Professional Documents
Culture Documents
Alves & Hall (2006) System Identification of A Concrete Arch Dam and Calibration of Its Finite Element Model
Alves & Hall (2006) System Identification of A Concrete Arch Dam and Calibration of Its Finite Element Model
SUMMARY
A magnitude 4.3 earthquake occurred near Pacoima Dam on 13 January 2001. An accelerometer array
that had been upgraded after the Northridge earthquake recorded the motion with 17 channels on the
dam and the dam–foundation interface. Using this data, properties of the rst two modes are found from
a system identication study. Modal properties are also determined from a forced vibration experiment
performed in 2002 and indicate a signicantly stier system than is estimated from the 2001 earthquake
records. The 2001 earthquake, although small, must have induced temporary nonlinearity. This has
implications for structural health monitoring. The source of the nonlinear behaviour is believed to
be loss of stiness in the foundation rock. A nite element model of Pacoima Dam is constructed
and calibrated to match modal properties determined from the system identication study. A dynamic
simulation of the 2001 earthquake response produces computed motions that agree fairly well with the
recorded ones. Copyright ? 2006 John Wiley & Sons, Ltd.
KEY WORDS: Pacoima Dam; system identication; forced vibration test; nite element analysis; modal
properties; Northridge earthquake
1. INTRODUCTION
Pacoima Dam is a 113 m high concrete arch dam located in Los Angeles County, California,
that was completed in 1928. The dam construction includes a thrust block that supports the
dam at the left abutment. The thrust block meets the dam at a contraction joint, one of 11
vertical joints throughout the body of the dam. An accelerometer site located on a ridge about
∗ Correspondence to: S. W. Alves, Lawrence Livermore National Laboratory, P.O. Box 808, L-126, Livermore,
CA 94551, U.S.A.
† E-mail: alves6@llnl.gov
‡ This work was performed as a doctoral student at the California Institute of Technology, and it is not aliated
with Lawrence Livermore National Laboratory.
§ Professor of Civil Engineering.
Contract=grant sponsor: California Strong Motion Instrumentation Program; contract=grant number: 1000-744
Received 23 June 2005
Revised 24 January 2006
Copyright ? 2006 John Wiley & Sons, Ltd. Accepted 31 January 2006
1322 S. W. ALVES AND J. F. HALL
3
1 4 2 5
13 16
12 6 7 8 15
14 17
Right Left
Abutment Abutment
11 10
9
Plan View
15 m above the thrust block recorded a peak acceleration of 1:25g during the magnitude 6.6
San Fernando earthquake in 1971, which had its epicentre about 8 km north of the dam.
During this event, a section of the upper left abutment rock slid a short distance, and the
joint between the dam and thrust block opened almost 1 cm. Work was done to repair this
damage, and an extensive array of 17 accelerometers was installed at Pacoima Dam in 1977
[1]. The array is illustrated in Figure 1, and it is also shown in CSMIP Report OSMS
01-02 [2]. Note that when viewing the dam from downstream, as in the gure, the left
abutment is actually to the right of the dam. Channels 1–8 are on the dam, 9–11 are at the
base, and 12–17 are on the abutments adjacent to the dam.
The array was in place on 17 January 1994 to record motion from the magnitude 6.7
Northridge earthquake, which had its epicentre about 18 km southwest of Pacoima Dam.
However, most of the recorded channels could not be fully digitized and processed due to
large amplitudes and high frequencies, and two of the channels did not record at all. The
upper left abutment rock slid again, and the joint at the thrust block opened 5 cm at the crest
this time. The damage to the dam and the rock was repaired [3, 4].
On 13 January 2001, a magnitude 4.3 earthquake with its epicentre about 6 km south of
the dam was recorded by the accelerometer array. The peak acceleration on the dam was
0:16 g. The records are presented in CSMIP Report OSMS 01-02 [2]. A subset of these
records is shown in Figure 2. Note the dierences in the motions on the abutments and at
the base of the dam.
In this paper, the dynamic behaviour of Pacoima Dam is investigated by system identica-
tion using the records from the 1994 Northridge earthquake and the magnitude 4.3 earthquake
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1323
4 5 6 7 8 9 10 11 12
Time (sec)
in 2001, by forced vibration tests, and by nite element studies. Of interest are modal fre-
quencies of the dam, the level of damping, the state of contact in the contraction joints, and
sources and degree of nonlinearity.
The modal characteristics of a structural system can be determined from earthquake records
using system identication tools such as MODE-ID [5, 6]. This computer program assumes a
linear system with classical normal modes. No detailed structural model is needed. The modal
parameters are estimated by nonlinear least-squares matching of the modelled response to the
measured response. Measured acceleration time histories are supplied as input to the model
and as the measured output to be matched. The output error function that is minimized by an
optimization procedure is given by
Tf
NR
J () = [w i (t) − y i (t; )]2 dt (1)
i=1 Ti
where w i and y i are the measured and modelled acceleration, respectively, at the ith out-
put response degree of freedom, NR is the number of output response degrees of freedom,
[Ti ; Tf ] is the time interval of the records to be matched and represents a vector of the
modal parameters being estimated. These parameters are the frequencies, damping, shapes and
participation factors for each mode. A pseudostatic matrix can also be estimated by MODE-ID
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1324 S. W. ALVES AND J. F. HALL
and consists of rows corresponding to the output response degrees of freedom and columns
corresponding to the input degrees of freedom. Each entry in the matrix can be interpreted as
the static displacement at an output degree of freedom if one of the input degrees of freedom
is displaced a unit amount while the others are held xed.
For Pacoima Dam, channels 1–8 are supplied as output accelerations and channels 9–17
are supplied as input accelerations. Including the extra free parameters associated with the
pseudostatic matrix gives MODE-ID too many parameters to t in this case, so an option to
hold the matrix xed throughout the optimization routine is utilized. The matrix entries are
calculated using a nite element model of Pacoima Dam.
If the excitation to a structure is completely dened by the input degrees of freedom,
then MODE-ID theoretically estimates the modal properties of the structure as if it were
supported on a rigid foundation. For an arch dam in a narrow canyon, the excitation is
spatially nonuniform, which means it can never be dened completely. In the present case, the
excitation is recorded only at the base of the dam and at one location on each abutment. This
means that while foundation interaction cannot be fully accounted for, the modal parameters
estimated by MODE-ID could still be inuenced to a signicant degree by interaction with
the foundation.
Symmetric, 4.73 Hz
Antisymmetric, 5.06 Hz
Figure 3. Symmetric and antisymmetric mode shapes estimated by MODE-ID with a 2-mode model
using the 2001 earthquake records. The open circles are the locations of the crest-level stations.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1325
200
D Recorded
0
Acceleration (cm/sec2)
Channel 1
150
0
Channel 2
150
0
Channel 5
4 5 6 7 8 9 10
Time (sec)
strongest motion of the earthquake, both modes are consistent with those identied using the
full 20-s records. As the earthquake motion dies out, the frequencies of both modes increase
slightly. This indicates that the response of Pacoima Dam to the 2001 earthquake was actually
somewhat nonlinear.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1326 S. W. ALVES AND J. F. HALL
Accel. (cm/sec2)
1000
Channel 1
0
0 1 2 3 4 5 6 7 8 9 10 11
Time (sec)
windows, the nonlinear nature of the response and the fact that the modes are closely spaced,
the identied modal parameters may not be very accurate. However, the measure-of-t in
MODE-ID is most sensitive to modal frequency [5], so the frequencies are robust quantities.
The results reveal that the frequencies had dropped to about 4 Hz on average, indicating that
the system was less sti after the arrival of the shear wave pulse than it was before, which is
consistent with the observation of damage. These results are similar to the ndings of others
that have used the Northridge records for system identication [10].
Forced vibration tests on Pacoima Dam from April 1980 determined the rst symmetric and
antisymmetric mode frequencies to be 5.45 and 5:60 Hz, respectively [11]. These frequencies
are signicantly higher than those estimated by MODE-ID using the 2001 earthquake records.
Modal damping estimated from the 1980 test results by the half-power method was 7.3% (sym-
metric) and 9.8% (antisymmetric), which also exceeds the MODE-ID estimates. However, the
data from the tests were of a quality that made it dicult to determine damping accurately [1].
To verify and update the 1980 tests, additional forced vibration tests were performed on
Pacoima Dam over 1 week in July=August 2002 [7]. An eccentric mass shaker that exerts
a unidirectional, sinusoidal force proportional to excitation frequency squared was used to
generate the input. The shaker was placed near the centre of the crest and frequency sweeps
were conducted from 2.5 to 11:0 Hz for shaking in both the stream and cross-stream directions.
The shaker force ranged from 2:72 kN (0.61 kips) at 2:5 Hz to 52:75 kN (11.86 kips) at
11:0 Hz. Kinemetrics SS-1 Ranger seismometers were used to measure the motion at ve
locations in two perpendicular, horizontal directions. The Rangers were placed near the existing
accelerometers at their three crest locations, oriented radially and tangentially, and at their
two locations along the right and left abutments about 24 m below the crest, oriented east–
west and north–south. The channels are numbered 1fv–10fv as shown in Figure 6, with the
crest station at the centre denoted by C. Notice that the numbering is dierent than for the
accelerometer array. Orientations of channels 1fv and 2fv were estimated to be N86E and
S04E, respectively, which are essentially stream and cross-stream and within 1◦ each of the
two shaking directions. The shaker was placed about 2:6 m northeast of Ranger location C.
The highest velocities recorded on the crest during the forced vibration were less than 1% as
large as velocities recorded on the crest during the 2001 earthquake.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1327
N Shaker 1fv
5fv
3fv C 2fv
4fv
6fv
7fv
9fv
8fv
10fv
Plan View
Figure 6. Locations of the Ranger seismometers and the shaker with orientations. Channels 7fv–10fv
are approximately 24 m below the crest.
x 10
6
Amplitude (Ranger Output / ω3)
2
2.5 3 4 5 6 7 8 9 10 11
Frequency (Hz)
Figure 7. Frequency response curve for channel 1fv from the stream-direction forced vibration test.
For a perfectly symmetric dam with the shaker at the centreline, shaking in the stream
direction excites only symmetric modes and shaking in the cross-stream direction excites only
antisymmetric modes. At Pacoima Dam, due to lack of sucient symmetry, the directions of
motion at location C for the rst symmetric mode and the rst antisymmetric mode are both
primarily in the stream direction. As a result, there is considerable interference between the
two modes for both directions of shaking, and this makes determination of modal frequencies
and, especially, damping dicult. Figure 7 shows the interfering resonances of the symmetric
and antisymmetric modes between 5 and 6 Hz in the channel 1fv response from the stream-
direction shaking. The quantity plotted in the gure is the amplitude of the Ranger output
divided by frequency cubed, which is proportional to the displacement of the dam per unit
shaker force.
One technique to eliminate interference between two modes is to align the direction of
shaking force perpendicular to the motion of one of the modes, which theoretically eliminates
the response of that mode, thus isolating the other one [12]. For the Pacoima Dam data, this
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1328 S. W. ALVES AND J. F. HALL
Symmetric
Antisymmetric
Figure 8. Symmetric and antisymmetric mode shapes determined from forced vibration testing in 2002.
The open circles are the locations of the crest-level Ranger seismometers.
was done mathematically by combining the results of the two shaking directions vectorially.
As a second step, the two perpendicular directions of motion recorded at a station were also
combined vectorially to further eliminate modal interference. Since the directions of motion
of the modes at the shaker and instrument locations are not known beforehand, iteration is
required. Another method employed to isolate the rst two modes involves summation and
subtraction of the responses of channels 3fv and 5fv in varying amounts until one mode
disappears. This is based on the two channels being in-phase for the symmetric mode and
out-of-phase for the antisymmetric mode. For more details on the modal isolation procedures,
see Reference [7].
From these eorts, the symmetric mode is determined to have a frequency between 5.35
and 5:45 Hz with damping between 4.0 and 7.0% of critical, and the antisymmetric mode
is determined to have a frequency between 5.65 and 5:75 Hz with damping between 4.5
and 5.5%. The mode shapes are shown in plan view in Figure 8 with the undeformed crest
plotted for reference. While there are some noticeable dierences, the shapes have a basi-
cally similar character to those estimated by MODE-ID using the 2001 earthquake records
(Figure 3). However, the frequencies of the modes determined from the forced vibration tests
are signicantly higher than those estimated by MODE-ID.
Modal properties of Pacoima Dam have been determined by system identication using earth-
quake records and by forced vibration testing. Similar symmetric mode shapes and antisym-
metric mode shapes, where the symmetric mode has the lower frequency, are identied from
the 2001 earthquake and the forced vibration tests. However, the frequencies and damping
of the identied modes from the two sources vary. The dierences in the modal parameters
are summarized in Table I, including the 1980 forced vibration tests, with the water level for
each event listed. The water level can be an important factor since the added mass of the
water aects the properties of the dam system. The variation in the modal damping is not
very signicant considering the diculty associated with estimating the damping because of
the closely spaced modes, but the frequency variations indicate a signicant dierence in the
stiness of the dam system.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1329
Table I. Modal frequencies and damping of the rst symmetric and antisymmetric modes of Pacoima
Dam from forced vibration tests and a system identication study using earthquake records.
Water level Symmetric mode Antisymmetric mode
Excitation Date below crest (m) frequency (damping) frequency (damping)
The frequencies from the 1980 tests and the 2002 tests are comparable. There are some
slight dierences, and the trend in the symmetric mode frequency is opposite to what would
be caused by the mass dierence from the change in water level. Perhaps this indicates that
the dam system had lost some stiness between 1980 and 2002. This could have happened
during the Northridge earthquake when the left abutment was damaged. The repairs after
the Northridge earthquake may not have completely returned the dam to its pre-earthquake
stiness. Another possible factor is that the additional hydrostatic pressure from the higher
water level in 1980 may have tightened the joints and thereby increased the stiness of the
dam compared to its state in 2002; such eects have been observed elsewhere [13].
The frequencies determined from the 2002 tests are signicantly higher than the frequencies
identied from the 2001 earthquake records by MODE-ID. However, there was no damage
observed after the 2001 earthquake, and no repairs were made that would have stiened
the system. Furthermore, the dierence in the water levels was only 5 m, and so should not
have signicantly aected the frequencies. In addition, had the system identication procedure
been able to fully account for foundation interaction, the MODE-ID frequencies based on the
2001 earthquake records would have been even lower. Thus, another explanation is needed to
account for the discrepancy between the two sets of frequencies. Such an explanation could
involve some form of nonlinearity that lowered the response frequencies during the 2001
earthquake.
The response of the dam to the 2002 forced vibration tests is about two orders of magnitude
smaller than to the 2001 earthquake. Therefore, while the response to the forced vibration tests
is probably linear, the 2001 earthquake must have induced some form of nonlinearity in the
response of Pacoima Dam that caused the stiness to be reduced, but did not lead to any
permanent damage. Using the square root relation between natural frequency and stiness, the
drop in stiness during the 2001 earthquake necessary to explain the dierence between the
two sets of frequencies is about 22%. This decrease in stiness is quite large considering that
the 2001 earthquake was magnitude 4.3 and only caused small amplitude motions in Pacoima
Dam (peak acceleration on the crest of only 0:16g).
The nonlinearity in the 2001 earthquake response could come from two sources. First is
the opening and closing of the dam’s contraction joints. The 2001 earthquake occurred in the
winter when cold temperatures could have left the joints in a partially opened state, even before
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1330 S. W. ALVES AND J. F. HALL
the earthquake began. Second is loss of stiness in the foundation, particularly in the upper left
abutment. This abutment was damaged and repaired in both the San Fernando and Northridge
earthquakes. While damage was not observed after the 2001 earthquake, perhaps the fractured
rock exhibited a lower secant stiness when subjected to this low-level earthquake excitation,
but its initial stiness was not permanently reduced because the rock was not stressed highly
enough. Perhaps, seismic waves travelling through the foundation induce this behaviour more
easily than mechanical forcing on the crest of the dam, but the dierence may simply be that
the forced vibration generates much smaller motion that is not strong enough to deform the
foundation rock signicantly.
Presented in Section 5 is a nite element study of Pacoima Dam. To the extent possible,
the above issues are investigated.
A nite element model of Pacoima Dam is constructed using Smeared Crack Arch Dam Anal-
ysis (SCADA), a nite element analysis program used for modelling concrete arch dams that
can include contact nonlinearities in the dam concrete [7, 14]. Interactions are considered with
an incompressible water reservoir and a massless rock foundation. The program incorporates
damping of the Rayleigh type using the dam=foundation stiness matrix and the dam mass
matrix. Water compressibility and rock mass are omitted to facilitate nonlinear analysis, but
this involves some error; for example, radiation damping mechanisms are not modelled ex-
plicitly. However, the Rayleigh damping coecients can be chosen to represent the overall
level of damping in the system. Additionally, for the low water levels considered in this
paper, water compressibility will not be important anyway. The program has been modied
to handle ground motion excitation that is spatially nonuniform.
The nite element meshes generated for Pacoima Dam, the water reservoir and the rock
foundation are shown in Figure 9, separated for clarity. The dam, water and rock are
modelled using 110 elements, 1320 elements and 728 elements, respectively. The thrust block
is modelled as part of the dam. The water mesh extends 206 m upstream or about 180%
of the dam height. The upstream direction of the reservoir is rotated by 20◦ to follow the
initial topography of the canyon upstream of the dam. The rock mesh extends 152 m away
from the dam and 73 m upstream and downstream of the dam at the dam–rock interface. To
approximate the topography immediately extending away from the dam crest, the rock slopes
up 50◦ over horizontal above the right abutment and extends horizontally away from the left
abutment.
5.1. Calibration
The process of calibrating a nite element model to parameters determined from actual
data can yield considerable insight into the appropriateness of the parameter values. For the
Pacoima Dam model, some of the properties are xed as follows: unit weight of concrete,
22:3 kN=m3 (142 lb=ft 3 ); water level, 38 m below the crest, which is close to the levels during
the 2001 earthquake and the 2002 forced vibration tests; Poisson’s ratio for concrete, 0.20;
and Poisson’s ratio for rock, 0.25. E values (elastic modulus) for the concrete and rock are
adjusted in the calibration.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1331
Figure 9. Finite element meshes of Pacoima Dam, the water reservoir and the rock foundation.
The reservoir is shown lled to crest level.
The frequencies determined from the 2002 forced vibration tests can be reproduced as
eigenfrequencies of the linear nite element model by choosing E for the concrete as 28:1GPa
(4080 ksi) and E for the rock as 13:9 GPa (2020 ksi). The resulting frequencies are 5:46 Hz
for the symmetric mode and 5:68 Hz for the antisymmetric mode, which compare well to the
forced vibration frequencies of 5.35–5:45 Hz for the symmetric mode and 5.65–5:75 Hz for
the antisymmetric mode. The selected value of E for the concrete is in the upper range of
typical values, and the rock modulus is within a large range of eld data [15]. If the two
values of E are shifted closer to each other, then the two computed frequencies move closer
to each other also, which worsens the t. If the two values of E are shifted further apart, then
the concrete modulus gets too high. Computed mode shapes are shown in Figure 10 in plan
view along with the undeformed crest for reference. The shapes are not identical to the ones
in Figure 8 from the forced vibration tests, but the dierences are not signicant, especially
considering the diculties associated with the closely spaced modes.
Calibration can also be made to the MODE-ID frequencies based on the 2001 earthquake
records. Since it is not known whether joint opening reduces the eective stiness of the
dam or if the overall stiness of the rock is reduced, or both, the previous E values are
scaled down equally by 22% to 21:9 GPa (3180 ksi) for the dam and 10:9 GPa (1575 ksi)
for the rock. Computed frequencies are now 4:82 Hz for the symmetric mode and 5:02 Hz
for the antisymmetric mode, close to the identied frequencies of 4:73 Hz for the symmetric
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1332 S. W. ALVES AND J. F. HALL
Symmetric
Antisymmetric
Figure 10. Symmetric and antisymmetric mode shapes computed from the linear nite
element model. The open circles are the locations of the crest-level stations and the
dots are the locations of nodes on the crest.
mode and 5:06 Hz for the antisymmetric mode. Mode shapes are the same as those shown in
Figure 10. The accuracy of this t suggests the MODE-ID frequencies may be more represen-
tative of the actual system than one in which the foundation is rigid. This can be investigated
further by replacing the exible rock foundation in the nite element model with a rigid one
and then recalibrating the model to the MODE-ID frequencies. This requires that E for the
concrete be reduced to about 16:6 GPa (2410 ksi), which seems too low. Also, the computed
modes reverse order so that the symmetric mode frequency is higher than the antisymmetric
one. These are more reasons why the MODE-ID frequencies may be more representative of
the actual system with its exible rock foundation.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1333
82
Temperature ( F)
o
66
50
Displacement (mm)
10
5
0
Joints omitted
Joints included
0 0.5 1 1.5 2
Time (years)
Figure 11. Stream-direction displacement (upstream is positive) of the centre of the crest relative to the
thrust block computed with the nite element model for cases with joints omitted and joints included.
The simulation covers a period of 2 years over which the temperature varies as shown.
because the joints open instead. Without joints in the model, the agreement is much better.
The centre of the crest displaces upstream when the dam is heated and downstream when
cooled, and the peak-to-peak displacement is about 14 mm relative to the thrust block. The
results of these analyses, which are shown in Figure 11, indicate that the contraction joints
of the dam are closed throughout the course of the year. Grout in the joints may supply
sucient tensile strength to prevent the joints from opening, or perhaps, creep of the concrete
over time has caused the dam to be in a more compressed state that requires larger tensile
stresses to pull the joints open.
The annual 17 mm peak-to-peak displacement of the centre of the crest revealed by the GPS
monitoring greatly exceeds the maximum dynamic displacements during the 2001 earthquake,
which were about 4 mm peak-to-peak. This suggests that in addition to the joints being
initially closed when the 2001 earthquake occurred, they should not have opened much during
the event. Therefore, temporary loss of stiness in the foundation rock during the earthquake is
the most likely explanation to account for the dierences between the MODE-ID frequencies
based on the 2001 earthquake records and the higher frequencies determined from the forced
vibration tests.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1334 S. W. ALVES AND J. F. HALL
225
Nonlinear Linear
Acceleration (cm/sec )
2
Channel 2
75
0
Channel 4
4 5 6 7 8 9 10
Time (sec)
Figure 12. Acceleration time histories corresponding to channels 2 and 4 computed from nonlinear and
linear nite element simulations of the 2001 earthquake response.
To simulate the 2001 earthquake response, channels 9–17 are applied to the nite element
model as free-eld motions at the dam–rock interface, which assumes that dam–rock inter-
action eects are not present in the records to any great extent. An accelerograph located
downstream of the dam in the canyon bottom failed to record; however, even if it had, the
motion there is very dierent than the free-eld motion along the dam–rock interface, which is
spatially nonuniform due to topographic eects. Alternatively, channels 9–17 could have been
applied as absolute motions, but this would have been inconsistent with the nding that the
MODE-ID frequencies, to which the nite element model is calibrated, may be more charac-
teristic of the actual system with its exible foundation rock. Since channels 9–17 sample only
two locations on the abutments and one at the base of the dam, interpolation=extrapolation is
necessary to dene the spatially nonuniform input at all nodes along the dam–rock interface
of the nite element model. This procedure is done in the frequency domain so that amplitude
and phase of the ground motion are accounted for [7].
To investigate further the degree to which joint opening could aect the dynamic response,
two cases are run: one in which the joints are omitted (linear case) and one in which they are
included (nonlinear case, no-tension joints). The computed horizontal motions are resolved
into radial and tangential components corresponding to the recorded motions. Computed accel-
eration histories corresponding to channels 2 and 4 are compared in Figure 12 for both cases,
and very little dierence is apparent. There is some high frequency motion in the acceleration
history corresponding to channel 4 for the nonlinear case that does not occur in the linear one.
Channel 4 is tangential, so this high frequency motion is from joint impacts. There is a small
decrease in the main oscillation frequency on the order of 2% when joints are present that
can be visually deduced from the computed time histories upon close inspection. The largest
joint opening is only 0:3 mm. Therefore, the nonlinear nite element model conrms that very
little of the hypothesized nonlinear behaviour during the 2001 earthquake can be attributed
to behaviour of the contraction joints. This is additional evidence that loss of stiness in the
foundation rock plays the major role.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1335
250
SCADA Recorded
0
Acceleration (cm/sec2)
Channel 1
225
0
Channel 2
200
0
Channel 5
4 5 6 7 8 9 10
Time (sec)
0.3
SCADA Recorded
0
Channel 1
Displacement (cm)
3
0.3
0
Channel 2
3
0.2
0
Channel 5
2
4 5 6 7 8 9 10
Time (sec)
The response of the nite element model for the linear case, which is believed to most
closely correspond to the behaviour during the 2001 earthquake, is compared to the recorded
accelerations and displacements on the dam for channels 1, 2 and 5 in Figure 13. The com-
puted acceleration histories overestimate the recorded accelerations somewhat. However, the
computed displacements agree better with the recorded motion, and the oscillatory behaviour
of the computed motions agrees well with the records. Overall, the agreement is fairly good.
Possible reasons for the dierences include inaccuracy of the interpolation=extrapolation pro-
cedure for the input, application of the records from channels 9–17 as free-eld motions, not
having the best choices for the E values of the concrete and rock, and underestimation of the
amount of damping actually present. Although the damping in the nite element model is based
on the MODE-ID results using the 2001 earthquake records, the actual damping could have
been somewhat higher depending on the degree to which the system identication includes
radiation from foundation interaction. On the other hand, the forced vibration tests indicate
lower damping values, and these results should fully include foundation radiation damping.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
1336 S. W. ALVES AND J. F. HALL
6. CONCLUSIONS
For Pacoima Dam, frequencies corresponding to the fundamental symmetric and antisymmetric
modes identied from records obtained during a magnitude 4.3 earthquake are signicantly
smaller than those determined from forced vibration testing. Even though the response of the
dam was two orders of magnitude larger for the small earthquake, the drop in stiness needed
to explain the decrease in the frequencies is over 20%, a surprisingly large factor considering
the low amplitudes of motion. Study indicates that this reduction in stiness is not the result of
initially open joints in the dam or the opening and closing of joints during the earthquake. The
most likely candidate is temporary loss of overall stiness in the abutment rock mass. This
stiness loss could have occurred in the upper left abutment, which had suered permanent
deformations in two previous earthquakes.
The high sensitivity of the dam’s frequencies to the level of response in the low-amplitude
range has implications for structural health monitoring. It would have been reasonable to
expect the response to the small magnitude 4.3 earthquake to be linear and exhibit the same
modal frequencies as observed during the forced vibration testing, but this is not the case. How
is the large drop in stiness to be interpreted? Is it or is it not associated with damage? What
should the base line be for Pacoima Dam, from which further stiness changes imply damage?
The present study to determine the dynamic properties of Pacoima Dam was hampered by
several inherent diculties. First, for an arch dam like Pacoima with geometry that departs
noticeably from perfect symmetry, modal interference from the closely spaced ‘symmetric’
and ‘antisymmetric’ modes can be very severe, especially complicating the estimation of
damping. Second, two recording stations on the abutments and one at the base of the dam are
not sucient to dene the spatially nonuniform excitation. For a system identication tool like
MODE-ID, if the input is incomplete, the foundation condition of the identied system is not
known, falling somewhere between being rigid and having the actual exible one. Also, for a
nite element simulation, the excitation must be interpolated=extrapolated over long distances
around the canyon on the dam–rock interface with the possibility of introducing considerable
error. Third, motions recorded near the dam–rock interface are contaminated to some degree
by dam–rock interaction. Moving a recording station away from the dam to try to obtain
free-eld motions introduces contamination from topographic eects.
One of the important factors when conducting seismic safety analysis of dams is the level
of damping. For the low response levels associated with the magnitude 4.3 earthquake and
the forced vibration tests, damping for the fundamental symmetric and antisymmetric modes
has been estimated to be in the range 4–7% of critical for Pacoima Dam with the water level
about 40 m below the crest.
ACKNOWLEDGEMENTS
This research was partially funded by the California Strong Motion Instrumentation Program of the
California Geological Survey in the Department of Conservation under contract No. 1000-744. The
authors also acknowledge James Beck for consultations regarding MODE-ID, and Ziyad Duron for
providing the eccentric mass shaker.
REFERENCES
1. Hall JF. The dynamic and earthquake behaviour of concrete dams: review of experimental behaviour and
observational evidence. Soil Dynamics and Earthquake Engineering 1988; 7(2):57–121.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337
SYSTEM IDENTIFICATION OF A DAM AND CALIBRATION OF ITS MODEL 1337
2. Processed data for Pacoima Dam—channels 1 through 17 from the M 4.3 Earthquake of 13 January 2001.
Report OSMS 01-02. California Strong Motion Instrumentation Program, 2001.
3. Morrison Knudsen Corporation. Report on Initial Assessment of the Eects of the January 17, 1994
Northridge=San Fernando Earthquake on Pacoima Dam, Phase 1. County of Los Angeles Department of
Public Works, 1994.
4. Hall JF (Technical ed.). Northridge earthquake of January 17, 1994 reconnaissance report, volume 1. Earthquake
Spectra 1995; 11(Supplement C):EERI 95-03.
5. Beck JL, Jennings PC. Structural identication using linear models and earthquake records. Earthquake
Engineering and Structural Dynamics 1980; 8(2):145–160.
6. Werner SD, Beck JL, Levine MB. Seismic response evaluation of Meloland Road Overpass using 1979 Imperial
Valley earthquake records. Earthquake Engineering and Structural Dynamics 1987; 15(2):249–274.
7. Alves SW. Nonlinear analysis of Pacoima Dam with spatially nonuniform ground motion. Report No. EERL
2004-11. Earthquake Engineering Research Laboratory, California Institute of Technology: Pasadena, CA, 2004.
8. Processed data for Pacoima Dam—channels 8 through 11 from the Northridge Earthquake of 17 January 1994.
Report OSMS 94-15A. California Strong Motion Instrumentation Program, 1994.
9. Phase 1 data for Pacoima Dam—channels 1–6, 12, 13 and 15–17 from the Northridge Earthquake of 17 January
1994. Report OSMS 95-05. California Strong Motion Instrumentation Program, 1995.
10. Bell DK, Davidson BJ. Response identication of Pacoima Dam for the 1994 Northridge earthquake. Eleventh
World Conference on Earthquake Engineering, Oxford, England, 1996 (Disc 2, Paper No. 774).
11. ANCO Engineers, Inc. Dynamic Testing of Concrete Dams. Prepared for National Science Foundation, 1982.
12. Duron ZH, Hall JF. New techniques in forced vibration testing. In Recent Advances in Structural Dynamics,
Proceedings of a Session Sponsored by the Aerospace Division of the American Society of Civil Engineers
in Conjunction with the ASCE Convention, Pardoen GC (ed.). ASCE: Seattle, WA, 1986; 16–33.
13. Darbre GR, Proulx J. Continuous ambient-vibration monitoring of the arch dam of Mauvoisin. Earthquake
Engineering and Structural Dynamics 2002; 31(2):475–480.
14. Hall JF. Ecient nonlinear seismic analysis of arch dams, user’s manual for SCADA (Smeared Crack Arch
Dam Analysis). Report No. EERL 96-01. Earthquake Engineering Research Laboratory, California Institute of
Technology, Pasadena, CA, 1996 (Modied July 1997).
15. Woodward-Lundgren & Associates. Pacoima Dam, Determination of In Situ Dynamic Elastic Foundation
Properties, Report of Investigations June–August 1971. Prepared for International Engineering Company, Inc.,
1971.
16. Behr JA, Hudnut KW, King NE. Monitoring structural deformation at Pacoima Dam, California using continuous
GPS. Proceedings of the 11th International Technical Meeting of the Satellite Division of the Institute of
Navigation, Nashville, TN, 1998; 59–68.
Copyright ? 2006 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2006; 35:1321–1337