Dso 06 03

RECLAMATION Managing Water in the West
Report DSO-06-03
Investigation of the Failure Modes

of Concrete Dams
Physical Model Tests
Dam Safety Technology Development Program
U.S. Department of the Interior

Bureau of Reclamation
Technical Service Center
Denver, Colorado December 2006
Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0188
The public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering
and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson
Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to
comply with a collection of information if it does not display a currently valid OMB control number.
PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To)
12-2006
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER

Investigation of the Failure Modes of Concrete Dams—Physical Model
Tests
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER

Harris, David W. and Fred Travers
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT

Bureau of Reclamation NUMBER
Technical Service Center DSO-06-03
Materials Engineering and Research Laboratory
Denver, Colorado
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
Bureau of Reclamation
Denver, Colorado
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
DSO-06-03
12. DISTRIBUTION/AVAILABILITY STATEMENT

National Technical Information Service, 5285 Port Royal Road, Springfield, VA 22161
13. SUPPLEMENTARY NOTES
14. ABSTRACT
This investigation focuses on two case histories to develop failure modes under loadings similar to earthquakes. The models are
intended primarily to produce cases which can be compared to computer analyses. The first is a 1 50 scale model of the Koyna
Gravity dam section, which cracked during an earthquake. This is a simpler model using a 2-dimensional cross section and no
water in the simulation. The second model is a simulation of an arch dam in a wide canyon with a reservoir in flight.
Conclusions are stated for trends observed.
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION 18. NUMBER 19a. NAME OF RESPONSIBLE PERSON
OF ABSTRACT OF PAGES
SAR
a. REPORT b. ABSTRACT a. THIS PAGE 19b. TELEPHONE NUMBER (Include area code)
UL UL UL
Standard Form 298 (Rev. 8/98)

Prescribed by ANSI Std. Z39.18
Mission Statements
The mission of the Department of the Interior is to protect and
provide access to our Nation’s natural and cultural heritage and
honor our trust responsibilities to Indian Tribes and our
commitments to island communities.
The mission of the Bureau of Reclamation is to manage, develop,

and protect water and related resources in an environmentally and
economically sound manner in the interest of the American public.
Acknowledgments
This research was sponsored by the Dam Safety Technology Development
Program of the Bureau of Reclamation.
iii
Contents
Page
Acknowledgments.................................................................................................. iii
Introduction............................................................................................................. 1
2-Dimensional Koyna Dam Cross Section Test ..................................................... 1
Background ....................................................................................................... 1
Physical Model Test.......................................................................................... 2
Concrete Mix Design and Material Properties............................................ 2
Model Construction and Instrumentation ................................................... 6
Input Motions.............................................................................................. 9
Test Results............................................................................................... 11
Model 1—Cracked Model .................................................................. 11
Model 2—Monolithic Model.............................................................. 15
Conclusions and Discussion ..................................................................... 19
3-Dimensional Arch Dam Simulation .................................................................. 20
Previous Work on Shake Tables ..................................................................... 20
Introduction..................................................................................................... 21
Experiment Setup and Procedure.................................................................... 23
Concrete Mix Design and Material Properties................................................ 24
Model Construction and Instrumentation ....................................................... 25
Results and Indications of the Models ............................................................ 27
In-Situ Tests for Modal Shape and Frequency ......................................... 27
Linear Versus Nonlinear Structural Behavior........................................... 27
Effect of Joints on Nonlinear Behavior .................................................... 29
Effects of a Wide Canyon ......................................................................... 30
Water in Joints .......................................................................................... 39
Conclusions........................................................................................................... 39
Recommendations................................................................................................. 40
References............................................................................................................. 40
Appendix—Investigation of the Failure Modes of Concrete Arch Dams Using

Physical Model Tests, by David W. Harris and Fred Travers
Tables
No. Page
1 Estimated concrete properties, the associated scale factors, and the model
material target values ...................................................................................... 2
iv
Contents
2 Model concrete mix components .................................................................... 3

3 Properties of model materials ......................................................................... 4
4 Averaged values of tested properties from dam core.................................... 23
5 Estimated concrete properties, the associated scale factors, and the model
material target values .................................................................................... 24
6 Model concrete mix components .................................................................. 24
7 Models with associated properties ................................................................ 25
8 Typical frequencies of dams ......................................................................... 26
Figures
No. Page
1 Stress-strain graphs. ........................................................................................ 5

2 Load vs. crack width in ⅓-point beam tension. ............................................. 5
3 Unload-reload test showing plastic behavior of the low-strength concrete. ... 6
4 First Koyna model mounted on the shake table. The shrinkage crack and
eventual failure plane is sketched in. .............................................................. 7
5 Instrument locations........................................................................................ 8
6 Second Koyna model failure plane. ................................................................ 9
7 First Koyna test—horizontal acceleration at the top of the model. .............. 10
8 The seismic record for upstream/downstream motion during the Koyna
event.............................................................................................................. 10
9 Koyna acceleration response spectrum at 5% damping................................ 11
10 First Koyna model—base acceleration of 0.5 g............................................ 12
11 First Koyna model—base acceleration of 2.25 g’s....................................... 12
12 First Koyna model—base acceleration of 2.5 g’s......................................... 13
13 First Koyna model—base acceleration of 2.75 g’s....................................... 13
14 First Koyna model—displacement at the top of the model. ......................... 14
15 Second Koyna model—frequency sweeps.................................................... 15
16 Second Koyna model—horizontal acceleration at the base of the model. ... 17
17 Second Koyna model—horizontal acceleration at the top of the model. ..... 17
18 Second Koyna model—vertical acceleration at the top of the model........... 18
19 Second Koyna model—displacement at the top of the model...................... 18
20 Final failure of Futatsuno arch dam model. .................................................. 21
21 ISMES wide arch dam model failure. Homogeneous dam shaken to failure
with earthquake simulation. .......................................................................... 22
22 ISMES test of narrow canyon dam model. ................................................... 22
23 Model 2—monolithic model accelerations................................................... 27
24 Horizontal-joint model accelerations............................................................ 28
25 Vertical-joint model accelerations. ............................................................... 28
26 17x2-joint model accelerations. .................................................................... 29
27 Initial crack normalized to stiffness.............................................................. 30
28a Monolithic model 2 initial cracking.............................................................. 31
28b Monolithic model 2 final crack..................................................................... 31
v
Investigation of the Failure Modes of Concrete Dams
29a Model 10—horizontal joint initial failure, south camera.............................. 32

29b Model 10—horizontal joint final failure....................................................... 32
29c Model 10—horizontal joint north view, initial cracking .............................. 33
29d Model 10—horizontal joint north view, final failure.................................... 33
30a Model 11—vertical joint south view, final failure........................................ 34
30b Model 11—vertical joint south view, initial cracking .................................. 34
30c Model 11—vertical joint north view, final failure........................................ 35
30d Model 11—north view, initial cracking........................................................ 35
31b Model 11—vertical joint south view, final failure........................................ 36
28a Monolithic model 2 initial cracking.............................................................. 31
29a Model 10—horizontal joint initial failure, south camera.............................. 32
32 Approximate location of all initial cracks in different models. .................... 37
33 Mode shapes for typical dam, measured in the field. ................................... 37
34 Final crack pattern and mode shapes. ........................................................... 38
35 Cracks accounting for wide canyon effects. ................................................. 38
vi
Introduction
One of the most studied cases of a dam subjected to earthquake loading is Koyna
Dam in India. This 338-foot (103-meter) high dam suffered cracking during a
magnitude 6.5 earthquake in 1967 (Chopra and Chakrabarti, 1971). During this
event, the ground acceleration in the stream direction reached 0.49 g with a total
duration of strong shaking lasting about 4 seconds. At the time of the event, the
reservoir was 37 feet below the crest.
Following the Northridge Earthquake in California on January 17, 1994 and the
Kobe Earthquake in Japan 1 year later on January 17, 1995, new consideration
has been given to the magnitude of the vertical acceleration of seismic events.
Continuing concerns about the performance of concrete dams subjected to severe
earthquakes has motivated investigation into ways to analyze and predict this
performance using nonlinear numerical analysis techniques (Donlon and Hall,
1991). In some cases, linear dynamic analyses indicate high stresses that can only
be further studied with nonlinear models.
2-Dimensional Koyna Dam Cross

Section Test
Background
Previous studies of the behavior of concrete dams subjected to seismic
accelerations have been conducted on single gravity dam monoliths (Niwa and
Clough, 1980; Norman, 1986; Tinawi et al., 1998). Chopra and Chakrabarti
(1971) and Donlon and Hall (1991) discussed development of a modeling
material that would maintain similitude with the prototype. Donlon and Hall’s
(1991) work compared to linear elastic analysis results. More recent studies have
been completed as centrifuge models (Plizzari, Saouma, and Waggoner, 1995;
Renzi, 1994). This more recent work was conducted to provide data for
comparison to numerical models.
The purpose of this investigation, conducted at the Bureau of Reclamation’s

(Reclamation) Materials Engineering and Research Laboratory, was to produce
results for comparison to nonlinear computer models. The geometry of the model
was scaled from Koyna Dam and follows previous work (Donlon and Hall, 1991;
Niwa and Clough, 1980). Because a comparison to numerical models that predict
failure was to be made, models were formulated that, to the extent possible,
maintained similitude relationships and yet were simple enough for direct
comparison with computer-predicted results. To this end, unlike previous studies
(Donlon and Hall, 1991; Niwa and Clough, 1980), similitude with reservoir
effects was not attempted. This eliminated the need to model coupling effects.
Two models were tested—a model with a natural but preexisting crack, and a
continuous model cracked during testing.
Physical Model Test

The scale chosen for this model was a 1 50 scale. Similitude requirements for
models have been summarized in other references (Krawinkler and Moncarz,
1980) and estimated properties of Koyna Dam have also been suggested (Donlon
and Hall, 1991; Niwa and Clough, 1980). These properties are summarized in
table 1.
Table 1.—Estimated concrete properties, the associated scale

factors, and the model material target values
Property Prototype estimate Scale factor Target value
2 2
E 4,000,000 lb/in 50 80,000 lb/in
2 2
(27,940,000 kN/m ) (558,800 kN/m )
2 2
fc= 4,000 lb/in 50 80 lb/in
2 2
(27,940 kN/m ) (558 kN/m )
2 2
ft 400 lb/in 50 8 lb/in
2 2
(2,794 kN/m ) (55.9 kN/m )
3 3
Density 150 lb/ft 1 150 lb/ft
c
εu 0.0025 1 0.0025
t
εu 0.00012 1 0.00012
Concrete Mix Design and Material Properties

In this study, a new, low-strength concrete mix was designed. Considerable work
has been done in previous studies (Donlon and Hall, 1991; Niwa and Clough,
1980; Donlon, 1989) to produce a similitude-appropriate concrete mix. As has
been suggested, curing and the associated shrinkage cracking can be problematic
in the use of concrete mixes with highly reduced properties. In addition, the use
of any lead product to meet density requirements poses special problems for the
handling, storage, and disposal of this hazardous substance. This latter problem,
in particular, limits the options for commercial mass production of the material
and complicates the disposal of it. When modeling nonlinear failures, additional
consideration must be given to ensuring that the correct failure mechanism is
reproduced at model scale.
2
2-Dimensional Koyna Dam Cross Section Test
The mix for this study used bentonite pellets as a component to reduce strength.
The use of bentonite pellets posed a problem logistically as saturation of the
bentonite is required prior to mixing. The mix components and proportions for
the initial laboratory-mixed concrete and the commercially mixed model concrete
are shown in table 2.
Table 2.—Model concrete mix components
Lab mix, Volume in mix, Model mix, Volume in mix,

3 3 3 3
Component lb/yd ft lb/yd ft
Air 0.14 0.52

(½% entrapped (½% entrapped
air assumed) air assumed)
Water 560 8.99 480 7.68
Cement 160 0.82 168 0.86
Bentonite 40 0.25 42 0.26
Sand 1366 8.4 1454 8.87
No. 4 —⅜" 553 3.36 0 0

gravel
⅜" - ¾" 829 5.04 1458 8.81

gravel
Note: water/cement = 3.5; bentonite/(bentonite+cement) = 20%
The trial mix was initially made in the laboratory with bentonite saturation
accomplished overnight. Based on the apparent success of this mix, both shake
table models were made using this design. Due to the volume required for a
shake table model (6 yd3), the actual model mix was supplied commercially. For
the commercially supplied concrete, it was assumed that saturation would take
place in the mixer drum during transit. Water was adjusted from the original
design at the plant to decrease sloshing in transit. On site, a slump of
approximately 7.5 inches was used as an indicator of a correct mix. Slump was
not a good indicator of strength as indicated by changes in the properties of the
concrete. Results from the two mixes are shown in table 3.
Breaks for all compressive cylinder tests failed in a classic shear plane typical of
concrete of approximately 65 degrees. Other materials were tested in the lab
based on lead/plaster combinations as trial mixes. These materials created failure
modes not typical of concrete such as horizontal layer crushing. It is clear that not
all parameters matched the similitude requirements simultaneously. Changes in
mix water had the largest effect. However, as was stated previously, the primary
intent was to produce calibration data for testing of computer models.
3
Table 3.—Properties of model materials
Apparent Apparent Apparent

scale scale scale
Lab (dimension- Model (dimension- Model (dimension-
trial al scale mix— al scale mix— al scale
Property results target) Koyna I target) Koyna II target)
Density, lb/ft3 133.1 0.9 (1.0) 135 0.9 (1.0)
E, lb/in2:
7-day 42,000 89 (50)
15-day 157,000 25 (50)
28-day 74,000 54 (50)* 55,000 72 (50)
Rapid loading: 113,000

15 days 93,457
28 days 80000
35 days
fc=
Static loading:
7-day 50 80 (50) 89 (50) 45 (50)
15-day 203 20 (50)
28-day 84 48 (50) 154 (50) 26 (50)
120-day 290
Rapid loading:
7-day 70
ft
Static:
15 days/split tension 27 15 (50)
15 days/beam tension 60
21 days/direct tension 14 29 (50)
28 days/split cylinder 12 33 (50) 20 20 (50)
Rapid loading:
15 days/split cylinder 52 8 (50)
28 days/split cylinder 22
εuc 0.004 2 (1) 0.005 2.5 (1) 0.004 2 (1)
*Density corrected by E = scale*density = 506
Laboratory testing was done in support of the tests, with standard tests being run
(e.g., typical static compression stress-strain data are shown in fig. 1), and
specialized tests were used to help assist in the calculation of parameters that may
be required in nonlinear computer material models. Typical fracture (crack width
versus load-beam test for a standard ⅓-point bending test) data are shown in
figure 2, and unload-reload data demonstrating plasticity of the material are
shown in figure 3. These tests were not intended to be an exhaustive set of all
tests required for published numerical models, but are believed to be
representative of the types of data needed.
4
Figure 1.—Stress-strain graphs.
Figure 2.—Load vs. crack width in ⅓-point beam tension.
5
Figure 3.—Unload-reload test showing plastic behavior of the low-strength concrete.
Model Construction and Instrumentation

The tests were performed in Reclamation’s Materials Engineering and Research
Laboratory. The Vibration Laboratory is used for large scale tests and has been in
existence at Reclamation since 1969 (McCafferty, 1970). For these experiments,
a shake table was constructed having movement constrained to a single axis
(horizontal only). The table was tested for its response modes and also tested in
motion with accelerometers to determine its capabilities for use at higher
frequencies. The table responded well for input frequencies below 22 Hz, which
was below the table=s lowest natural frequency of 30 Hz, but higher frequencies
were eliminated for testing. Response of the table was clearly best at frequencies
of 26 Hz and below. For this reason, a similitude simulation of an earthquake
motion was not used. Rather, for practical reasons associated with the table, and
for simplicity in numerical model calibration, a sinusoidal motion was selected.
The model is shown in figure 4 on the shake table. The 1 50 scale model resulted
in an 8.5-foot tall model, figure 5, weighing 7,850 pounds. A slab, representing a
foundation, was cast monolithically with the model to provide a fixed lower
boundary at the base of the dam. Instrumentation measured displacements and
accelerations of the model and input motion of the actuator. The general
instrumentation locations are shown in figure 5 and the accompanying table.
6
Figure 4.—First Koyna model mounted on the

shake table. The shrinkage crack and eventual
failure plane is sketched in.
The first model was cast lying down on its side. Form construction and concrete
placement were much easier with access to an entire face and only a 1-foot 9-inch
depth of material. After approximately 20 days, a small shrinkage crack appeared
in the exposed face. At this time, tension tests were run that may be useful in
modeling the onset of shrinkage. At approximately 28 days, the model was
positioned on the shake table, and the forms were removed. The shrinkage crack
was evident on the side of the model, and the sloped face and was assumed to
extend through the entire model. The plane of the crack had an inclination of
approximately 20 degrees from horizontal toward the side of the model. After
approximately 1 additional week, the surface had dried sufficiently to apply
instrumentation, and the test was run.
The second model (fig. 6) was cast upright on the shake table to avoid the
shrinkage cracking experienced in the first model. By testing earlier, the onset of
shrinkage cracking was avoided, and the second model produced a material
failure under dynamic loading. Earlier testing also held the model concrete
7
Figure 5.—Instrument locations.
Instrument ID Type Orientation Height from base
Accelerometer 1 Acceleration Horizontal, x-direction 0
Accelerometer 2 Acceleration Horizontal, y-direction 0
Accelerometer 3 Acceleration Vertical, z-direction 0
Accelerometer 4 Acceleration Horizontal, x-direction 0.66 m (2.17 ft)
Accelerometer 5 Acceleration Horizontal, x-direction 1.22 m (4.00ft)
Accelerometer 8 Acceleration Vertical 2.6 m (8.5 ft)
LVDT 1 Displacement Horizontal, x-direction 0
LVDT 2 Displacement Horizontal, x-direction 0.97 m 3.17 ft)
LVDT 3 Displacement Horizontal, x-direction 1.69 m (5.54 ft)
8
Figure 6.—Second Koyna model failure plane.
strengths lower. Laboratory testing was performed on test specimens of the same
material immediately following the breaking of the model.
Input Motions
Numerical analysis predicted that the frequency of the fundamental mode of the
model was approximately 14 Hz, but this fundamental mode was out of the plane
of the test, that is, side to side in the 2-D model. The cantilever mode, mode 2 of
the model but the first mode in plane of the test, was predicted at approximately
28 Hz. Modal sweeps were run on the model at frequencies starting at 2 Hz and
increasing to 28 Hz with a constant input acceleration of 0.1 g. The results are
shown in figure 7. The first input frequency that showed an increase of
acceleration above the input of 0.1 g was 14 Hz. The effect was demonstrated in
the plane of testing. Higher frequencies did produce a more dramatic effect. A
sinusoidal motion of 14 Hz (approximately 2 Hz prototype) was chosen for the
input for all subsequent calibration tests as this lowest response frequency was
believed to be the easiest for numerical simulation and calibration. The
earthquake record for upstream/downstream motion of the Koyna event (see
fig. 8) is believed to have a primary component at 2.4 Hz, a 0.42-second period..
This is more readily seen in the response spectrum of figure 9. With this set
frequency of 14 Hz, the accelerations that developed in a single horizontal
direction (upstream and downstream to the model) were increased until failure
occurred.
9
Figure 7.—First Koyna test—horizontal acceleration at the top of the model.
Figure 8.—The seismic record for upstream/downstream motion during the Koyna event.
10
Figure 9.—Koyna acceleration response spectrum at 5% damping.
Test Results
As mentioned above in the discussion of the input motion, the chosen input
motion for the model was a 14-Hz sinusoid.
Model 1—Cracked Model

Four typical acceleration plots are shown in figures 10 through 13 for model 1. In
figure 10, the acceleration of the base of the dam and the acceleration at the base
of the known crack were measured to be 0.5 g, while the acceleration of the crest
of the model measured nearly 2 g’s. This magnification of acceleration with
height from the base of approximately 4 times, is similar to tests reported in the
literature (Donlon and Hall, 1991; Niwa and Clough, 1980). The model did not
show failure characteristics at this acceleration, which corresponds to the field
case.
As this model was to be used with computer programs to model sliding failure
mechanisms, testing was continued. Base accelerations were increased while
maintaining the 14-Hz input motion. From this point, the constant input
frequency has the advantage of seeing changes in response as model
11
Figure 10.—First Koyna model—base acceleration of 0.5 g.
Figure 11.—First Koyna model—base acceleration of 2.25 g’s.
12
13
characteristics change. At about a 2-g acceleration at the base, a puffing of

material from the crack was observed. This was caused by a rocking motion of
the top piece of the model (the block above the crack), which acted as a bellows,
blowing worn material from the cracked surface.
The next increment in acceleration, at a base acceleration of 2.25 g’s showed a

change in response of the portion of the dam above the crack. As can be seen in
figure 11, the magnification of acceleration from base to top attenuates, showing a
maximum of 3.75 g’s or a magnification factor of 1.6 times. There is evidence of
a phase shift of motion between the top and bottom at this time in the testing.
The next ramp of acceleration was to an acceleration of 2.5 g’s of the base. As
can be seen in figure 12, with this acceleration, the top and bottom of the model
show nearly equal acceleration, with a full 180-degree phase shift between the
pieces. As can be seen in figure 14, which shows displacement at the top of the
model, the top of the dam is sliding along the base by this time.
Finally, at a base acceleration of 2.75 g’s, the bottom motion is at a higher

acceleration than the top of the dam (fig. 13). By this time, the displacement of
the top piece is well under way, approximately 2 inch (fig. 14), and the base
Figure 14.—First Koyna model—displacement at the top of the model.
14
motion is not readily transferred to the top section. The cross section maintained
stability, and sliding progressed slowly during the input. The top block could be
observed to be progressively sliding down the preexisting shrinkage crack
surface.
Model 2—Monolithic Model

As with the first model, a modal sweep was completed first. Accelerations,
normalized to the base motion and recorded during the sweep, are shown in
figure 15. In comparing the modal responses with those for the first model,
shown in figure 7, some differences are noticed in the response frequencies. In
model 1, 24 Hz seemed to indicate the first cantilever mode. In model 2, the first
modes that show an amplification factor above the input 0.1 g are 20 Hz and
22 Hz. Review of the directional component shows that this mode appears to
emerge at 22 Hz. In both models, another increased response with respect to the
input motion occurs at 28 Hz. These differences are believed to be inherent
differences in the two models as built, but generally the two models appear
similar in their modal response.
Figure 15.—Second Koyna model—frequency sweeps.
15
Model 2 was tested using the same strategy as the first model, that is, a 14-Hz
sinusoidal motion was used and ramped in acceleration until failure. The total test
duration using this method was almost 8 minutes. The final failure occurred at
the change in slope of the model on a flat downward plane sloping toward the
upstream face. The angle of the failure plane was 53 degrees from horizontal,
which was 90 degrees from the lower slope in the bottom of the model, beginning
at the invert. This angle is consistent with previous studies (Niwa and Clough,
1980) but was a single flat surface.
The model was videotaped during testing. Review of the tape revealed that the
crack was not visible in one frame and had propagated completely by the next
video frame. The standard video frame rate is approximately 1 30 of a second.
With the input motion of 14 Hz, the period for 2 cycle would be 1 28 of a second.
This indicates that the crack developed and propagated in less than 0.03 second,
either during a stroke, or more probably, at the reversal of a stroke. The base
acceleration at the time of failure was 2.2 g’s.
Analysis of the test data revealed anomalous behavior beginning approximately

330 seconds into the test (figs. 16-19). This behavior is most prominently
displayed in figure 18, which is the vertical acceleration of the model measured at
the top of the structure. It can be seen that up to the 330 second point in the test,
the vertical acceleration increases linearly with increasing horizontal input
acceleration. This response is as expected and is attributed to a slight flexing of
the shake table frame. At around 330 seconds, the vertical acceleration starts
increasing dramatically and continues to increase throughout the duration of the
test. This increase is accompanied by a corresponding decrease in the horizontal
acceleration of the top of the structure as seen in figure 17. Figure 19 shows a
rather abrupt decrease in the displacement of the top, which would correspond
with the decreased acceleration. These phenomena are not believed to be related
to the failure of the dam portion of the model, but rather appear to be a failure in
the base of the model, which acted as the foundation of the structure.
The conclusion from these data is that the material around the all-thread
embedded in the base started failing at around 330 seconds and allowed the model
to rock. As more material failed, the rocking increased, which resulted in the
increasing vertical accelerations and decreased acceleration of the top, initially.
Eventually, the material failure around the all-thread was severe enough that the
entire model could slide back and forth a small amount in the direction of the
excitation. This is evidenced by the spikes in base acceleration shown starting
around 400 seconds in figure 16. This indeterminate boundary condition would
be nearly impossible to model on a nonlinear analysis time-step basis. It is
believed that general comparisons can still be made based on the final
acceleration and the material properties presented.
16
Figure 16.—Second Koyna model—horizontal acceleration at the base of the model.
Figure 17.—Second Koyna model—horizontal acceleration at the top of the model.
17
Figure 18.—Second Koyna model—vertical acceleration at the top of the model.
Figure 19.—Second Koyna model—displacement at the top of the model.
18
It was noted that after initiation of the crack, the top of the model began to slide
before toppling occurred. The top portion toppled from the model approximately
1 second (14 cycles) after crack propagation.
Conclusions and Discussion

1. A new low-strength concrete mix is proposed that shows promise for use in
similitude testing. The mix, which uses bentonite as the media to reduce
strength properties, is readily adjusted to various scales. The components
may be mixed en masse and can be provided by commercial producers
because no hazardous materials are used. Disposal is also easily
accomplished by conventional methods.
2. The new mix produced strength and stiffness characteristics that nearly
matched the similitude requirements. More importantly, for nonlinear
modeling of the failure mechanism, the mix fails in a shear plane almost
identical to that of conventional concrete.
3. The initially cracked model and the monolithic model showed general mode
shapes and damping that were similar for small accelerations.
4. The kinematically nonlinear model (sliding model) demonstrated that there

was some initial bond on a typical shrinkage crack, even a crack visible to
the eye on multiple faces, which needs to be overcome before sliding can be
initiated.
5. Once sliding starts, the nonlinear effect creates very large changes in the
dynamic response under a constant frequency sinusoidal input motion. The
amplitude of the acceleration of the piece above the crack in this model
actually becomes less than the amplitude of the acceleration of the base, and
the response is phase shifted. Put simply, the base can slide back and forth
beneath the top with the motion being nearly uncoupled.
6. The monolithic model failed with a material failure that was characteristic of
previous models and believed to be characteristic of cracks in the field.
7. During the monolithic test, a nonlinear change in the base fixed-boundary

condition created a highly nonlinear and indeterminate boundary condition.
This nonlinear change also showed large changes in the dynamic response of
the model, which are easily seen when compared to the constant input
motion. Unfortunately, this same boundary condition change makes exact
time history matching of numerical models impossible.
8. Both models failed at approximately 2.2 g’s of acceleration. In the kinematic

model, the crack allowed a slow progressive sliding during the cyclic motion.
In the materially nonlinear model, a crack was initiated in less than 1 30 of a
second and sliding occurred for a number of cycles before the top of the
19
model toppled. The toppling is inconsistent with previous models and is

believed to be related to the vertical accelerations caused by the boundary
condition change.
9. Laboratory tests of the material were performed in conjunction with the

shake table models to provide parameters typically needed in nonlinear
numerical material models.
10. Results from the kinematic failure model (sliding) can conceivably be time
step matched to verify nonlinear models. Results from the materially
nonlinear models can be verified in a general manner to verify the cracking
pattern and acceleration required for failure.
3-Dimensional Arch Dam Simulation
Previous Work on Shake Tables

The arch section of the Techi Dam, Taiwan, has been modeled at a scale of 1150
(Niwa and Clough, 1980). A primary purpose of that study was to model the
opening of joints; thus the dam was articulated. The model was tested with
motion in two planes, upstream/ downstream and cross-canyon. Vibration mode
frequencies were tested by suspending a weight from the model and subsequently
cutting the weight loose to produce face vibrations. The El Centro Earthquake
record was used in the shake table model test with a time reduction of 1150 .
Intensity was increased until the model collapsed. Significant joint degradation
occurred at the arch end, probably due to local crushing at one end. In biaxial
excitation the arch collapsed with 1.34-g acceleration in the upstream/downstream
direction and 0.91 g in the cross-canyon direction. The collapse occurred
surprisingly close to the end of the excitation.
The same authors (Niwa and Clough, 1982) noted that the test results showed
significant nonlinear behavior. Significant influence of the joints generated
increasingly poor correlations with the numerical analysis. Crushing in joints is
suggested to improve analytical results.
Other tests have been conducted with shake table models (Oberti, and Castoldi,
1981) on smaller scale models at a scale of 1100 and were tested to failure. A
sinusoid input motion was used to determine characteristic frequencies of the
structure. Next, similitude-corrected time histories of seismic events were used at
increasing amplitudes to induce failure. Natural frequencies appear to be
produced correctly when the foundation is carried 1 to 2 times the height of the
dam both depthwise and in the lateral directions. The length of the reservoir
20
shows no significant effect in the failure modes with reservoir length being
2 times the dam height.
Test models have been completed to model sliding only (Mir and Taylor, 1996).
Typically, these models were 1,000 mm in height. A short reservoir tank with
absorbent rubber was used to eliminate the hydrodynamic effect (the tank was
0.4 m in length). The input motions used were (1) a 7.5-Hz sine wave (this
waveform was hard to match exactly because some free vibration was present in
table), (2) 5 Hz ramped up in 5 cycles, held for 10 cycles, then ramped down in
5 cycles, and (3) a simulated earthquake of 12 seconds duration with the input
ramped until slipping occurred.
Several models have been constructed to simulate failures in dams. One such
model was the Futatsuno Arch Dam (Yoshida and Baba, 1965). This 76-meter
high, 210-meter crest dam was modeled at 1 50 scale. In this model, the first crack
appeared at 0.27 g and 32 Hz in a spillway pier, and on the dam at 0.41 g and
30 Hz. Final collapse occurred at 0.69 g, and 17 Hz. The final failures are shown
in figure 20.
Testing of models run to failure has been conducted at the ISMES facility in
Italy with the results shown in figures 21 and 22 for wide and narrow canyons,
respectively (Oberti and Castoldi, 1980; Oberti and Lauletta, 1967).
Introduction
All of the models chosen for this study were based on a similar geometry. Models
were run in a sequence of (1) a monolithic dam, (2) 1 horizontal joint at
approximately mid-height, (3) 1 vertical joint at the mid-point of the valley, and
Figure 20.—Final failure of Futatsuno arch dam model.
21
Figure 21.—ISMES wide arch dam model failure. Homogeneous dam shaken to failure
with earthquake simulation.
Figure 22.—ISMES test of narrow canyon dam model.
22
(4) 17 vertical joints and 2 horizontal joints (one model was run with 17 vertical
joints as a test of construction method). All models were run with a sinusoidal
input motion at 14 Hz beginning at 0.25 g and increasing every 30 seconds by
0.25 g until a structural collapse was created in the model.
Experiment Setup and Procedure

The scale chosen for this model was 1150 . Similitude requirements for models
have been summarized in other references (Krawinkler and Moncarz, 1980).
Properties of mass concrete in dams have previously been published (Harris,
Mohorovic, and Dolen, 2000) and are summarized in table 4. For this study,
typical values of properties were chosen as target parameters for the model and
are summarized in table 5. A summary of this work is contained in the appendix
and in this presentation it is suggested that the model values fall in appropriate
ranges compared to field values.
Table 4.—Averaged values of tested properties from dam core
Dynamic compressive Dynamic modulus of Dynamic tensile

2 2 2
Dam strength, lb/in elasticity, Mlb/in strength, lb/in
Pine Flat 5280 3.43 Unknown
Deadwood 5930 3.83 690
Stewart Mountain 5350 3.99 515
Roosevelt 4090 4.21 755

Saturated 1a
Roosevelt 4600 4.09 485

Air dried 1b
Roosevelt 6430 4.84 840

Saturated all other
Roosevelt 4850 4.10 840

Air dried all other
Roosevelt 3730 5.70 575

12" diameter
Air cured
Hoover 8040 4.33 975
Folsom 4760 4.50 510
Monticello 4870 6.12 505
Englebright 6660 4.63 585
23
Table 5.—Estimated concrete properties, the associated scale

factors, and the model material target values
Property Prototype estimate Scale factor Target value
2 2
E 5,200,000 lb/in 150 35,000 lb/in
2 2
36,322,000 kN/m 244,475 kN/m
2 2
fc= 4,500 lb/in 150 30 lb/in
2 2
31,432 kN/m 209 kN/m
2 2
ft 450 lb/in 150 3 lb/in
2 2
3,143 kN/m 21 kN/m
3 3
Density 150 lb/ft 1 150 lb/ft
3 3
2,403 kg/m 2,403 kg/m
c
εu 0.001 1 0.001
t
εu 0.0001 1 0.0001
Concrete Mix Design and Material Properties

In this study, a new low-strength concrete mix was designed. Considerable work
in this area has been accomplished in previous studies to produce a similitude-
appropriate concrete mix (Donlon and Hall, 1991; Niwa and Clough, 1980;
Donlon, 1989). Work from Harris, et al. (2000) was used previously with
2-dimensional model studies in Reclamation’s Engineering Materials and
Research Laboratory. The advantages of this design were discussed previously.
This design was modified for the 1150 scale models of this study; the mix
components and proportions are shown in table 6.
Table 6.—Model concrete mix components
3
Component Lab mix, lb/yd Volume in mix
Air 0.14 (½% entrapped air assumed)
Water 739.8 11.86
Cement 156.6 0.8
Bentonite 63 0.39
Sand 2245.41 13.81
Note: water/cement = 4.72; bentonite/(bentonite+cement) = 28.7%
Early tests with this material demonstrated that all scaled parameters desired
could not be met simultaneously. Specifically, the modulus of the material was
on the order of 3,500 lb/in2, and density was 118 lb/ft3. With these constants,
24
coefficients for calculating the required similitude relations were derived

following the modeling theory developed by Houqun et al. (1994). The
coefficients used were: CE = 4,500,000/3,500; Cρ = 150/118; and Cs = 150, where
CE represents the coefficient for modulus, Cρ represents the coefficient for
density, and Cs is the scale factor. Table 7 introduces the strength and stiffness
and the series of models produced for this study.
Table 7.—Models with associated properties
Failure
Avg. Avg. Avg. acceleration, g
Age, comp, split, beam,
Model Type Date days lb/in2 lb/in2 lb/in2 E Initial Final
crack failure
M-1 Monolith 3/12/99 7 28.1 3.2
M-2 Monolith 3/31/99 7 23.25 4 2302.6 0.75 5.0
M-4 Monolith—1st 7/28/99 6 33.64 3761

pulse
M-5 Monolith 8/12/99 6 37.5 4.83 3.1 3088.4
M-6 Vertical joint 8/27/99 7 25.6 4.67 0.7 0.85

(saloon door)
M-7 Vertical joint 5/2/00 8 40.3 1.5 1.5
M-8 Vertical joint 6/21/00 7 28.2 1.5 1.5
M-9 Horizontal joint 7/19/00 7 36 0.95 1.75
M-10 Horizontal joint 8/2/00 7 52.1 5172.1 1.65 1.65
M-11 Vertical joint 8/22/00 6 41.97 4.4 3758.8 1.2 1.2
M-12 17 vertical 10/3/00 5 27.01 2146.6 0.6 0.6

joints
M-13 17 vertical 4/10/01 5 23.2 3.58 3461.5 0.5 0.5

joints
M-14 2 horizontal, 17 4/24/01 5 19.625 3.09 3099 1 1.25

vertical joints
M-15 2 horizontal, 17 5/1/01 5 21.4 4.05 3948 0.75 0.75

vertical joints
Model Construction and Instrumentation

As discussed in the previous section, sinusoidal loadings were used as input.
The selection of the frequency of the motion was based on the similitude
coefficients and typical values of natural frequencies of dams tested onsite. The
frequency conversion similitude coefficient becomes:
Cρ
Cf =CS
CE
25
For this model, the scale (Cs) is 1150 . Considering the range of modulus values of
the materials in various models (CE), a range of frequency coefficients would be
from 4 to 5.36. Table 8 shows typical measured values of natural frequencies for
specific dams. A typical first mode natural frequency of 3.3 Hz was chosen for a
wide valley structure. Applying the range of similitude coefficients to this
frequency yields a range of model scale frequencies from 12 to 16 Hz and hence
the chosen excitation frequency of 14 Hz. In the summary discussion in the
appendix, a discussion of appropriate values is based on ranges of values from
field cases.
Table 8.—Typical frequencies of dams (Takahashi, 1964; Rouse and Bouwkamp, 1967; Duron
and Hall, 1988; Oberti and Castoldi, 1981; Houqun et al., 1994)
Frequency (Hz) Damping (%)
Sym. Asym. Sym. Asym.

Reservoir
Dam level st nd rd st nd st nd rd st nd
1 2 3 1 2 1 2 3 1 2
Kamishiba Full 3.8 5.8 8.7 4.3 7.2 5 4 4.5 4 4.5
Low - 6.3 9.7 4.7 8.0 - 4 4.5 4 4
Sazananigawa Full 5.5 6.8 - 4.3 8.7 2 3.7 - 3 2
Low 6.7 - - 5.5 - 1.8 - - 1.8 -
Monticello - 3.13 4.68 7.60 3.55 6.00 2.7 2.5 2.4 2.2 2.1
Morrow Point - 2.95 3.95 5.40 3.30 6.21 4.0 3.9 4.3 1.5 3.3
Alpe Gera 78% 3.25 - - 4.56 - 5.4 - - 5.12 -

gravity
Empty 3.47 6.16 - 4.72 7.43 4.4 4.50 - 4.50 3.43
Fiastra gravity 88% 4.72 7.87 - 5.97 9.72 3.27 2.38 - 2.46 2.50
72% 4.29 7.34 - 2.56 9.16 3.30 2.80 - 7.34 2.50
Place Moulin 95% 2.03 3.63 - 2.03 2.96 1.15 1.20 - 1.18 1.22
arch
Talvacchi arch 90% 3.8 5.35 - 3.68 6.7 3.5 3.5 - 3.5 3.5
Barcis arch Full 10.1 15.3 - 7.6 16.3 4 4 - 7.0 3.5
Ambiesta arch Full 4.27 7.3 - 3.90 - 3.02 6.65 - 2.15 -
94% 4.7 - - 2.0 - 1.9 - - 4.7 -
26
Results and Indications of the Models

In-Situ Tests for Modal Shape and Frequency
Tests were conducted on the model using frequency sweeps at very low
accelerations, 0.5 g, to physically test for the model modes. The low acceleration
was chosen to ensure that damage would not occur during the frequency sweeps.
Unfortunately, the low excitations did not produce sufficient displacement for a
conclusive determination of the model’s dynamic behavior.
Linear Versus Nonlinear Structural Behavior

The models all acted in a structurally linear manner until the onset of any
cracking. This was evidenced by a characteristic mode shape and increase of
acceleration from the base of the dam to the top of the dam. This pattern is
shown for the monolith, horizontal-only joint, vertical-only joint, and 17x2-joint
models in figures 23 through 26, respectively. These figures all show time plots
of acceleration over a fairly broad time. From the plots, the increase between top
acceleration versus the base of the dam is shown in the overlay of data. The
initiation of cracking causes a nonlinear effect, which results in a change in the
pattern. A good example of this is shown in figure 24 with an obvious change in
response. In figure 25, another model is shown with an expanded scale time.
Figure 23.—Model 2—monolithic model accelerations.
27
Figure 24.—Horizontal-joint model accelerations.
Figure 25.—Vertical-joint model accelerations.
28
Figure 26.—17x2-joint model accelerations.
What is clear in all the figures is that there is a linear increase of acceleration with
height until the initiation of cracking. When cracking occurs, this linear behavior
changes rapidly. This sudden and extremely different nonlinear behavior is
highly dependent on joint type as described in the next section.
An investigation of the linear effect for different types of models is shown in

figure 27, with the critical, or acceleration to cause cracking, normalized to a the
stiffness within each model. In this figure, it can be seen that, for this model,
cracks generally occurred at an acceleration of 0.70 g. As all models are linear
until the initiation of cracking, this parameter is unaffected by model type.
Effect of Joints on Nonlinear Behavior

Photos of the different models are shown in figures 28 through 31. These models
employed the different jointing patterns of monolith, horizontal joint, vertical
joint and multiple joint, respectively. From these figures, it is easily seen that the
joint patterns have a great effect on the initial and final cracking patterns.
Figure 32 shows all initial cracking patterns overlain on the same picture. The
joints control the initial cracking pattern. The monolith breaks initially into one
fairly large piece from approximately ¼ of the distance across the canyon, down
to about the ¼ of the height, and then up the centerline. All other patterns are
controlled by the joints. The vertical joint creates a deep crack along the joint, the
horizontal joint forming the predominant pattern in the first crack. In the 17x2-
29
1 .8
1 .6 1 H o r iz o n ta l
1 .4
1 .2 1 V e r tic a l
In i ti a l C r a c k A c c e l e r a ti o n B e fo r e
1 17x2
N o r m a l i z a ti o n
0 .8
M o n o lith
0 .6 1 7 V e rt ic a l J o in t s
0 .4
0 .2
0
0 0 .2 0 .4 0 .6 0 .8 1
N o r m a l i ze d A c c e l e r a ti o n o f F i r s t C r a c k
Figure 27.—Initial crack normalized to stiffness.
joint case, the upper horizontal joint forms the predominant pattern with three
vertical joints in the initial pattern.
The final crack pattern is different for each model, once again showing the
influence of the constructed joints. Figure 33 shows expected inflection points for
typical linear analysis. Figure 34 shows the final pattern for the 17x2 model with
the mode inflection points superimposed.. The pattern of the formation of five
large blocks in the failure mode is somewhat common in all models and might be
suggested as a pattern that is consistent with the modes of the structure. The final
failure in all models occurred after considerable time, on the order of 20 to
30 seconds of shaking following the initial crack. This is consistent with
comments by other authors (Niwa and Clough, 1982) and is most likely
associated with the need to abrade joints before there is sufficient space for the
blocks to snap through in the downstream direction.
Effects of a Wide Canyon

Figure 35 shows the initiation of failure in the dam at the ¼ point across the
canyon and the centerline of the dam. Two differences are noted:
1. The onset of nonlinearity occurs first in the centerline and about 2 second
later at the quarter point. Although this seems like a small difference,
accounting for the time similitude of 12.8 times for the model to field, this
30
Figure 28a.—Monolithic model 2 initial cracking.
Figure 28b.—Monolithic model 2 final crack.
31
Figure 29a.—Model 10—horizontal joint initial failure, south camera.
Figure 29b.—Model 10—horizontal joint final failure.
32
Figure 29c.—Model 10—horizontal joint north view, initial cracking.
Figure 29d.—Model 10—horizontal joint north view, final failure.
33
Figure 30a.—Model 11—vertical joint south view, final failure.
Figure 30b.—Model 11—vertical joint south view, initial cracking.
34
Figure 30c.—Model 11—vertical joint north view, final failure.
Figure 30d.—Model 11—north view, initial cracking.
35
Figure 31a.—Model 11—vertical joint south view, final failure.
Figure 31b.—Model 11—vertical joint south view, final failure.
36
Figure 32.—Approximate location of all initial cracks in different models.
A
C
Figure 33.—Mode shapes for typical dam, measured in the field.
37
A B C
Figure 34.—Final crack pattern and mode shapes.
Figure 35.—Cracks accounting for wide canyon effects.
38
Conclusions
difference is 6.4 seconds later in real time, which is a significant portion of

an earthquake record.
2. The degradation of the acceleration of the ¼ point is almost immediate,

whereas there is a time period of approximately 2 seconds of increased
motion in the centerline before the degradation of acceleration is observed.
Water in Joints
One additional issue for consideration is the ability of water to penetrate joints
during the earthquake. Water was observed squirting from cracks in the dam as
they opened and closed after failure. It is noted that no considerations were made
to model the viscosity of the water for the model scale. Nevertheless, water is
clearly seen traveling through the dam in this sequence.
Conclusions
1. The Koyna Model gave results similar to previous studies and to what
actually happened in the field. The 3-D arch dam model compares with
previous models and linear measurements such as response frequencies made
in the field.
2. All models show the onset of sudden cracking and pronounced structural
nonlinearity following cracking. This nonlinearity is characterized by the
bottom of the dam being able to slip back and forth beneath the top of the
dam.
3. The arch dam model demonstrates a critical acceleration of 0.70 g’s for first
cracking of this specific model independent of joints in the models.
4. The crack pattern in the models is dominated by the joint patterns.
5. The time to final full failure when converted to full scale times exceeds the
duration of any recorded earthquake.
6. Final failure is a push through of the dam into downstream. This failure
mechanism appears to require abrasion in the joint before it can be
established.
7. Water does pass through cracks in the model in approximately 1 of a

10
second.
39
Recommendations
1. Physical models can be used to find extreme cases needed for the evaluation
of critical structures. The results can be used to improve understanding of
numerical modeling or to develop and improve capability. The results can be
used to develop initial and terminal failure modes for issues such as risk
analyses.
2. Some data needed for nonlinear modeling has been gathered using
preliminary testing methods. Additional work is required to establish
methods to find necessary parameters.
3. Wide canyon dams are clearly affected by the cross-canyon mode shapes.
Some studies need to be conducted for narrow canyons to investigate the
initiation of cracking and failure mechanisms.
References
Bureau of Reclamation, Vibration Studies of Monticello Dam, Research Report
No. 9, United States Department of the Interior, 1967.
Chopra, A.K. and P. Chakrabarti, The Koyna Earthquake of December 11, 1967
and the Performance of Koyna Dam, Report No. EERC 71-1, Earthquake
Engineering Center, University of California, Berkeley CA, 1971.
Donlon, W.P., Experimental Investigation of the Nonlinear Seismic Response of

Concrete Gravity Dams, Report No. EERL 89-01, Earthquake Engineering
Research Laboratory, California Institute of Technology, Pasadena, 1989.
Donlon, W.P. and J.F. Hall, “Shaking Table Study of Concrete Gravity Dam
Monoliths,” Earthquake Engineering and Structural Dynamics, Vol. 20, 1991,
769-786.
Duron, Z.H. and J.F. Hall, “Experimental and Finite Element Studies of the
Forced Vibration Response of Morrow Point Dam,” Earthquake Engineering and
Structural Dynamics, Vol. 16, 1988.
Harris, David W., C.E. Mohorovic, and T.P. Dolen, “Dynamic Properties of Mass
Concrete Obtained From Dam Lines,” ACI Materials Journal, Vol. 97, No. 3,
American Concrete Institute, 2000.
Harris, David W., Nathan Snorteland, Timothy Dolen, and Fred Travers,
“Shaking Table 2-D Models of a Concrete Gravity Dam,” Earthquake
Engineering and Structural Dynamics 2000, Vol. 29, 2000, 769-787.
40
References
Houqun, Chen, et al., Model Test and Program Verification on Dynamic Behavior
of Arch Dams with Contraction Joints, Institute of Water Conservancy and
Hydroelectric Power Research, Report No. SVL-94/02, July 1994.
Krawinkler, H. and P.D. Moncarz, Similitude Requirements for Dynamic Models,

American Concrete Institute, v SP-73, 1980.
McCafferty, R.M., Test Facilities USBR Vibration Test System, Shock and
Vibration Bulletin, Bulletin 41, December 1970, 109-117.
Mir, R.A. and C.A. Taylor, “An Investigation Into Base Sliding Response of
Rigid Concrete Gravity Dams to Dynamic Loading,” Earthquake Engineering
and Structural Dynamics, Vol. 25, 1996, 79-98.
Niwa, A. and R.W. Clough, Non-Linear Seismic Response of Arch Dams,

Earthquake Engineering and Structural Dynamics, Vol. 10, 1982, 267-281.
Niwa, A. and R.W. Clough, Shaking Table Research On Concrete Dam Models,
Report No. UCB/EERC 80-05, Earthquake Engineering Research Center,
University of California, Berkeley, CA, 1980.
Norman, C.D., Dynamic Failure Tests and Analysis of a Model Concrete Dam,
Technical Report SL-86-33, U.S. Army Waterways Experiment Station,
Vicksburg, MI, 1986.
Oberti, G. and A. Castoldi, “The Use of Models In Assessing the Behavior of

Concrete Dams,” Dams and Earthquakes—Proceedings of a conference held at
the Institution of Civil Engineers, London on 1-2 October 1980, Thomas Telford
Limited, London, 1981.
Oberti, G. and E. Lauletta, “Structural Models for the Study of Dam Earthquake
Resistance,” Ninth International Congress on Large Dams, Istanbul, Turkey,
Sep. 4-8, 1967.
Plizzari, Saouma, and Waggoner, “Centrifuge Modeling and Analysis of Concrete

Gravity Dams,” Journal of Structural Engineering, Vol. 121, No. 10, October
1995, 1471-1479.
Renzi, Ferrara, and Mazza, Cracking In a Concrete Gravity Dam: A Centrifugal

Investigation, International Workshop on Dam Fracture and Damage, Chambery,
France, 1994.
Rouse, George C. and Jack G. Bouwkamp, Vibration Studies of Monticello Dam,

Research Report No. 9, Bureau of Reclamation, United States Department of the
Interior, 1967.
41
Takahashi, T., Results of Vibration Tests and Earthquake Observations on

Concrete Dams and Their Considerations, ICOLD Congress No. 8., Edinburgh,
Report 14, Vol. II, 1964, 239-260.
Tinawi, R., P. Leger, M. Leclerc, and G. Cipolla, Shake Table Tests for the
Seismic Response of Concrete Gravity Dams, Eleventh European Conference on
Earthquake Engineering, Paris, September 1998.
Yoshida, T. and K. Baba, “Dynamic Response of Dams,” Proceedings of the

Third World Conference in Earthquake Engineering, Auckland, Vol. 2, 1965,
748-764.
42
Appendix
Summary Paper:
Investigation of the Failure Modes of Concrete Arch Dams Using Physical Model
Tests, by David W. Harris and Fred Travers
Investigation of the Failure Modes of Concrete Arch Dams using Physical Model Tests
By
David W. Harris1, Fred Travers2
Summary
This paper describes the design and testing of physical models of concrete arch dams.
The models are approximately 1/150 scale of a typical wide canyon dam. Material properties are
adjusted to the model scale, a reservoir and foundation are included. The models utilize different
joint patterns of a monolith, single horizontal and vertical joints, and multiple vertical and
horizontal joints. Models are tested to collapse of the structure. Conclusions are that initial
behavior of the structure is not influenced by the joints. Initial cracking of the models is
influenced by the joints. Final collapse is a push through of sections of the dam downstream and
is controlled by the joints both with the pattern of the failure and with the acceleration required
to fail the structure. Water was observed passing through the model during the loading.
1
Group Manager, Materials Engineering and Research Laboratory
U.S. Bureau of Reclamation, D-8180
PO Box 25007
Denver, CO 80225
303-445-2375 Phone
303-445-6341 FAX
Email: dwharris@do.usbr.gov
2
Electronics Engineer, Materials Engineering and Research Laboratory
Key Words: Concrete dam, Earthquake, Physical Model, Arch dam, Failure
1
I. Introduction
The earthquake event in 1967 and subsequent cracking of Koyna Dam led to
advanced studies of dams and the effects of large loads.1 Pacoima Dam showed a opening in a
contraction joint of nearly a centimeter following the San Fernando Earthquake of 19712.
Following the Northridge Earthquake in California on January 17, 1994 and the Kobe Earthquake
in Japan 1 year later on January 17, 1995, new consideration has been given to the magnitude of
the vertical acceleration of seismic events. Additional damage occurred in the Pacoima Dam
along the abutment of the dam in the Northridge Earthquake. An extensive overview of work
done in the investigation of dam performance, dam properties based on field observations, and
laboratory testing was presented by Hall in 19882. The field reference values from this work are
used in this study to configure parameters for a physical model. Continuing concerns about the
performance of concrete dams subjected to severe earthquakes has motivated investigation into
ways to analyze and predict this performance using both physical modeling and nonlinear
numerical analysis techniques3.
Previous model studies of the behavior of concrete dams subjected to seismic
accelerations have been conducted on gravity dam monoliths3,4,5,6. In references 3 and 4 attention
was given to developing a modeling material which maintained similitude with the prototype. In
this paper a new modeling material is suggested for models on the scale of 1/150. More recent
studies have been completed as centrifuge models7,8. In the centrifuge model of Toktogul Dam9
cracks appeared along the abutments of the model and along a horiziontal line similar to the
results which are found in this paper.
2
Arch dams have also been tested as physical models. The arch section of the Techi Dam,
Taiwan has been modeled at a scale of 1/1504. A primary purpose of that study was to model the
opening of joints, thus the dam was articulated. The model was tested with motion in 2 axes,
upstream/downstream and cross-canyon. Vibration mode frequencies were tested by suspending
a weight from the model and subsequently cutting the weight loose to produce face vibrations.
The El Centro Earthquake record was used in the shake table model test with a time reduction of
1 / 150 . Intensity was increased until the model collapsed. Significant joint degradation
occurred at the arch end, probably due to crushing of the material at the end. In biaxial excitation
the arch collapsed with 1.34 g acceleration in the upstream/downstream direction and 0.91 g in
the cross-canyon direction. The collapse occurred close to the end of the excitation. It was
noted10 that the test results showed significant non-linear behavior. Significant influence of the
joints generated increasingly poor correlations with the numerical analysis. A method to account
for crushing in joints was suggested to improve the analytical results. The physical model in this
study included joints as discontinuities. The input motion is sinusoidal, with collapses being
induced by an increase in the intensity of the record. As stated in this paper, collapses occurred
near the end of the testing record, similar to the previous study.
Other failure tests of models11 have been conducted with shake tables at a scale of 1/100. A
sinusoid input motion was used to determine characteristic frequencies of the structure. Next,
similitude-corrected time histories of seismic events were used at increasing amplitudes to induce
failure. Natural frequencies appeared to be produced correctly when the foundation was carried 1
to 2 times the height of the dam both depthwise and in the lateral directions. The length of the
reservoir showed no significant effect with reservoir length being 2 times the dam height. Fot the
3
models in this study a rigid (stiff) foundation is used which does not represent the field case.
However, the frequency contact does appear to match the field case. A reservoir is included with
the length of about the depth of the reservoir; a dissipating wall was used to reduce reflection
effects.
Another model12 study was the Futatsuno Arch Dam. This 76 m high, 210 m crest dam
was modeled at 1/50 scale. In this model the first crack appeared at 0.27 g and 32 Hz in a
spillway pier. On the dam cracking initiated at 0.41 g and 30 Hz. Final collapse occurred at 0.69
g, and 17 Hz. Accelerations which induce failures are similar in this paper. Models have also
been tested at the ISMES facility in Italy11,13.
A series of Experimental Investigations was conducted by Zhou, et al14, to investigate
high arch dams. The model scale of 1:350 was used with a modeled compressive strength
material of 0.3 Mpa (45 psi), or 15,750 psi in the field case. Joints were modeled and were
formed by cutting the mold with a thin saw and using asphalt pads in the joint. The softness of
the joint padding was shown to have an effect on the response of the model. Final cracking
patterns were shown to be influenced by the joints, with a monolithic dam showing cracking in
the uppermost part of the dam, and jointed models showing cracking deeper in the dam. The
model reported in this paper uses a model material with a simulated field strength of
approximately 34 Mpa (5,000 psi). Joints are cast into the model in various patterns, creating
various versions of tests similar to this reported case. Water is used in a reservoir behind the
modeled structure. The relationship of crack locations is similar to this reported case.
4
Testing has been completed to model sliding of dams on the foundation contact only15.
Typically, these models were 1000 mm in height and had a short reservoir tank (0.4 m in length)
with absorbent rubber to eliminate the hydrodynamic effect. The input motions used were: a) a
7.5 Hz. sine wave (this waveform was hard to match exactly because some free vibration was
present in table), b) a 5 Hz sine wave ramped up in 5 cycles, held for 10 cycles, then ramped
down in 5 cycles, c) simulated earthquake of 12 seconds duration with the input amplitude
ramped until slipping occurred. The input motion used in the models in this paper uses a ramped
approach to increase the acceleration to failure, each load is held 30 seconds and then ramped by
0.25 g’s.
The purpose of the investigation described in this paper is to produce failures in arch dam
models for comparison with predictions by nonlinear computer models16,17. Unlike many of the
previous investigations, a specific field site is not being modeled using precise similitude
relationships. Rather, lowest value properties, and worst case loadings are allowed and the
model is loaded to absolute collapse in the test. The models were designed to the extent possible
to maintain similitude relationships of typical properties found in field tests, and in standard
laboratory tests, and yet be simple enough for direct comparison with the computer-predicted
results. This testing was conducted at the Bureau of Reclamation, Materials Engineering and
Research Laboratory18. This paper attempts to describe in detail the physical modeling, the
material properties of the models, similitude comparisons, and the models’ behavior. The
computer predictions and comparisons with these physical models have already been
published17,18
5
II. Experiment Overview
All of the models chosen for this study were based on a similar geometry. The field case
selected was a typical arch dam of approximately 90 m. (300 feet) in height in a wide canyon.
The scale chosen for this model series was 1/150. A model that was 0.6 m (2 feet) in height,
with a cross canyon width of 1.7 m. (5 feet 6-1/2 inches) was constructed on a 1-dimensional
(upstream-downstream) shake table . The arch was constructed as a single curvature arch with a
vertical upstream face and tapering with increasing height. A reservoir was constructed behind
the dam and filled with water for the tests. At the initiation of shaking the reservoir height was
approximately 1.2 cm (1 inch) from the top of the dam.
Different models were constructed: 1) A monolithic dam, 2) One horizontal joint at
approximately mid height, 3) One vertical joint at the mid point of the model dam, and 4) 17
vertical joints and 2 horizontal joints (one model was run with17 vertical joints as a test of
construction method), see Table 1. All models were run with a sinusoidal input motion at 14 Hz
beginning at 0.25 g and increasing every 30 seconds by 0.25 g until a structural collapse was
created in the model.
III. Similitude considerations
Considerable effort was made in the early model considerations to match each property in
the model using a similitude modeling approach. Early in the material development it was clear
that it would be difficult to model all parameters with a perfect similitude match. For that
reason, the following priority of modeling considerations was made: 1. Failure mechanism, 2.
6
Tensile strength, 3. Frequency content, 4. Strain rate, 5. Compressive strength, 6. Modulus, 7.
Density. Table 2 summarizes the model similitude comparisons.
Creating a modeling concrete which failed in the same manner as the full strength field
case was considered the most important modeling consideration. This was due to the desire to
allow failure through the dam, and not assume a prior that all cracks occur in joints. For the
monolithic dam, any failure mechanism would be initiated as a material failure, as no joints are
present. In the models with only horizontal, or vertical joints, a collapsing failure would also
require failure through the material. In the models combining vertical and horizontal blocks, it
was possible for large blocks to form from combined joints and fail, but the possibility of a
material failure instead of a failure through the joints only was made possible in the model with a
modeled material. Thus, a modeling material which simulated full scale behavior was developed.
The comparison of failure mechanisms in concrete was made by comparing the shear
failure produced in a compression tests. Direct tension failures, which occur on a flat plane, were
not tested but were assumed to occur in a typical flat plane based on the similar nature that beam
tests which were run failed in the same manner of full strength mixes. That is, the beams failed
in a vertical plane. Split cylinder tests also were identical in their failure mechanism. A typical
failure, in compression, is shown in figure 1. On the left is a failed cylinder from the model mix,
while a typical 50 year old cored sample is shown on the right. Note that a typical shear plane of
approximately 65 degrees is shown in both failures. That is, the plane of failure that occurs in
shear, in both the model mix and a typical full strength, aged concrete core occur in the same
manner.
7
Properties of mass concrete in dams has previously been published19,20. These results
represent tests run from approximately 500 cores obtained from dams and used to find various
standardized properties. All tests were run in accordance with ASTM21 C39 - Standard Test for
Compressive Strength of Cylindrical Specimens, ASTM21 C 469 Standard Test for Modulus of
Elasticity and Poisson’s Ratio in Compression, ASTM21 C496 Splitting Tension Strength of
Cylindrical Concrete Specimens. Tests for Dynamic Properties were done using similar
procedures and were tested at strain rates appropriate for earthquake loadings (see later
discussion) owing to the rate sensitivity of concrete. It is not suggested that these are the only
properties that can be measured from core, only that these standard methods serve as a
comparison of other properties reported in the literature. For this model study, values of
properties from the model mix, as measured, are compared to properties which have previously
been measured from dam cores. One additional standard test was used, the Three-point bending
test - ASTM21 C-78, with specimens cast into standard beam molds from the model mixed
material.
The compared material property values and a ratio to compare similitude ratios are shown
in Table 2. Tensile strength was considered high in importance to assure that failures initiated in
a loading range similar to field cases. Values of split tension strength range in the model mix
from 20 to 33.3 kPa (3.09 to 4.83 psi), with an average value of 28.3 kPa (4.1 psi). The range in
the model mix is approximately 35%; this range of values is less than a typical range of values
tested from cores obtained from existing dams and tested in the lab using the split cylinder test
(see Table 2). A lower range is considered a positive in the research environment to lower the
parameter variation. The average value of 28.3 KPa (4.1 psi) if scaled to the prototype using a
value of 150 would represent a prototype tension strength of 4.24 MPa (615 psi). Values
8
obtained from core recovered from operating dams range from 1.5 MPa (215 psi) to 5.4 MPa
(785 psi). This modeled value of tension strength is well within this range of measured values
and therefore is considered to represent a proper value within similitude considerations as shown
in Table 2.
Compression strength of the mix was considered a lower priority for these simulations.
The average value for compression strength, as shown in Table 1 is 0.2 MPa (31 psi). Scaled by
the model ratio of 150 this would produce a modeled value in the field of 32 MPa (4650 psi).
Values measured from cores obtained from dams range from approximately 9.6 MPa (1400) psi
to 58 MPa (8400 psi). An average of all core tested is 30.5 MPa (4425 psi). Thus this parameter
is considered in the range and near the average of all tests completed and a proper value within
similitude considerations, as shown in Table 2.
The average Modulus of Elasticity for the model mixes as shown in table 1 was 23.5 MPa
(3415 psi). This value scaled by the model scale of 150 would compare to field values of
approximately 3620 Mpa (525,000 psi). The lowest measured Modulus of Elasticity measured
from core taken from dams is 5170 MPa (750,000 psi). Thus, this parameter does not fall within
values from the field, being 0.7 of the lowest measured value or 30% less than a perfectly
modeled similitude parameter, this is noted in Table 2. The lower Modulus would result in
larger displacements, and higher strains, under dynamic loading than would be expected in the
field. Thus failures reported in this study may occur at lower, or more conservative loads.
Comparisons with numerical predictions which use the measured values as input into the
numerical models can still be compared directly.
9
The average density for the modeling mix was 1890 kg/m3 (118 lbs/ft3). Reclamation
data22 for average weight of fresh concrete varies from 2250 kg/m3 (141 lbs/ft3) with a 3.8 cm.
(1 ½ inch) MSA (Maximum size of aggregate) and 4.5% air content to 2515 kg/m3 (157 lbs/ft3)
with 15 cm. (6 inch MSA) and 3% air content. Density is not a parameter which is scaled and
therefore should fall into this range. The density of the model mix does not fall into this range, it
shows a ratio of 0.85 from the reference minimum value, or about 15% low from measured field
values, as noted in Table 2.
The frequency content of an earthquake when loading a dam dynamically plays an
important role, as the structural responses that can be excited produce a more critical situation. A
typical earthquake spectrum used in dam analysis is shown in figure 2. Values of modal periods
for a typical dam2 (which is similar to this model) tested in the field using a shaker device are
superimposed on Figure 2a. A reference line has been added to show that the first principle
period in the earthquake spectrum would relate to the lower mode shapes of the dam, in this case
the spectrum would occur near the 3rd mode shape of the dam. The modal periods for the model
were calculated using a finite element analysis. These modal periods are shown in figure 2b,
along with a reference line which represents the input motion of a 14 Hz. sinusoid. Comparing
the two figures it can be observed that the frequency content of the model input excites most
closely the 3rd modal frequency (see reference line)and is similar to the first period motion in the
field situation which excites the structural frequencies between the second and third structural
frequencies. This comparison demonstrates that the frequency content in the model and field
cases are similar, as noted in Table 2.
10
The advantages of using a sinusoidal input have been discussed previously23. With a
regular input pulse, the onset of nonlinear behavior may be detected in the instrument
measurements before the effect is visible in the model. Figure 3 shows the instrument readings
for the model with one horizontal joint. In this graph the base accelerations of the model are
plotted in white (on a white background) which for clarity in the figure masks the accelerations
of the accelerometer at the 3/4 height position on the model (plotted in black). From the figure it
can be seen that initially the accelerations in the upper part of the dam grow with the input
motion in the first part of the record. The onset of nonlinear behavior is seen at approximately
time 317 seconds in the record; there is a shift in the record from centering around zero. At
approximately time 320 seconds in the record, the accelerations begin to grow rapidly and a
spike in the record indicates when the crack in the model cracks passes through the model cross
section. It is this time when the failure is visible to the eye, about 3 seconds after the effect
begins. Following the crack, the top motion can not keep up with the base, in other words - the
base can slide back and forth under the top, and the acceleration is masked in the graph by the
foundation motion plotted in white. Following time 330, the acceleration is increased, as per the
test protocol, and as can be seen the collapse of the structure occurs as indicated by the high
accelerations and erratic behavior of the instruments.
As mentioned, the strain rate to which the model material was subjected during loading
was given a high priority in the modeling considerations. This is due to the inherent increase in
concrete’s strength at failure, as the strain rate is increased. An earthquake record produces a
typical strain rate for concrete of 10-3 (inches per inch per second)24. Data from cores taken
from actual dams and tested at a strain rate of approximately 10-3 showed a compression strength
ratio of increase from static to dynamic strength of 1.07 and a increase of tension strength of
11
1.44. Data from other tests25 suggest an increase of 1.1 for compression strength in the same
order strain rate, and 1.56 for tension. At an order higher strain rate (10-2) data25 suggests an
increase of 1.15 for compression and 1.7 for tension. For computation purposes26, a generally
linear trend of approximately a 10% increase in strength for each order of magnitude increase in
strain rate. The strain rate comparison from field cases to the model was made by comparing a
typical earthquake strain rate of 10-3 in the field to an estimated strain at failure(tension),
0.0001 in the model, at a frequency of 14 Hz; yielding a strain rate of 1.4 10-3 which would yield
an expected increase of 4% in strength due to an increased rate effect. As the strain at failure is
conservatively estimated, the strain rate is believed to be accurately modeled. An increase in the
speed of the time history plot using a similitude ratio for time of the square root of the scale (150)
would yield a time suppression ratio of 12.25. This increase would result in a corresponding
apparent increase in tension strength of about 12.25%. Although this additional strength
resisting failure would be within a reasonable range of tension strength values actually measured
from cores retrieved from dams, it may lead to nonconservative assumptions as the dam will not
fail until higher loads are reached, relative to a slower strain rate.
Summarizing the model to measured data comparisons: The modeling material fails in
the same shearing manner as the field material, the frequency content is similar in the field and
model cases, the strain rate effect is a possible increase of approximately 4% in the tension
strength, the model static tension and compression strength values demonstrate similitude ratios
within the range of expected variation from cores measured from dams, the Modulus value is
approximately 30% lower than measured values - which will allow greater displacements and
12
strains at lower loads, the density is approximately 15% low compared to laboratory measured
values of fresh mixes.
IV. Concrete Mix Design
For this study a new low-strength concrete mix was designed. Considerable work in this
area has been accomplished in previous studies3,4,11 to produce a similitude-appropriate concrete
mix. Work with 2 dimensional model studies here in the same laboratory and the advantages of
this mix design are discussed in reference 23. This mix design was modified for the 1/150 scale
models of this study. The mix components and proportions are shown in Table 3.
Preparation of the mix required special care due to the need to saturate the bentonite in
the mix. Bentonite and water for the mix were placed in a paddle mixer and mixed continuously
for approximately 5 hours. When the mix appeared visually to be smooth and consistent, the
remaining ingredients were added, mixed for a period of a few minutes. The mix was emptied
into containers suitable for forklift transport and poured into the model mold immediately. The
shake table was used to vibrate the mold slightly with a few strokes 2 or 3 times as the mold was
filled. Any type of vibratory device or tapping of the mold had been eliminated in previous
experiments as this changes the water/cement ratio due to bleeding in a material this weak and
causes a change in strength.
As was noted with Table 1 and the above discussion, properties were obtained from each
mix for the different models to be used as input when numerical methods were attempted as a
13
comparison to the model results. A direct measurement of the damping characteristics was not
attempted due to the fact that the model shape was difficult to suspend independently, the
mechanical effects of the table are difficult to isolate, and the material was weak for such a test.
Some early attempts were made using an input amplitude of 0.05 g’s to measure and predict
modal frequencies. Although this loading was quite low, 2 models were lost as the motion was
hard to energize at this low level.
V. Model Construction, Loading, and Instrumentation
The bottom thickness of the arch was 17.75 cm. (7 inches) with the top thickness being 2.54 cm.
(1"). Each model was poured using a form to produce the arch shape. Figure 4 shows one model
ready for testing (note that the reservoir is covered with plastic behind the dam).
The dam model was made of a modeled concrete material and was placed into an
abutment made of concrete of 27 MPa (4,000 psi) compression strength. The abutment was the
same height as the arch with heavily reinforced blocks on each side which extended a total of 56
cm. (22 inches) in the upstream/downstream direction and were 35 cm. (14 inches) wide from the
dam interface to the edge of the model. The foundation was approximately 13 cm. (5 inches)
thick at its thinnest location in the center of the dam. All thread rods which were 2.54 cm (1") in
diameter were used to securely fasten the abutment block to the shake table. The modeling intent
was to produce a foundation that was rigid relative to the dam. This simplified the direct
comparison of numerical prediction results17,18 since nonlinear material modeling was not
necessary in the foundation An offset key was used in the foundation to receive the model.
14
Joints when used in the model were made horizontally by placing 1 layer of sheet plastic
on top of a lift and then proceeding to pour. Vertical joints were constructed using a bag of
plastic (2 layers) with a rubber strand on the downstream sloped face. The rubber strand was
pulled tight to force the plastic bag against the face. A forsenic piece of a model is shown in
Figure 5 to show the joints in a failed and nonfailed portion of the dam (showing the bags). Note
the horizontal line on the concrete in the model which points out the horizontal joints, and that
failure did not occur in the joint plane but through the modeling material.
The upstream extent of the reservoir from the dam was approximately 0.76 m.(2.5 feet),
or slightly more than 1 dam height. The reservoir was constructed on the shake table and thus
was inflight with the model. A gravel wall was used at the upstream wall to aid in dissipating
energy and reflecting it back into the reservoir and towards the dam. The reservoir fluid was
plain water, it was applied directly to the upstream concrete face of the model.
For these experiments a shake table was constructed having movement constrained to a
single axis (horizontal only in the upstream/downstream direction). The table was tested for its
response modes and also tested in motion with accelerometers to determine its capabilities for
use at higher frequencies. The table responded well for input frequencies below 22 Hz, which
was below the table’s measured lowest natural frequency of 30 Hz. Higher frequencies were
eliminated for testing. Response of the table was clearly best at frequencies of 26 Hz and below.
A sinusoidal input motion was easily applied within these limitations.
15
Instrumentation in all models consisted of accelerometers on the base of the dam and at
1/4 points throughout the height of the dam at the centerline, and at the predicted2 inflection
points along the length of the dam. In earlier models, LVDT’s were used downstream in the
same locations as the accelerometers on the centerline; these were later abandoned to assure that
the push through of the final failure was not braced or influenced by the instruments. A highly
accurate pressure gage was submerged upstream of the dam in early models but did not register
any hydrodynamic pressure changes and was later abandoned.
VI. Results and Indications of the Models
The models all acted in a structurally linear manner until the onset of any cracking. This
was evidenced by a characteristic mode shape and increase of acceleration from the base of the
dam to the top of the dam. As noted in the discussion previously, this is shown in Figure 3. This
pattern was shown for the Monolith, Horizontal-Only Joint, Vertical-Only Joint, and 17x2-Joint
models.
Figure 5 shows all initial visible cracking patterns overlaid on an outline sketch of the
model. Initial cracks appear in the video of the model test in a single frame, that is the crack
becomes visible in 1/30 of a second. The joints control the initial cracking pattern. The
monolith breaks initially into one fairly large piece starting approximately 1/4 of the distance
across the canyon, down to about the 1/4 of the height, and then up the centerline. As noted in
Table 1, this initial crack occurs at an acceleration of 0.75 g’s. (Note that all model results are
shown in table 1, two of the monolithic models were lost attempting modal sweeps to measure
16
structural frequencies of the dam; these models are noted as “ 1st pulse” in the table). The first
crack in the horizontal joint only model is along the horizontal joint and along the abutments to
the top of the dam. Model M-10 had a Modulus value significantly different from other models,
and the acceleration of first cracking is higher and discounted from this discussion. Model M-9
showed a first crack at acceleration of 0.95. The model with a single vertical joint creates an
initial deep crack along the joint and then cracks along the abutment to the top. Four vertical joint
models were completed, since M-6 created a failure mode which was not typical of all other
models it is discounted from the conclusions. Models M-7, M-8, and M-11, which all failed in a
similar manner, are used to draw conclusions.. These models initially cracked at an approximate
acceleration of 1.5 g’s (with M-11 being slightly different at 1.2). In the 17x2 joint case, the
upper horizontal joint forms the predominant pattern along with 3 vertical joints in the initial
crack.. The initial acceleration at cracking is an average of 0.82. The variation of initial crack
accelerations 0.75 to 1.5 for all cases used for conclusions is related to the variation of the model
material properties in the production of each model. A modulus range of E=2302 to E=3948, a
42% difference in material property creates, a nearly corresponding 50% difference in predicted
acceleration, with higher moduli or strength resulting in higher accelerations. The models
generally act in a similar manner before the onset of cracking.
Models with 17 vertical joints only failed at the lowest acceleration, an average of 0.55.
In these models Modulas is similar to other cases, but the split tension strength is lower which
may account for the lower acceleration to initial cracks.
17
The final crack pattern is different for each model, showing the influence of the
constructed joints, see Figure 6. The final failure of the monolith, Figure 6a was the formation
of 5 predominant blocks and a subsequent downstream “snap through” of the centermost 2
blocks (this failure mode is similar and may be more clearly seen in Figure 6c). Along the lower
part of the model a shear plane was formed through the dam to create the final failure. The final
failure of the single horizontal joint produced a similar pattern to the monolith but the lower
portion of the final crack was located on the preexisting joint (Figure 6b) and the plane of failure
was along the preformed joint. The single vertical joint final failure, Figure 6c, produced cracks
almost to the bottom of the model and again demonstrated 5 predominant blocks in the final
failure, a shear angle was present in the failure plane in the lower portion of the dam. The model
with 17 vertical joints and 2 horizontal joints failed in a pattern of 5 blocks above the upper
horizontal joint, and some cracking along vertical joints between the upper and lower horizontal
joint (Figure 6d). These latter failures occurred through the material rather than the joint, as
shown in Figure 5. The final accelerations in Table 1 show that only the monolithic model
required a significant increase in acceleration to create the final collapse. Most models failed by
continuing the sinusoidal input once the model had cracked.
This testing series suggests that any significant joint will influence the final failure
acceleration value when compared to a monolith. As shown in Table 2 the monolith failed at 5
g’s, the horizontal crack at approximately 1.7 g’s, the vertical joint only at 1.5 g’s, and the 17
vertical by 2 horizontal joint model at an average of 1.0 g’s. Generally more significant joints
lower the final failure acceleration. The final failure in all models occurred after considerable
time, on the order of 20-30 seconds of shaking following the initial crack. This is consistent with
18
comments by other authors10 and most likely results from the need to abrade joints before there
is sufficient space for the blocks to be pushed downstream by the water in the reservoir.
VII. Water in Joints
One additional issue for consideration is the ability of water to penetrate joints in a dam
during an earthquake.. It is noted that the fluid used is water, without adjustment for viscosity
similitude. Nevertheless, water was clearly seen traveling through the dam in these tests during
the onset of cracking and throughout the test duration.
VIII. Conclusions
1. Models act linearly before the onset of cracking.
2. All models show the sudden onset of cracking followed by pronounced structural nonlinearity.
This nonlinearity is characterized by the bottom of the dam being able to slip back and forth
beneath the top of the dam.
3. The crack pattern in the models is dominated by the joint patterns.
4. The time to final failure when converted to full scale times exceeds the duration of any
recorded earthquake. In this paper final failure is defined as a total collapse; note that significant
cracking does occur before the final collapse.
5. Final failure is a push through of sections of the dam downstream. This failure mechanism
appears to require abrasion in the joints before it can occur.
6. Water does pass through cracks formed in the model by shaking.
19
IX. Acknowledgments
This research was sponsored by the Dam Safety Program, U.S. Bureau of Reclamation.
X. REFERENCES
1. A.K. Chopra and P. Chakrabarti, “The Koyna Earthquake of December 11,1967 and the
Performance of Koyna Dam,” Report No. EERC 71-1, Earthquake Engineering Center,
University of California, Berkeley, CA 1971.
2. Hall, J.F., “The dynamic and earthquake of concrete dams: review of experimental behaviour
and observational experience,” Soil Dynamics and Earthquake Engineering, Vol. 7, No. 2, pp58-
121.
3. W.P. Donlon and J.F. Hall, “Shaking Table Study of Concrete Gravity Dam Monoliths,”
Earthquake Engineering and Structural Dynamics, Vol 20, 769-786 (1991).
4. A. Niwa and R.W. Clough, “Shaking Table Research On Concrete Dam Models,” Report No.
UCB/EERC 80-05, Earthquake Engineering Research Center, University of California, Berkeley,
CA, 1980.
5. C.D. Norman, “Dynamic Failure Tests and Analysis of a Model Concrete Dam,” Technical
Report SL-86-33, U.S. Army Waterways Experiment Station, Vicksburg, MI 1986.
20
6. R. Tinawi, P. Leger, M. Leclerc, and G. Cipolla, “Shake Table Tests for the Seismic Response
of Concrete Gravity Dams,” Eleventh European Conference on Earthquake Engineering, Paris,
September 1998.
7. Plizzari, Saouma and Waggoner, “Centrifuge Modeling and Analysis of Concrete Gravity
Dams,” Journal of Structural Engineering, v 121 n 10, p. 1471-1479, Oct. 1995.
8. Renzi, Ferrara, and Mazza, “Cracking In a Concrete Gravity Dam: A Centrifugal
Investigation,” International Workshop on Dam Fracture and Damage,” Chambery, France, 1994.
9. Lyatkher, V.M., Kapstan, A.D., Semenov, I.V., “Seismic Stability of Toktogul Dam,
Hydrotechnical Construction, No.2, pp 8-14, 1977.
10. Niwa, A. and R.W. Clough, , “Non-Linear Seismic Response of Arch Dams,” Earthquake
Engineering and Structural Dynamics, Vol. 10, 267-281, 1982.
11. Oberti, G. and A. Castoldi, “The Use of Models In Assessing the Behavior of Concrete
Dams,” Dams and Earthquakes - Proceedings of a conference held at the Institution of Civil
Engineers, London on 1-2 October 1980, Thomas Telford Limited, London, 1981.
12. T. Yoshida and K. Baba, “Dynamic Response of Dams”, Proc. 3rd World Conf. Earthquake
Eng., Auckland, Vol. 2, pp. 748-764 , 1965.
21
13. G. Oberti and E. Lauletta, “Structural Models for the Study of Dam Earthquake Resistance,”
Ninth International Congress on Large Dams, Istanbul, Turkey, Sept 4-8, 1967.
14. Zhou, J., Lin G., Zhu,T., Jefferson, A.D., Williams, F.W. 2000, “Experimental investigations
into seismic failure of large arch dams,” ASCE Journal of Structural Engineering, Vol. 126, No.
8, 2000, pp. 926-935.
15. Mir, R.A. and C.A. Taylor, “An Investigation Into Base Sliding Response of Rigid Concrete
Gravity Dams to Dynamic Loading,” Earthquake Engineering and Structural Dynamics, Vol. 25,
79-98 (1996).
16. Payne, T.L., “Shaking Table Study to Investigate Failure Modes of Arch Dams,” 12th
Europeon Conference on Earthquake Engineering, Paper Reference 145, Elsevier Science, Ltd.,
2002.
17. D.W. Harris, T.L.Payne, and L.K. Nuss, “Comparison of the Nonlinear Behavior of Concrete
Arch Dams using Physical and Numerical Models,” Presented at Third US/Japan Workshop on
Advanced Research on earthquake Engineering for Dams, San Diego, CA., June 22-23, Hosted
by US Army Corps of Engineers, ERDC, 2002.
18. R. M. McCafferty, “Test Facilities USBR Vibration Test System,” Shock and Vibration
Bulletin, Bulletin 41, p. 109-117, December 1970.
22
19. Harris, David W., C. E. Mohorovic, and T. P. Dolen, “Dynamic Properties of Mass Concrete
Obtained From Dam Cores,” ACI Materials Journal, Vol. 97, No. 3, American Concrete Institute,
2000.
20. Harris, D.W., T.P. Dolen, C.E. Mohorovic, and P. Mitchell, “ Properties Obtained from Dam
Cores,” Presented at Third US/Japan Workshop on Advanced Research on earthquake
Engineering for Dams, San Diego, CA., June 22-23, Hosted by US Army Corps of Engineers,
ERDC, 2002.
21. Annual Book of ASTM Stanadards, 1998, V. 4.02, Concrete and Aggregates, American
Society of Testing and Materials, West Conshohocken, Pa., 1998.
22. U.S. Department of Interior, Concrete Manual 8th ed., United States Government Printing
Office, 1988.
23. Harris, David W., Nathan Snorteland, Timothy Dolen, and Fred Travers, “Shaking Table 2-D
Models of a Concrete Gravity Dam,” Earthquake Engineering and Structural Dynamics 2000; 29:
769-787.
24. Bishoff, P.H. and S.H Perry, “Compressive Behavior of Concrete at High Strain Rates,”
Materials and Structures, V. 24, pp. 425-450, 1991.
23
25. Ross, C.A., Kuennen, S.Y., Tedesco, J.W., “Effects of Strain Rate on Concrete Strength,”
Session on Concrete Research in the Federal Government, ACI Spring Convention, Washington,
D.C., March 1992.
26. Lin, J.I., DYNA3D: A Nonlinear, Explicit, Three Dimensional Finite Element Code for
Solid and Steuctural Mechanics: User Manual, Methods Develpment Group, Lawrence
Livermore National Laboratory, p. 2002.
24
Table 1 - Models with associated properties.
Model Type Date Age Avg Comp Avg Split Avg Beam E Failure Acceleration, g
(days) (lb./in.2) (lb./in.2) (lb./in.2) Initial Crack Final Failure
M-1 Monolith - Stripping crack 3/12/99 7 28.1 3.2
M-2 Monolith 3/31/99 7 23.25 4 2302.6 .75 5.0
M-4 Monolith - 1st Pulse 7/28/99 6 33.64 3761
M-5 Monolith - 1st Pulse 8/12/99 6 37.5 4.83 3.1 3088.4
M-6 Vertical Joint (saloon door) 8/27/99 7 25.6 4.67 0.7 0.85
M-7 Vertical Joint 5/2/00 8 40.3 1.5 1.5
M-8 Vertical Joint 6/21/00 7 28.2 1.5 1.5
M-9 Horizontal Joint 7/19/00 7 36 0.95 1.75
M-10 Horizontal Joint 8/2/00 7 52.1 5172.1 1.65 1.65
M-11 Vertical Joint 8/22/00 6 41.97 4.4 3758.8 1.2 1.2
M-12 17 Vertical Joints 10/3/00 5 27.01 2146.6 0.6 0.6
M-13 17 Vertical Joints 4/10/01 5 23.2 3.58 3461.5 0.5 0.5
M-14 2 Horizontal, 17 Vertical Joints 4/24/01 5 19.625 3.09 3099 1 1.25
M-15 2 Horizontal, 17 Vertical Joints 5/1/01 5 21.4 4.05 3948 0.75 0.75
25
Table 2 - Comparison of Model Mix Properties and Properties Measured from Dam Cores
Property Properties Measured from Dam Core or Dam Model Value Similitude Ratio
Minimum Maximum Measured Scaled by 150
Failure Mechanism
Shear angle in Compression 65 65 na 1.0
Tension Strength (psi) 215 785 4.1 615 1.0
Frequency Input
Field
Shake tests - Periods 0.13 0.17 0.22 0.28 0.32
First Period of quake 0.25
Model 0.07 0.09 0.10 na

Calculated Periods 0.072 1.0
Input motion
Strain Rate approx 1 x 10-3 approx na 1.0 to 1.04

1.4 x 10-3
Compression Strength (psi) 1400 8400 31 4650 1.0
Young’s Modulus(psi) 750,000 7,500,000 3415 525,000 0.7
Density(lb/ft**3) 141 157 118 na 0.85
26
Table 3 - Model Concrete mix components.
Component Lab Mix
lb./yd.3 Volume in mix

Air 0.14 (1/2% entrapped air assumed)
Water 739.8 11.86
Cement 156.6 0.8
Bentonite 63 0.39
Sand 2245.41 13.81
Note: water/cement = 4.72 Bentonite/(Bentonite+Cement)= 28.7%
27
Figure 1 - Failure of Modeled Material and 50 Year recovered Core
28
Acceleration versus Period of a Typical Earthquake with
Model Period Characterisitics
Dam Mode Shapes
1. 6 14 Hertz
1.6
Response Spectra 1. 4
1.4 Model Calculated
1.2 Modes - Typical 1. 2
Modes
Acceleration
Acceleration
1
0.8 0. 8
0.6 0. 6
0.4
0. 4
0.2
0. 2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0
0 0.02 0.04 0. 06 0.08 0. 1 0. 12
P eriod P eriod
Figure 2 - a. Response spectra and 3 mode shapes measured on a typical dam b. Calculated mode shapes and model input motion
29
5.00E+00
4.00E+00
3.00E+00
2.00E+00
1.00E+00
Accel - g
0.00E+00
2.70E+02 2.80E+02 2.90E+02 3.00E+02 3.10E+02 3.20E+02 3.30E+02 3.40E+02 3.50E+02
-1.00E+00
-2.00E+00
-3.00E+00
-4.00E+00
-5.00E+00
T im e - s e c
Figure 3 - Instrumentation data for Horizontal Joint Model Base (White) and 3/4 Height (Black) accelerometer
30
Figure 4 - Shake table model ready for testing, note reservoir behind dam (covered in plastic)
31
Figure 5 Model with joints
following failure
Figure 5 - Illustration of initial cracks in models
32
Figure 5a. Initial crack
of monolithic model Figure 5b. Initial crack
of horizontal model
Figure 5c. Initial crack of vertical joint model Figure 5d Initial crack of 17 vertical, 2 horizontal joint model
33
Figure 6 - Final Failure of Models
Figure 6b - Horizontal Joint Model

Figure 6a - Monolithic Model
Figure 6d - 17 vertical, 2 horizontal

Figure 6c - Vertical joint model
joint model
34

Dso 06 03

Uploaded by

Copyright:

Available Formats

Dso 06 03

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dso 06 03

Uploaded by

Copyright:

Available Formats

RECLAMATION Managing Water in the West

Investigation of the Failure Modes

Dam Safety Technology Development Program

U.S. Department of the Interior

4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT

12. DISTRIBUTION/AVAILABILITY STATEMENT

13. SUPPLEMENTARY NOTES

15. SUBJECT TERMS

Standard Form 298 (Rev. 8/98)

The mission of the Bureau of Reclamation is to manage, develop,

Appendix—Investigation of the Failure Modes of Concrete Arch Dams Using

2 Model concrete mix components .................................................................... 3

1 Stress-strain graphs. ........................................................................................ 5

29a Model 10—horizontal joint initial failure, south camera.............................. 32

2-Dimensional Koyna Dam Cross

The purpose of this investigation, conducted at the Bureau of Reclamation’s

Physical Model Test

Table 1.—Estimated concrete properties, the associated scale

Property Prototype estimate Scale factor Target value

Concrete Mix Design and Material Properties

Table 2.—Model concrete mix components

Lab mix, Volume in mix, Model mix, Volume in mix,

Air 0.14 0.52

Water 560 8.99 480 7.68

Cement 160 0.82 168 0.86

Bentonite 40 0.25 42 0.26

Sand 1366 8.4 1454 8.87

No. 4 —⅜" 553 3.36 0 0

⅜" - ¾" 829 5.04 1458 8.81

Note: water/cement = 3.5; bentonite/(bentonite+cement) = 20%

Table 3.—Properties of model materials

Apparent Apparent Apparent

Density, lb/ft3 133.1 0.9 (1.0) 135 0.9 (1.0)

Rapid loading: 113,000

εuc 0.004 2 (1) 0.005 2.5 (1) 0.004 2 (1)

*Density corrected by E = scale*density = 506

Figure 1.—Stress-strain graphs.

Figure 2.—Load vs. crack width in ⅓-point beam tension.

Figure 3.—Unload-reload test showing plastic behavior of the low-strength concrete.

Model Construction and Instrumentation

Figure 4.—First Koyna model mounted on the

Figure 5.—Instrument locations.

Instrument ID Type Orientation Height from base

Accelerometer 1 Acceleration Horizontal, x-direction 0

Accelerometer 2 Acceleration Horizontal, y-direction 0

Accelerometer 3 Acceleration Vertical, z-direction 0

Accelerometer 4 Acceleration Horizontal, x-direction 0.66 m (2.17 ft)

Accelerometer 5 Acceleration Horizontal, x-direction 1.22 m (4.00ft)

Accelerometer 6 Acceleration Horizontal, x-direction 1.47 m (6.67 ft)

Accelerometer 7 Acceleration Horizontal, x-direction 2.03 m (6.67 ft)

Accelerometer 8 Acceleration Vertical 2.6 m (8.5 ft)

LVDT 1 Displacement Horizontal, x-direction 0

LVDT 2 Displacement Horizontal, x-direction 0.97 m 3.17 ft)

LVDT 3 Displacement Horizontal, x-direction 1.69 m (5.54 ft)

Figure 6.—Second Koyna model failure plane.

Density corrected by E = scaledensity = 506