martin et al. 2011), riverine flow
(Stockstill and Vaughan 2009; Stockstill et al. 2010), and dam breach (Savant et al. 2011).

The two-dimensional shallow water (SW2) equations are used for the application presented
here. This module solves the conservative form of the SW2 equations, allowing for local and
global mass conservation. ADH is a temporally and spatially adaptive code. For temporal
adaption, the code utilizes a variable time step such that failure to reach convergence at a
given time step size does not end the simulation; rather, the time step size is reduced and the
solve is attempted again (Savant et al. 2011). ADH also has the ability to allow continuous
wetting and drying such that the flood front is computed and visualized accurately (Berger
and Lee 2004; Savant and Berger 2011). For this model simulation, a maximum time step of
10 seconds was specified with smaller time steps being performed automatically as needed.
Model Setup
All hydrodynamic numerical models require certain basic input data to perform simulations.
These data are bathymetric and topographic elevations for defining the model domain, inflow
hydrographs or water elevations, as well as other boundary information such as rainfall and
ocean water surface elevation (if the area of interest is in a tidal zone) for driving the model,
as well as roughness characteristics and the ground surface and other parameterizations
depending upon the objective of the modeling application. In the following sections, these
domain construction data, driving data, and parameterization needs are addressed.
Model Domain
The bathymetric and topographic information for this workshop were provided from the
ICOLD Theme C Formulation Team in the form of a DEM with 9.4 meter resolution. Two
DEMs were provided, one representing the domain without the dam, and the other with the
dam. The DEM covered an area that was approximately 250 km
2
. The hypothetical dam was
located at coordinate 4499.66 meters and 6681.57 meters and has a crest width of 24 meters
and a crest length of 360 meters. The river bed elevation at the base of the dam is 211 meters
and the crest elevation is 272 meters with 3H:1V embankment slopes. The maximum storage
capacity of the reservoir is approximately 38 million cubic meters.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
295
The ADH mesh was created using the Surface-water Modeling System (SMS) (Aquaveo
2009) meshing algorithms and consists of 13,790 triangular elements comprised of 7,138
nodes at the triangle corners. Both the catastrophic model and the gradual breach model used
the same mesh. The dam was built into the mesh at the provided coordinates specifying the
width of the elements along the dams crest and determining the upstream and downstream
length of the embankment. Element areas range from approximately 250 m
2
to 25,000 m
2
,
with the finest resolution located at the dam and along the channel and the larger areas located
downstream in the lake. After the mesh was constructed, the separate DEMs were
interpolated to the mesh. The DEM without the dam was interpolated to the mesh for the
catastrophic model and the DEM with the dam was for the gradual breach model. Figure 1
shows the mesh at the dam site, the model domain and bathymetry.
Model Boundary Conditions and Parameterization
Numerical modeling of any type requires the specification of driving boundary information.
For the dam breach model, this driving force is maximum volume of the reservoir at the crest
elevation, 272.0 meters. This is achieved by creating initial depth conditions at each node.
For the nodes representing the reservoir, the water surface elevation was entered and the
initial water depth was determined from the mesh bathymetry. The rest of the mesh was


Figure 1: Model Domain.

given a -0.1 meter water depth in order to simulate a dry mesh. In an attempt to prevent back-
up of the flow which may cause artificial inundation not created by the flooding, an
evaporation boundary was applied to the downstream boundary that allows the water to be
transported out of the domain. However, after the model was complete, it was observed that
after the flood wave hits the boundary, the wave is reflected back into the model domain.
This only occurred in the area of the lake and it was determined that it would not affected the
area of interest.

Bed roughness was an additional user specified parameter that defines the frictional
resistance. ICOLDs Theme C Formulation Team provided a gridded dataset representing
the land use/land cover of the region. This data was used to identify areas of similarity and
applied to the mesh as material types. Figure 2 shows the various material types used in the
model. The elements and nodes associated with each material type will have the same
material properties specified in the boundary conditions. Literature search describing the
varying roughness coefficients for differing floodplains were used to determine the bed
roughness for each material type (Arcement, G.J., Schneider, V.R., and Oregon Department
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
296
of Transportation, 2005). Table 1 shows the material types used in the model along with its
roughness value.
Table 1: Bed Roughness
Material Type Mannings n Value
Channel 0.018
Forest 0.100
Developed 0.020
Shrub/Cropland 0.050
Shrub/Forest 0.070
Lake 0.018
Reservoir 0.018
Dam Crest 0.018
Upstream Dam Face 0.018
Downstream Dam Face 0.018
Downstream Boundary 0.018
(Arcement, G.J., Schneider, V.R., ) (Oregon Department of Transportation, 2005)

In the ADH numerical code, the Mannings value is converted to an equivalent roughness
height that depends on the instantaneous water depth at a particular node. The friction
formulation is then derived from the logarithmic velocity profile based on open channel flow.
Thus, while the user-specified frictional value (Mannings number) is constant over the
specified material type, the actual applied resistance is spatially varying according to the
water depth and velocity magnitude at each node in the mesh. This results in higher frictional
values for locations of shallower depths as is experienced in nature and can be extremely
important when estimating flood extents.
Figure 2: Material Types
For the catastrophic model, the dam does not exist in the mesh therefore the models initial
conditions at time 0.0 simulated a wall of water at a depth of 61 meters.
For the gradual breach, the dam does exist in the mesh. Equations developed by Froehlich
(1995a) were used to determine the average breach width and the breach formation time and
are shown below. Failure of the dam was assumed to be caused by overtopping resulting in
80-percent of the dam height removed or approximately 48.8 meters. Therefore, the
maximum width of the breach was approximately 141 meters and the breach formation time
was approximately 48 minutes (2940 seconds). (The height of the breach and the formation
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
297
time were rounded to 49 meters and 49 minutes). To achieve the gradual breach, elevations
of the nodes representing the dam were gradually lowered at specified time intervals. At the
end of the breach, the elevations of the nodes representing the breach were 223 meters.

(1)

(2)

Where: B
ave
= Average Breach Width (m)
K
o
= Constant (1.4 for overtopping failures, 1.0 for piping)
V
w
= Reservoir volume at time of failure (m
3
)
h
b
= Height of the final breach (m)
t
f
= Breach formation time (hours)
Model Results
Model Simulations
The ADH model simulations for both the catastrophic and the gradual breach were performed
on the SGI Altix Ice machine (Diamond) at the USACE ERDC (ERDC DSRC 2011). This
machine uses an SLES (Linux) 11 operating system with a 2.8 GHz Intel Xeon Nehalem-EP
Processor. The catastrophic model was run on 16 processors and simulated 6 hours 40
minutes (24,000 seconds) and required approximately 60 minutes to complete (wall time).
The gradual breach was run on 12 processors and simulated 12 hours (43,200 seconds) and
required approximately 2 hours to complete. Both models had output intervals of 100
seconds.

In accordance to ICOLDs Theme C Formulation Team, the information obtained for both
the catastrophic event and the gradual breach were discharge hydrographs for the dam failure
and along five cross-sections specified by the Formulation Team. The model also provided
contours of the peak flood depths, peak unit flow rate, and the flood wave arrival times.
Because ADH uses a mesh, the output results are also in the mesh format. In other words,
each node in the mesh contains all the information for each time step. To determine the total
population at risk including the age demographics as well as the flooded area, the ADH depth
file were converted to a raster for each output time interval. The raster was then converted to
a shapefile which contained the depth at 1 meter intervals. The Formulation Team specified
0.5 meter intervals, however due to the constraints of this methodology, only integer values
could be used. The census block shapefile provided by the Formulation Team was clipped to
the depth shapefile and the ratio of the clipped area compared to the entire area determined
the total population potentially affected by the flood.
Catastrophic Model Results
As mentioned previously, the catastrophic model simulated the worst case scenario where the
entire dam is removed instantaneously. The maximum discharge at the time of the breach
was approximately 84,000 cms. At each cross section, the maximum discharge varied
between approximately 61,000 cms and 81,000 cms. Figure 3 shows the discharge
hydrographs for each location. The total volume at the reservoir was approximately 38.93
km
3
and at each of the cross sections, the volume varied between 39.4 km
3
and 40.9 km
3
.
Table 2 shows the maximum discharge with the associated times of occurrence at the dam and
for each of the 5 cross-sections along with the total volume for each location.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
298
Though ADH conserves fluid and constituent mass to machine precisions, for wet-dry
problems this conservation is dependent on user specified tolerance parameters. These
parameters control how accurately ADH solves the continuity and momentum equations using
the Newton-Raphson iterative approach for solving a system of implicit equations. Very small
tolerance parameters will drive the solution to conserve mass to machine precision but can
significantly increase run time, large tolerances will significantly reduce run time but will not
provide mass conservation or the correct results. A balance has to be achieved between the
requirements of a particular project or simulation and whether exact mass conservation is
required. For the simulations presented in this document the mass is conserved to 3-percent
for catastrophic breach. ADH utilizes a diffusive wave type shock capturing scheme in wet-
dry areas, this scheme, for extreme wet-dry scenarios, can hide fluid mass in dry cells and is
an additional, though rare, source of mass errors. All elements/nodes are included in
computations at all times and no minimum include in computation depth is used by ADH.
The error shown in Table 2 shows how well ADH conserved the mass at each cross-section.
Table 2: Maximum Discharge for Catastrophic Event


Location
Time of Maximum
Discharge
(seconds)
Maximum
Discharge
(cms)

Total Volume
(km
3
)


Error
At the Dam 1 83,828 38.93 -
X-Section 1 26 80,855 39.37 1.01
X-Section 2 207 66,027 39.47 1.01
X-Section 3 293 60,738 39.69 1.01
X-Section 4 385 62,089 39.65 1.00
X-Section 5 546 67,166 40.91 1.03



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
299
Figure 3 Maximum Discharges for Catastrophic Event.

The reservoir was completely emptied in approximately 50 minutes (3000 seconds). The
flood wave was completely at the lake after 2.7 hours. As one would expect, the greatest
depths occur within the channel. Maximum depths in the floodplain range between 2 and 12
meters. The unit discharge in the floodplain is less than 50 m
2
/s. Figures 4 and 5 show the
maximum peak depth for the catastrophic model and the peak unit flow rate, respectively.
The flood wave reaches the floodplain about 5 minutes after the dam is breached and reaches
the downstream boundary at approximately 1.67 hours (100 minutes). Figure 6 shows the
arrival time in 15 minute intervals.
Figure 4 Peak Flood Depth in meters for Catastrophic Event

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
300
Figure 5 Peak Unit Flow Rate (m2/s) for Catastrophic Event
Figure 6 Flood Wave Arrival Times for Catastrophic Event
The total flooded area at peak depths, excluding the area of the lake, was approximately 41.1
km
2
and affected well over 28,000 people. The total population at risk after 180 minutes is
over 26,000. These differences are the results of taking a single snap shot of the population in
time. As the flood wave travels downstream, the inundated area covers a finite area. For the
next time, the inundated area covers a completely different area. Sometimes these areas
overlap which result in duplicating the population at risk and at other times there is a gap
resulting in the population at risk not being accounted for. Therefore, when assessing loss of
life and economic consequences, it is best to use the peak flood depth values. Figure 7 shows
the clipped total population at risk for the peak depths. Table 3 shows the population at risk
at 15 minute time intervals and includes age demographics and Table 4 shows the flooded
areas at 1 meter intervals along with the total population at risk and the age demographics.
For the 12-year and under age demographic, the total population at risk after 3 hours is over
4,300 and about 4,600 for the maximum flooded area. The 65-year and over age demographic
has over 4,700 at risk after 3 hours and around 5,200 for the maximum flooded area. The
total flooded area with depths greater than 8 meters is over 3.35 km
2
which affects a total
population at risk of less than 2,000. For the 12-year and under demographics, the total
population at risk affected by depths greater than 8 meters is slightly over 400 and over 130
for the 65-year and over demographics.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
301
Table 3: Total Population at Risk for Catastrophic Event
Time Intervals
(minutes
Total Population
at Risk
12-yr and Under
Population at Risk
65-yr and Over
Population at Risk
15 3,179 815 225
30 12,516 1,936 2,487
45 9,807 1,449 1,910
60 629 97 132
75 33 7 1
90 19 4 1
105 13 2 0
120 12 2 0
135 12 2 0
150 11 2 0
165 11 2 0
180 11 2 0
Total 26,253 4,320 4,759
Table 4: Flooded Area Population at Risk for Catastrophic Event
Peak Flood
Depth Range
(meters)
Flooded Area
excluding lake
(km
2
)

Total Population
at Risk
12-yr and Under
Population at
Risk
65-yr and Over
Population at
Risk
>/= 1 41.09 28,802 4,641 5,208
>/= 2 32.37 21,273 3,429 3,994
>/= 3 17.55 6,274 1,207 1,044
>/= 4 9.73 3,494 835 327
>/= 5 5.94 2,886 710 220
>/= 6 4.56 2,598 635 202
>/= 7 3.82 2,201 535 167
>/= 8 3.35 1,710 413 136
Total 41.09 28,802 4,641 5,208



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
302
Figure 7 Total Populations at Risk for Catastrophic Event
Gradual Breach Model Results
Failure of the dam for the gradual breach scenario was assumed to be caused by overtopping
resulting in 80-percent (49 meters) of the dam height removed. The maximum width of the
breach was 141 meters and the breach formation time was 49 minutes (2940 seconds).
Elevations of the nodes representing the dam were gradually lowered by 1 meter at 1 minute
intervals. At the end of the breach, the elevations of the nodes representing the breach were
223 meters.

The maximum discharge at the dam location was over 17,000 cms and occurred 33.33
minutes (2000 seconds) after the initial time of breach. The maximum discharges for the 5
cross-sections ranged from between 16,400 cms and 20,500 cms. Table 5 shows the
maximum discharges at the dam and the 5 cross sections along with the calculated volumes of
each hydrograph. Figure 8 shows the discharge hydrographs for each location. Since the
lowest elevation of the final breach is 223 meters, the reservoir is assumed to retain a volume
of approximately 0.36 km
3
. The volumes at the cross sections range from 38.5 km
2
to 42.2
km
2
. As previously mentioned, for wet-dry problems, the conservation of mass is dependent
on the tolerance parameters specified by the user which controls how accurately ADH solves
the continuity and momentum equations. For the gradual breach simulation, the mass was
conserved to 8-percent. The error in Table 5 shows how well ADH conserved the mass at
each cross section.

The maximum peak depths for the gradual breach inundated area are over 20 meters and
occur in the channel. The floodplain depths range from 0.5 meters to less than 6 meters, with
a few scattered pockets of slightly higher depths. The peak unit flow rates in the floodplain
are less than 10 m
2
/s and in the channel, the peak unit flow rate is over 120 m
2
/s. The flood
wave arrives at the floodplain about 45 minutes after the initial breach and reaches the
downstream boundary after a little over 140 minutes. The flood wave was completely in the
lake after 3.3 hours. Peak flood depth, peak unit flow rate and the flood arrival times are
shown in figures 9 through figure 11, respectively.
Table 5: Maximum Discharge for Gradual Breach Event


Location
Time of Maximum
Discharge
(seconds)
Maximum
Discharge
(cms)
Total
Volume
(km
3
)


Error
At the Dam 2000 17,357 38.93 -
X-Section 1 1690 16,717 38.54 0.99
X-Section 2 2246 16,839 39.03 1.0
X-Section 3 1029 16,447 40.19 1.03
X-Section 4 1174 20,527 40.82 1.05
X-Section 5 1409 18,185 42.21 1.08
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
303

Figure 8 Maximum Discharges for Gradual Breach Event.



Figure 9 Peak Flood Depth in meters for Gradual Breach Event.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
304
Figure 10 Peak Unit Flow Rate in square (m
2
/s) for Gradual Breach Event.

Figure 11 Flood Wave Arrival Times for Gradual Breach Event.
The total flooded area, excluding the lake, was slightly over 20 km
2
. The total population at
risk from this inundation is over 9,000 people. The highest population at risk occurs within
the first 75 minutes of the breach and affects over 15,500 people. As mentioned previously,
as the flood wave travels downstream, the single snap shots in time may over lap resulting in
the duplication of the number of population at risk. This is the case for the gradual breach
which results in the total population at risk at the end of 3 hours being double the total
population at risk from the peak flooded area. Again, the more reasonable assessment of the
total population at risk should be based on the peak depth flooded area. For the 12-year and
under age demographic, the total population at risk after 3 hours is over 3,500 and half that,
about 1,600 for the maximum flooded area. The 65-year and over age demographic has over
3,000 at risk after 3 hours and around 1,700 for the maximum flooded area. The maximum
inundated area affected by depths greater than 8 meters is 1.99 km
2
. The total population
affected by these depths is less than 300, with the 12-year and under population at risk
slightly over 50 and under 30 for the 65-year and over demographics.
Table 6: Total Population at Risk for Gradual Breach Event
Time Intervals
(minutes)
Total Population
at Risk
12-yr and Under
Population at Risk
65-yr and Over
Population at Risk
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
305
15 217 32 29
30 2,871 738 208
45 2,966 760 223
60 4,795 882 890
75 4,812 728 1,024
90 1,664 233 395
105 800 114 199
120 182 18 62
135 82 6 28
150 32 3 8
165 30 3 8
180 38 4 10
Total 18,489 3,523 3,084

Table 7: Flooded Area Population at Risk for Gradual Breach Event
Peak Flood
Depth Range
(meters)
Flooded Area
excluding lake
(km
2
)

Total Population
at Risk
12-yr and Under
Population at
Risk
65-yr and Over
Population at
Risk
>/= 1 20.02 9,226 1,600 1,699
>/= 2 10.00 4,107 892 541
>/= 3 5.85 2,768 699 220
>/= 4 3.99 2,121 531 166
>/= 5 3.05 1,302 321 101
>/= 6 2.49 326 142 48
>/= 7 2.21 401 86 34
>/= 8 1.99 272 52 27
Total 20.02 9,226 1,600 1,699



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
306
Figure 12 Total Populations at Risk for Gradual Breach Event
Conclusion
The results from the catastrophic breach event showed a total flooded area, excluding the
lake, of 41.1 km
2
which affects over 28,800 people. The total number of 12-year and under
affected by the maximum inundated area is over 4.600 and the total number of 65-year and
over is over 4,200. The flooded area with depths greater than 8 meters is 3.3 km
2
and affects
about 1,700 individuals, over 400 are 12-year and under and over 130 are 65-year and over.
The greatest number of the population at risk occurs within the first hour of the dam breach
resulting in over 26,000 being affected. Three hours after the initial dam breach, the majority
of the flood wave has reached the lake and no longer affects the city of Hydropolis.

The gradual breach scenario assumed breaching was caused by overtopping resulting in 80-
percent of the dam being removed. The time to reach the maximum breach and the maximum
width were determined using equations developed by Froehlich (1995a). The results showed
a total flooded area excluding the lake of over 20 km
2
which affects over 9,200 people. The
number of 12-year and under affected by the maximum flooded area is 1,600 and the number
of 65-year and over is 1,699. The flooded area with depths greater than 8 meters is 1.99 km
2

affects less than 300 individuals with over 50 being 12-year and under and less than 30 being
65-year and over. The majority of the population at risk occurs within the first 90 minutes of
the dam breach; however the total number is skewed because the inundated area for each time
interval overlaps.

These results represent the impacts of a dam breach, whether that breach is catastrophic or
gradual. Though loss of life was not calculated using the described model, the population at
risk was significant. Additionally, economic impacts were not considered, as this is not part of
the applied model. Even without these values available for comparison, there are many data
elements that will be useful in comparing the ADH model to other models presented at this
workshop. The use of the virtual test bed should prove to be a valuable tool, both now and in
the future, for comparison of such models.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
307
References
[1] Aquaveo (2009). Surface-water Modeling System Version 11.0, Aquaveo.
http://www.aquaveo.com/pdf/SMS_11.0.pdf.
[2] Arcement, G.J., Schneider, V.R. Guide for Selecting Mannings Roughness
Coefficients for Natural Channels and Flood Plains U.S. Geological Survey Water-
supply Paper 2339. http://www.fhwa.dot.gov/bridge/wsp2339.pdf
[3] Berger, R.C. and Lee, L.M. (2004) Multidimensional Numerical Modeling of Surges
Over Initially Dry Land. Coastal and Hydraulics Engineering Technical report,
ERDC/CHL TR-04-10. Vicksburg, MS: U.S. Army Engineering Research and
Development Center.
[4] Berger, R.C., Tate, J.N., Brown, G.L., and Savant, G. (2010). Adaptive Hydraulics:
User Manual. Vicksburg, MS: U.S. Army Engineering Research and Development
Center. https://adh.usace.army.mil/.
[5] Engineer Research and Development Center (ERDC) DoD Supercomputing Resource
Center (DSRC), http://www.erdc.hpc.mil/docs/diamondUserGuide.html, accessed
December 5, 2011.
[6] Martin, S.K., Savant, G., and McVan, D.C. (2010). Lake Borgne Surge Barrier Study.
Coastal and Hydraulics Engineering Technical Report, ERDC/CHL TR-10-10.
Vicksburg, MS: U.S. Army Engineering Research and Development Center.
[7] Martin, S.K., Savant, G. and McVan, D. (2011) Two Dimensional Numerical Model of
the Gulf Intracoastal Waterway Near New Orleans: Case Study, Journal of Waterway,
Port, Coastal and Ocean Engineering, doi:10.1061/(ASCE)WW.1943-5460.0000119.
[8] McAlpin, T.O., Berger, R.C., and Henville, A.M. (2009). Bush Canal Floodgate Study.
Coastal and Hydraulics Engineering Technical Report, ERDC/CHL TR-09-09.
Vicksburg, MS: U.S. Army Engineering Research and Development Center.
[9] Oregon Department of Transportation (2005). Hydraulics Manual (Part 1); Chapter 8,
Appendix A. Engineering and Asset Management Unit Geo-Environmental Section.
ftp://ftp.odot.state.or.us/techserv/Geo-
Environmental/Hydraulics/Hydraulics%20Manual/Chapter_08/Chapter_08_Appendix_A/
Chapter_08_Appendix_A.pdf
[10] Savant, G., Berger, R.C., McAlpin, T.O., and Tate, J.N. (2011). An Efficient Implicit
Finite Element Hydrodynamic Model for Dam and Levee Breach. Journal of Hydraulic
Engineering, ASCE. Vol. 137, No. 9, pp: 1005-1018.
[11] Savant, G and Berger, R.C. (2011) Wetting and Drying Characteristics of Adaptive
Hydraulics. USACE, ERDC Technical Note, USACE/ERDC-11-DRAFT, In Progress.
[12] Stockstill, R.L. and Vaughan, J.M. (2009). Numerical Model Study of the Tuscarawas
River below Dover Dam, Ohio. Coastal and Hydraulics Engineering Technical Report,
ERDC/CHL TR-09-17. Vicksburg, MS: U.S. Army Engineering Research and
Development Center.
[13] Stockstill, R.L., Vaughan, J.M., and Martin, S.K. (2010). Numerical Model of the
Hoosic River Flood-Control Channel, Adams, MA. Coastal and Hydraulics Engineering
Technical Report, ERDC/CHL TR-10-01. Vicksburg, MS: U.S. Army Engineering
Research and Development Center.
[14] Tate, J.N., Lackey, T.C., and McAlpin, T.O. (2010). Seabrook Fish Larval Transport
Study. Coastal and Hydraulics Engineering Technical Report, ERDC/CHL TR-10-12.
Vicksburg, MS: U.S. Army Engineering Research and Development Center.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
309
Assessment of ICOLD Benchmark Case Study using
Flood2D-GPU and HEC-FIA
B. A. Thames
1
and A. J. Kalyanapu
2

1
US Army Corps of Engineers, 801 Broadway, Nashville, TN 37203, USA
2
Department of Civil & Environmental Engineering, Tennessee Technological University,
1020 Stadium Drive, Prescott Hall 334 Box 5015, Cookeville, TN 38505, USA
E-mail: brantley.a.thames@usace.army.mil
Abstract
This paper presents a benchmark case study of a dam failure using Flood2D-Graphics
Processing Unit (GPU), a two-dimensional hydraulic modeling software, coupled with the
Hydrologic Engineering Centers Flood Impact Analysis (HEC-FIA) software, a flood
consequence model, to evaluate flood risk in a hypothetical test bed environment. The
defined objective of this study is to present the findings prescribed by the ICOLD Theme C
Formulation Team, which include inundation mapping, population at risk, loss of life, and
direct economic impacts. The increased computational capability afforded by Flood2D-GPU
allows for a much deeper analysis of the hypothetical scenario. Many studies have been
conducted to evaluate the uncertainty in the estimation of breach parameters (breach width
and time of failure); however, how breach parameter estimation uncertainty affects the
uncertainty in consequence estimation has been rarely evaluated. A secondary objective of
this study is to determine the affects of breach parameter estimation uncertainty on
consequence estimation uncertainty through a comparative analysis of four breach parameter
estimation (BPE) methods and a sensitivity analysis of the main breach parameters such as
breach depth, breach width, time of failure, and breach side slope. Through these analyses,
breach depth, typically assumed using engineering judgment, is found to be the most sensitive
breach parameter. Similar breach formation times are computed for three of the BPE
methods, Froehlich (1995a), Froehlich (2008), and Von Thun and Gillette, evaluated while
varying breach widths are computed; however, the consequence results are similar despite the
varied breach width computations highlighting the importance of breach formation time.
Introduction
With the increased awareness of the infrastructure deficiencies within the United States and
world-wide, dam safety and understanding the impacts of dam failures has become much
more emphasized due to the large number of dams world-wide with population centers within
the destructive path of dam failure discharges. Studies such as these strive to better define the
hydraulics of a dam failure and the resulting flood impacts.

Many US federal agencies, such as the US Army Corps of Engineers (USACE), the
Department of Homeland Security (DHS), and the Bureau of Reclamation (USBR), are
working diligently through research and practical analysis to better understand the mechanics
of a dam failure. This case study provides insight into hydraulic and consequence modeling
tools available to simulate a dam failure and evaluate possible impacts from flooding.



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
310
Methodology
Various modeling software are used to obtain the risk assessment results prescribed by the
ICOLD formulation team for Theme C, which consist of an array of hydraulic and
consequence-based products. To evaluate the proposed problem, a dam breach discharge
hydrograph is first developed using the unsteady flow component of the Hydrologic
Engineering Centers River Analysis System (HEC-RAS) version 4.1. The dam breach
discharge hydrograph is a necessary input into Flood2D-GPU, which leverages the
computers graphics card for much improved computational capabilities of the model input
[8]. Output from Flood2D-GPU, which includes inundated areas, depth grids, and flood wave
arrival time grids, serve as inputs into HEC-FIA, which provides the consequence assessment
for the solution to the problem statement. HEC-FIA is a single event GIS-based software that
determines impacts from flooding, such as population at risk (PAR), economic damages, and
loss of life (LOL) on a structure-by-structure basis [9].

Ultimately, a dam failure within the proposed hypothetical test bed is evaluated using this
combination of modeling software through a comparative analysis of the four suggested BPE
techniques and a sensitivity analysis of four breach parameters to evaluate the range of
possible PAR, LOL, and economic damage outcomes and the sensitivity of these outcomes to
breach parameter estimations.
Development of Dam Failure Hydrograph (HEC-RAS)
HEC-RAS is utilized to generate dam failure discharge hydrographs for input into Flood2D-
GPU. HEC-RAS is one-dimensional hydraulic model capable of performing unsteady flow
dam failure simulations utilizing user-provided hydraulic inputs (cross-section and inline
structure geometry, boundary conditions, Mannings n-values, etc.) and breach parameters
(breach opening dimensions and location, breach weir coefficients, breach formation time,
etc.) [13].

The HEC-RAS model developed for this evaluation uses the information provided by the
formulation team including a digital elevation model (DEM), a land use layer, and
information about the dam embankment and failure scenario. Because the only purpose of the
HEC-RAS model is the generation of a dam failure discharge hydrograph, the extent of the
model is limited to the reach just adjacent to the proposed dam.

The HEC-RAS model geometry is developed by processing the DEM using HEC-GeoRAS, a
data pre- and post-processing software developed for HEC-RAS. The HEC-RAS model
consists of a storage area to represent the reservoir above the dam, an inline structure
representing the dam and the necessary bounding cross sections upstream and downstream of
the dam. The stage-volume curve provided by the formulation team is utilized in the HEC-
RAS model to represent the storage characteristics of the reservoir. The cross section
geometry is developed directly from the DEM and imported into HEC-RAS using HEC-
GeoRAS. The inline structure geometry is developed using a combination of information
provided by the formulation team in the problem statement and the DEM.

The most critical input into the HEC-RAS model is the determination of the breach geometry
and parameters that include the final breach bottom width (W
b
), final breach invert elevation
(El
b
), breach side slopes (SS), and breach formation time (t
f
). [16] conducted an uncertainty
analysis of embankment dam breach parameters for all of the proposed BPE approaches
provided by the formulation team with the exception of Froehlich (2008), which did not exist
at the time of publication. This study focuses on four BPE methods: Froehlich 1995a [3],
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
311
Froehlich 2008 [5], the MacDonald and Langridge-Monopolis (MLM) [10], and the Von
Thun and Gillette (VTG) [15] methods, and the calculated or assumed values are presented in
Table 1. [16] shows through an uncertainty analysis, that in predicting breach width, breach
failure time, and peak flow through the breach, the Froehlich (1995a) technique produced the
lowest mean prediction error and the narrowest width of uncertainty band for all three
estimations. The only exception to this is the width of the uncertainty band for the breach
width estimation which is only slightly larger than the VTG technique. Based on these
results, the Froehlich (1995a) BPE technique emerges as the best compromise; however, the
current study compares the four methodologies across the range of sensitivity for all methods
to highlight the range of possible outcomes in terms of hydraulic and consequence results. In
addition, this evaluation provides a better understanding of how breach parameter sensitivity
affects consequence estimation uncertainty through a sensitivity analysis of four breach
parameters, SS, W
b
, El
b
, and t
f
, for each BPE method. The sensitivity analysis approach
involves computing these four breach parameters using each BPE method (Table 1) and
adjusting these values to create a sensitivity range. SS, W
b
, and t
f
are increased and
decreased by 25 and 50%. Because the initial assumption is that the breach height formation
will extend to the toe of the dam for all BPE methods, the sensitivity range for El
b
is
developed by increasing the base assumption of 211m by 7.625m or one eighth of the total
breach depth (h
b
) until the value reaches half of the total breach depth.

Using the computed values from Table 1, a base condition breach failure hydrograph is
developed for each BPE method using HEC-RAS serving as the basis for the comparative
analysis. In addition, breach hydrographs are developed for the range of sensitivity values.
Sixty eight simulations are computed to produce breach hydrographs for input into Flood2D-
GPU. In interest of space, the resulting breach hydrographs for the four base condition
simulations are presented in Figure 1.
Table 1: Computed Breach Parameter Estimates using the Four BPE Methods
Parameter Froehlich (1995a) Froehlich (2008) MLM (1984) VTG (1990)
SS 1.4 1.0 0.5 1.0
W
b
62 m 110 m 12 m 207 m
h
b
61 m 61 m 61 m 61 m
t
f
0.7 hrs 0.6 hrs 2.0 hrs 0.9 hrs
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
312

Figure 1: Computed Discharge Breach Failure Hydrographs from unsteady flow HEC-RAS
Model for Base Simulations
Two-Dimensional Hydraulic Modeling (Flood2D-GPU)
In this study, a GPU flood model, named Flood2D-GPU, developed in NVIDIA's CUDA
programming environment is used [8]. The modelling framework uses the 2D unsteady
numerical flood model that solves the non-linear hyperbolic shallow water equations using a
first-order accurate upwind difference scheme. These equations are developed from the
Navier-Stokes equations by integrating the horizontal momentum and continuity equations
over depth often referred to as the depth averaged or depth integrated shallow water equations
(i.e., Saint Venant equations). The non-conservative form of the partial differential equations
is [7;8;12]:







where, h is the water depth, H is the water surface elevation, u is the velocity in the x-
direction, v is the velocity in the y-direction, t is the time, g is the acceleration due to gravity,
S
fx
is the friction slope in the x-direction, and S
fy
is the friction slope in the y-direction. The
upwind finite difference numerical scheme is used to discretize governing equations (1-3), as
it yields non-oscillatory solutions, through numerical diffusion [2;11]. A staggered grid
computational stencil is used to define the computational domain with the water depth (h) in
the centre of the cell and u and v velocities on the cell edges. The model requires a digital
elevation model (DEM) to represent topography, Mannings n for surface roughness
representation and a flow hydrograph. The CPU environment in this module consists of the
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
50000
1 1.5 2 2.5 3 3.5
D
i
s
c
h
a
r
g
e

(
C
M
S
)

Time (hrs)
Froehlich 1995a
Froehlich 2008
MLM
VTG
(1) Equation Continuity 0 =
c
c
+
c
c
+
c
c
y
vh
x
uh
t
h
(1) Equation Continuity 0 =
c
c
+
c
c
+
c
c
y
vh
x
uh
t
h
(3) direction - y in Equation Momentum 0
(2) direction - in x Equation Momentum 0
= +
c
c
+
c
c
+
c
c
+
c
c
= +
c
c
+
c
c
+
c
c
+
c
c
fy
fx
gS
y
H
g
y
v
v
x
v
u
t
v
gS
x
H
g
y
u
v
x
u
u
t
u
(3) direction - y in Equation Momentum 0
(2) direction - in x Equation Momentum 0
= +
c
c
+
c
c
+
c
c
+
c
c
= +
c
c
+
c
c
+
c
c
+
c
c
fy
fx
gS
y
H
g
y
v
v
x
v
u
t
v
gS
x
H
g
y
u
v
x
u
u
t
u
(3) direction - y in Equation Momentum 0
(2) direction - in x Equation Momentum 0
= +
c
c
+
c
c
+
c
c
+
c
c
= +
c
c
+
c
c
+
c
c
+
c
c
fy
fx
gS
y
H
g
y
v
v
x
v
u
t
v
gS
x
H
g
y
u
v
x
u
u
t
u
(3) direction - y in Equation Momentum 0
(2) direction - in x Equation Momentum 0
= +
c
c
+
c
c
+
c
c
+
c
c
= +
c
c
+
c
c
+
c
c
+
c
c
fy
fx
gS
y
H
g
y
v
v
x
v
u
t
v
gS
x
H
g
y
u
v
x
u
u
t
u
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
313
random number generator, and storage of input data and parameters and source code to
calculate future time step using Courant condition, and implement hydrograph update. The
GPU environment contains the computational engine that solves continuity and momentum
equations, and boundary condition implementation. A 2D model is chosen because of its
better representation of flood flow (especially in floodplains) simultaneous flood extent
delineation and instantaneous flood velocities at all nodes in the computational domain.
Flood2D-GPU is validated for accuracy and found to provide significantly reduced
computational time (up to two orders of magnitude) compared to the same flood model
implemented serially in a CPU-based environment [8].

Application of Flood2D-GPU in this study uses HEC-RAS dam breach hydrographs,
Mannings n map generated from the land use/cover data provided along with the digital
elevation model (with dam burnt into the topography). The methodology of dam breach
hydrograph is explained in previous section. To develop spatially variable Mannings n
surface map, the National Land Cover Dataset (NLCD) Land cover (LC) codes were used to
estimate approximate Mannings n values based on recommendations from [6]. Table 2
provides the LC code and its assigned Mannings n value and Figure 2 depicts the resultant
spatially variable Mannings n map.
Table 2: Estimated Mannings values for National Land Cover Dataset
NLCD LULC Class Name LULC Code Manning's n
Open Water 11 0.025
Developed, Open Space 21 0.016
Developed, Low Intensity 22 0.030
Developed, Medium Intensity 23 0.030
Developed, High Intensity 24 0.050
Barren Land (Rock/Sand/Clay) 31 0.025
Deciduous Forest 41 0.120
Evergreen Forest 42 0.200
Shrub/Scrub 52 0.070
Grassland/Herbaceous 71 0.050
Pasture/Hay 81 0.030
Cultivated Crops 82 0.035
Woody Wetlands 90 0.160
Emergent Herbaceous Wetlands 95 0.110

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
314

Figure 2: Land Use Land Cover Derived Mannings Map for the case study

Consequence Assessment (HEC-FIA)
HEC-FIA is used to determine the consequences, such as PAR, economic damages, and LOL,
using the results of the Flood2D-GPU simulations. [9] presents that HEC-FIA is a stand-
alone, GIS enabled model for estimating flood impacts due to floodingthat generates
required economic and population datato compute urban and agricultural economic flood
damage, area inundated, number of structures inundated, population at risk, and loss of life.
Input requirements for HEC-FIA include a stream alignment, terrain grid, depth grid, arrival
time grid, a structure inventory, and timing information for the purpose of evaluating a single
event [9]. Based on the information provided by the formulation team, several assumptions
and derivations from the provided data are made to develop the required inputs into HEC-
FIA.

ArcHydro Tools, an extension of ArcGIS Desktop, is used to create a stream centerline by
processing the DEM. The depth and arrival time grids for all simulations are derived from the
results of the Flood2D-GPU simulation. The depth grid required for HEC-FIA is the max
depth grid; however, additional depth grids for the specified time intervals were derived for
the required PAR, LOL, and economic damage results. The arrival time grid is assumed to be
the time after failure at which the flood depth will reach two feet of depth.

For this study, the structure inventory must include a structure name, occupancy type, damage
category, structure value, and population information in order to derive the prescribed results.
HEC-FIA utilizes two separate databases, a structure occupancy and global inventory table, to
define the structure inventory against which all consequence estimations are made.

The structure occupancy table consists of general occupancy types, such as residential,
commercial, and industrial, and, for this study, is populated using occupancy types using the
database that comes with Federal Emergency Management Agencys (FEMA) HAZards
United States Multi-Hazard (HAZUS-MH) Tool [1]. The structure occupancy type table
includes the content-to-structure value ratio for defining content values for each structure,
links to depth-damage curves for structure and content values, and risk elevations for defining
at what elevation an individual is at risk of fatality based on age. HEC-FIA only considers
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
315
two age groups, those under and over the age of 65; however, to ascertain consequence results
for populations under the age of 14, a separate HEC-FIA model is created for each simulation.
The assumption is that anyone under the age of 14 would be accompanied by an adult and
either carried vertically to safety or is mobile enough to move vertically to safety.

The global inventory table consists of each structure, or parcel for this study, and specific
parameters related to each structure such as depth-damage curves for the structure and
contents, structure and content value, first floor elevations, and number of stories. This
information is either user-defined or derived from other sources such as the DEM or through a
relationship with the structure occupancy type table. For instance, the depth-damage curves
for the structure and contents are adopted from the structure occupancy type table by way of
the general occupancy type assigned to each structure by the user. The structure appraised
values, population values for the various age groups, structure name, and number of stories is
defined by the user in ArcGIS Desktop and imported into HEC-FIA. First floor elevations are
derived from the DEM and increased by an assumed three feet to account for foundation
height.

The most time intensive process is related to the development of the global inventory.
Spatially joining the parcel dataset layer with the census block layer in ArcGIS Desktop to
compute the number of parcels per census block provides the ability to uniformly distribute
the necessary population values for the age groups of interest, under 14, between 14 and 65,
and over 65 years of age, into each parcel. The uniformly distributed population values are
spatially joined back to the parcel dataset to attain an age group specific population value for
each parcel.

With no structure values and limited information regarding structure condition and square
footage provided by the formulation team, the structure values are assumed for each
occupancy type provided in the parcel dataset through collaboration with a US Army Corps of
Engineers (USACE) economist to achieve reasonable values for each structure occupancy
type (Table 3). Each structure is assigned a generic name based on the occupancy type. The
structure name, population, structure values, and HAZUS occupancy type are compiled within
ArcGIS Desktop and imported into HEC-FIAs global inventory.

Table 3: Estimated Structure Value Assumptions
Zone Description
Estimated
Value
($1000)
A-1 Rural Agriculture - 1-ac Minimum Zone 50
C-1 Neighborhood Commercial Zone 650
C-2 Community Commercial Zone 650
O-1 Office and Institutional Zone 1000
P Parking Zone 750
R-1 Single Family Residential Zone 125
R-2 Residential Zone: Houses, Townhomes and Medium Density Apartments 1000
R-3 Residential Zone: Houses, Townhomes and High Density Apartments 2500
RA-1 Residential and Agricultural Zone, Semi-Urban Area 75
R-D Residential and Related Uses Zone, Developing Area 100
R-LT Residential Zone: Houses and Limited Townhomes 750
R-T Residential Zone: Houses and Townhomes 500
SU-1 Special Use Zone 1500
SU-2 Special Neighborhood Zone, Redeveloping Area 125

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
316
A time window is needed for each HEC-FIA simulation, which defines a start, end, and
warning issuance time for each simulation. The start time is based on the initiation of the
breach. The assumption of no warning is made based on the scenario described of no cell
phone service and no emergency action plan. [9] assert that warning time plays a critical role
in preventing loss of life and property, and as such, is a very sensitive parameter in the HEC-
FIA software. The sensitivity of warning time also emphasizes the importance of a robust
warning system especially in situations of high risk dams just upstream of major population
centers.

PAR, LOL, and economic damages at specific time intervals are required results for this
study. Because of the manner in which HEC-FIA treats incremental changes in PAR, LOL,
and economic damages, depth grids, derived from Flood2D-GPU results, for each of these
time intervals are simulated individually through HEC-FIA to develop these consequence
results and post-processed to determine the PAR, LOL, and economic damages incrementally
for each time interval. Because this is a labor intensive task, this time interval analysis is only
conducted for the four base BPE method simulations.

With all of the necessary information pre-processed and imported into HEC-FIA, two
simulations, one for over and under the age of 65 and one for under the age of 14, are set up
and run for each BPE method simulation, each sensitivity range, and for each required time
interval totaling 184 HEC-FIA simulations. The results from these simulations are presented
in the Results and Discussion section, and electronic copies of the required models, modeling
outputs, GIS results, and spreadsheet results are provided supplemental to this document.
Results and Discussion
Two-Dimensional Hydraulic Modelling Results
In this report, the two-dimensional hydraulic modeling results using the base condition
Froehlich 1995a BPE method are presented results and used in comparisons. Also, a
sensitivity analysis of four breach parameters is evaluated and results are compiled and
discussed. Additional results from these analyses and from other BPE methods are provided
supplemental to this report.

The Froehlich 1995a base simulation resulted in flood depths ranging from 0.5 m to nearly 8
m in the downstream areas of the dam. Figure 3 presents the map of the standard deviation in
the peak flood depths from the multiple breach hydrographs which were derived from the
sensitivity analysis of the four breach parameters. A high variation in the peak flood depths is
noticed along the canyon immediately downstream of the dam. In the downstream region
away from the mountainous terrain, smaller standard deviation in simulated peak flood depths
is noticed that indicates less sensitivity due to the breach parameters. Thus it is likely that
variability in consequence estimates including direct flood damages, LOL, PAR in the
downstream region may be smaller compared to the upstream region of the floodplain. A
similar trend is also observed in the maximum flood velocities for these breach hydrographs.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
317

Figure 3: Standard deviation map of the peak flow depths from the dam breach hydrographs
developed for the four breach parameters using Froehlich 1995a BPE method

Figure 4 presents the simulated flood discharge hydrographs at the five required cross-section
locations. For the Froehlich 1995a BPE method, the variability of flood discharge from all the
sensitivity runs are represented by the error bars. It is observed that a large variability in the
magnitude of flood discharge is observed at the beginning of the dam breach. This is
attributed to the influence of breach parameter sets used for generating the hydrographs. This
is a clear indication that the consequence analysis, especially the LOL, PAR, direct economic
damages during the initial 60 minutes of the breach, will be significantly affected by the
choice of breach parameters used. With progression of time, the variability decreases at all
cross-section locations.


Figure 4: Simulated flood discharge plots at the five cross-section locations for the Froehlich
1995a BPE method. Error bars indicate the variability of discharges due to various breach
parameter sets.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
318
Consequence Assessment Results
For the purposes of presenting results and making comparisons, the consequence results using
the base condition Froehlich 1995a BPE method is considered the basis of the comparative
analysis. In addition to this comparative analysis, a sensitivity analysis of four breach
parameters are evaluated for each BPE method and results are compiled and discussed in this
report. Additional results from these analyses are provided supplemental to this report.

The Froehlich 1995a base simulation results in a PAR of 13976 persons with 2223 persons
(about 16% of total PAR) under the age of 14 and 2695 persons (about 19% of total PAR)
over the age of 65. HEC-FIA estimates a fatality total of 1914 persons under these simulation
conditions, which is almost 14% of the total PAR. In reality, the total LOL seems to be
inflated and is probably overestimated when considering the resilience and resourcefulness of
a typical population. However, the assumption of no warning time appears to be the largest
source of the elevated LOL totals, and a sensitivity of this assumption could provide a large
range of possible outcomes for LOL. For instance, a warning just 15 minutes in advance of
the breach reduces the total estimated LOL by 55% to 862 persons. A warning 1 hour in
advance of the breach results in an estimated LOL of 511 persons or a 73% reduction.

Direct economic damages are estimated using HEC-FIA using a depth of flooding derived
from the depth grid and a first floor elevation for each structure in combination with a depth-
damage curve. The total economic damages estimated by HEC-FIA are approximately $361
million dollars; however, this value is based strongly on the structure value assumptions made
based on occupancy type and the content-to-structure value ratio derived from the FEMA
HAZUS-MH Tool [1].

For the Froehlich 1995a base simulation, Table 4 shows incremental PAR, LOL, and direct
economic impacts for specified time intervals. The PAR does not become affected by
flooding until 30 minutes after the initiation of the dam breach based on flood wave
propagation estimated by Flood2D-GPU and the population provided by the census block
dataset. However, during the 30-60 minute time interval following the breach a large
percentage of the PAR becomes affected and the greatest LOL occurs during this period. The
greatest percentage of the PAR actually becomes affected in the 120-180 minute time interval
due to an area of low population in the center of the damage reach. Interestingly, the LOL
during the 120-180 minute time interval is not substantial even with the larger PAR due to
lower flood depths allowing individuals to retreat vertically to safety within each structure.
The decreased flood depths also explain why the direct economic damages are also greatest in
the 30-60 minute time interval as opposed to the 120-180 minute time interval.
Table 4: Incremental Time Interval Results for the Froehlich
1995a Base Simulation
Time
Interval
(min)
Total
PAR
14-yr
and
Under
PAR
65-yr
and
Over
PAR
LOL
Direct Economic
Impact ($US)
0-15 0 0 0 0 $0.00
15-30 0 0 0 0 $0.00
30-60 2915 755 210 1549 $128,971,979.93
60-90 512 122 31 329 $77,632,731.34
90-120 1410 168 347 5 $42,877,015.97
120-180 7638 956 1835 7 $85,161,297.09
> 180 1501 222 273 24 $26,714,802.91
Total 13976 2223 2695 1914 $361,357,827.24
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
319

The majority of the flooding by area occurs at the lower depths of flooding (0 to 1.5m) with a
majority of the PAR contained within these same lower depths of flooding. Table 5 illustrates
the distribution of PAR and flooded area across the various depth ranges.
Table 5: Incremental Depth Range Results of PAR
and Flooded Area
Peak
Flood
Depth
Range
(m)
Flooded
Area
(m)
Total
PAR
14-yr
and
Under
PAR
65-yr
and
Over
PAR
0 - 0.5 3498463 2786 447 464
0.5 - 1 6433454 4993 664 1079
1 - 1.5 5310192 2056 267 488
1.5 - 2 3706161 887 89 286
2 - 2.5 2874378 424 49 126
2.5 - 3 2379332 262 35 71
3 - 3.5 1600888 107 29 8
3.5 - 4 756803 68 20 4
4 - 4.5 427789 69 19 5
4.5 - 5 337724 115 32 8
5 - 5.5 372655 217 59 15
5.5 - 6 378761 249 71 17
6 - 6.5 252956 155 43 11
6.5 - 7 247030 282 78 22
7 - 7.5 200874 228 61 16
7.5 - 8 198091 245 61 12
>8 2305519 827 199 62
Total 31281068 13976 2223 2695

Supplemental results for the various simulations are provided electronically.

When initially choosing a BPE method, the question of which technique provides the most
accurate results is raised driving the need for a comparative and sensitivity analysis. The goal
of this evaluation is to understand how the uncertainty of these methods and the sensitivity of
the critical breach parameters affect the results of a consequence analysis. Table 6
summarizes the consequence results of both the comparative and sensitivity analysis.

For the comparative analysis, the results from the Froehlich 1995a, Froehlich 2008, and VTG
base simulations are extremely similar while the results of the MLM base simulation are
much lower than the other three simulations, which is expected based on the dam failure
hydrographs computed using HEC-RAS.

The mean results for all parameters is typically only slightly less than the base simulation
results. In addition, the base simulation results are typically much closer to the higher end of
the range of values from the sensitivity analysis. A closer look at the results of the breach
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
320
invert sensitivity analysis shows that consequence results are very sensitive to the breach
invert assumption. These breach invert simulations are skewing the range of consequence
results on the lower end of the range while also lowering the mean of the sensitivity
simulations. The large values for standard deviation for all consequence results can also be
attributed to the breach invert simulation runs. For the Froehlich 1995a sensitivity analysis,
removing the breach invert sensitivity simulations from the statistical calculations produces a
much tighter range for all consequence results with a 66% reduction in total PAR standard
deviation, a 68% reduction in LOL standard deviation and an 80% reduction in direct
economic impacts standard deviation, which indicates that these reductions in standard
deviation are also due to the sensitivity of the breach invert parameter.

In terms of sensitivity, breach invert is, above all, the most sensitive breach parameter;
however, when comparing the breach hydrograph and consequence results of the various BPE
methods, time of formation also drives the shape and peak of the breach hydrograph, which is
directly related to flood consequence results. In fact, the difference in breach hydrograph
peak and shape between the MLM and other three BPE methods, which is quite significant, is
primarily due to the longer time of formation computed for MLM.

In interest of space, many of the results discussed are not presented in this document;
however, the results are provided supplemental to this report electronically.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
321



T
a
b
l
e

6
:

H
E
C
-
F
I
A

B
a
s
e

S
i
m
u
l
a
t
i
o
n

R
e
s
u
l
t
s

f
o
r

C
o
m
p
a
r
i
s
o
n

a
n
d

S
e
n
s
i
t
i
v
i
t
y

S
i
m
u
l
a
t
i
o
n

S
t
a
t
i
s
t
i
c
s


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
322
Conclusions
This paper presents a solution to the problem statement using Flood2D-GPU coupled with
HEC-FIA offering possible hydraulic and consequence outcomes under the hypothetical
scenario provided. The comparative analysis of various BPE methods provides recognition of
the similarities and differences between the methods. The sensitivity analysis of the breach
parameters demonstrates the importance of specific parameters by illustrating the effects of
these parameters on not only the hydraulic outputs but also the consequence results. These
analyses combined begin the conversation into utilizing the most appropriate BPE method and
focusing in on which breach parameters are the most critical to approximate the most accurate
hydraulic and consequence results in a study of this nature.

The results of the comparative analysis suggests that similar consequence results can be
expected when using the Froehlich 1995a, Froehlich 2008, and VTG breach parameter
estimation methods but, without additional case studies using this same approach, relative
replication of these results is not substantiated. However, all of these approaches are based on
straightforward regression relationships and should produce similar results. The MLM BPE
method is similar in many aspects to the other three BPE methods with the exception of the
inclusion of a volume of material eroded term. Additional case studies could provide insight
into the differences of results experienced during this study.

In general, the results of the sensitivity analysis follow the expected trends. As breach depth,
width, and side slopes increase and the time of breach formation decreases, the breach
discharge hydrograph shape becomes more compressed resulting in higher peak discharges,
which ultimately results in more significant flood impacts and consequences. Several
conclusions can be drawn from these analyses.

Final breach invert is very difficult to estimate and is an assumed value not calculated by any
of the BPE methods. Unfortunately, the results of this analysis show that the breach
hydrograph and consequence results are extremely sensitive to the breach invert assumption.
When conducting a risk-based assessment of a dam failure, special care should be taken to
ensure that the best estimation of breach invert is assumed based on the sensitivity of breach
hydrograph and consequence outcomes to the estimation of this parameter. Ultimately,
additional research could provide a better approach to estimating this parameter drastically
reducing the uncertainty in these types of analyses.

When comparing the breach parameter estimations of SS, El
B
, W
b
, and t
f
, using the three BPE
methods that produced similar consequence results, with the results of the sensitivity
simulations, the breach formation time proves to be a critical parameter in the development of
breach discharge hydrographs. Even with a W
b
varying between 49 and 146 m, the resulting
breach hydrograph shape and peaks are similar due to the similarity in formation time, which
ranges from 0.6 to 0.9 hours.
References (arranged in alphabetical order)
[1] Department of Homeland Security, Federal Emergency Management Agency (2007),
Multi-Hazard Loss Estimation Methodology Flood Model HAZUS-MH MR3 Technical
Manual, Washington, D. C.
[2] Ferziger, J.H., and Peric, M. (2002). Computational methods for fluid dynamics, 3rd
edition Springer, New York.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
323
[3] Froehlich, D. C. (1995a). Embankment dam breach parameters revisited. Water
Resources Engineering, Proceedings of the 1995 American Society of Civil Engineers
Conference on Water Resources Engineering, New York, 887891.
[4] Froehlich, D. C. (1995b). Peak outflow from breached embankment dam. Journal of
Water Resources Planning and Management, American Society of Civil Engineers,
121(1), 9097.
[5] Froehlich, D. (2008). Embankment Dam Breach Parameters and Their Uncertainties.
Journal of Hydraulic Engineering, 134(12), 17081721.
[6] Hossain, AKM Azad, Yafei Jia, and Xiabo Chao. (2009). "Estimation of Manning's
roughness coefficient distribution for hydrodynamic model using remotely sensed land
cover features." Geoinformatics, 2009 17th International Conference.
[7] Judi, D. R., Burian, S. J., and McPherson, T. N. (2010). Two-dimensional fast-response
flood modeling: Desktop parallel computing and domain tracking. Journal of Computing
in Civil Engineering, in press.
[8] Kalyanapu A. J., Shankar S., Stephens A., Judi D. R., Burian S. (2011). Assessment of
GPU computational enhancement to a 2D flood model Journal of Environmental
Modeling and Software, 26, 1009-1016.
[9] Lehman, W. and, Needham, J. (2009). Consequence estimation for dam failures.
Hydrologic Engineering Center, US Army Corps of Engineers.
[10] MacDonald, T. C., and Langridge-Monopolis, J. (1984). Breaching characteristics of
dam failures. Journal of Hydraulic Engineering, 110(5), 567586.
[11] Patankar, S.V. (1980). Numerical heat transfer and fluid flow, Taylor and Francis.
[12] Tingsanchali, T. and Rattanapitikon, W. (1999). 2-D modeling of dambreak wave
propagation on initially dry bed. International Journal of Science and Technology, 4(3),
28-37.
[13] United States Army Corps of Engineers. (2010). HEC-RAS River Analysis System -
Users Manual, Hydrologic Engineering Center, US Army Corps of Engineers.
[14] United States Army Corps of Engineers. (2012). HEC-FIA Flood Impact Analysis - Users
Manual, Hydrologic Engineering Center, US Army Corps of Engineers.
[15] Von Thun, J. L., and Gillette, D. R. (1990). Guidance on breach parameters. Internal
Memorandum, U.S. Dept. of the Interior, Bureau of Reclamation, Denver, 17.
[16] Wahl, T. L. (2004). Uncertainty of Predictions of Embankment Dam Breach
Parameters. Journal of Hydraulic Engineering, 130(5), 389-397.




ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
325
Dam Failure and Consequence Assessment with
Standardized USACE MMC Procedures
International Commission on Large Dams 12
th
International
Benchmark Workshop on Numerical Analysis of Dams (Theme C:
Computational Challenges in Consequence Estimation for Risk
Assessment)
D. Williams
1
and K. Buchanan
2

1
U.S. Army Corps of Engineers, Tulsa, Oklahoma, USA
2
U.S. Army Corps of Engineers, Huntington, West Virginia, USA
Email: david.j.williams@usace.army.mil
Abstract
The U.S. Army Corps of Engineers (USACE) Modeling, Mapping and Consequences
(MMC) Production Center has submitted an entry for the International Commission on Large
Dams (ICOLD) 12
th
International Benchmark Workshop on Numerical Analysis of Dams
(Theme C: Computational Challenges in Consequence Estimation for Risk Assessment). As
part of this exercise, standard MMC processes were used to develop hydraulic modeling of a
dam break inundation as well as the associated consequences. Modeling techniques included
one- and two- dimensional numerical models, use of geographic information systems, and
economic and life loss estimation. Breach parameters were developed from standard
regression equations. Based on the rules and regulations that USACE has established for the
estimation of risk, economic consequences and life loss modeling results from this breach
analysis corresponded to a high hazard dam. This analysis demonstrates the standardized
MMC process for consistently evaluating all USACE dams while providing the flexibility to
analyze projects that range from small and simple to large and extremely complex.
Introduction
The Modeling, Mapping and Consequences (MMC) Production Center supports the U.S.
Army Corps of Engineers (USACE) Institute for Water Resources (IWR) Risk Management
Center (RMC) and the HQ USACE Office of Homeland Security, Critical Infrastructure
Protection and Resilience (CIPR) program by analyzing the potential consequences of dam
and levee infrastructure failures. The mission of the MMC includes production of hydraulic
models, consequences estimates, and inundation maps for USACE dams and levees in support
of these programs. All products are developed in accordance with MMC standard operating
procedures. The intent of MMC products is to support a risk-based assessment, prioritization,
and management framework for USACE CIPR, Dam Safety, and Levee Safety programs.
Analysis of the risk associated with dam and levee failure events allows decision makers to
prioritize investments to reduce the risk to life and property.

MMC dam safety analysis is typically based on five failure and five non-failure dam breach
scenarios needed for CIPR and Dam Safety program evaluations. Breach parameters for
earthen dams (e.g. average breach width, breach formation time, and side slopes) are
calculated using the following empirical equations: MacDonald and Langridge Monopolis
(1984) [1], Von Thun and Gillette (1990) [2], Froehlich (1995) [3], and Froehlich (2008) [4].
Results from all of these methods are compared, and results from a single empirical equation
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
326
is selected for the analysis. Parameters from multiple equations are not considered to be
interchangeable.

Hydraulic modeling is performed using either HEC-RAS [5] (one-dimensional) or FLO-2D
[6] (two-dimensional) unsteady flow analysis. The use of FLO-2D in the MMC process is
generally restricted to areas where the one dimensional flow assumption utilized in HEC-RAS
is not appropriate; such as coastal plains, alluvial fans, flat dense urban areas, and other large
flat unconfined floodplains. Both models provide unsteady flow simulation. Steady flow
models are not permitted within the MMC process.

Following the development of the hydraulic model, the output from the dam failure analysis is
processed using ArcGIS [7]. Inundation depth grids and inundation boundary shapefiles are
generated, and these files are used in conjunction with structure and population data from tax
parcels, point shapefiles, or the Federal Emergency Management Agency (FEMA) HAZUS
[8] dataset to develop a HEC-FIA (Flood Impact Analysis) [9] model. The structure
information is compared with the extent and timing of the flooding to estimate damages and
life loss. This information (in the form of a consequence assessment report), along with an
atlas of the downstream inundation, is provided by the MMC to USACE district offices for
emergency action planning.
Project Description
The MMC is participating in the International Commission on Large Dams (ICOLD) 12
th

International Benchmark Workshop on Numerical Analysis of Dams (Theme C:
Computational Challenges in Consequence Estimation for Risk Assessment). Participants in
Theme C are free to select the type and sophistication of modeling used to develop flood
inundation areas and consequences estimates. When possible, the standard MMC process
[10] was followed in the development of the dam break modeling and consequence estimation
for this project.
The numerical problem proposed for the workshop consists of estimating the consequences of
failure of a dam in a mountain valley above the fictitious city of Hydropolis. The hypothetical
dam, which was constructed for flood control, is 3.5 kilometers upstream from the city.
Sandy clays and clayey sands are the predominant material within the rolled earth fill dam.
With a 61-meter vertical distance from toe to crest, the structure has a high head differential
when the reservoir is completely filled. The corresponding volume of the reservoir at the dam
crest elevation is 38,276,000 cubic meters [11] (Figure 1).


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
327

Figure 1: Storage-elevation curve for the reservoir.
This exercise examines the downstream consequences of a catastrophic failure of the dam
resulting from overtopping. Expected outcomes from this study include gridded data (peak
flood depth, flood wave arrival time, peak unit flow rate, population at risk, loss of life, flood
severity, direct economic impact), tabular data (flooded area, population at risk, loss of life,
direct economic impact), and hydrographs (dam failure discharge, cross-sectional
hydrographs). All of these products except for the peak unit flow rate are standard
deliverables in a MMC dam breach analysis.
Model Development
Modeling for this exercise was developed using both one- and two-dimensional techniques
that are consistent with MMC standard operating procedures. Initially, only a HEC-RAS
hydraulic model was planned for this study. This model computes flood wave parameters at
cross-sectional nodes along a one-dimensional channel. Inspection of the terrain grid
provided for this study revealed that the use of HEC-RAS for the entire downstream reach
would be inappropriate since Hydropolis is located within a relatively flat plain below the
base of the mountain range. In this reach, it was determined that a two-dimensional model
would be more appropriate; therefore, FLO-2D was also used.

Existing terrain downstream from the dam was provided by the Theme C formulation team.
This dataset was provided in a gridded format with a cell width of 9.476 meters. No
projection was assigned to the terrain dataset, but all geospatial data provided by the
formulation team was in the same unknown coordinate system.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
328
The terrain dataset was used to extract cross sections in ArcGIS 10 (HEC-GeoRAS [12]) for
use in a HEC-RAS model. The spatial extent of this model was restricted to 200 meters
below the dam (modeled as an inline structure) as it was used to generate a dam breach
hydrograph for a FLO-2D inflow boundary condition (Figure 2). The HEC-RAS model was
run as an unsteady flow simulation with a computation interval of 5 seconds. A breach
hydrograph was written to file with a 1-minute output interval, which was sufficient for
defining the dam breach hydrograph.

Figure 2: HEC-RAS model geometry for breach hydrograph determination.
Parameters from the MacDonald and Langridge Monopolis empirical breach equation were
not specifically considered in this analysis since it is considered to be an envelope equation,
although parameters from the other equations were checked to ensure that they fell within the
envelope range. The Froehlich (1995), Froehlich (2008), and Von Thun and Gillette
equations were carried forward as unique scenarios in the HEC-RAS model. Since each of
the three methods estimated a short breach formation time, a computational time step of 5
seconds was used in the HEC-RAS model in order to capture the peak discharge.
Following the initial model development with the Froehlich (1995), Froehlich (2008), and
Von Thun and Gillette breach parameters, the Froehlich (1995) equation was carried forward
as the preferred method. Froehlich (1995) analyzed 63 earthen, zoned earthen, earthen with a
core wall, and rockfill dam failures [3]. Data from these events was used to develop
regression equations for average breach width

(meters) and breach formation time t


f
(hours):



(1)



(2)
For this exercise, K
0
= 1.4 (overtopping failure). The reservoir volume V
w
(38,276,344 m
3
)
and breach height h
b
(61 meters) were provided by the organizers of the benchmark
workshop. This method provided a short breach development time coupled with a relatively
wide breach width (Table 1):
Table 1: Breach parameters for hypothetical dam
Breach Equation
Formation Time
(hours)
Bottom Width
(m)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
329
MacDonald and Langridge-Monopolis 1.8 5
Froehlich (1995) 0.7 62
Froehlich (2008) 0.6 39
Von Thun and Gillette 1.5 177

A breach hydrograph generated in HEC-RAS was then used as the inflow boundary condition
in a FLO-2D model (Figure 3). The model assumed a static pool elevation of 272 meters with
no inflow. Overtopping of the dam crest was the prescribed mode of failure. Unlike the one-
dimensional flow model (HEC-RAS), FLO-2D computes flow across a network of grid cells
with eight possible directions of flow into or out of each cell. This type of flow distribution
makes FLO-2D well-suited for analysis of the flood wave downstream from the dam.


Figure 3: Breach hydrograph resulting from Froehlich (1995) parameters
Faster computation time of the FLO-2D model required a grid-cell width larger than 9.476
meters. By choosing an integer multiple of the terrain grid-cell size, conversion between the
two could be simplified when resampling. Therefore, a computational grid-cell dimension of
94.76 meters (square) was selected. Elevation values were interpolated for the larger
computational grid dimensions when imported into FLO-2D.
Surface roughness values were developed by converting a National Land Cover Database
(NLCD) shapefile [13] to corresponding Mannings n values [14]. The NLCD shapefile was
provided by the Theme C formulation team. Once a new surface roughness shapefile was
created, the values were interpolated for the computational grid dimensions when imported
into FLO-2D.
The FLO-2D model was run with a simulation time of 12 hours and an output interval of 5
minutes. Since the downstream reach is both short and steep, the simulation time was more
than sufficient for routing of the breach hydrograph through the entire study area. An output
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
330
interval of 5 minutes was necessary to capture the peak of the flood wave at downstream
locations.
A depth grid, flood wave arrival time grid, and flood severity grid were generated by using
MAPPER [6], a post-processing tool for FLO-2D models. Peak unit flow rate is not a FLO-
2D output. This variable was therefore determined by taking the maximum flow rate for each
grid cell and dividing by the computational grid-cell width. Cross-sectional hydrographs,
which were required at five locations downstream from the dam, were developed by setting
those locations as outflow nodes in sequential FLO-2D model runs. The time-series discharge
for each cross-sectional outflow grid cell was then summed to construct a representative
hydrograph for each key location. This was necessary because FLO-2D does not generate
time-series discharge for grid cells within the computational domain if no channel element is
used; it does write this output to file for outflow nodes, however. Hydraulic modeling results
were then used for the determination of consequences based on a structure inventory.
A structure inventory was developed using three datasets provided for the workshop [15].
These included a parcel shapefile with zoning information, a census block shapefile with
population and employment information, and a land use grid. Typically, the economic and
life loss analysis in a MMC study includes the utilization of a FEMA HAZUS dataset
supplemented by imagery, land use, or parcels. Since this was not provided for the
benchmark exercise, non-standard methods were used to estimate consequences. The
following assumptions were required for the creation of a structure inventory with population
values in HEC-FIA:

- Since the dam fails at 11:00pm on a Saturday, all residential populations are in their
homes.
- Due to the day and time, only 10% of the workers in certain job types are working.
These included the following jobs: Mining, Quarrying, and Oil and Gas Extraction;
Utilities; Manufacturing; Transportation and Warehousing; Administrative and
Support and Waste Management; Health Care and Social Assistance; Arts,
Entertainment, and Recreation; and Accommodation and Food Services.
- All workers come from census blocks outside the inundation area (not enough data to
model movements from one census block to another).
- All workers are under the age of 65.
- All data was up to date and accurate (parcels, blocks, and land use).

Once the structure inventory and hydraulic data for the dam breach had been developed, they
were used to develop the HEC-FIA model. Depth grids and arrival time grids were developed
from FLO-2D output and imported into the model. The arrival time grids represented the
time at which the water depth at each cell reached 0.6 meters (approx. 2 feet).

Impact areas were created by reclassifying the arrival time grid into arrival time zones
required for reporting results (0-15min, 15-30min, etc.). A polygon was created from that
file. Some features had to be manually merged to create a unified impact area polygon for the
HEC-FIA model. The structure inventory was created from the parcel file with populations
and values developed in the pre-modeling phase. Structure placement was at the centroid of
each parcel and was done within HEC-FIA.

For estimates of necessary evacuation time for each structure, a hazard boundary shapefile
was created representing the boundary where the inundation becomes 0.6 meters (2 feet) or
less, which is considered a safe zone. Evacuation was modeled as a path from the structure to
the nearest point on this boundary at an average speed of 16 kilometers per hour.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
331
The warning time used for the loss of life calculation was set at 30 minutes after the breach
initiation. The problem statement specified that the failure was sudden and unexpected, and
that witnesses within the canyon had no mobile phone service. A warning would not be issued
until after the water reached the urban area at the end of the canyon, approximately 20
minutes after the failure. The warning curve used was the HEC-FIA default curve for the
United States Emergency Alert System, and the mobilization curve was the default HEC-FIA
curve included in the program.

The HEC-FIA program evaluates damages with depth-damage curves based on occupancy
type. Population distribution is also performed by occupancy type. To determine occupancy
types, a cross walk table was created that assigned each parcel zone description to a specific
occupancy type typically used in a HEC-FIA structure inventory.

Populations and jobs were provided for each census block and broken down by category. To
facilitate the distribution to individual parcels, the populations were combined into the
following categories by census blocks:

- Resident Population under age 14
- Resident Population under age 65
- Resident Population over age 65
- Working Population (10% of the workers from industries potentially in operation at
11pm on a Saturday)

In order to determine the amount of population for each structure, each occupancy type was
assigned a number of residential households or working households. Then for each census
block, the residential populations were divided by the total number of residential households
in the block and the working populations were divided by the total number of working
households in the block. The resulting residential and working household sizes per census
block were then used to create the populations for each parcel based on the number and type
of households assigned to the occupancy type of the parcel. For example, if the residential
household size in a census block was 2.5, then a single family residential structure in that
block would get 2.5 people while a five to nine unit multiple family dwelling in the same
census block would get 7 households, or 17.5 people.

Table 2 shows the final occupancy types used and the number of assigned households. In the
beginning there were more occupancy types included, but the ones listed below represent all
of the types included in the final inventory after being clipped to the inundation boundary.
The methodology, including the number of households assigned to each occupancy type, is
representative of a typical HAZUS based structure inventory used by the MMC.
Table 2: Population distribution by occupancy type
Occupancy
Type
Occupancy Type Description Residential
Households
Working
Households
COM1 Average Retail - 1
COM4 Average Prof/Tech Services - 1
GOV1 Average government services - -
RES1-1SNB Single Family, 1 Story no Basement 1 -
RES1-2SNB Single Family, 2 Story no Basement 1 -
RES1-3SNB Single Family, 3 Story no Basement 1 -
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
332
RES3AI Multi Family Dwelling, Duplex 2 -
RES3CI Multi Family Dwelling, 5-9 Units 7 -
RES3DI Multi Family Dwelling, 10-19 Units 14.5 -

No data was provided on structure values in the parcel dataset. A standard USACE dam
failure consequence estimate will often use the National Structure Inventory (NSI) developed
by the USACE Hydrologic Engineering Center and based on HAZUS and Census data. The
team decided to use average values from the NSI for each occupancy type to determine
structure values. The NSI data is broken down by county, so a national average would have
required significant effort. Instead of using a national average, the team reviewed data on
housing values from the 2011 American Community Survey and determined that the state of
Maine had the closest housing value to the national average. Data from the NSI for each
county in Maine was combined and average values for each occupancy type were calculated.
Vehicle and content values were also assigned (Table 3).
Table 3: Average values by occupancy type
Occupancy
Type
Occupancy Type
Description
Average
Structure
Value
Average
Content Value
Average
Vehicle Value
COM1 Average Retail $545,888 $545,888 $21,819
COM4
Average Prof/Tech
Services
$550,027 $550,027 $22,707
GOV1
Average government
services
$481,998 $481,998 $27,020
RES1-
1SNB
Single Family, 1 Story
no Basement
$122,061 $61,030 $14,825
RES1-
2SNB
Single Family, 2 Story
no Basement
$121,957 $60,528 $14,766
RES1-
3SNB
Single Family, 3 Story
no Basement
$127,135 $63,568 $15,175
RES3AI
Multi Family
Dwelling, Duplex
$206,625 $103,312 $19,795
RES3CI
Multi Family
Dwelling, 5-9 Units
$753,610 $376,805 $26,195
RES3DI
Multi Family
Dwelling, 10-19 Units
$1,288,765 $644,383 $29,931

Since the parcel shapefile appeared only to have information about zoning and not specific
information about structures, the team assumed that not every parcel had a structure on it.
Also, some parcels had unknown zone information so those parcels were removed from the
final inventory. Information in the census blocks and the land use file also needed to be
evaluated to inform the structure inventory development.

One particular issue involved a large number of parcels in the middle section of the study
area. These parcels appeared to be subdivided into mostly single family home lots. However,
the census block data showed no residential population in those areas. Figure 4 shows census
blocks labeled with total residential population with the parcels overlain on top in blue. The
red census blocks have zero population.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
333

Figure 4: Comparison of parcels and census block populations

The land cover file showed no development in the area in question except for the roads
themselves; no development is shown to indicate structures. Figure 5 shows development as
red or pink. The roads in the center area are listed as Developed-Open Space and everything
other than roads is listed mostly as grassland or shrub (brown and pale green colors).



Figure 5: Land use downstream of the dam

This data can be interpreted in several ways. For example, it may be a situation where a
subdivision has been planned for future development, and the land has been parceled off and a
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
334
road system developed but no houses have been built and occupied. Alternatively, it could be
a scenario where the parcel data was recently updated while the land cover and census data
are several years old and do not reflect the recent development. With no way to ground truth
the data through aerial imagery or site visits the team chose to adopt the first interpretation,
assuming that all data was current and structures had not yet been built in that area. The large
group of central parcels was removed from the final inventory by deleting all residential
parcels in census blocks with no residential population.

In addition, all parcels more than 100 meters outside the dam failure inundation area were
deleted. This minimized the parcels to only those necessary for the HEC-FIA structure
inventory.
Results and Discussion
The hydrograph that was computed from Froehlich (1995) breach parameters had a peak
discharge of 30,917 m
3
/sec at the dam site. The time to peak discharge occurred 31 minutes
after breach initiation began. Nearly all of the reservoir volume was evacuated within 90
minutes of the breach initiation.

Discharge immediately downstream from the dam remained confined to the mountain valley.
This part of the reach, in which the flood wave was limited to a maximum width of 400
meters, continues downstream for approximately 3,000 meters. Minimal attenuation of the
flood wave occurred through the valley, which was verified by developing cross sections
downstream to the base of the higher terrain (Figure 6). At the lower end of the valley, the
peak discharge of the cross-sectional hydrograph had only decreased to 26,474 m
3
/sec.

Figure 6: Breach hydrographs at selected locations downstream from the dam
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
335
When the flood wave reached the more highly populated area below the base of the mountain,
a change in the slope and layout of the surrounding terrain resulted in a more pronounced
attenuation and lateral dispersion of the inundated area. This behavior of the flood wave is
consistent with the transition from a steep mountain valley to a relatively flat, unconfined
plain such as an alluvial fan or coastal area. This attenuation was highlighted by peak unit
flow rates, which ranged from 300 m
2
/sec immediately downstream from the dam to less than
1 m
2
/sec for most of the area in the broader flood plain.
Peak depths of the flood wave ranged from 24.5 meters immediately downstream from the
dam to less than 1 meter at the downstream computational boundary (Figure 7). Much of the
populated area fell within the 1- to 3- meter range, although some structures in the lower part
of the mountain valley were inundated by depths as high as 12 meters. Arrival times for the
flood wave were only 30 minutes for the lower part of the mountain valley. The population
residing within the broader flood plain below the valley fell within an arrival time zone of 1 to
1.5 hours. The most distant reach of the modeled area coincided with an arrival time of 4
hours.

Figure 7: Peak inundation depths (meters) below the dam
Initial results from the HEC-FIA model are displayed below in Tables 5 and 6 (by arrival time
and by depth of flooding) as required by the benchmark reporting requirements.
Table 4: Consequences by arrival time
Time
Interval
(min)
Total
Population
At Risk
14-yr and Under
Population at
Risk
65-yr and Over
Population at
Risk
Loss of
Life
Direct
Economic
Impact ($US)
0-15
- - - - -
15-30
3,951 1,002 282 1,844 $142,504,368
30-60
7,612 933 1,855 34 $266,968,912
60-90
12,672 1,952 2,166 11 $248,115,360
90-120
- - - - -
120-180
- - - - -
> 180
- - - - -
Total
24,235 3,887 4,303 1,889 $657,588,640

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
336
Table 5: Consequences by depth
Peak Flood
Depth Range
(m)
Flooded
Area (m)
Total Population
At Risk
14-yr and Under
Population at Risk
65-yr and Over
Population at Risk
0 - 0.5 3,699,606 2,749 399 511
0.5 - 1 11,646,576 4,249 660 687
1 - 1.5 23,248,253 9,162 1,401 1,626
1.5 - 2 12,625,355 3,183 446 716
2 - 2.5 6,151,044 1,353 160 373
2.5 - 3 2,990,216 540 85 122
3 - 3.5 2,128,172 153 13 55
3.5 - 4 1,670,211 129 28 19
4 - 4.5 1,068,576 173 49 15
4.5 - 5 511,839 133 32 9
5 - 5.5 377,144 259 71 17
5.5 - 6 448,981 299 78 27
6 - 6.5 233,470 250 67 15
6.5 - 7 152,654 136 33 9
7 - 7.5 287,348 286 69 19
7.5 - 8 296,328 502 131 43
> 8 1,562,455 679 167 43
Total 69,098,227 24,235 3,887 4,303
Conclusion
This exercise demonstrates the practices and procedures that the MMC uses to analyze the
large inventory of dams that are owned and operated by USACE. Most of these dams are
larger and more complex than the hypothetical dam that was analyzed in this project. Some
include complex reaches that extend for hundreds of miles downstream with travel times that
last for days or weeks. Elements of each of these dams and their downstream reaches are
unique, and it is important to emphasize that the sophistication of MMC models, while
incorporating a standardized process, varies by project.

Since the reach in this exercise was both steep and short, flood wave travel times to the
downstream boundary of the model were very quick. As a result of these conditions,
comparisons of the inundation areas resulting from the Froehlich (1995), Froehlich (2008),
and Von Thun and Gillette equations were very similar. Modeling with dam breach
parameters from all three equations resulted in a flood wave with a steep peak and fast travel
time through the confined mountain valley below the dam followed by lateral attenuation of
the flood wave in the broad plain below the base of the mountain range.

Breach parameters from Froehlich (1995) were carried forward in the modeling used for
consequences analysis. Estimation of economic damages and life loss was challenging
because FEMA HAZUS data, a standard input for MMC HEC-FIA models, was not available
for the benchmark exercise. Therefore, average values were used, which may or may not be
representative of Hydropolis. Total damages of $657M and a life loss of nearly 2,000 persons
were estimated for this dam breach.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
337
Based on the rules and regulations that USACE has established for the estimation of risk,
these results correspond to a high hazard dam. The procedures that were used in developing
the hydraulic modeling and consequences estimation for the benchmark exercise were
consistent with USACE MMC processes. These guidelines, while providing consistency in
the analysis of all USACE owned and operated dams, also provide enough flexibility to
analyze projects that range from small and simple to large and extremely complex.
Acknowledgments
Funding for this project was provided by the U.S. Army Corps of Engineers, Office of
Homeland Security. Internal review of the numerical modeling and consequences estimation
results was provided by the U.S. Army Corps of Engineers, Hydrologic Engineering Center.
References
[1] MacDonald, T.C., and Langridge-Monopolis, J. (1984). Breaching characteristics of dam
failures. Journal of Hydraulic Engineering, 110(5): 567-586.
[2] Von Thun, J.L. and Gillette, D.R. (1990). Guidance on breach parameters. Unpublished
internal memorandum, U.S. Bureau of Reclamation, Denver, CO, March 13, 1990.
[3] Froehlich, D.C. (1995). Peak outflow from breached embankment dam. Journal of Water
Resources Planning and Management, 121(1): 90-97.
[4] Froehlich, D.C. (2008). Embankment dam breach parameters and their uncertainties.
Journal of Hydraulic Engineering, 134(12): 1708-1721.
[5] U.S. Army Corps of Engineers (2010). HEC-RAS River Analysis System (Version 4.1)
[Computer software]. Davis, CA: Hydrologic Engineering Center.
[6] OBrien, J.D. (2009). FLO-2D (Version 2009) [Computer software]. Nutrioso, AZ: FLO
Engineering.
[7] ESRI (2011). ArcGIS Desktop (Release 10) [Computer software]. Redlands, CA:
Environmental Systems Research Institute.
[8] Federal Emergency Management Agency (2012). HAZUS-MH (Version 2.1) [DVD].
Washington, DC: FEMA.
[9] U.S. Army Corps of Engineers (2012). HEC-FIA Flood Impact Analysis (Version 2.2)
[Computer software]. Davis, CA: Hydrologic Engineering Center.
[10] U.S. Army Corps of Engineers (2013). MMC standard operating procedures for dams.
Vicksburg, MS: Modeling, Mapping and Consequences Production Center.
[11] International Commission on Large Dams (2013). Part 1 Hydraulic Modelling and
Simulation. Graz, Austria: 12
th
International Benchmark Workshop on Numerical
Analysis of Dams (Theme C).
[12] U.S. Army Corps of Engineers (2011). HEC-GeoRAS Extension for ArcGIS 10 (Version
10.0) [Computer software]. Davis, CA: Hydrologic Engineering Center.
[13] Homer, C.G., Huang, C.C., Yang, L., Wylie, B., and Coan, M (2004). Development of a
2001 national landcover database for the United States. Photogrammetric Engineering
and Remote Sensing, 70(7): 829-840.
[14] Mattocks, C., and Forbes, C. (2008). A real-time, event-triggered storm surge forecasting
system for the state of North Carolina. Ocean Modelling, 25: 95-119.
[15] International Commission on Large Dams (2013). Part 2 Consequence Estimation.
Graz, Austria: 12
th
International Benchmark Workshop on Numerical Analysis of Dams
(Theme C).


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
339



OPEN THEME



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
341
Behavior of an arch dam under the influence of
creep, AAR and opening of the dam/foundation
contact
E. Robbe
1

1
Hydro Engineering Center of EDF,Savoie-technolac, 73373 Le Bourget du Lac, FRANCE
E-mail: emmanuel.robbe@edf.fr
Abstract
This paper summarizes the behavior analysis of a thin, 37m high, arch dam built in 1956. At
first, the behavior of this dam was quite difficult to understand: irreversible displacements
occur during the years following the first filling, and then, from 1960 until now, others
displacements were observed and difficult to read considering usual behavior of arch dam. In
order to understand it and to evaluate the stress level of the concrete, finite-element analyzes
are leading. First linear, then non-linear analysis taking increasingly into account
dam/foundation contact opening, swelling, creep and damage are used to answer the problem
and understand the historic behavior of this dam
Introduction
The subject of this paper is an arch dam with a particularly complex behavior observed. The
purpose of the study is to understand this behavior and to evaluate the level of stress of the
dam. Increasingly complex finite-element analyzes are used (from linear to non-linear) and
calibrated on the observed displacements of the dam. Stress levels are also compared to crack
patterns observed on the structure.
In this presentation, geological conditions and safety evaluation of the gravity abutment will
not be discussed.
Presentation of the dam

Figure 1: view of the dam
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
342
The studied dam is a thin arch, with a gravity abutment on the right side. 37 m high above its
foundation made of gneiss, the dam is divided into 6 cantilevers (about 17m wide) for the arch
part and 2 cantilevers for the right abutment. Figure 1 shows the design of the dam: the
thickness goes from 2 m on the crest to 5.4 m at the bottom, with a sloping upstream face in
the lower part only. The valley is quite wide (L/H about 3) and clearly asymmetrical.
The dam has been built from 1954 to 1956. In 1961, works has been done in order to grout 2
parallels faults discovered on the left abutment and to build an apron to protect the foundation
in case of overflowing.
Observed behavior
This part describes the behavior of the dam monitored with points on the downstream face of
the dam from the first filling in 1956 until 2001. These points are represented on Figure 2.
From 2001 until now, pendulums are used in order to follow the dams behavior.
A statistic analysis of the monitoring data is used in order to evaluate the part of the
hydrostatic and the seasonal effects on the dams displacements. Once theses reversible
effects evaluate, the irreversible effects can be estimate. Regarding these data, the behavior
can be separated in 2 phases:
- During the 5 years after the first filling of the dam (1956-1960)
- 40 years from 1961 until 2001
Behavior after first filling
The analysis of the radial displacements of the dam shows the following behavior:
- At the crest, the central part moves downstream while the top of the sides cantilevers
move upstream,
- All the points at the central cantilever records downstream displacements.
Long-term behavior
The radial displacements of the monitored points located on the downstream face of the dam
are represented on the Figure 2 for approximately 40 years.
Theses displacements can be described the following way:
- The crest of the dam moves upstream continuously (0.3 mm/year for the central
cantilever)
- The lower part of the arch goes first downstream during 10 years then upstream after
10 years of stabilization. The velocity of the downstream displacement is close to the
one observed on the crest.
A first global interpretation of these observations is that the concrete is under the influence of
swelling, which can explains the upstream displacements of the crest. For the lower part,
creep is more important during the first 10 years, that explains the first downstream
displacements, and decreases gradually and therefore swelling become leading.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
343

Figure 2: irreversible radial displacements (from 1956 to 2001)
Linear analysis
A finite-element linear analysis of the dam is realized with Code_Aster [1]. The physical
properties for concrete and rock are chosen by comparison between the monitored data and
the results of the FE analysis. The calibration is done considering hydrostatic load and
thermal load in order to represent correctly the behavior of the dam in winter and summer for
example.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
344


Figure 3: Mesh of the dam and foundation

The linear analysis shows that the particular design leads to an unconventional behavior of the
dam: the lower part works as cantilever while arch-effects appear in the upper part. That leads
to high vertical tensile stress at the heel of the dam (close to 8 MPa in winter) and in the
middle of the downstream face where seepage are already observed on a lift joint. Because of
theses high tensile stresses, the FE is upgraded in order to take into account the opening of the
dam/foundation contact and the lift joint, and to evaluate their consequences.
Non-linear analysis including dam/foundation contact
The Figure 4 presents the joint elements introduced in the mesh in order to simulate the non-
linear behavior:
- dam/foundation contact under the 6 cantilevers,
- dam/foundation contact under the right gravity abutment,
- horizontal lift joint at the change of inertia of the dam,
- contraction joints.
Joint elements open under normal tensile stress but share behavior is still elastic-linear
whatever the opening-state of the joint. Except for vertical contraction joints, uplift is taking
into account. In particular, spread of the uplift under the dam with joint opening is used.

Figure 4: mesh of the joint elements
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
345


Figure 5: comparison of the deformed shapes at the crest (elastic, with joint elements)

The introduction of the joint elements modifies the deformation at the crest of the dam (Figure
5) Theses irreversible displacements are equivalent to the ones observed during the first years
after the first filling of the dam and presented earlier. The analysis shows that the
dam/foundation contact is open at the heel of the dam, particularly in winter (5 mm). That
leads to an increase of the uplift under the central cantilever. This is confirmed by a recently
installed piezometer which records the almost full uplift during winter.
Non-linear analysis including AAR and creep
If the previous section evaluates the behavior of the dam after few years, the recorded
displacements during the following 40th years need to be taken in account in order to estimate
correctly the stress state of the dam. Considering the displacements recorded, the behavior of
the dam is under the influence of swelling and creep and therefore, the model used have to be
able to represent correctly these phenomena. In this purpose, the model developed by Grimal
[2] for AAR on concrete dam is particularly adequate. This visco-elasto-plastic orthotropic
damage model includes chemical pressure induced by AAR and takes into account the
influence of creep on the behavior. It has been used to evaluate the behavior of Chambon's
dam [2] and showed the models capability to reproduce displacements with acceptable
accuracy. The main developments brought by this model concern interactions between AAR
pressure and long term strain (creep) on the one hand, and the swelling anisotropy induced by
oriented cracking on the other hand.
Here, this model is coupled to the joint elements presented earlier. The calibration of the
model is realized the following way:
- first, the kinetic of the chemical pressure induced by AAR and the volume of AAR gel
are chosen in order to fit the displacement monitored on the crest of the dam
- then, the parameters for creep are chosen in order to fit displacements of the lower part
of the dam
The result of this calibration is presented on Figure 6.




With joint element
Elastic behaviour
Initial shape
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
346

Figure 6: irreversible radial displacement estimated and calculated

Considering the complexity of the dam's behavior, the calibration is considered acceptable
despite the fact that the central cantilever's behavior isn't perfectly represented. However, the
model is able to represent the change of direction of the lower part of the dam (displacements
downstream then upstream) while the crest is moving continuously upstream.

The influence of the long term behavior of the dam is analyzed regarding the evolution of the
dam/foundation contact opening (Table 1) and of the stress state of the concrete (Figure 7).
For the first point, the model shows a closure of the dam/foundation contact, particularly in
summer while an opening is still likely possible (3.3 mm). That can explain why the full uplift
is recorded only in winter under the dam. The analysis of the stress state of the dam shows an
important increase of the compression stress close to the left abutment: from 5 to 10 MPa
between 1960 and today. On the right side, the gravity abutment brings more flexibility,
which leads to lower stress values. In-situ stress measurements with flat jack are under
consideration in order to confirm the stress values calculated.
Table 1: evolution of the opening of the dam/foundation contact with swelling
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
347

Opening of the
dam/foundation contact
(mm)
1961
Opening of the
dam/foundation contact
(mm)
2010
Hydrostatic load +
Summer
2.8 1
Hydrostatic load + Winter 4.8 3.3


Figure 7: main arch stress on the upstream face in summer: without (1961) and with swelling
(2011)
Conclusion
At the beginning of this study, the dam's behavior was difficult to understand: multiple
mechanisms were involved and therefore, the problem could not be solved with basic
interpretations. As often, model increasingly complex allow to separate each phenomena
(opening of the dam/foundation contact, swelling, creep) and to understand their influences on
the general behavior of the dam. The last model is used to evaluate the impact of the long-
term behavior on the stress state of the dam and to decide what need to be done to assure the
safety of the structure.
References
[1] Code_Aster software, www.code-aster.org
[2] Grimal E., Bourdarot E (2013). Modeling AAR on concrete dam. ICOLD 81th Annual
Meeting Symposium, Seattle, 2013, pp. 99-109.
5 MPa
10 MPa

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
349
Need for transient thermal models, with daily inputs,
to explain the displacements of arch gravity dams
I. Escuder-Bueno
1
, D. Galn
2
and A. Serrano
3

1
Instituto de Ingeniera del Agua y Medio Ambiente. Universitat Politcnica de Valncia
(Spain) http://www.ipresas.upv.es
2
Divisin de Seguridad de Presas. Canal de Isabel II Gestin (Spain)
http://www.gestioncanal.es
3
iPresas, SPIN-OFF UPV. web: http://www.ipresas.com
E-mail: iescuder@hma.upv.es
Abstract
La Acea Dam, with 66 meters of maximum height, belongs to the typology of concrete
gravity dams, and is operated as part of the water supply infrastructures of a Spanish major
city. Many instrumentation records were available but among them, the ones provided by four
direct pendulums outstood by its quality and consistency. The range of the values of the
movements registered by those pendulums, of almost 4 cm and of totally elastic nature
(showing no irreversible movements), set some interpretation challenges. The apparent
incapacity of the models to reproduce the observed behaviour was used as a starting point for
the diagnosis of the main sources of uncertainty: the nature of the foundations and the state of
the joints, among others. These aspects have constituted the developmental axis of a series of
works that have led to a number of effective and efficient actions on the dam during the last
years. However, an updated transient thermal model with daily inputs on external and internal
measured temperatures, coupled with a mechanical model has shown an spectacular
improvement in the explanation of the displacements recorded by the instrumentation.
Software and constitutive models
The first numerical model of global simulation of the stress-deformation behaviour of the dam
of la Acea was done as part of the works First review and general safety analysis, carried
out by the consultancy firm OFITECO in 2005 for CANAL DE ISABEL II, owner of the
dam. The model was done with SAP2000NL and intended, among others, to capture some
existing radial movements that have been registered by the four direct pendulums and that
were higher that expected.
Based on the results of these works, Francisco Blazquez Prieto and Ignacio Escuder presented
the case as one of the problems to be tackled by independent calculation teams during the
Ninth International Benchmark Workshop on Numerical Analysis of Dams, organized by
the Dam Calculation Committee of the International Commission on Large Dams (ICOLD) in
Saint Petersburg, Russia, in June 2007.
According to Escuder and Blazquez (2007), despite the great diversity of software (SAP
2000NL, CANT-SD, COQ-EF, SOFiSTiK, MERLIN, DIANA y ANSYS) and thermal
models (three of which were transient), none of the seven teams that worked on the records of
the radial movements tracked by the pendulums from January 1999 till September 2001, was
able to predict the magnitude of the recorded displacements when using realistic values of
stiffness of the dam body and foundations.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
350
In 2012, a re-evaluation of the model was done, which is presented in the present paper. The
distinctive feature and more promising fact of this re-evaluation when adopting different
software and a new behaviour model, was the fact having a much longer list of data (now
running until 2011) that included daily temperature measurements (whereas at the time of the
Benchmark Workshop, the experts had to work on monthly measurements given that daily
measurements were missing in 1999 and an important part of 2000).
With regard to thermal modelling, it is worth mentioning that the 2005 model, though done
with a SAP 2000NL Licence, a very appropriate code for structures in general, could not
guarantee to make the most of the data, mainly due to the limitations of the thermal module of
the package (stationary nature).
Upon consideration of this aspect, it was decided to develop the numerical models with a
different software package, FLAC (Itasca, 2004), which would allow a transient analysis of
temperature transmission and thus provide the best-adjusted results realistically attainable.
FLAC (Itasca, 1994) is a finite differences code (explicit scheme) that allows the simulation
of the behaviour of soils, rocks, concrete structures, etc. as well the interaction among these
elements. The program is based on a scheme of Lagrangian calculation in which each element
behaves accordingly to a specifically prescribed stress-displacement relation, as a response to
the applied forces and the existing constrains in its boundaries.
FLAC is doted with an internal programming language (FISH) that allows the definition of
each calculation organization (i.e., complex construction sequences) and calculations of very
different nature (i.e., for each constitutive relation). The main potential of the base code is the
capacity of the software to model tensional states through the use of pre-set constitutive
models and others purposely defined by the user. For la Acea case, a mechanical elastic and
linear model was adopted and coupled with a transient thermal model.
The stiffness parameters that define the mechanic model are Youngs Modulus (E) and
Poissons Ratio (v), which are somehow equivalent and that can express the strain-
displacement relations accordingly to the Volumetric Deformation Coefficient (K) and Shear
Deformation Coefficient (G).
With regard to the thermal module of FLAC 3D, it must be noticed it incorporates diffusion
models as well as advection ones. The simulation of the first type of model, used for the case
of la Acea, allows the analysis of heat transmission through the materials along time and the
development of thermally induced stresses and displacements.
The process of heat diffusion that the program solves through the optional thermal module is
defined by the following differential equation:

(1)

Where is the density of the medium, C the specific heat, k is the conductivity and q
v
is the
source of hydration heat.

In the development of the model, all involved parameters (density, conductivity and specific
heat) have been assumed as being constant and independent of the temperature. Indeed, since
the variations of concrete properties with temperature are very small, they can be neglected
without concern. With respect to the boundary conditions, the external parameters of the dam
are considered as convective surfaces, that is, as behaving accordingly to the following
relation:
v
q
z
T
y
T
x
T
k
t
T
C =
|
|
.
|

\
|
c
c
+
c
c
+
c
c

c
c
2
2
2
2
2
2

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
351

(2)
Where q
n
is the component of normal flux pointing towards the outside of the considered
boundary expressed in W/m
2
, h is the convective heat transfer coefficient expressed in
W/m
2
C, T is the temperature of the boundary surface and T
e
is the temperature of the
surrounding fluid, with both temperatures expressed in C.
In the coupled thermal-mechanical model, the solution of the thermal stresses requires a
reformulation of the incremental relations of stress-displacements that FLAC 3D achieves
through a subtraction of the part of movement due to the change of temperature from the total
incremental deformation.
Finally, the laws of movement defined in a continuous medium are transformed through the
former approximation into a system of equations that are applied to the nodes that represent
the structure. These equations correspond to the laws of Newton applied to these discretized
nodes and result in a system of ordinary differential equations that can be solved numerically
from the explicit method of finite differences with respect to time.
Summing up, it can be stated that the code FLAC gathers the necessary conditions to tackle
successfully the study of La Acea dam behaviour, taking advantage of the existence of
thermal and movements records of a level of detail and quality very unusual in such
structures.
Simulation methodology
The present analysis simulates the stress-deformation behaviour of the dam of La Acea in
the period of 1999-2011, which is the one for which movement data were available through
pendulum readings. In particular, two radial measurements were available (crest and
horizontal gallery) from 4 pendulums (pendulums from 1 to 4).
The mechanical properties used in the model are the same employed in former studies of the
dam and are shown hereafter (Table 1):
Table 1: Dam and concrete mechanical properties
Youngs Modulus (dam) 20 10
9
Pa
Poissons Ratio (dam) 0.2
Density (dam) 2405.7 kg/m
Youngs Modulus (
(foundation)
10 10
9
Pa
Poissons Ratio (foundation) 0.2
Density (foundation) 2242.6 kg/m
The main difference between the current analysis and those carried out in the past is that
better data are now available. In particular:
1. Daily records of the external temperatures in the dam since October 2000. Before,
only average monthly temperatures were available.
2. Daily data from the three thermometers located in the concrete since January 2008.
Prior to this date there was no data about dam concrete temperatures.
3. Data on the reservoir water temperature (year 2011, monthly periodicity).

( )
e n
T T h q =
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
352
Higher periodicity (daily) of pendulums readings since July 2004 thanks to the installation of
automatic readers.

In order to understand the effect the different hypotheses have on the results, several models
of the dam have been done, through the combination of the following options:
- Geometry: 2d and 3d
- Foundation: With and without foundation
- Temperatures: simplified hypothesis and improved hypothesis

Since 2008 there exists a register of the temperatures measured in three points inside the
concrete that have helped to improve the original thermal hypothesis in the terms indicated
hereafter, and that correspond to what has been named improved hypothesis:
- The temperature of the concrete in contact with the air is equal to the ambiance
temperature plus the effect of solar radiation (T = 10 C).
- The temperature of concrete in contact with the water has been assumed to be constant
and equal to 5 C except for the superior layer that acts as a transition element
between the ambiance temperature and water.
- Specific heat: 3.5 W/m K, thermal conductivity: 650 J/kg K, convection coefficient:
16 W/m K and linear thermal expansion coefficient: 10
-5


Figure 1 shows the global geometry of the model and Figure 2 the location of the pendulums
whose movements, measured in the crest, were used to check the goodness of the adjustment.
Figure 3 represents the whole of the external variables (water level and external average
temperature) as well as the pendulums measurements.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
353

Figure 1: Complete geometrical model

Figure 2: Location of the four direct pendulums
FLAC3D 3.10
Itasca Consulting Group, Inc.
Minneapolis, MN USA
(c)2006 Itasca Consulting Group, Inc.
Step 27477654 Model Perspective
12:17:42 Fri Sep 21 2012
Center:
X: 8.750e+001
Y: 8.750e+001
Z: 1.275e+003
Rotation:
X: 40.000
Y: 0.000
Z: 30.000
Dist: 5.714e+002 Mag.: 1
Ang.: 22.500
Block Group
Live mech zones shown
b11
b9
b7
b5
b3
b1
b0
b2
b4
b6
b8
b10
b12
cimiento
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
354

Figure 3: External variables (level in the reservoir and average ambiance temperature) and
movements measured by the pendulums

1250
1260
1270
1280
1290
1300
1310
1320 -15
-10
-5
0
5
10
15
20
25
e
n
e
-
9
9
m
a
y
-
9
9
s
e
p
-
9
9
e
n
e
-
0
0
m
a
y
-
0
0
s
e
p
-
0
0
e
n
e
-
0
1
m
a
y
-
0
1
s
e
p
-
0
1
e
n
e
-
0
2
m
a
y
-
0
2
s
e
p
-
0
2
e
n
e
-
0
3
m
a
y
-
0
3
s
e
p
-
0
3
e
n
e
-
0
4
m
a
y
-
0
4
s
e
p
-
0
4
e
n
e
-
0
5
m
a
y
-
0
5
s
e
p
-
0
5
e
n
e
-
0
6
m
a
y
-
0
6
s
e
p
-
0
6
e
n
e
-
0
7
m
a
y
-
0
7
s
e
p
-
0
7
e
n
e
-
0
8
m
a
y
-
0
8
s
e
p
-
0
8
e
n
e
-
0
9
m
a
y
-
0
9
s
e
p
-
0
9
e
n
e
-
1
0
m
a
y
-
1
0
s
e
p
-
1
0
e
n
e
-
1
1
m
a
y
-
1
1
s
e
p
-
1
1
e
n
e
-
1
2
Tmed
Cota
-10
-5
0
5
10
15
20
25
30
35
e
n
e
-
9
9
m
a
y
-
9
9
s
e
p
-
9
9
e
n
e
-
0
0
m
a
y
-
0
0
s
e
p
-
0
0
e
n
e
-
0
1
m
a
y
-
0
1
s
e
p
-
0
1
e
n
e
-
0
2
m
a
y
-
0
2
s
e
p
-
0
2
e
n
e
-
0
3
m
a
y
-
0
3
s
e
p
-
0
3
e
n
e
-
0
4
m
a
y
-
0
4
s
e
p
-
0
4
e
n
e
-
0
5
m
a
y
-
0
5
s
e
p
-
0
5
e
n
e
-
0
6
m
a
y
-
0
6
s
e
p
-
0
6
e
n
e
-
0
7
m
a
y
-
0
7
s
e
p
-
0
7
e
n
e
-
0
8
m
a
y
-
0
8
s
e
p
-
0
8
e
n
e
-
0
9
m
a
y
-
0
9
s
e
p
-
0
9
e
n
e
-
1
0
m
a
y
-
1
0
s
e
p
-
1
0
e
n
e
-
1
1
m
a
y
-
1
1
s
e
p
-
1
1
e
n
e
-
1
2
P1 Rad Coronacin
P2 Rad Coronacin
P3 Rad Coronacin
P4 Rad Coronacin
-10
-5
0
5
10
15
20
25
30
35
e
n
e
-
9
9
m
a
y
-
9
9
s
e
p
-
9
9
e
n
e
-
0
0
m
a
y
-
0
0
s
e
p
-
0
0
e
n
e
-
0
1
m
a
y
-
0
1
s
e
p
-
0
1
e
n
e
-
0
2
m
a
y
-
0
2
s
e
p
-
0
2
e
n
e
-
0
3
m
a
y
-
0
3
s
e
p
-
0
3
e
n
e
-
0
4
m
a
y
-
0
4
s
e
p
-
0
4
e
n
e
-
0
5
m
a
y
-
0
5
s
e
p
-
0
5
e
n
e
-
0
6
m
a
y
-
0
6
s
e
p
-
0
6
e
n
e
-
0
7
m
a
y
-
0
7
s
e
p
-
0
7
e
n
e
-
0
8
m
a
y
-
0
8
s
e
p
-
0
8
e
n
e
-
0
9
m
a
y
-
0
9
s
e
p
-
0
9
e
n
e
-
1
0
m
a
y
-
1
0
s
e
p
-
1
0
e
n
e
-
1
1
m
a
y
-
1
1
s
e
p
-
1
1
e
n
e
-
1
2
P1 Rad G.Horiz
P2 Rad Man GH
P3 Rad G.Horiz
P4 Rad G.Horiz
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
355
Summary of the results and conclusions
The main conclusions that can be extracted from the current analysis of the stress-deformation
behaviour of the dam of La Acea through the simulation methodology described along the
former sections, established in order to make the most of the whole set of records of thermal
monitoring and movements, are:
- The elaborated models reproduce in a reasonably satisfactory way the thermal
behavior of the structure as well as the movements observed in the dam.
- The fact of having carried out a transient thermal simulation for each day, capturing
the impact of the direct sun exposure on the downstream side of the dam and
estimating with accuracy the water temperature have been key elements to the
goodness of the obtained numerical model.

Figure 4 shows a summary of the average errors incurred by the different models. As it can be
observed, the best model is the so called reference one (3D, with foundation and the
hypothesis of improved temperature).

Figure 4: Average errors in mm for each model (shown for each pendulum, from P1 in darker
red to P4, in a lighter grade.)
It can be stated that thanks to the recently developed numerical models a new tool is at hand
to interpret the strain-deformation behaviour of the dam of La Acea. This tool can
complement the auscultations that are currently being done on the dam and that have provided
an excellent level of information about both the thermal and movement behaviour. Also, the
numerical tool has permitted the satisfactory and consistent reproduction of the stress-
deformation behaviour auscultated until the end of 2011 and can act in the future as a way of
contrasting the impact of new actions planned on the dam as well as a form of detecting
potential behaviour change trends.

Figure 5 shows the comparison between the reference model (with the hypothesis of improved
temperature) and the model with the original hypothesis of temperature.
0
0.5
1
1.5
2
2.5
3
3.5
Modelo de referencia Sin cimiento 2d Hiptesis de temperatura
inicial
Mezcla 2d - 3d
P
r
o
m
e
d
i
o

d
e

e
r
r
o
r
e
s

a
b
s
o
l
u
t
o
s
P1
P2
P3
P4
Reference
model
Impact of
no
cosidering
foundation
Impact of
no
cosidering
3D
Impact of
no
updating
thermal
parameters
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
356

Figure 5: Extent of the improvement of the thermal model (reference model in red, non-
updated thermal model in blue)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
357
Acknowledgements
Some of the calculation routines have been developed within the following Research Project:
Integration of the components of anthropic risk in the systems of dams and reservoirs global
safety management (Incorporacin de los componentes de riesgo antrpico a los sistemas
de gestin integral de seguridad de dams y embalses BIA2010-17852), funded by the Spanish
Ministry of Science and Innovation and with funds FEDER.
References
[1] I. Escuder and F.Blazquez, Analysis of the Elastic Behaviour of an Arch Dam,
Hydropower and Dams, Issue 5, (2007)



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
359
The rehabilitation of Beauregard Dam:
the contribution of the numerical modeling
A. Frigerio
1
and G. Mazz
1

1
Ricerca sul Sistema Energetico RSE S.p.A., via R. Rubattino 54, Milan, ITALY
antonella.frigerio@rse-web.it
Abstract
The paper describes the case-history of the Beauregard dam (Italy), a concrete arch-gravity
structure 132 m high built between 1951 and 1960 on the Dora di Valgrisenche River. The
design reservoir volume was about 70 mil. m
3
. The geological and geotechnical investigations
carried out since the dam construction and deepened in the last decade have underlined that a
Deep-Seated Gravitational Slope Deformation (DSGSD) is located on the left slope. Since the
first fillings of the reservoir, the interaction between the DSGSD and the dam was recognized
to have relevant implications on the dam structural safety. For that reason, the Italian Dam
Authorities in 1969 prescribed a limitation of the reservoir level with a corresponding
reduction of the its volume to 6.8 million m
3
, about 1/10 of the initial design volume.
The studies, which include a detailed analysis and thorough interpretation of the monitoring
data over a time span of more than 50 years, have allowed to gain insights into the
understanding of the DSGSD behavior and its interaction with the dam.
To interpret the already experienced effects of the bank movements against the dam and to
forecast the future possible trends, the numerical modeling activities have played a key role.
The solution chosen to guarantee a long-term safety operation foresees the demolition of the
upper part of the dam in order to drastically reduce the cracking pattern of the dam body
caused by the compression of the vault for the DSGSD movement.
Introduction
The Beauregard dam, located in the Aosta Valley, Italy (Figure 1) was completed in 1958.
Operated by the Italian Electric Energy Company (ENEL) up to July 2001, the entire scheme
was then acquired by CVA (Compagnia Valdostana delle Acque).


Figure 1: Location of the Beauregard dam on the Italian territory and aerial view of the dam

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
360
The 132m high arch-gravity dam has a crest length of 408m, it is 45.6m thick at the
foundation level and 5m thick at the crest level. With a maximum operating design level of
1770m asl, the total reservoir volume was estimated in 70 mill. m
3
. The filling of the reservoir
was undertaken in stages between 1958 and 1968.
Since the dam construction, the monitoring system installed on the left slope abutment
showed the presence of a clear relationship between the reservoir level and the rate of
movement of the Deep Seated Gravitational Slope Deformation (DSGSD) located on this left
slope (Figures 2, 3, and 4).


























Figure 2: Displacement vectors measured from surface targets
Figure 3: Horizontal displacement distribution
along the plumb-lines installed in the lower
portion of the slope close to the dam
Figure 4: Zones of maximum shear strain
rate: overall slope (safety factor = 1.40-
1.45)
Coherently with the left bank movement, the dam was observed to deflect upstream, due to
the trust of the slope against the vault, having as a consequence the appearance of cracks on
the downstream face (Figure 5). As a consequence, in 1969 the operational reservoir level was
lowered down to 1710m asl, corresponding to a reservoir volume of 6.8 mill. m3, as enforced
by the Italian Dam Authorities.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
361

Figure 5: The dam was observed to deflect upstream with cracks appearing on the downstream
face and openings of some upstream vertical joints
The continuous monitoring of the DSGSD and of the dam, carried out by ENEL, first, and
by CVA, in recent years, has allowed to operate the hydroelectric power scheme under
closely controlled conditions. Moreover, some works carried out to control the superficial
water flows due to melting snow and rain, in order to limit their filtration into the
sliding body, have allowed to reduce the yearly speed of the slope from 1-2 cm to few
mm. However, the interaction between the DSGSD and the dam has been recognized to
have relevant effects on the long-term dam behavior as the cracking pattern was continuously
progressing in time. Figure 6 shows the results of an on-site investigation carried out with a
tomographic system that put into evidence the areas of the dam where cracking is
particularly important, i.e. along the peripheral joint and close to the downstream dam toe.








Figure 6: P-wave velocity tomogram: (left) downstream face; (right) main cross section
The strong interaction between the DSGSD and the dam has been recognized to have relevant
possible implications from the civil protection as well as energy production points of view,
and posed important territorial and environmental issues. For this reason the Owner decided
to start a further study, making also reference to the support offered by numerical modeling,
with the aim to find the most suitable and long-term solution of the problem.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
362
The contribution of the numerical modelling
Synthetically, the main aims of the numerical model were:
- The interpretation of the dam behavior experienced since its first fillings,
- The calibration of the mechanical parameters of the dam-rock system in terms of
comparison between the observed dam behavior and model results,
- The forecast of the future dam behavior at short-middle term adopting the calibrated
model,
- The support to the designer for the analysis of possible technical solutions to
guarantee the long term operation of the dam.

The numerical simulations of the dam were carried out using the FEM code
ABAQUS/Standard [1]. A three-dimensional finite element model of the concrete arch dam
including a proper portion of the surrounding rock mass foundation was generated in order to
carry out numerical analyses (Figure 7).
In the rock mass foundation the active portion of the DSGSD has been taken into account as
well as the shear surface, modeled in terms of an interface whose behavior is described by a
frictional law. The nodes where the DSGSD movements were imposed are outlined with a red
point in Figure 7.























Figure 7: Finite Element mesh of the dam and the surrounding rock mass
A linear elastic isotropic constitutive law was assumed for the rock mass foundation whose
mechanical parameters were defined on the basis of the results provided by the more recent
geotechnical studies.
The Concrete Damaged Plasticity (CDP) constitutive law, available in ABAQUS library, was
adopted to describe the concrete behavior (Figure 8). The model assumes that the uniaxial
tensile and compressive responses of concrete are characterized by damaged plasticity. Under
uniaxial tension the stress-strain response is linear-elastic until the failure value is attained;
beyond this value a strain softening relationship follows. In compression the softening
behavior occurs after an initial stress hardening. On the basis of the results of the indirect
tensile tests, two different sets of material parameters were assigned to the concrete of the
dam body and the "pulvino".

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
363

Figure 8: The Concrete Damage Plasticity constitutive law (Fenves, 1998) has been assumed
for concrete and pulvino of the dam
Several preliminary analyses were carried out to identify the numerical model (calibration
phase) in order to reproduce the real dam behavior dealing with the interaction between the
structure and the slope sliding.
The movement of the left slope abutment was imposed by appropriate displacement boundary
conditions applied at the nodes marked in red (as explained above, Figure 7) and the
hydrostatic load was firstly considered taking into account the variation relevant to the first
fillings; later, a constant reservoir level at 1710 m asl was assumed, according to the present
operational conditions.
The identification process was mainly based on:
- The measured displacements along the dam crest (Figure 9);
- The monitoring points of two plumb lines, located inside the dam body in the main
cross section and in the fourth block towards the left abutment respectively;
- The crack pattern of the structure as determined by tomographic investigations (Figure
10, to be compared with Figures 5 and 6).

As shown in Figures 14-15, the identified numerical model exhibits a good agreement with
the deformation of the dam crest as well as the.



Figure 9: Deformation of the dam crest: (left) based on monitoring; (right) based on
computations

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
364

Figure 10: Localisation of damage computed with the 3D numerical model: (left) on the
dowstream face of the dam; (right) on the main cross section
Afterwards, the identified numerical model was used to predict the future behavior of the
structure considering a continuous increase of the landslide movement according to the trend
foreseen from the past behavior.
The main purpose of the analyses was to assess if the dam might undergo to local or global
instability phenomena such as snap-back.
The curve nr. 1 in Figure 11 allows to exclude snap-back instability for the overall structure
because beyond the peak value the reaction force on the main cross section of the dam
decreases gradually as the landslide movement increases.
Although the measured displacement towards the upstream direction of the middle point of
the dam crest (about 0.21 m) is close to the peak value of the reaction force curve, it has to be
bear in mind that the structure is subjected to an imposed deformation rate due to the landslide
movement. For this reason the structure will be able to follow the softening branch of the
reaction force curve avoiding sudden failure when the peak value will be attained.























Figure 11: Reaction force on the main cross section of the dam vs the upstream-downstream
displacement of the middle point of the crest dam

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
365
The dam rehabilitation
Different possible long-term solutions were considered by the designer (Studio Marcello,
Milan) who was in charge of the rehabilitation program.
Numerical modeling was widely adopted in order to analyze the different solutions which
main aim was the reduction of the trust effect of rock sliding against the dam.
Among the different solutions, the two more widely analyzed were those shown in Fig. 12.

Figure 12: Two of the possible long-term solutions adopted for the rehabilitation of
Beauregard dam
The partial demolition solution (right picture of Figure 12) was the one finally decided by the
rehabilitation designer. The demolition of the upper part of the dam will lower the crest
elevation from 1772m to 1720m asl.
The estimated volume of the demolished concrete will be about 150,000 m
3
. The demolished
material will be used to fill the two volumes located at the upstream and downstream dam
toes (Figure 13).
Additional works have been carried out in the frame of the rehabilitation interventions, among
which the improvement of the hydraulic scheme with the construction of a gated spillway and
a discharge tunnel.























Figure 13: Cross section of the dam: final configuration after the demolition

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
366
The methodology chosen for the demolition is blasting. The demolition started in the Spring
of 2013 and will end in 2014.
Figures 14 and Figure 15 show some phases of the dam demolition that is in progress.
Just to give some figures about each demolition step, the case shown in Figure 15 is
relevant to an explosion carried out with about 850 ton of explosive located in 120 boreholes
with a diameter of about 8 cm and a length of 6m.
The demolished concrete shown in the picture regards a volume of about 6m high x 7m thick
x 50m wide.



Figure 14: Different sequences of the dam demolition: the dam before the starting of works
(up-rigth); the preparation of one explosion step (up-left and down)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
367


Figure 15: Different sequences of the dam demolition
Conclusion
The case history of Beauregard dam is an emblematic case that has posed to the engineers in
charge of operation and safety very critical choices, considering that the construction of new
dams, and structures in general, represents their main mission.
In the Beauregard case it has been deemed necessary to proceed, as sometimes happen for the
human life, with a sort of a limb amputation in order the save the dam life.
The decision has been difficult and it has been necessary to investigate very deeply and for
long time before all possible aspects affecting the system, deciding which could have been the
most reasonable solution from the engineering point of view.
The choice has been taken considering the different facets of the problem: safety, mainly,
environmental protection, energy production.
Environmental as well as economic and financial aspects are not in the aim of the present
paper and could be widely shown in a future paper. Analogously, also some other technical
aspects relevant to the civil works that are under construction at the downstream toe
and facing to check the presence of possible leakages, will be discussed in further papers.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
368
The present paper has the aim to put into evidence the support that numerical modeling can
give to decision makers in a very complex problem like the one proposed by Beauregard dam.
Acknowledgements
The Authors wish to express their gratitude to Lorenzo Artaz and Morena Colli of CVA who
has given the permission to present the present work.
References
[1] Hibbitt, et al, ABAQUS/Standard Theory Manual, Release 6.6, 2005.
[2] G. Barla et al., The Beauregard dam (Italy) and the deep-seated gravitational deformation
on the left slope. Hydropower 2006, Kunming, China, October 2006.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
369
Earthquake Assessment of Slab and Buttress Dams
H. B. Smith
1,2
and L. Lia
2

1
Norplan AS, Oslo, Norway
2
Department of hydraulic and environmental engineering, NTNU, Trondheim, Norway
E-mail: herman.bjorn.smith@norplan.com
Abstract
This paper outlines work on the examination of the dynamic properties and seismic safety of
slab and buttress dams [1]. The work has been carried out with a linear elastic numerical
model established in the Finite Element Method program Abaqus.

One single dam section in a typical Norwegian slab and buttress dam with heights of 12 and
25 meters has been modeled in regards to varying reservoir water level and lateral bracing.
Abaqus has been used for the frequency analysis and the dynamic time-history analysis.

The natural vibration modes of the dam represent movements in separate directions. Through
a seismic event the greatest response will be represented by the buttress deflection in direction
along the dam axis (axial direction). The resistance towards earthquakes will depend
significantly on individual stability of elements rather than the global stability. In particular,
the tensile stresses occurring in the buttress when deflecting in the axial direction are found to
be a potential failure mechanism.

Providing lateral bracing by struts positively influences the response of the slab and buttress
dam in seismic events. When lateral bracing is provided, the ability of the dam to transfer the
inertial forces to the abutment is important.
Introduction
The American-Norwegian engineer Nils Ambursen developed the slab and buttress dam, the
Ambursen dam, in the early 20th century. Several of these dams were built until the 1970s,
especially in North America and Norway. Figure 1 presents the Hen Dam, a typical
Norwegian slab and buttress dam with a structural height of 12 meters. [2] [3]


Figure 1: The Hen Dam, view of downstream side with insulating wall

The reinforced concrete structure consists of the buttresses and an upstream inclined slab. The
typical section and dimensions of Norwegian slab and buttress dams are presented in Figure
2. Insulating walls and lateral bracing, such as struts, are not always present. The upstream
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
370
slab is often constructed as continuous over two or three sections, with joints where the
moment is zero. The resulting moment diagrams are as presented in Figure 3.
Earthquake safety regulations were
introduced for Norwegian dams in 2010.
Due to the recently updated regulations
and the low number of slab and buttress
dams in earthquake prone regions, there
is uncertainty regarding the dynamic
assessment of existing slab and buttress
dams in Norway.

The slab and buttress dam is proven to
be highly efficient for static loading.
Based on the experiences from
registered dam breaches, earthquake
incidents and upgrades, the slab and
buttress dam needs to be checked for
earthquake loads in both horizontal
directions. Structural slenderness of
buttresses and slabs makes the structure
particularly vulnerable to axial loading.
Load distribution and interaction between
the various elements could cause the
strength and stability of individual
elements to be more critical than the
global stability. When dynamic loads are
considered, several possible failure
mechanisms are identified.
Finite Element Model
Based on a literature review, existing computational methods and site inspections of slab and
buttress dams, a linear elastic numerical model was established in the Finite Element Method
(FEM) program Abaqus. The importance of various structural details in a seismic context has
been analysed and utilised for the modeling of a single dam section in typical Norwegian slab
and buttress dams with heights of 12 and
25 meters. Two models have been used for
the analysis regarding height variation,
reservoir water level and presence of
lateral bracing. Dynamic properties, global
response and impact on the structural
elements have been analysed through
eigenfrequency analysis and dynamic
response-history analysis. For convenience
the results presented in this paper focus on
the base case of a 25 meters high dam
section.

The dam section was modeled, as
presented in Figure 4, as a single buttress
and continuous slab with symmetrical
Figure 2: Typical cross section [4]
Figure 3: Moment distribution in continuous
slabs [4]
Figure 4: Illustration of the model and generated
mesh
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
371
boundary conditions in the slabs axial direction at the mid-span.
The dimensions of the modeled section are:

- 5 meters c/c between buttresses
- Crest width of 2.5 meters
- 5V:4H inclination of the upstream slab
- 5V:1H downstream inclination of the buttress
- Both the slab and buttress thickness increases 20 mm per meter from t = 300 mm, at
the dam crest

The model was partitioned in order to provide nodes in the locations of all lateral bracing. The
generated mesh gave a total of 2676 shell elements and 2863 nodes.

The foundation rock is expected to be of good quality and is therefore idealised with no mass
and with great stiffness compared to the slender slab and buttresses. The slab and buttress
dams are normally designed without transmission of reinforcement to the foundation or
between the structural elements. The joint between slab and buttress is designed with a shear
key to avoid displacement in the axial direction. As these joints will allow some rotation, all
connections are modeled as hinged.
Lateral Bracing
When present, struts and insulating walls will provide
lateral bracing for the buttresses in addition to the upstream
slab. Insulation walls are introduced to reduce the
temperature gradient through the slab. With typical
thickness of t = 120 mm lightly reinforced, low concrete
quality, the actual contribution of the insulation wall in a
seismic event is not included. Only the effect of lateral
bracing on the dynamic properties of the slab and buttress
dams by struts is therefore investigated.

Struts are normally simply supported between the
buttresses, i.e. only compression forces are transferred
through the struts. The struts are modeled by massless
elastic springs that ensure the stiffening effect in the axial
direction. Working both in compression and tension, the springs were modeled only on one
side of the buttress in the chosen locations presented on Figure 5.
Loads
The applied static loading includes gravity and hydrostatic water pressure. Uplift, sediment
loads and downstream water pressure are neglected, in addition to temperature loads.

Seismic loads are applied according to Eurocode 8 [5] regulations. This gave a maximum
peak ground acceleration for mainland Norway of 1.6 m/s
2
horizontally and 1.0 m/s
2

vertically. The applied acceleration time histories, presented in Figure 6, are developed for
ground type A and adjusted for Norwegian geological and seismic conditions.

The acceleration time histories are applied along the foundation in all three directions
simultaneously. The acceleration time history in the axial direction is also applied along the
edges of the slab.
Figure 5: Selected placement of
struts
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
372
Because of the upstream inclination, the horizontal hydrodynamic pressure is simulated
according to Zangars theory of added mass [7]. The overlaying masses of water are assumed
to be directly accelerated by the dams vertical accelerations [8]. The applied hydrodynamic
added mass was decomposed from the horizontal and vertical contributions to work
perpendicular to the upstream slab, and applied as inertia point masses at different levels,
according to the presented distribution from Figure 7. Because of the flat slab, no
hydrodynamic interaction was taken into account in the axial direction.



a) Applied in the upstream-downstream direction
b) Applied in the axial direction
c) Applied in the vertical direction
Figure 6: Utilized acceleration time histories [6]
Figure 7: Distribution of applied hydrodynamic added mass
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
373
Material Properties, Strengths and Prescribed Damping
The concrete is modeled with elastic properties corresponding to a B25 concrete, described in
Eurocode 2 [9]. The struts are modeled as elastic springs, with corresponding stiffness
k = EA/L, acting both in tension and compression.

The concrete strength was determined in accordance with Eurocode 2 [9] regulations, adjusted
for a 50 % increased dynamic tensile strength, according to Raphael [10]. The struts capacity
was determined by calculations of buckling and compression capacity.

Viscous Rayleigh damping, embedded in the concrete material properties and the
hydrodynamic point masses, are used to model all damping in the system, including dam-
water-foundation interaction and frictional dissipation. The additional stiffness proportional
damping from the struts was not taken into account. Because the model represents the
structures elastic behaviour, a damping ratio of two percent was considered appropriate. It
should be noted that an increased damping ratio would reduce the systems response.
Results
Frequency Analysis
The frequency analysis described the structures natural vibration modes with movements in
separate directions. In all analysed cases, the first mode is governing for the axial direction.
Figure 8 describes the vibration mode shape and variation of period for oscillations in the slab
and buttress dam section with water at crest level for different section heights and with or
without lateral bracing.

Figure 8: The first vibration mode a) Mode shape without lateral bracing b) Mode shape with
struts present c) Variation of the vibration period
The first mode describing movement in the upstream-downstream and vertical direction was
also governing in these directions for all cases. Figure 9 describes the mode shape and
variation of period for the slab and buttress dam section with or without presence of water at
crest level, for different section heights without lateral bracing.
a) b)
c)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
374

Figure 9: The first vibration mode describing movement in the upstream-downstream
direction a) Mode shape without water b) Mode shape with water at crest level c) Variation of
the vibration period
Dynamic Response-History Analysis
Through the seismic event, the models were never exposed to compressive stresses close to
the concrete design capacity. The maximum principal stresses are therefore presented. It is
assumed that cracking of concrete will occur when the design dynamic tensile strength
f
ctd,dyn
= 1,5 MPa, is exceeded, marked by gray in the graphic presentations.

The typical deflected shape and maximum principal stress state during a seismic event are
presented in Figure 10, for the 25 meters high dam section with water at crest level and with
or without lateral bracing in the form of struts.

a) Without lateral bracing b) With lateral bracing

Figure 10: Typical deflected shape during seismic event, deformation scale factor of 100

a) b) c)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
375
The maximum principal stresses on each side of the buttress with water at crest level, are
presented on Figure 11 without lateral bracing and on Figure 12 with struts.


Figure 11: Maximal principal stresses on both sides of the buttress without lateral bracing
during a seismic event

Figure 12: Maximal principal stresses on both side of the buttress with lateral bracing during a
seismic event.
The time-history plot of transferred inertia forces to the highest loaded strut in the dam section
of 25 meters height is presented in Figure 13. Representing both the struts on each side of the
buttress, the plot describes the absolute value of the forces through the given elastic spring.

Figure 13: Time-history plot of the highest loaded strut in the 25 meters high dam section.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
376
Discussion
Dynamic Properties
The distinct differences in the structures vibration modes in different directions, combined
with the variation of the vibration period, leads to the following observations:

Dam height An increase in the section height results in a greater mass while the
dam stiffness decreases both in the axial and the upstream-
downstream direction. Thus, a greater vibration period in all
directions is obtained.
Water level At higher water levels, the additional hydrodynamic mass increases
the vibration period in the upstream-downstream direction.
Lateral bracing Increased stiffness in the axial direction due to lateral bracing
reduces the vibration period in the same direction.
Seismic Safety
Dam safety requirements for slab and buttress dams include overturning and sliding stability
in addition to the required structural strength for all cross sections.

The dam safety should be considered in a seismic event and retain its integrity during the
earthquake. The seismic safety has been evaluated based on global stability and occurring
stress levels in the buttress. The struts are additionally considered. As the upstream slab is
designed for tensile stresses, it is not further considered.

Global Stability
The maximum fluctuations of the reaction forces acting in different directions on the dams
sections would, in combination with one another, constitute a danger to the system's global
stability. Nevertheless, the maximum fluctuations occur at various times throughout the event
and are only applicable in very short time periods. The system will not have time to react, and
there will be no danger of collapse even if the theoretical factor of safety against sliding or
overturning in a given time goes below 1.0. The consequences may be minor deformations if
exceedance of capacity is repeated.

The theoretical safety factor against sliding through the seismic event for the 25 meters high
dam section with water at crest level is presented in Figure 14.


Figure 14: Variation in theoretical safety factor against sliding for a 25 meters high dam
section with water at crest level.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
377
Also considering the rest capacity when assuming a frictional factor of tan| = 1, the risk of
sliding is not the critical case. Nevertheless, the theoretical safety factor against sliding for
slab and buttress dams is reduced as the dam height increases.

The upstream inclination keeps the dam stable against overturning throughout the seismic
event, as tensile stresses never occur along the buttress foundation.

Stresses in the Buttresses
A lower section height reduces both the response and stress level of the buttress. The presence
of lateral bracing prevents the buttress deflection out of its own plane, and largely reduces the
stresses. When struts are present, the buttress is bent around its connection points, and the
maximum stresses are found in those locations.

The occurring tensile stresses in the numerical model have to be seen in relation with the
presence of reinforcement. Before 1955 it was common to construct the slab and buttress
dams without reinforcement in the buttresses [11]. In addition to tensile stresses, construction
joints would endanger the buttress stability in such a case.

For dams constructed later than 1955, a minimum amount of reinforcement will always be
present. Even though tensile stresses can cause cracking of the concrete, the strength and
ductile nature of reinforcement will nevertheless increase the safety level compared to an
unreinforced buttress.

Struts
The impact on the struts from a single section must be seen in relation to the struts
compression capacity, found to be dimensioning over the buckling capacity. The transferred
inertia forces are accumulated towards the abutments through the struts from neighbouring
sections. When beams are not continuous over the dam length, unfavourable moments can be
introduced into the buttresses.

Other Potential Failure Mechanisms
The following potential failure mechanisms should also be taken into consideration; the
connections between elements, shear capacity and punching shear capacity of the buttresses,
stress accumulation around cutouts, strength of the foundation and bearing capacity of the
struts.
Uncertainties Sources of Error
Because only one single dam section has been modeled, several assumptions have been
required. Among others, the influence of adjacent sections, especially of different heights, is
of importance.

As the results show a non-linear response and because reinforcement is not taken into
account, the response and maximum occurring tensile stresses in the numerical model can
only be indicative.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
378
Conclusions
The results demonstrate how the dam vibration modes represent movements in separate
directions. An increased dam section height results in an increased response because of the
greater oscillating mass. Meanwhile, a raised water level in the reservoir increases the period
for oscillation in the upstream-downstream direction.

Through a seismic event, the greatest response will be represented by the buttress deflection
in the axial direction. The resistance towards earthquakes will be mainly dependent on the
individual stability of elements rather than the global stability. In particular, the tensile
stresses occurring in the buttress, as a result of deflection in the axial direction, is found to be
a potential failure mechanism.

Providing lateral bracing by struts positively influences the response of the slab and buttress
dam in seismic events. For the Norwegian slab and buttress dams in particular, the
contribution of the existing insulating walls is considered to be small but positive. When
lateral bracing is provided, the ability to transfer the inertial forces to the abutment is
important.

In addition to the assessment of simplified analytical methods, it is recommended that further
numerical studies should focus on the slab and buttress dams nonlinear response, on the
importance of global dam geometry, and verification of the numerical input parameters by
physical vibration tests of an existing slab and buttress dam.
References
[1] Smith. Earthquake Assessment of Slab and Buttress Dams, Master thesis at the
Department of Hydraulic and Environmental Engineering, NTNU, Trondheim, Norway,
2013.
[2] Various authors. Ambursen dams - Selected articles from the Ambursen Hydraulic
Construction Company. R. M. Rudolph, Approximate Date 1912.
[3] NORUT Narvik AS. Innovativ forvaltning av betongdammer, 2009. In Norwegian.
[4] Guttormsen. TVM4165 - Vannkraftverk og vassdragsteknikk. Tapir akademiske forlag,
2006. Compendium of the Department of hydraulic and environmental engineering,
NTNU. In Norwegian.
[5] Standards Norway. Eurocode 8: Design of structures for earthquake resistance - Part 1:
General rules, seismic actions and rules for buildings, 2004. NS-EN 1998-
1:2004+NA:2008.
[6] Kaynia. Communicated by. NTNU and NGI, Oslo, Norway.
[7] Zangar. Hydrodynamic pressures on dams due to horizontal earthquake effects.
Engineering monographs, No. 11, 1952.
[8] Scott et al. Dam Safety Risk Analysis, Best Practices Training Manual. U.S. Department
of the Interior, Bureau of Reclamation in cooperation with the U.S. Army Corps of
Engineers, 2011. Version 2.2.
[9] Standards Norway. Eurocode 2: Concrete structure - Part 1-1: General rules and rules for
buildings, 2004. NS-EN 1992-1-1:2004+NA:2008.
[10] Raphael. Tensile strength of concrete. Journal of the American Concrete Institute, 2:158
165, 1984.
[11] Kleivan, Kummeneje and Lyngra. Concrete in Hydropower Structures. Hydropower
Development. Norwegian Institute of Technology, Division of Hydraulic Engineering,
1994.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
379
Solution of dam-fluid interaction using
ADAD-IZIIS software
V. Mircevska
1
, M. Garevski
1
, I. Bulajic
2
and S. Schnabl
3

1
Institute of Earthquake Engineering and Engineering Seismology,
univ. Ss. Cyril and Methodius, Box 101, 1000 Skopje, R.Macedonia
2
Mining institute, Batajnicki put 2, 11000 Belgrade, Serbia
3
Faculty of Civil and Geodetic Engineering, University of Ljubljana,
1000 Ljubljana, Slovenia
E-mail: violeta@pluto.iziis.ukim.edu.mk
Abstract
Fluid dam interaction has a remarkable impact on the dynamic response of dams and could
play an important role in assessment of their dynamic stability. This is particularly
emphasized when dams are subjected to strong seismic excitations. The phenomenon has been
for the first time physically explained and mathematically solved by Westargaard. The very
first dynamic analysis based on application of Added Mass Concept; underestimated the
random earthquake nature in assessment of the hydrodynamic effects. In the recent years,
various BEM-FEM and FEM-FEM techniques have been developed to account for many
significant parameters that influence the accuracy of calculated hydrodynamic effects. This
paper presents a BEM-FEM orientated solution based on the use of the matrix of
hydrodynamic influence as a very effective tool for analyses of extensive domains of fluid-
dam-foundation rock systems for two major reasons: the computation time is far more
effective than that in direct or iterative coupling methods and stability of the solution. The
presented analyses are based on the use of genuine software originally written for static and
dynamic analysis and design of arch dams.
Introduction
The ADAD-ver.3 computer program [1], originally written for static and dynamic analysis
and design of arch dams, is under development for the last several years. It implements an
analytical procedure for the three-dimensional dynamic analysis of arch dams including the
effects of dam-water interaction (water incompressibility), soil-structure interaction and the
nonlinear behavior of the of contraction joint manifested by partial joint opening and closing
as well as tangential displacement. The process of generation of mathematical model runs
parallel and interactively with the process of design of the particular dam. The program gives
an option for computer design of the dam body [2,3], whose embedment is in accordance with
topology of the terrain. Program offers automatic pre-processing for generation of finite
element mesh of dam and part of the foundation mass to account for the dam-foundation
interaction phenomenon, as well as effective way of generation of boundary element mesh,
sufficiently accurate in following the topology of the terrain to account for the fluid structure
interaction.
The program use sensitivity search analysis to detect the most adequate location of the
truncation surface [4], where non-reflecting truncation boundary conditions should be
imposed. The truncation surface should be located in a way to define the required
completeness of the wave field where expansion of P compressive and dilatational waves
takes place followed by scattering and radiational effects. Its further displacement away from
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
380
the dam should have a negligible impact on the calculated magnitude of the hydrodynamic
effects.
The conducted analyses are based on an original and simple FEMBEM fluidstructure
interaction solution embedded in the ADADIZIIS software. This solution eliminates the
difficulties of direct and iterative coupling methods by analyzing independently the two
physically coupled sub-domains. The interaction effect is obtained in an uncoupled way
computing the matrix of hydrodynamic influence by applying the concept of virtual work of
unit accelerations. The suggested method does not belong to direct coupling or to iterative
coupling methods, yet with its computational steps it offers a two-way coupling by
transferring the fluid forces to the structure and the structural accelerations back to the fluid.
The paper presents BEM-FEM oriented solution of the fluid-dam interaction along with the
boundary element discretization of the reservoir domain. The dam was subjected to El Centro
earthquake excitation with duration of 7 sec., scaled to the pick acceleration of 0.3g. The dam
properties are the following: Dam height H=130m; Youngs modulus E = 31.5 GPa; mass
density = 2450 kg/m
3
; Poissons ratio = 0.2; the acoustic wave velocity in water c = 1440
m/s.
Numerical Model of Arch Dam and Fluid Domain
The process of generation of the discrete mathematical model of the arch dam and the fluid
domain runs parallel with the process of design of the dam body. This process starts by
digitization of topographic data of the terrain and the shape of the main central cantilever,
Figure 1. The developed pre-processing within the ADADIZIIS software enables each
topographic isoline to be mathematically presented by a set of equations of a second order,
i.e., curves passing through three neighboring digitalized points on it. Each isoline is stored
in the computer by means of a certain number of polynomials. The program gives an option
for computer design of the dam body [2,3], whose embedment is in accordance with topology
of the terrain. Arch dam body can be modeled in a form of few circular segments as well as in
the form of a parabola. During modeling, it is possible to observe the shape of the arches at all
elevations along with the corresponding tables containing their geometrical parameters and to
observe their mutual position in order to control overlapping. Figure 2, presents the
mathematical model of the dam generated automatically, using adopted shapes of the arcs at
all selected elevations in accordance with the topology of terrain. The model is formed by 199
substructures. The substructures are automatically digitized into a certain number of finite
elements that are not presented in the figure. The model contains 6294 finite elements and
11170 external nodes. The bend of the system is 3111. The model posses 2013 contact
elements that are involved at the connections between substructures blocks. The contact
elements are generated automatically and the contact element mesh refinement is in
accordance of the model analyst request.
ADAD-IZIIS software offers a very efficient and accurate modeling of the 3D fluid
boundaries according to the downstream topology of the canyon terrain. The boundary
elements at the extrados of the arch dam are directly extracted from the general FE model
while the boundary element mesh that represents the boundaries of the reservoir, i.e., both
banks, the reservoir bottom, the water mirror and the reservoir end, very accurately follows
the shape of the topographic isolines. Generation is simply by giving the number of planes
that intersect the terrain along with their distances from the uppermost point of the crown
cantilever. In case of highly irregular and twisted terrain, more section planes and a more
refined mesh should be used in order to model the complexity of the terrain in the most
accurate way, which is undoubtedly easily feasible and with shorter computational time if
BEM technique is engaged. However, for the concrete configuration of the terrain and
detected most adequate location of the truncation surface situated at the downstream
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
381
distance of 210m from the dam, the number of boundary elements used in the model for
accurate modeling of the 3D fluid boundaries is 1600, figure 3.

Figure 1: a) Topology of the terrain b) Shape of the central cantilever

Figure 2: Mathematical model of a dam (substructures and construction joints)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
382
Figure 3: Boundary element mesh of the fluid domain

Fluid-structure interaction is affected by the irregularity of the terrain in the near surrounding
of the dam-fluid interface. The topology of the terrain dictates the most adequate location
of the truncation surface where non-reflecting truncation boundary conditions should be
imposed. The detection of the most adequate location of the truncation surface is an
important task in development of a reservoir model due to the fact that the HDP intensities on
the dam-fluid interface are sensitive to the extent and type of waves generated by the
boundaries. The truncation surface should be located in a way that its further displacement
away from the dam has a negligible impact on the calculated HDP intensities.
The program offers sensitivity search analysis to detect the most adequate location of the
truncation surface. For the presented BE model in figure 3., the elapsed CPU time for
performing such analysis is approximately 15 min. The numerical procedure is based on the
conventional BEM. Laplace differential equation that governs the incompressible and inviscid
fluid motion is used. The procedure is conducted over a rigid dam-canyon-walls assemble. It
follows a horizontal acceleration of 1g applied in downstream direction. The acoustic elastic P
waves were generated as a result of the vibration of the considered upstream dam face and the
rigid canyons walls. The expansion of generated waves and the way of their propagation as
compressive or dilatation waves depends not only on the specified boundary conditions but on
the shape of the reservoir boundaries in respect to the direction of the seismic excitation.
Three different types of truncation boundary conditions were considered: a) stationary type of
truncated boundary conditions, i.e., perpendicular acceleration at all the points on the
truncation surface is set to zero; b) hydrodynamic pressure at all points on the truncation
surface set to zero; and c) non-reflecting boundary condition. Sixteen different location of the
truncation surface were considered and analyzed. The curve associated with the TBC that
allows dissipation of the outgoing waves shows mostly decreasing trend until reaching the
meeting point of the curves of TBC type (a) and TBC type (b), at distance of L=180 m
(L=1.6-1.7Hw) away from the dam. Further on, this curve remains almost horizontal,
indicating unaltered value of HDP. This means that it is irrelevant whether TS is positioned
at a greater distance than L=180m, since the effect of the amount and type of generated waves
is negligible with further increase of the model length. However, this effect is remarkable
along length L<180m, wherefore placing TS closer to the dam means overestimation of the
HDP. In accordance with the applied direction of ground acceleration and due to the irregular
configuration of the terrain in the vicinity of the dam, the right bank generates mostly
dilatation waves while the left bank generates compressive waves. For the concrete
topological conditions, the considered depth of impounded water, according the results
presented in [4], the most adequate location of the truncation is selected at a distance of
L=1.9Hw=210m.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
383

Figure 4: Variation of the normalized hydrodynamic pressure magnitudes as a function of the
considered 16 locations of the truncated surface and different truncation boundary conditions
(selected node at the bottom and at the middle of the crown cantilever)
BEM-FEM Solution of Fluid-Dam Interaction
Fluidstructure interaction is the interaction of a moveable and/or deformable structure that is
immersed in a fluid and/or contains a fluid. A model that captures such an interaction must
use two-way coupling model, where the fluid motion affects the structures motion and the
structures motion affects the fluids motion. Coupling should provide compatible link of both
media which means equilibrium and correct transition of the physical variables at the
interface. There exist various algorithms for coupling the merits of both BEM and FEM
numerical methods, direct [5-8] and iterative coupling methods [9-12]. ADAD-IZIIS software
[1] is based on BEM-FEM oriented solution of the coupled structure and the incompressible
and inviscid fluid. The solution of the coupled system is accomplished by calculating in
advance the matrix of hydrodynamic influence utilizing the concept of virtual work of unit
accelerations. This matrix is stored in the system and recalled in any time step of the
dynamic response of the dam. Hence the solution of the coupled systems is actually separated
and mutually independent. Hydrodynamic forces are obtained as a product of the matrix of
hydrodynamic influence and the vector of manifested total accelerations along the normal at
any interface node. The interaction effect at the fluidsolid interface is enforced by adding the
matrix of hydrodynamic forces to the classic equation of dynamic motion of the dam, eq. (3).
The governing equation for solving the small amplitude irrotational motion of the impounded
incompressible and inviscid fluid is governed by the three-dimensional Laplaces equation as
follows:
0
2
2
2
2
2
2
=
c
c
+
c
c
+
c
c
z
W
y
W
x
W

(1)
where W(x,y,z) is a function of the potential in the fluid domain. The equation (1) has to be
amended by the specified essential and natural type of boundary conditions that exists at
the boundaries of the analyzed fluid domain. Applying BEM technique, Brebbia [13], the
discretization of boundary surfaces is by an assemble of eight nodded quadratic Serendipity
type of boundary elements as follows:

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
384
0 ) ( ) (
2
1
2
2
1
1
1
1
= I
c
c

c
c
+ I
c
c

c
c
+
I
=
I
=
}

d
n
p
W p
n
W
d
n
p
W p
n
W
W
NEL
nel
NEL
nel
i

(2)

where: i=1,NBE; NBE is a number of nodes in the boundary ele ment model.
The differential equation of motion of a discrete system written in an incremental form for the
"i"-th time increment is the following:

(3)


where:
i
F
HD
A

is a vector of hydrodynamic force increment and


i
P A

is a vector of seismic
force increment.
In eq. (3), the vector of hydrodynamic force increment eq, 4 is calculated by use of previously
defined matrix of hydrodynamic influence and directly added to the vector of seismic force
increment.

| |
tot
nj ij
HD
a W
i
F A = A

NPT j i , 1 , =
(4)
where: | |
ij
W is a matrix of hydrodynamic influence;
tot
nj
a A is the absolute acceleration along
the boundary element normal; NPT is the total number of nodes at dam-fluid, bottom-fluid
and banks-fluid interfaces, where the natural type of boundary condition exists
Figure 5, presents the time history response of relative displacement velocity and acceleration
for the selected node at the dam crest, top of the crown cantilever, where the extremes of the
response occurred. Obviously, fluid structure interaction based on BEM-FEM numerical
solution modifies the extreme of the response acceleration at the dam crest by 38% in respect
to the dam responce with empty reservoir. Additionally calculated are hydrodunamic effects
by use Westergaard added mass concept. It gives lower modification of the dam responce, i.e,
the exterem of the relative responce acceleration at the dam crest is modified by 31% Nothe
that, the time of extreme occurance is not coinside. Westergaard added mass concept do not
give recogition to the impact of the dam flexibility on the amount of generated energy in the
fluid domain and therefore on the intensity of the manifested FSI effects. The flexibility
property of the dam and the influence of the reservoir domain alter the behavior of the fluid
significantly and consequently the coupled system has a stronger response.
The figure 6 shows the izolines of distribution of the principal stress G3 that acts along the
arches with and without included hydrodynamic effects, whereat hydrodynamic effects are
calculated according to added mass method and coupled BE-FE method. The stress extreme is
increased by 15% if added mass method is used and 49% if coupled BE-FE method is used.
Despite the influence of the terrain irregularities on the amount of energy transferred to the
fluid domain, this effect so far, has not been analyzed in detail by exsisting software
regardless whether boundary element method (BEM) or finite element method (FEM) is used
for fluid discretization. ADAD-IZZIS software gives an opption for taking into account the
influence of terrain irregularities on the magnitide of calculated hydrodymanic effecs.
Figure 7., presents a snapshot of hydrodynamic pressure distribution over the interface, at
time T=4.95sec. It is obtained under the assumptions that the topology of the canyon has a
regular shape as indicated in the drawing.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
385

Figure 5. Modification of the dam response at the dam crest, relative displacement in (m) ;
relative velocity in (m/sec) and relative acceleration in (m/sec
2
). FSI effects defined by use of
BEM-FEM solution and Westergaard added mass method


a)
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
386

b)



c)
Figure 6: Distribution of principal stress G3 at the extrados face at t=2.42 s
a) empty reservoir b) FSI using BEM-FEM regular terrain
c) FSI using added mass method

Figure 7: Snapshot of hydrodynamic pressure distribution at time T=4.95 sec
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
387
Conclusion
ADAD-IZIIS software is based on BEM-FEM oriented solution of the coupled structure and
incompressible and inviscid fluid domain. The software offers process of generation of the
mathematical model that runs parallel and interactively with the process of design of the arch
dam body. Automatically are generated the finite element mesh of the dam and part of the
foundation mass for accounting the phenomena of dam-foundation interaction, and boundary
element mesh that presents the boundaries of the fluid domain for accounting the fluid
structure interaction. The solution of the coupled system is accomplished by use of matrix of
hydrodynamic influence utilizing the concept of virtual work of unit accelerations.
Comparison of the calculated hydrodynamic effects using both BEM-FEM solution and added
mass method is made. The added mass method provides acceptable results only in the range
of Westergaard restricted hypothesis. Since it neglects the dam flexibility and water
compressibility and does not require any discretization of the reservoir domain wherever these
features have an impact on the magnitude of hydrodynamic effects there will be discrepancy
of the obtained resultants.
References
[1] Mircevska V., Bickovski V. (2008). ADAD-IZIIS software: Analysis and Design of Arch
Dams, IZIIS- univ. Ss. Cyril and Methodius, Users Manual.
[2] Houard L., Boogs (1997). Guide for Design of Arch Dams. A Water Resources,
Technical Publication, Engineering Monograph No 36., 1-17.
[3] Wiliam, P., Greager, and Joel D., Justin (1984). Engineering for Dams, John Wiley &
Sons
[4] Mircevska V., Bickovski V., Aleksov I., and Hristovski V. (2013). Influence of irregular
canyon shape on location of truncation surface. Engineering Analysis with Boundary
Elements, Vol. 37, pp. 624636.
[5] Estorff von O., Prabucki M.J. (1880). Dynamic response in the time domain by coupled
boundary and finite elements. Computational Mechanics Vol. 6, pp. 3546.
[6] Pavlatos G.D. Beskos D.E. (1994). Dynamic elastoplastic analysis by BEM/FEM.
Engineering Analysis with Boundary Elements Vol. 14, pp. 5163
[7] Czygan O., Estorff von O. (2002). Fluidstructure interaction by coupling BEM and
nonlinear FEM. Engineering Analysis with Boundary Elements Vol. 26, pp. 773779.
[8] Yu G. Mansur, W.J., Carrer J.A.M., Lie S.T. (2001). A more stable scheme for
BEM/FEM coupling applied to two-dimensional elastodynamics. Computers &
Structures Vol. 79, pp. 811823.
[9] Lin C-C, Lawton EC, Caliendo JA, Anderson LR.(1996). An iterative finite element-
boundary element algorithm. Computers & Structures Vol. 39(5), pp. 899909.
[10] Feng YT, Owen DRJ. (1996). Iterative solution of coupled FE/BE discretization for plate-
foundation interaction problems. International Journal for Numerical Methods in
Engineering Vol. 39(11), pp. 889901.
[11] Soares Jr D, von Estorff O, Mansur WJ. (2004). Iterative coupling of BEM and FEM for
nonlinear dynamic analyses. Computational Mechanics Vol. 34, pp. 6773.
[12] Estorff von O., Hagen C. (2006). Iterative coupling of FEM and BEM in 3D transient
elastodynamics. Engineering Analysis with Boundary Elements Vol. 30, pp. 611622
[13] Brebbia C.A. (1978).The boundary element method for engineers, Pentech Press



ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
389
Influence of Surface Roughness on Sliding Stability
Tests and numerical modeling
. Eltervaag
1
, G. Sas
2
and L. Lia
3
1
Department of Structural Engineering, Faculty of Engineering Science and Technology,
Trondheim, Norway
2
Northern Research Institute, Narvik, Norway
3
Department of Hydraulic and Environmental Engineering, Faculty of Engineering Science
and Technology, Trondheim, Norway
E-mail: oystein.eltervaag@sweco.no
Abstract
Lightweight concrete dams slide when the shear capacity of one or more sliding planes in the
dams structure or foundation is exceeded. Several laboratory shear tests were carried out on
concrete rock samples. The two materials were mated trough teeth-sawed interfaces with
different inclination profile. This paper presents the results of the numerical modeling of those
tests.
The results of the shear tests were compared to the predictions of the model used in the
Norwegian guidelines. It has been found that the model used in the guidelines do not predict
the shear capacity accurately. Through finite element analyses a better representation of the
tests has been achieved, especially regarding the influence of roughness.
Introduction
In Norway, the stability of a dam is reconsidered every 15 to 20 years, depending on the
consequence class for the given dam. In the past 40 - 50 years, since the large hydropower
development poque in Norway, the level for required safety has increased. Consequently
some dams constructed during that period no longer are considered to have the sufficient
safety and expensive upgrading is needed.


Figure 1: Profiles of test samples [1]
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
390
Two of the samples with geometry 1 and 4 were casted with bond between the rock and
concrete. The remaining ten samples had no bond. During casting of concrete a plastic film
was attached on the rock surface of those ten samples to prevent bonding.
Theory
According to the Norwegian regulations [2] the sliding stability is assessed using the shear
friction method, and is expressed by a factor of safety (FS). For a horizontal sliding plane this
factor is found from the following equation.

tan c A V
FS
H
| +
=


(1)
In equation (1) c is the cohesive parameter, A is the shearing plane area, V and H are the
vertical and horizontal forces acting on the plane respectively. The frictional factor of the
plane surfaces is expressed by tan (). Thus, the factor of safety is calculated from the
averaged normal stress and friction factor of the sliding plane. The shear friction method is
based on the Mohr-Coulomb criterion [3] for describing the shear capacity of the sliding
plane. This criterion states that the shear capacity, , is linearly dependent on the applied
normal stress, , trough a material specific frictional parameter tan() plus a cohesive
parameter, c. This can be expressed as follows:

tan c t o | = +
(2)

In the 1960s it was recognized that the failure envelope for rock mass (rocks with joints and
faults) was curved. One of the major contributions to this understanding was when Patton [4]
derived a bi-linear failure criterion from experiments with saw-toothed rock specimens
shown in Figure 2. Patton observed that sliding occurred at lower levels of normal stress than
what was needed to cut off the saw-toothed geometry. This failure mechanism can be
described on the form:
tan( )
b
i t o | = + (3)

Were
b
is the material specific friction angle, simply denoted in the Mohr-Coulomb
relation, equation (2), and i is the angle of the asperities, called the asperity inclination or
dilation angle. When the normal stress exceeded a certain value the saw teeth were cut off at
their base. Patton [4] explained this as a change of governing failure mode, from a sliding
failure along the material interface, to a failure in the material itself. The shear capacity
regarding this material failure is described by:

tan( )
x r
c t o | = + (4)

Where c
x
is the bond of the failing material, and
r
is the residual friction angle of the failure
plane. The shear envelope obtained from these expressions is shown in Figure 2 .
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
391

Figure 2: Patton's bi-linear failure criterion [4]
According to Patton [4] the bi-linear failure envelope illustrates that there are two possible
failure mechanisms for the rock specimens studied. The first mechanism is sliding over the
asperities (saw-teeth) and occurs at low normal stresses. The second mechanism is shearing
through the asperities and occurs at relatively high normal stresses.
These different failure modes have been further described by Johansson [5]. Johansson [5]
developed a conceptual model to describe sliding failure of one idealized quadratic asperity.
Three failure modes were identified; sliding along the side of the asperity facing the load,
shear-failure along the base of the asperity, and tensile failure in the rock-base of the asperity.
A sketch of an idealized asperity is shown in Figure 3.


Figure 3: Principle sketch of an idealized 2D asperity [5]

To describe sliding along the loaded face of the asperity (the left side in Figure 3) Johansson
[5] uses Pattons formulation, eq. (4), for shear capacity for low normal stresses (transformed
from stresses to forces).

tan( )
b
T N i | = + (5)

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
392
The equation for shear failure along the base of the material is derived based on the Mohr-
Coulomb criteria, eq. (2) and Pattons equation for high normal stress, eq. (4).

2
tan( )
i sp i
T c L N | = + (6)

For a tensile failure to occur in the rock-base, the vertical tensile stresses must exceed the
tensile capacity of the rock. To calculate the average tensile stress in the rock Johansson [5]
assesses the moment-equilibrium about point O in Figure 3.

2
(3 4 )
2 tan( )
ci ti asp
L
T
i
o o +
=

(7)

Where
ti
is the tensile strength and
ci
is the compressive strength of the rock.
Method
For this work the finite element software ATENA-Science developed by Cervenka Consulting
has been used [6]. The geometry of tested samples is modeled as a 2D plane stress problem,
left side of Figure 5. To avoid numerical instability when applying the loads directly on
concrete, the surrounding steel shear box was modeled also, right side and medallion in
Figure 4. In Atena the boundary conditions were applied as fixed lines along the outside of
the lower steel frame, the loading was introduced by a vertical line load along the top of the
steel frame, and horizontal displacement along the left edge of the upper steel frame.
Monitoring points were added at the left edge of the upper steel frame and at the right edge of
the concrete material to enable load/displacement curves.

Figure 4: 2D geometrical model of a test sample with ten degree asperity angle
ATENA uses interface elements to model contact between two parts of a model. To model the
behavior of the interface elements ATENA uses the Penalty Method [6]. The physical
properties of these interface elements are governed trough the Mohr-Coulomb criterion
presented in eq. (1). To obtain input values of the interface properties the results from the tests
are compared to the failure criteria described above. However, these classic formulations
present large errors compared to the test results. Reliable input parameters were obtained by
applying a hybrid formulation where the shear capacity is determined from the actual failure
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
393
mechanism for the tests; sliding over or shearing off the asperities, eqs. (5) or (6). This hybrid
formulation for shear capacity of the samples can be expressed as follows,

tan( )
cos (1 tan tan )
b
b
c A
V N i
i i
|
|

= + +


(8)

where i is the residual inclination angle of the asperities, which may differ from the original
inclination angle if the asperities are sheared off. The basic friction angle is material constant
and therefore equal for all the samples. Thus, the basic friction angle can be found from a
sample without asperities, and is determined to be 34.4. The cohesive parameter is found by
inserting the known values from the results of each of the tests into eq. (8). These values are
listed in Table 1. The physical explanation of this parameter is believed to be wear of the
interface as cut-off of micro asperities (micro roughness) as shown in eq. (6). The cut-off
shows a decreasing effect as the samples slide against each other. To account for this a
softening behavior of the cohesive parameter is introduced. In ATENA the effect of the
asperities (macro roughness) is included trough the geometry of the interface.
In ATENA the loading is applied incrementally. First the normal pressure was applied as a
vertical line-load. The line-load for the different tests has been found by distributing the
applied experimental vertical force presented in [1] over the length of the samples (240mm).
The applied line-loads are presented in Table 1.
Table 1: Interface and loading input parameters
Test results [1] Input used in ATENA
Sample i []
Hmax

[mm]
N
[kN]
i []
H

[mm]
Cohesion
[kPa]
Q [kN/m]
1.1 40 15.23 27.52 0 20 1245 114
1.2 40 15.44 46.91 0 20 1720 195
1.4 40 26.34 68.49 0 20 2371 285
2.1 20 24.99 27.57 20 20 282 115
2.2 20 21.98 47.77 20 20 781 199
2.3 20 20.07 68.36 20 20 1552 285
3.1 10 33.16 24.82 10 20 309 103
3.2 10 32.94 45.98 10 20 382 192
3.3 10 33.23 67.65 10 20 536 282
4.2 0 33.23 67.23 0 20 62 280

Then the horizontal force is applied trough prescribed deformation of ten millimeter at the left
edge of the upper steel frame, see Figure 4. To achieve plane horizontal displacement (avoid
rotation) of the upper part master-slave boundary conditions are introduced along the upper
edge in this interval. Interval three and four are identical to interval two, except that the
master-slave boundary conditions are removed. The standard incremental and iterative
NewtonRaphson method for material nonlinear structural analysis was used in the numerical
simulations, based on the finite element method. The specimens were modeled with a mesh of
8-node serendipity plane stress finite elements. A Gaussian integration scheme with 2 x 2
integration points was used for all the concrete elements, [6,7].The mesh is shown in Figure 5.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
394

Figure 5: Mesh of test 2.2
Both the concrete and rock materials are modeled using a fracture-plastic model that
combines constitutive sub-models for tensile and compressive behavior, as presented in the
ATENA user manual [6], see Figure 6. This fracture model employs the Rankine failure
criterion and exponential softening, with the hardening/ softening plasticity component based
on the Mentrey William failure surface [6]. The concrete post-cracking tensile behavior was
simulated by a softening function in combination with the crack band theory [6].

Figure 6: Constitutive model for concrete (left) and softening function (right) [6]
The decision for modeling the rock as concrete is based on the behavior of the actual rock
samples during compressive standardized tests. It was observed that this behavior is more in
line with the available material models for concrete than rock. The steel shear box was
modeled as a linear elastic material. Theoretical background of the above mentioned
constitutive models are given in the ATENA Theory manual [7]. Input parameters for the rock
and concrete material are given in Table 2.


ef
ef
t f
0
E
c
t
c
0
c
eq
c
c
ef
c f
U
c
d
loading
unloading
o
c
G
f
w
w = 5.14
G
f
ct
f
c
c
f
ct
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
395
Table 2: Material parameters for concrete material models
Material Concrete Rock
Youngs Modulus [GPa] 37 100
Poissons Ratio 0.2 0.2
Tensile strength [MPa] 4.1 2.317
Compressive Failure Stress [MPa] 58 280
Fracture Energy [MN/m] 1.03x10
-4
1.05x 10
-5
Plastic strain -0.00147 -0.000296
Onset of crushing [MPa] -8.61 -1.54
Critical compressive displ. [m] -0.0005 -0.0005
Density [kN/m
3
] 23 23

In Table 1it is seen that the geometry of tests 1.1 to 1.4 is modeled with zero inclination
angle. The reason for this is that these tests showed a shearing failure mode, where the
asperities were cut off, therefore there were no sliding between asperities but material failure.
A total shearing failure mode has been hard to obtain from the numerical models. The main
reason for this is that the analyses crash after the material starts to crack due to instability
problems with zero or negative Jacobian for the stiffness matrix. This error message is an
indication of an ill-conditioned system of equations [8]. Perhaps a more refined mesh with
more integration points could have been used. Due to low computational capacities this was
not possible at the time when this work was carried out. In Figure 7 the concrete material
failure is shown. The blue field in the left part indicates that the tensile strength of the
material is reached.


Figure 7: Material failure for test 1.4.
Due to the numerical instability problems only the peak shear capacity was obtained from
these analyses. However, this capacity is governed by the concrete material parameters alone,
not by the interface parameters. In the test report [1] only the compressive strength was given
for the rock and concrete materials. Thus, the tensile strength and onset of crushing have been
generated automatically by ATENA. To overcome these problems the tests in series 1 have
been modeled using the residual inclination angle of the asperities (zero) according to eq. (8).
Results
The results from the numerical analyses have been monitored by recording the horizontal
displacement and reaction-force along the loaded edge of the model. The load-displacement
curves from the analyses (labeled ATENA) are presented in the same diagrams as the curves
from the actual tests (labeled LTU after the lab where the tests were conducted). In total,
eleven of the twelve shear tests were analyzed. Due to space limitation in Figure 8 below, the
graphical results are shown for selected analyses only. Full description of all results are
presented in [9].
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
396








Figure 8 Load displacement diagrams showing tests and numerical results
In general the procedure of assuming cut-off of micro roughness as cohesion gives a good
representation of the peak shear capacity of the actual tests. When coupled with a softening
behavior of the cohesive parameter, the capacity along the failure development and the
residual shear capacity are represented sufficiently accurate. Since the input parameters are
found from each specific test the finite element models should produce results that match both
the peak and residual shear capacity from the tests exact. However, studying the results, it
becomes evident that some errors still occur. The difference between the peak shear capacity
from the laboratory tests and the numerical analyses are listed in Table 3.




ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
397
Table 3: Error in peak shear capacity from the numerical models.
Sample V
max
[kN] test V
peak
[kN] ATENA Error
[kN] [%]
1.1 90.53 89.91 0.62 1
1.2 131.19 125.80 5.39 4
1.4 183.42 173.21 10.41 6
2.1 50.74 50.81 0.07 0.1
2.2 100.59 100.57 0.02 0.02
2.3 162.83 162.08 0.75 0.5
3.1 35.38 34.65 0.73 2.
3.2 58.70 57.56 1.14 2
3.3 85.43 83.65 1.78 2
4.1H 240.00 225.75 14.25 6
4.2 49.56 49.52 0.04 0.1

The error-% is calculated with respect to the actual value from the tests [1]. The average error
is approximately 2% with a maximum of 6% for two analyses. Compared to the results from
the Mohr-Coulomb and Patton criteria in Table 4 this is a significant improvement.
Table 4: Error in peak shear capacity from hand calculations.
From tests [1] Mohr-Coulomb Patton
Sample I [] N [kN] V [kN V [kN] Error [%] V [kN] Error
[%]
1.1 40 27.52 90.53 361.60 332 113.73 26
1.2 40 46.91 131.19 380.99 190 193.86 48
1.3 40 68.49 183.42 402.57 119 283.04 54
2.1 20 27.57 50.74 20.32 59 41.19 18
2.2 20 47.77 100.59 35.21 65 71.89 28
2.3 20 68.36 162.83 50.39 69 102.88 37
3.1 10 24.82 35.38 18.29 48 26.06 26
3.2 10 45.98 58.70 33.89 42 48.28 18
3.3 10 67.65 85.43 49.86 42 71.04 17

Although the errors in Table 3 are not large, their cause needs to be addressed. One possible
source is the inaccurate recording of the residual capacity in the tests. Due to the fact that
when concrete was cracking the analysis was not always stable, the residual capacity only
describes the capacity as long as the interface asperities are intact. This explains why the
residual capacity is not well represented for test series 2 where some deformation of the
asperities was registered in the tests. For test series 1 this source of error is avoided as the
interface is modeled with the inclination of the residual sliding plane from the actual tests.
It is hard to determine why stable results after material failure in the concrete was not
obtained, but it is believed that a fine mesh and more refined material models would improve
this. Not all the required input parameters for the material models were obtained from the
physical tests, thus some default values were used in the modeling. With lacking parameters,
a full parametrical study would be needed to refine the materials.
The results from the analyses show that the actual shear capacity of the tests are better
represented through numerical modeling than the formulations available for hand
calculations. Especially the opportunity to represent the softening behavior after the peak
shear capacity is reached, is a large benefit.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
398
Conclusion
Through finite element modeling, more sophisticated analyses regarding stability towards
sliding might be obtained. This allows more sustainable design of structures subjected to
sliding instability. However, it must be noted that due to the limited amount of samples
analyzed and the scale effects encumbered with shear capacity, further work is needed to
utilize the potential of finite element modeling of stability towards sliding for full scale dams.
A full parametric study is needed to improve the method.
Acknowledgements
Acknowledgements are given to the professor Kjell H. Holthe at the Norwegian University of
Engineering Science and Technology and Dr. Dobromil Pryl at Cervenka Consulting.
Acknowledgements are also given to EnergiNorge for financial support.
References
[1] Liahagen, S., (2012) Stabilitet av betongdammer - Ruhetens pvirkning p
skjrkapasiteten mellom betong og berg. Masters Thesis. NTNU,IVM, Trondheim,
unpuplished.
[2] NVE, Norwegian Water Resources and Energy Directorate, (2005), Retningslinje for
betongdammer.
[3] Coulomb, C. A. (1776). Essai sur une application des regles des maximis et minimis a
quelquels problemesde statique relatifs, a la architecture. Mem. Acad. Roy. Div. Sav.,
vol. 7, pp. 343387.
[4] Patton, F.D (1966). Multiple modes of shear failure in rock. 1
st
ISRM Congress,
September 25 October 1, 1966, Lisbon, Portugal. LNEC
[5] Johansson, F., (2009). Shear Strength of Unfilled and Rough Rock Joints in Sliding
Stability Analyses of Concrete Dams. Ph.D. Thesis. Royal Institute of Technology
(KTH), Stockholm.
[6] Cervenka, V. Cervenka, J. Zednek, J. Pryl, D. (2013) ATENA program Documentation,
Part 8: Users Manual for ATENA_GiD Interface.
http://www.cervenka.cz/assets/files/atena-pdf/ATENA-Science-GiD_Users_Manual.pdf
[7] Cervenka, V., Jendele, L., Cervenka. J., (2012) ATENA Program Documentation, Part 1:
Theory. http://www.cervenka.cz/assets/files/atena-pdf/ATENA_Theory.pdf
[8] Pryl, D., Cervenka, V. (2013) ATENA Program Documentation, Part 11:
Troubleshooting manual. http://www.cervenka.cz/assets/files/atena-pdf/ATENA-
Troubleshooting.pdf
[9] Eltervaag, . (2013) Sliding stability of Lightweight Concrete Dams Development of
numerical models. Masters Thesis. NTNU, IKT, Trondheim. Unpublished.

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
399
Seismic Analysis of a Concrete Arch Dam
Considering Concrete Heat Generation Damage Effects
P. Dakoulas
1

1
University of Thessaly, Volos, GREECE
E-Mail: dakoulas@uth.gr
Abstract
This study evaluates the seismic performance and safety of Tavropos arch dam (Greece)
considering the dynamic canyon-dam-water interaction. Tavropos dam has a height of 83 m
and crest length of 220 m. The numerical model of the arch consists of 17 concrete cantilevers
separated by vertical joints, whereas the foundation-abutment model represents a large region
of the canyon. The interface behavior at the vertical joints and the horizontal base joint is
modeled in three different ways: as bonded surfaces, as surfaces in contact with frictional
strength, and as a combination of bonded and contact surfaces. To account for possible prior
damage, the study simulated the thermo-mechanical phenomena that took place during dam
construction, related to heat generation due to concrete hydration. The seismic analysis is
performed against a MCE having a magnitude of 7.5 and peak ground acceleration of 0.47g.
The seismic evaluation is based on an extensive parametric investigation of the effects of
various key factors, including concrete strength, joint interface behavior, water level,
hydrodynamic pressures, ambient temperature, canyon rock flexibility, spillway geometry,
earthquake excitation characteristics, etc. It is concluded that the expected overall stability
and performance of Tavropos concrete arch dam is satisfactory.
Introduction
Lake Plastiras was created after construction of an arch dam (Fig. 1) in Tavropos river, a
tributary of Acheloos river in Northern Greece [4]. The double-curvature Tavropos arch dam
has a height of 83 m and crest length of 220 m. Construction started in 1955 and was
completed in 1959, during which the dam was impounded. Today, Lake Plastiras provides
drinking water to the city of Karditsa and nearby villages, and irrigation to the region near
Larissa. Its electric power is 130 MW, whereas the lake is an attractive tourist destination with
more than 120,000 tourists annually.

Figure 1: Tavropos Arch Dam in Lake Plastiras, Greece
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
400
The objective of this work is to conduct a seismic performance and safety evaluation of the
data, considering the dynamic canyon-dam-water interaction and current data about the local
seismicity based on recent research [11][3][8][9][10].

Figure 2: Finite element discretization of (a) concrete arch dam geometry consisting of 17
parts and (b) Part No.10 geometry at the middle section


Figure 3: Finite element discretization of the canyon geometry (a) model A and (b) model B
consisting of full quadratic finite elements for the entire canyon region.
Numerical Model
Dam geometry
The study is based on a detailed numerical model of the dam and canyon using the code
ABAQUS [1]. Fig. 2a illustrates the numerical discretization of the arch dam body. The dam
body consists of 17 parts, having a width equal to 12 m, except of the central part, which has
width equal to 16 m. The 17 parts are connected with vertical joints. Fig. 2b shows Part 10,
located at the mid-section of the dam. As all parts near the central region of the arch, Part 10
consists of 4 concrete blocks: A, B, C and D. Blocks B, C and D at the lower upstream side
were designed as to increase the stability of the dam in the case of a potential strong
earthquake during the phase of construction. They are all attached to main block A, but there
are horizontal joints between blocks B and C, as well as between blocks C and D to reduce
seismic tensile stresses. To improve numerical efficiency, only 2 out of the 4 actual horizontal
joints of each part were discretized, without any loss of accuracy. All dam parts are
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
401
discretized using solid hexahedral elements C3D8, having 8 nodes and 8 integration points. A
total number of about 4400 elements were used for the modeling of the arch dam geometry.
Canyon geometry
The discretized canyon geometry has length of 600 m, width of 600 m and depth of 250 m. It
consists of two parts: (a) the first part discretizes the irregular geometry near the foundation of
the dam. It is discretized by using quadratic, reduced integration hexahedral elements
C3D20R, each having 20 nodes, whereas the irregular rock surface and dam foundation
excavation, are discretized using quadratic modified tetrahedral elements C3D10M, which are
robust, accurate and suitable for frictional contact behavior; (b) for the rest of the canyon, two
different models have been utilized:
(1) model A (Fig. 3a) consists of hexahedral solid elements C3D8, having 8 nodes
(2) model B (Fig. 3b) consists of quadratic hexahedral elements C3D20R, having 20 nodes.
About 30500 and 27000 elements were used for canyon models A and B, respectively.
Damage plasticity model for concrete
The plastic-damage constitutive model for cyclic loading by Lee and Fenves [6] is used here
for modeling the behavior of concrete. The model takes into account the effects of strain
softening, distinguishing between the damage variables for tension and compression. It
incorporates a degradation mechanism that represents the effects of damage on the elastic
stiffness and the recovery of stiffness after crack closure. Three values of compressive
strength have been adopted, based on large cubic specimens (20cm x 20cm x 20cm) obtained
during construction and tested after 365 days. The use of large size specimens is important to
avoid the negative effect that the large diameter aggregates may have on strength prediction.
The three values of compressive strength are: (A) the mean value
c
f = =46 MPa (B) the
mean value minus three standard deviations, 3
c
f o = 42 MPa and (C)
c
f =32 MPa, i.e. an
unlikely, much lower value to account for possible additional uncertainties. For
c
f =46 MPa,
the stress-strain behavior during uniaxial loading-unloading-reloading of a cube of concrete,
subjected separately to compression and tension tests is given in Figure 4. For the cyclic
tension tests, the softening behavior and the effect of the accumulated damage are taken into
account, whereas for the cyclic compression, these effects are ignored in the present study.
Table 1: Concrete Properties
Property Case A Case B Case C
Density, [kg/m
3
]

2350 2350 2350
Compressive strength, f
c
[MPa] 46 42 32
Dynamic tensile strength, f
td
[MPa] 6.4 6.0 5.3
Static Youngs modulus, E [GPa] 31 31 31
Dynamic Youngs modulus E
d
, [GPa] 42 42 42
Poissons ratio, 0.15 0.15 0.15
Coefficient of thermal expansion, a [C
-1
] 10
-5
10
-5
10
-5

Specific heat, c [J/(kg C)] 879 879 879
Conductivity, k [W/(m C)] 2.5 2.5 2.5
Heat transfer coefficient [W/(m
2
C)] 16 16 16
Hydration heat, q(t) [kcal/kg
cement
] (t : in days) ( ) /(0.0653 0.0175 ) q t t t = +
Joint behaviour
The behavior at the vertical joints and the horizontal joint at the dam base is modeled in three
different ways: as bonded surfaces, as surfaces in contact with frictional strength, and as a
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
402
combination of bonded and frictional surfaces. The bonded joints are considered to have the
tensile strength of concrete of the adjacent blocks. The surfaces with only frictional strength,
have a coefficient of friction equal to 0.5 for vertical joints and 0.8 for the horizontal joints.

Figure 4: Damage plasticity concrete model: stress-strain relation during loading and
unloading in uniaxial (a) tension and (b) compression.
Canyon rock properties
The canyon rock material is fine-layered limestone containing sparse intermediate layers of
radiolarites. During the static and dynamic analyses of the dam-canyon system, the canyon
rock is considered to behave as a linearly elastic material. Table 2 summarizes the canyon
rock properties. Two scenarios are examined for the dynamic analyses, namely, case R1 and
case R2, having shear wave velocities equal to
s
V = 2800 and 2000 m/s, respectively.
Table 2: Canyon Rock Properties
Property Case R1 Case R2
Density, [kg/m
3
]

2450 2450
Static Youngs modulus, E [GPa] 16 16
Dynamic Youngs modulus, E
d
[GPa] 48 25
Poissons ratio, 0.25 0.25
Shear wave velocity, V
s
[m/s] 2800 2000
Hysteretic damping,
r
0.03 0.03
Effect of heat generation during construction
It is of interest to examine concrete behavior immediately after construction, especially for
high performance concrete, as cracking may develop due to deformation caused by heat that is
generated during cement hydration reactions [7][2]. Potential cracking generated during or
immediately after construction of various dam segments might have developed weak areas,
which could affect its seismic performance during earthquakes. To examine the effects of
premature cracking, a numerical simulation of the dam construction process is conducted,
modeling the cement-hydration heat generation, radiation, convection, conduction and
ambient temperature variation for a period of about 24 months. Fig. 5 plots the distribution of
the temperature in a section of the dam after 200 days, whereas Fig. 6 plots the temperature
variation with time at points A, B, C, D and E, as well as the ambient temperature. The
maximum temperature occurs in the middle point (C) of the dam wall and has a value of 50.5
C at 200 days from the construction start. Examination of the major principal plastic strain
caused by the transient temperature differences within the dam showed that the concrete-
hydration heat did not cause any significant cracking within the main body of the dam. Any
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
403
such cracking within the dam body is practically either zero or negligible. Some thermal
cracking may have developed within the upstream aseismic concrete blocks (see Fig. 2b), but
it has absolutely no effect on the performance and stability of the dam in its current state.

Figure 5: Distribution of temperature within the dam body at 200 days since construction start


Figure 6: Variation of temperature at points A, B, C, D and E during construction versus time
Seismic analysis
Dynamic characteristics of the dam
The fundamental natural frequency of the dam for full reservoir and canyon rock with shear
wave
s
V =2800 m/s is 3.97 Hz, whereas for
s
V =2000 m/s, it reduces to 3.81 Hz.

Earthquake excitation
The seismicity of the region has been studied by Panagiotopoulos and Papazachos [10]. For
the present study, a Maximum Credible Earthquake of magnitude M ~ 7.5, with an epicentral
distance from the dam site about R = 15 km, is considered. The Maximum Probable
Earthquake is based on a magnitude of M ~ 7.0 and minimum distance R =15 km. The seismic
excitation considered consists of three different acceleration records. The first two have been
recorded at rock sites (Lucerne and Pacoima Dam records). The third record is synthetic and
has response spectra that match the Eurocode 8 design spectra for rock sites. All records have
been base-line corrected and scaled to a peak horizontal outcrop-rock acceleration equal to
0.47g and peak vertical outcrop-rock acceleration equal to 0.30g. Fig. 7 plots the horizontal
and vertical acceleration time histories of the synthetic record. Fig. 8 plots the acceleration
response spectra for the horizontal component and the Eurocode spectra for rock site. The
input excitation at the canyon base is computed through de-convolution of the selected
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
404
motions. To be able to account for the radiated seismic energy, appropriate dashpots are
placed at the canyon boundary and the excitation is imposed in terms of seismic stresses.


Figure 7: Synthetic excitation at outcrop rock (a) horiz. acceleration (b) vertical acceleration

Figure 8: Synthetic horizontal acceleration spectra and Eurocode spectra for rock sites

Hydrodynamic pressures
The hydrodynamic pressures acting on the upstream side of the dam have a significant effect
on the earthquake behavior of the dam. They are considered here by using the added-mass
formulation proposed by Zangar [11], which was implemented in code ABAQUS [1]. The
amount of mass b at a node is equal to:

1
(2 ) (2 )
2
m w
z z z z
b h C A
h h h h

(
= +
(

(1)

where h = depth of the reservoir,
m
C = coefficient based on the angle u of upstream surface to
the vertical, z = depth of node below water surface,
w
= water density and A = area around
node contributing to hydrodynamic forces.
Results and discussion
The objective of the parametric analysis is to investigate the effect of various factors on the
seismic performance of the Tavropos Arch Dam. More specifically, the parametric study
examines effect of concrete strength variation, seasonal temperature variation, reservoir water
level, canyon rock flexibility, joint behavior, hydrodynamic pressures, spillway geometry, and
earthquake excitation characteristics. Due to lack of space, only representative results of the
most likely scenario are presented in this article, whereas additional results will be published
elsewhere. The results presented below are based on a concrete compressive strength of 46
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
405
MPa, dynamic tensile strength of 6.4 MPa, critical summer temperature variation, maximum
reservoir water level (792 m), and canyon rock S-wave velocity 2800 m/s. Both models A and
B (shown in Fig. 3) are utilized for the analysis. First, the overall response and the stability of
the dam are considered.
Relative displacements
Results from two scenarios are presented here for the evaluation of the response and stability
of the dam. The first scenario assumes that the vertical joints and the base joint behave as
bonded surfaces. Figs. 9a and 9b plot the relative u/d displacement at mid-crest of the dam
subjected to the synthetic record for bonded joints, using the two canyon models A and B,
respectively (Fig. 3). The computed relative displacement by the two models is rather similar,
with the peak value about 4.5 cm. There is no residual relative displacement at the end of
shaking. The peak relative displacement for the other two excitations (not shown here) are
about 4 cm. Fig. 10a plots the distribution of the u/d relative displacement for bonded joints
using model A at a moment of peak response (t = 6 s). The second scenario is quite
conservative as it assumes that the joints have zero tensile strength and behave as surfaces in
contact having only frictional strength. Fig. 10b plots the residual relative u/d displacement
for the case of un-bonded frictional joints, allowing opening and sliding during shaking. The
maximum relative displacement in this case is less than 5 cm, which is considered as safe
with regard to the stability of the dam for this very conservative scenario.


Figure 9: Time history of relative u/d displacement at mid-crest assuming bonded joints
obtained from (a) model A and (b) model B


Figure 10: (a) Peak relative u/d displacement at t = 6 s assuming bonded joints and (b)
Residual relative u/d displacement at the end of shaking assuming un-bonded frictional joints.
Stresses and strains
Fig. 11 shows the distribution of major principal stress,
1
o , at the moment of maximum
response of the dam subjected to the Synthetic record based on Model A (Fig. 3). As shown in
the figure, high tensile stresses develop at the upper middle part of the arch dam, reaching in
some elements the concrete tensile strength for a very small time increment (e.g. 0.01s) and
causing some limited plastic tensile strain. These tensile stresses occur when the upper part of
the dam has a relative displacement towards upstream with respect to its base. In this case, the
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
406
combined effect of dam and water inertia (negative hydrodynamic pressure) cause the upper
part of the dam to deform towards upstream. This upstream relative displacement becomes
maximum at t=5.32 s for the Synthetic record (see point A in Figure 9a). Such deformation
causes an extension of the arch dam along its length, which is maximum at the upper-middle
area of the arch, resulting to relatively high, short-duration tensile hoop stresses. It is noted
that the tensile stresses become maximum at the upstream surface of the central crest area of
the arch, whereas they are quite smaller on the downstream side. Figs. 12a and 12b plot the
major principal stress time history at mid-crest, evaluated at the upstream surface element
using the two models A and B, respectively. The results show that
1
o reaches the tensile
strength momentarily in Fig. 12a, but it remains below the tensile strength in Fig. 12b.

Figure 11: Distribution of major principal stress
1
o

at a moment of maximum tension of
the dam subjected to the Synthetic record excitation (time t = 5.32 s)


Figure 12: (a) Major principal stress and (b) minor principal stress time histories at mid-
crest (upstream surface element)


Figure 13: Accumulated major principal plastic strains at the end of shaking
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
407

Despite the large tensile stresses that may develop momentarily during shaking, the
accumulated major principal plastic strain
1
p
c at the end of shaking is very limited. Fig. 13
plots the distribution of
1
p
c at the end of shaking for the Synthetic record based on model A.
Its maximum value is only about 5 10
-5
, located at the upstream face of the mid-crest
(concrete part 10). In reality, it is expected that the value of the tensile plastic strain
1
p
c will be
even smaller than the values shown in Fig. 13. This is because the actual tensile strength at
the joints is expected to be about 0.7
t
f , where
t
f is the tensile strength of the dam concrete.
Thus, in areas of higher tensile stress concentration, as at the mid-crest area, any cracks most
likely will occur at the weakest points, i.e. the vertical joints. As shown in other analyses not
presented here, if the entire vertical joints rupture before shaking (i.e. the joint tensile strength
is zero), the tensile stresses developing within the mid-crest area of the arch dam are
significantly reduced, due to the instantaneous, local opening of the vertical joints. Therefore,
the accumulated major principal plastic strains near the mid-crest area are much smaller.
Note that the results in Figs. 11, 12a and 13 are based on model A. By using the more refined
model B, it is shown that the tensile stresses developing in the mid-crest area are slightly
smaller and do not exceed the tensile strength. Thus, no plastic strains develop in the dam
body for the presented case of concrete with compressive strength equal to the mean value
c
f =46 MPa.
Figure 14 plots the distribution of the minor principal stress
3
o (maximum compression) at a
moment of maximum response, corresponding to time t = 6 s (point B in Figure 9a). This
occurs when the dam base and canyon moves upstream, while the dam is pushed by the
concrete and water inertia (positive hydrodynamic stresses) towards downstream, thereby
increasing the compressive stresses in the mid-crest area of the arch. The maximum value of
the compressive stresses at the mid-crest area during shaking is less than 12 MPa, i.e. the
compressive stresses are very small compared to the compressive strength of concrete in
uniaxial load (
c
f =46 MPa). In this case, the minimum available factor of safety against
compressive failure
1 3
/( )
c c
FS f o o = during shaking is larger than 2.5. Thus, no plastic
deformations occur due to compressive stresses. The compressive stresses for the other two
records are smaller than those obtained for the Synthetic record.


Figure 14: Distribution of minor principal stress
3
o

at a moment of maximum compression
of the dam subjected to the Synthetic record excitation (time t = 6.00 s)
Conclusions
Evaluation of the results of all parametric studies leads to the general conclusion that, for the
maximum earthquake intensity anticipated, the dam is safe against any instability and its
overall performance is satisfactory. More specifically, the main conclusions are the following:
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
408
1. The concrete-hydration heat generated during construction did not cause any significant
cracking within the main body of Tavropos dam.
2. The maximum seismic relative displacement of the dam for the case of grout-bonded
joints is small (about 3 to 4.5 cm). There are no permanent relative displacements at the
end of shaking.
3. For the extreme scenario in which the vertical joints and the joint at the base are already
ruptured before the earthquake, the relative displacement of the dam at the end of shaking
is limited to values less than 5 cm. Thus, even for this extreme scenario, the stability of
the 17 cantilever dam parts is fully assured.
4. Based on the results of all analyses, the maximum compressive stresses developing during
seismic shaking range between -10 MPa to -16 MPa and do not cause any plastic strains.
5. The magnitude of tensile stresses within the concrete parts depends on the behavior of the
vertical joints. The results of the parametric analysis show that the tensile plastic strains
(or cracking) developing within the dam body are very small. Therefore, the performance
of the dam with respect to the development of tensile cracking during shaking is
satisfactory.
6. The summer environmental temperature condition is the most critical one, as it results to
higher tensile plastic strains within the dam body.
7. Among the four reservoir water elevations investigated, the most critical one is the
maximum water level (792 m), which results to relatively higher tensile plastic strains
within the dam body.
8. The stiffer rock, having a wave velocity
s
V =2800 m/s, yielded consistently larger values
of tensile plastic strains compared to those obtained for
s
V =2000 m/s.
Acknowledgment
The financial support by the Public Power Corporation is gratefully acknowledged.
References
[1] ABAQUS (2012). Users Manual, Simulia, Pawtucket, Rhode Island.
[2] Azenha, M., Faria, R., Ferreira, D. (2009), Identification of early-age concrete
temperatures and strains: Monitoring and numerical simulation, Cement & Concrete
Composites, 31, 369378.
[3] Chopra, A. (2008). Earthquake analysis of Arch Dams: Factors to be considered, 14th
World Conference on Earthquake Engineering, October, 2008, Beijing, China.
[4] Dakoulas, P. Seismic analysis of Tavropos Arch Dam, Research Report to Public Power
Corporation, University of Thessaly, Volos, Greece
[5] Escuder, I.B. and Blazquez, F.P. (2007). Lessons learned in the analysis of the elastic
behavior of an arch-gravity dam performed by seven independent engineering teams, 9th
Bench. Work. Num. Analysis of Dams, ICOLD, St. Petersburg, Russia, June 22-24.
[6] Lee J., and Fenves, G.L (1998). A plastic-damage concrete model for earthquake analysis
of dams, Journal of Earthq. Eng. & Struct. Dynamics, 27: 937-596.
[7] Meghella, M. and Frigerio, A. (2009). Theme A: Initial strain and stress development in a
thin arch dam considering realistic construction sequence, 10th Benchmark Workshop on
the Numerical Analysis of Dams, ICOLD, Paris, France, Sept. 16-19.
[8] Mills-Bria, B., Nush, L., and Chopra, A. (2008). Current Methodology at the Bureau of
Reclamation for the Nonlinear Analysis of Arch Dams Using Explicit FE Techniques,
14th World Conference on Earthquake Engineering, October, 2008, Beijing, China.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
409
[9] Mojtahedi, S. and Fenves, G. (2000). Effect of contraction joint opening on Pacoima
Dam in the 1994 Northridge earthquake, California Strong-Motion Instrumentation
Program Data Utilization Report, CSMIP/00-05 (OSMS 00-07), Sacramento, Calif.
[10] Panagiotopoulos, D. and Papazachos, K.B. (2008). Active tectonics of Thessaly and
seismicity of Karditsa, 1st Conference on Development, Karditsa, Greece.
[11] US Bureau of Reclamation (2006). State-of-Practice for the Nonlinear Analysis of
Concrete Dams at the Bureau of Reclamation, USBR Report, Colorado, USA.


ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
411
Earthquake safety assessment of arch dams based on
nonlinear dynamic analyses
S. Malla
1

1
Axpo Power AG, Parkstrasse 23, CH-5401 Baden, SWITZERLAND
E-mail: sujan.malla@axpo.com
Abstract
For the earthquake safety assessment of two major arch dams in Switzerland, various linear
and nonlinear dynamic analyses were performed. For these analyses, a safety evaluation
earthquake (SEE) with a return period of 10,000 years was considered in each case. Even
though the input motion has a PGA of less than 0.2 g, the crest region would be subjected to
amplified accelerations significantly larger than 1.0 g. Moreover, high tensile stresses
occurring in the crest region are likely to exceed the tensile strength of the vertical contraction
joints and the horizontal lift joints. Consequently, these joints could open during the
earthquake shaking, possibly resulting in an upper portion of a dam block becoming fully
detached from the rest of the dam. The dynamic analysis of possibly detached portions in the
crest region showed that they could undergo some sliding and rocking motions. However,
they would remain stable during and after the earthquake. Based on the results of the
nonlinear dynamic analyses, it is concluded that the earthquake loading could cause some
limited damages, but an uncontrolled release of reservoir water is unlikely to occur.
Therefore, the investigated dams satisfy the safety requirements for the SEE.
Introduction
A program of systematic assessment of earthquake safety of all significant dams in
Switzerland is scheduled to be completed by the end of the current year 2013. A document
containing state-of-the-practice guidelines for this investigation program was issued by the
Swiss Federal Office of Energy in 2003. The primary goal of this program is to ensure the
safety of the downstream population against loss of life and property damage (Panduri et al.,
2012). For this purpose, it must be shown that the safety evaluation earthquake (SEE) would
not cause a dam failure resulting in an uncontrolled release of reservoir water. Depending on
the height and the reservoir volume, a dam is categorized as class 1, 2 or 3 and it has to be
checked for the SEE ground motion with a return period of 10,000, 5,000 or 1,000 years,
respectively.

At the time of the design of most of the existing dams, the earthquake action was usually
considered as a pseudo-static loading and typically a horizontal seismic coefficient of 0.1 g
was assumed. This is clearly inadequate even in regions with relatively low seismicity. In
practically all major arch dams, the crest region would experience horizontal accelerations
significantly exceeding 1.0 g in the event of the SEE, even when the ground motion at the
rock level has a horizontal peak ground acceleration (PGA) of the order of 0.20 g only.

In a linear elastic analysis, the earthquake shaking would typically cause high horizontal
stresses in the arch direction in the crest region of the dam. As the tensile strength of the
vertical contraction joints is usually quite low, these joints are likely to open during the
earthquake shaking. Once the vertical joints are open, the dam blocks would behave as
vertical cantilevers, leading to high vertical tensile stresses. Since the tensile strength of
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
412
horizontal lift joints is usually substantially lower than that of monolithic mass concrete, a
horizontal crack may form, causing the upper portion of a dam block to become fully
detached from the rest of the dam. Such a detached portion would be subjected to rather high
accelerations and it could thus undergo sliding and rocking motions during the earthquake
shaking. As long as a possibly detached portion remains stable during and after the
earthquake, an uncontrolled release of reservoir water could be ruled out and the safety
requirements for the SEE would be satisfied.

To investigate the dynamic stability of possibly detached upper portions in the crest region
and to analyze the behavior of vertical contraction joints in an arch dam during an earthquake,
nonlinear dynamic analyses have to be carried out. In this paper, two illustrative examples are
presented to show how earthquake safety of arch dams can be assessed on the basis of results
of 2D and 3D nonlinear dynamic analyses.
Methodology for simplified dynamic stability analysis
A simplified analysis of the seismic stability of possibly detached concrete blocks in the
central upper portion of an arch dam can be performed in the following steps (Malla and
Wieland, 2006; Wieland and Malla, 2012):
(i) Linear elastic dynamic time history analysis of a three-dimensional (3D) finite element
(FE) model of dam-reservoir-foundation system;
(ii) Selection of detached concrete blocks for dynamic stability analysis based on envelopes
of principal dynamic tensile stresses and absolute accelerations;
(iii) Obtaining time histories of radial and vertical components of absolute acceleration at
the base of each detached concrete block from the results of analysis step (i);
(iv) Setting up 2D FE model of each detached concrete block using contact elements to
simulate the cracked lift joint (gap elements may be employed to prevent the concrete
block from moving beyond the downstream face of the dam in view of the geometrical
constraints in an arch dam);
(v) Nonlinear dynamic analysis of rocking-sliding response of each detached concrete block
subjected to input base acceleration obtained in step (iii);
(vi) Evaluation of the maximum sliding movement and the maximum crack opening
displacements at the upstream and downstream edges of the base of each detached
concrete block for at least three different earthquake ground motions; and
(vii) Assessment of the dynamic stability of detached concrete blocks based on results of
step (vi).

The seismic safety of the Roggiasca and Gigerwald arch dams in Switzerland was checked by
employing this simplified procedure. The main results of this analysis are presented and
discussed in the following sections. Furthermore, nonlinear 3D FE analysis of the Gigerwald
arch dam was performed to investigate contraction joint openings during the earthquake.

All the dynamic calculations were performed with the help of the general-purpose FE
software ADINA (ADINA R & D, 2008). For the dynamic analyses, the three components of
earthquake excitation were simulated by artificially-generated spectrum-compatible
acceleration time histories. The hydrodynamic pressure was modeled as added masses acting
perpendicular to the upstream face of the dam.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
413
Earthquake safety assessment of Roggiasca arch dam
The 68 m high Roggiasca arch dam was completed and first impounded in 1965. With a crest
thickness of 2.5 m and a maximum thickness at the base of 7.5 m only, the dam has a
relatively high Lombardi slenderness coefficient of 21. At the location of this dam, the
10,000-year SEE has a horizontal PGA of 0.17 g.

The linear seismic response of the Roggiasca dam was analyzed using a 3D FE model, which
comprised the dam, the foundation rock, the sediment deposit on the upstream side and the
soil fill on the downstream side. The main results of this 3D analysis performed for 3 different
ground excitations (designated as earthquakes 1, 2 and 3) are listed in Table 1. The largest
accelerations and dynamic stresses were obtained at the middle and also at the two quarter
points of the dam crest. The results showed that the upstream sediment deposit and the
downstream soil fill would not play a significant role in the dynamic behavior of the dam.
Table 1: Main results of 3D linear elastic earthquake analysis of Roggiasca dam subjected to
SEE ground motion under full reservoir condition
Dynamic response (envelope) Earthquake 1 Earthquake 2 Earthquake 3
Relative crest displacement (mm)
- Along-stream direction
- Across-stream (left-right) direction
- Vertical direction

16.2
8.4
2.0

15.9
7.9
2.1

18.7
8.2
2.1
Absolute crest acceleration (g)
- Along-stream direction
- Across-stream (left-right) direction
- Vertical direction

1.45
0.62
0.41

1.34
0.64
0.39

1.27
0.72
0.35
Principal tensile stress (MPa) 6.1 5.6 6.6
Principal compressive stress (MPa) -6.3 -6.0 -6.2

The magnitude of the largest dynamic tensile and compressive stresses in the crest region of
the dam is about 6 MPa (see Table 1 and Figure 1). Even higher elastic stresses are computed
at the upstream and downstream edges of the dam-rock interface due to the stress singularities
at these reentrant corners.


Figure 1: Principal stress vectors due to earthquake 1 (without static loads) at time t = 7.19 s
when the highest tensile stress occurs under full reservoir condition
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
414
The largest compressive stress in the dam body under the combination of the static and
dynamic loads is about -10 MPa, which is not a problem for the dam concrete with a static
compressive strength of 46 MPa. In spite of the compressive stresses due to the static loads
(self-weight and hydrostatic pressure), relatively high horizontal tensile stresses of up to about
4 MPa oriented in the arch direction still occur in the crest region during the earthquake
shaking. If the thermal stresses in winter and the effect of the ongoing chemical expansion of
the dam concrete would also be considered, the tensile stresses would be even higher.

In order to investigate the stability of possibly detached concrete blocks during the earthquake
excitation, simplified 2D FE models shown in Figure 2 were employed. As the Roggiasca
arch dam is rather thin, any detached block in the crest region would be quite slender.



Figure 2: 2D FE models of 7 m and 13 m high detached blocks in crest region of Roggiasca
arch dam (gap elements prevent any movement beyond the downstream face)

In a dynamic analysis involving rigid body motions, it is considered prudent to use only the
stiffness-proportional part of the Rayleigh damping model, as the mass-proportional part
corresponds to external viscous dampers connected to the nodes of the model (Hall, 2006).
Three different Rayleigh damping models were considered in the stability analysis (see Table
2). Model A corresponds to that used in the 3D linear dynamic analysis. In model B, only the
stiffness-proportional part is kept. In model C, the stiffness-proportional part is further
substantially reduced, which is very conservative.
Table 2: Rayleigh damping models used for 2D dynamic stability analysis of detached
cantilever blocks in crest region of Roggiasca dam
Rayleigh
damping
model
Parameter
o
(s
-1
)
Parameter
|
(s)
A 2.20 0.00124
B 0.00 0.00124
C 0.00 0.00020
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
415
The results of the dynamic stability analysis listed in Table 3 show that the detached portions
in the crest region have a tendency to undergo only rocking with virtually no sliding, a
behavior explained by the slender form. Even the rocking motion is quite small and causes
dynamic crack openings of only a few millimeters at the base of the detached portion.
Table 3: Main results of 2D dynamic stability analysis of detached portions in crest region of
Roggiasca dam subjected to SEE ground motion under full reservoir condition
Rayleigh
damping
model
Earthquake
Max. horiz.
crest
displace-
ment (mm)
Max. sliding
displace-
ment
(mm)
Max. crack
opening,
u/s face
(mm)
Max. crack
opening,
d/s face
(mm)
(a) 13 m high detached block above crack at level 942 m a.s.l.
A 1 -4 0.0 0.0 0.7
B 1 -5 0.0 0.0 0.8
C 1 -6 0.0 0.0 1.1
B 2 -11 0.0 0.0 2.2
C 2 -14 0.0 0.0 3.0
B 3 -11 0.0 0.0 2.2
C 3 -17 -0.2 0.0 3.8
(b) 7 m high detached block above crack at level 948 m a.s.l.
A 1 -15 -0.1 0.0 5.6
B 1 -16 -0.1 0.0 6.1
C 1 -79 -0.2 0.1 30.7
B 2 -22 -0.1 0.0 8.5
C 2 -128 -0.4 0.0 49.8
B 3 -25 -0.1 0.0 9.7
C 3 -171 -0.7 0.1 66.7

The results of the dynamic stability analysis show that the detached portions remain
dynamically stable even when subjected to peak horizontal accelerations about twice as large
as the pseudo-static overturning acceleration. This behavior can be explained by the fact that
acceleration spikes would have a relatively high frequency of about 5 Hz corresponding to the
dominant natural frequency of the dam. Hence, such a spike would exceed the pseudo-static
overturning acceleration only for a very short duration of less than one-tenth of a second,
which is too short to produce any significant block rotation. A review of literature on dynamic
overturning of rigid blocks also confirms that acceleration peaks would have to be many times
larger than the pseudo-static overturning acceleration for a block with dimensions of the order
of a few meters to be toppled by a dynamic base excitation at such a frequency, as illustrated
in Figure 3 (Shi et al., 1996; Zhang and Makris, 2001).
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
416



Figure 3: Boundaries between stable and unstable regions of dynamic rocking motion based
on analytical solutions for half and full sinusoidal base excitations (Shi et al., 1996, Zhang
und Makris, 2001)
Note: A ground excitation with an amplitude of 1.5 g and a frequency of 5 Hz corresponds to the
red square (e
p
/p = 22.4 und a
p
/og = 3.2) and the red triangle (e
p
/p = 29.6 und a
p
/og = 4.5) in the
case of, respectively, the 7 m and 13 m high detached portions.
Earthquake safety assessment of Gigerwald arch dam
The 147 m high Gigerwald dam was completed and first impounded in 1976. The dam
thickness varies from 7 m at the crest to about 22 m at the base.

The main results of the linear elastic dynamic analysis of a 3D FE model of this double-
curvature arch dam for 3 different earthquakes are listed in Table 4. The crest region would be
subjected to amplified horizontal accelerations as high as about 2.0 g during the 10,000-year
SEE with a horizontal PGA of 0.19 g.
Table 4:Main results of 3D linear elastic earthquake analysis of Gigerwald dam subjected to
SEE ground motion under full reservoir condition
Dynamic response (envelope) Earthquake 1 Earthquake 2 Earthquake 3
Relative crest displacement (mm)
- Along-stream direction
- Across-stream (left-right) direction
- Vertical direction

48.0
15.3
6.6

45.6
16.0
7.4

44.5
18.0
6.9
Absolute crest acceleration (g)
- Along-stream direction
- Across-stream (left-right) direction
- Vertical direction

1.83
0.84
0.75

2.04
0.86
0.78

1.98
0.80
0.71
Principal tensile stress (MPa) 8.7 9.8 8.7
Principal compressive stress (MPa) -10.3 -9.5 -9.6

ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
417
The earthquake shaking produces dynamic tensile stresses of nearly 10 MPa in the central
crest region of the dam. Even after the combination with the compressive stresses due to the
hydrostatic water load, high horizontal tensile stresses of up to about 6 MPa would still
remain in the arch direction. In reality, such tensile stresses cannot develop due to the
presence of the vertical contraction joints, whose tensile strength is normally quite low.

The next step was to perform the dynamic analysis of a 3D FE model in which all 23 vertical
contraction joints were simulated as frictional contact surfaces and the dam concrete was
assumed to be linear elastic (uncracked). This analysis was performed for 3 different input
motions. Figures 4 to 6 depict some results obtained for earthquake 2. This analysis showed
that the seismic shaking would cause relative sliding displacements of nearly 1 cm between
the adjacent blocks and the contraction joints would open by maximum about 4 mm.

The largest compressive stress in the dam obtained in the nonlinear analysis approaches
nearly -19 MPa (excluding the corner singularity at the dam-rock interface), which is not a
problem for the dam concrete. In comparison, the maximum compressive stress in the case of
the linear analysis is about -16 MPa. In spite of the joint displacements, the dynamic
displacements and accelerations of the dam computed in the nonlinear analysis do not deviate
significantly from those in the corresponding linear analysis. The main difference lies in the
absence of any significant horizontal tensile stresses in the arch direction in the central crest
region in the case of the nonlinear analysis owing to the presence of the contraction joints.
During the brief openings of the contraction joints, the upper portion of a dam block acts
temporarily as a cantilever, due to which relatively high transitory vertical tensile stresses
exceeding 6 MPa appear on the downstream face of the dam, as depicted in Figure 6. Hence,
horizontal cracks are likely to form, especially at the lift joints, possibly resulting in the
detachment of the uppermost portion of a central block from the rest of the dam body.


Figure 4: Time histories of opening and sliding displacements of vertical contraction joint
11/12 at crest level due to earthquake 2 under full reservoir condition
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
418

Figure 5:Maximum opening and sliding displacements of vertical contraction joints at crest
level due to earthquake 2 under full reservoir condition


Figure 6: Stresses due to cantilever behavior during opening of vertical contraction joints
The safety of possibly detached portions in a central dam block subjected to the SEE shaking
was assessed using simplified 2D models (see Figure 7). This analysis showed that an 8 m
high detached portion could slide by up to about 50 cm towards the reservoir during the SEE
and the rocking motion would result in crack opening displacements of up to about 7 cm.
However, the detached block would remain stable during and after the earthquake and the
earthquake damage would not lead to an uncontrolled release of water.
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
419

Figure 7: 2D FE model of 8 m high detached portion of a central block of Gigerwald arch
dam (gap elements prevent any movement beyond the downstream face)
For a less conservative dynamic stability assessment, a nonlinear analysis was also performed
using a 3D model in which the uppermost 8 m of the central dam block was assumed to be
detached from the rest of the dam along the vertical contraction joints at the sides and an
assumed horizontal crack along a lift joint. As shown in Figure 8, the 3D analysis showed that
such a detached block would slide by up to 16 cm towards the reservoir, which is only about
one-third of the result obtained in the corresponding more conservative 2D analysis. The
substantially smaller sliding displacement in the 3D analysis can be attributed mainly to the
additional frictional resistance at the vertical contraction joint on each side, an effect that
could not be taken into account in the simplified 2D approach. The maximum crack opening
also decreased to about 3 cm in the 3D analysis.


Figure 8: Sliding displacement of 8 m high detached concrete block in central cantilever of
Gigerwald arch dam subjected to earthquake 2 under full reservoir condition
ICOLD - 12
th
INTERNATIONAL BENCHMARK WORKSHOP ON NUMERICAL ANALYSIS OF DAMS
420
Conclusions
The results of the earthquake safety assessment of the Roggiasca and Gigerwald arch dams
can be summed up as follows:
1. The ground excitation during the 10,000-year SEE produces relatively high accelerations
and tensile stresses in the central upper portion of a large arch dam.
2. The tensile strength of the vertical contraction joints and the horizontal lift joints is
usually significantly lower than that of the monolithic concrete. Thus, these joints may
open during the earthquake shaking, possibly resulting in the upper portion of a dam
block becoming fully detached from the rest of the dam.
3. Any detached uppermost part of a dam block in the investigated dams may be subjected
to some sliding and rocking motions, but it will remain stable during and after the
earthquake.
4. The earthquake loading could cause limited damages in the analyzed dams, but it will not
lead to an uncontrolled release of reservoir water. Hence, the investigated dams satisfy
the safety requirements for the SEE.
References
[1] ADINA R & D (2008). ADINA User Interface, Command Reference Manual, Vol. I:
ADINA Solids & Structures Model Definition, Report ARD 08-2, Watertown,
Massachusetts, USA.
[2] Hall, J.F. (2006). Problems encountered from the use (or misuse) of Rayleigh damping.
Earthquake Engineering and Structural Dynamics. Vol. 35, pp. 525-545.
[3] Malla, S. and Wieland, M. (2006). Dynamic stability of detached concrete blocks in arch
dam subjected to strong ground shaking. Proceedings of the 1st European Conference on
Earthquake Engineering and Seismology (ECEES), Geneva, Switzerland, 3-8 September
2006.
[4] Panduri, R., Droz, P., Malla, S., Radogna, R., Wieland, M. and Darbre, G.R. (2012).
Ongoing seismic safety assessment of Swiss dams. Proceedings of the 24th International
Congress on Large Dams, ICOLD, Kyoto, Japan, 6-8 June 2012.
[5] Shi, B., Anooshehpoor, A., Zeng, Y. and Brune, J.N. (1996). Rocking and overturning of
precariously balanced rocks by earthquake. Bulletin of the Seismological Society of
America, Vol. 86, No. 5., pp. 1364-1371.
[6] Wieland, M. and Malla, S. (2012). A simple method for seismic stability analysis of
detached concrete blocks and estimation of contraction joint openings in large arch dams.
Hydropower and Dams, Vol. 19, Issue 3, pp. 113-119.
[7] Zhang, J. and Makris, N. (2001). Rocking response of free-standing blocks under
cycloidal pulses. Journal of Engineering Mechanics, ASCE, Vol. 127, No. 5, pp. 473-483.




12
th
International
BENCHMARK WORKSHOP
ON NUMERICAL ANALYSIS OF DAMS, 2
nd
- 4
th
October, 2013, Graz AUSTRIA
Edited by
Gerald Zenz and Markus Goldgruber
Published by the Austrian National Committee on Large Dams
with the support of the Austrian Reservoir Commission (Staubeckenkommission)
I C O L D
proceedings
Austrian National Committee on Large Dams
Stremayrgasse 10/II, A-8010 Graz, AUSTRIA, Phone: ++43/316/8861, secretary@atcold.at, www.atcold.at
H
e
f
t

/

V
o
l
u
m
e

3
9







p
r
o
c
e
e
d
i
n
g
s

I
C
O
L
D

1
2
t
h

b
e
n
c
h
m
a
r
k

w
o
r
k
s
h
o
p

You might also like