The Proceedings of the 4
th
China-Japan-US Symposium on St
ructural Control and Monitoring
Oct. 16-17, 2006
1
Using embedded wired and wireless
seismic networks in the mo-
ment-resisting steel frame Factor bu
ilding for damage identification
Monica Kohler
1
, Thomas Heaton
2
, Ramesh Govindan
1,3
, Paul Davis
1,4
, and Deborah Estrin
1,5
(
1
Center for Embedded Networked Sensing, University of Califo
rnia at Los Angeles, Los An
geles, California, USA 90095-1596
)
(
2
Department of Civil Engineering, California Institute of Technology, Pasadena, California, USA 91125
)
(
3
Department of Computer Sciences, University of
Southern California, Los
Angeles, California, USA
90089-0781
)
(
4
Department of Earth and Space Sciences, University of Califo
rnia at Los Angeles, Los A
ngeles, California, USA 90095-1567
)
(
5
Department of Computer Sciences, Uni
versity of California at Los Angeles,
Los Angeles, Califo
rnia, USA 90095-1596
)
E-mail: kohler@ess.ucla.edu
Abstract:
Ideally both spectral and tim
e domain data could be used to compute the
total building response and to make predic-
tions of damage patterns based
on various input scenarios. The combination of
frequency change informa
tion coupled with that
provided by wavefield properties can pinpoin
t the time and location of damage more accurately, especially for densely instru-
mented structures such as the 17-story UCLA Factor building. The 72-sensor embedded seismic array in the Factor building,
recording continuous waveforms at 500 Hz, ma
kes it possible to observe subtle changes
in dynamic characteristics between pairs
of floors and to relate the measurements to
system properties such as changes in s
tiffness due to a column failure. The high dy
-
namic range of the 24-bit digitizers allows both strong motions and ambient vibrations
to be recorded with reasonable sig-
nal-to-noise ratios. Temporary decreases in frequencies of Fact
or building modes of vibration ha
ve been correlated with moder-
ate-to-strong shaking, and spectral amplitudes of ambient vibrat
ions have clear daily and weekly patterns that correlate with
working hours, wind speeds, and non-seismic vi
brations. Waveform data from the Factor array are also being used in comparison
with finite element calculations for predictive damage behavi
or. A three-dimensional model of
the Factor building has been
developed based on structural drawings. Ob
served displacements for 20 small and mode
rate, local and regional earthquakes were
used to compute the impulse response functions of the building by
deconvolving the subbasement records as a proxy for the free
field. It can be shown that small but significant changes in the travel times, mode shapes, and frequencies are observed in the
simulation results for strong ground shaking an
d for modifications to the structural m
odel for hypothesized damage patterns suc
h
as broken welds on a particular floor. Wire
less untethered devices whose design is guided by data analysis and simulations such
as
these can significantly increase the spatia
l resolution of structural response to ear
thquakes. A Mica-Z mote network controlled
by
Wisden software that monitors a local area such as a building is being assembled and tested. The software system addresses some
of the challenges associated with high sample rates and lim
ited radio bandwidth, yet allows structural data acquisition from a
relatively large networ
k of wireless sensors.
Key words:
Wireless sensors, Structural
monitoring, Damage dete
ction, Wave propagation
INTRODUCTION
We are using data analysis and predictive
modeling of the wired UCLA Factor building
seismic array to guide the design of a wireless
network for structural health monitoring. De-
velopments in wireless sensing for structural
monitoring are becoming increasingly sophis-
ticated (Lynch and L
oh, 2006); however, there
remain serious hardware and software limita-
tions in the technology that preclude the
measurement and detection of many types of
useful signals. Common limitations are that
digitizer resolution is
16-bit or lower and
cannot record useful low-amplitude vibrations.
Typically MEMS-type sensors have noise
levels that are too high, especially for
The Proceeding of 4
th
China-Japan-US Symposium on Structural Control and Monitoring,
Oct.16-17, 2006
2
low-frequency signals from which to infer
global structural properties. The vibration data
often cannot be calibrated or validated because
there are no analogous wired systems with
which to compare the wireless data. Commu-
nications and power limitations often limit the
number of nodes, components and sample rates
of waveform data bein
g recorded continuously
for long periods of time.
The wired Factor building network is a
testbed for observational and predictive mod-
eling that is being used
to design a structural
monitoring wireless network. Changes in the
mode shapes and frequencies are observed in
the data and modeling re
sults for strong ground
shaking and for hypothesized damage patterns.
Wireless untethered devices whose design is
guided by the data analysis can significantly
increase the spatial resolution of structural re-
sponse to earthquakes. The wireless network
uses a first-generation software system that
allows continuous sampling and reliable log-
ging of time-synchronized structural response
data. The scaling provides flexibility in de-
ployment in which nodes self-organize into a
multi-hop network, and they can be inserted
into and removed from the network dynami-
cally. The result is that the installation can be
faster and lower-cost.
In this paper we use
the wired data to show
that low-amplitude data provide a wealth of
information that can be used in structural
monitoring system design. We are particularly
concerned with the pr
oblem of identifying
inelastic behavior, especi
ally if this behavior
can be used to detect structural damage. Our
general approach is to look for ways to detect
deviations from linear behavior of buildings.
This can only be accomplished if we have a
detailed understanding of the linear response of
a building. However, if we have any hope of
examining small-to-moderate sized shaking
events, we need to deploy better than 16-bit
devices and low-noise sensors for
low-frequency signals. Wireless deployments
should include a high density of nodes with
flexible placement, three components at each
location, and sampling rate
s of at least 500 sps.
These requirements point to the need for
multi-scale analysis made possible by software
that accommodates multi-scale deployments.
Specifically, we show th
at by looking at spec-
tral properties, wave
propagation properties,
and predictive modeling that low-amplitude
data are being used in non-traditional ways to
understand the nonlinear response of buildings.
We investigated spectral and wave
propagation behavior in
the Factor building by
comparing finite element simulation results
with data from the building. The Factor
building is a 17-story
moment-resisting steel
frame structure with an embedded 72-channel
accelerometer array that is continuously re-
corded by 24-bit data loggers. The large spatial
density of this seismic
array presents a unique
opportunity to develop ha
zard analysis tools
that analyze the entire
wavefield in both the
space-time domain and the normal mode do-
main. Spatial aliasing prevents such analysis in
most other instrumented structures. Further-
more these data are being used to calibrate and
validate the wireless recording package.
The Factor Seismic Array
The Factor accelerometer array is com-
posed of four horizontal
channels per floor and
an additional two vertical channels on the two
bottom floors (Fig. 1). The horizontal sensors
are oriented north-south and east-west along
the mid-sections of each floor. Nine 24-bit
Quanterra 4128 digitizers record the continu-
ous 72-channel data in two data streams: one at
100 sps for long term archiving and one at 500
The Proceedings of the 4
th
China-Japan-US Symposium on St
ructural Control and Monitoring
Oct. 16-17, 2006
3
sps from which major events can be extracted.
The Factor building array is complemented by
two borehole seismometers consisting of a
shallow Episensor installed in a ~100 meter
deep borehole and one at the wellhead, both
approximately 25 m east of the Factor building.
Structural health monitoring software is being
used to monitor and archive the continuous 100
sps Factor building data
. The array is being
used to record weak
and strong motions from
local earthquakes. Based on our recordings of
Factor data to date, th
e array is recording sev-
eral dozen local and regional earthquakes each
year with good signal-to-noise ratios as well as
ambient vibrations from which building re-
sponse has been determined.
Figure 1: The Factor building and its seis-
mic array. Arrows show locations and po-
larities of sensors on each floor.
DATA ANALYSIS AND MODELING
Spectral Properties
As part of the vibration monitoring of the
Factor building done to da
te, a large quantity of
24-hour/day ambient vibration data, and sev-
eral small but significant local earthquakes
were recorded that show temporal changes in
vibration mode characte
ristics. Our real-time
monitoring program illustrates how changes
could be continuously m
onitored to detect sig-
nificant damage or breakage in a structure. It is
generally assumed that
nonlinear behavior is
small unless a structure has experienced severe
shaking from a large event. Our observations
show that measurable nonlinear effects are
occurring for small earthquakes due to changes
in the stiffness of the building or soil when
amplitudes get larger (Kohler et al., 2005).
Upon inspection of hundreds of ambient vi-
bration records for calm vs. windy days as well
as for earthquakes (e.g., the 9/3/02 M
L
=4.7
Yorba Linda and the 9/28/04 M
L
=6.0 Parkfield
earthquakes), a decrease in frequencies is ob-
vious in the raw spectral data (Kohler et al.,
2005; Skolnik et al., 2006). The frequencies
return to previous am
bient vibration levels
within seconds of the high amplitude motions.
The decrease in frequencies is also obvi-
ous for wind excitation. Fig. 2 shows stacked
data for two 24-hour periods for one calm day
(12/25/04) and one windy day (11/28/04) for
which the first few modal frequencies are ob-
viously decreased. The climate data were col-
lected by the UCLA Atmospheric Sciences
Department at the nearby Math Sciences
building that includes the maximum wind
speed recorded during 10-minute intervals
(similar to gust recordings). During the
11/28/04 period, average wind speeds were up
to 25 mph and gusts were nearly 40 mph. It
remains to be determined how the wind source
excitation function may have influenced the
frequency changes.
The Proceeding of 4
th
China-Japan-US Symposium on Structural Control and Monitoring,
Oct.16-17, 2006
4
Figure 2: Ambient vibration Fourier am-
plitude spectra recorded on 12/25/04 (a low
wind speed day: solid curves) and 11/28/04
(a high wind speed day: dashed curves).
Wave propagation properties
It is increasingly common to consider the
wave propagation response
of buildings (e.g.,
Safak, 1999; Todorovska et al., 2001a; Snieder
and Safak, 2006), especially when determining
whether damage is de
tectable through moni-
toring of those effects (e.g., Todorovska et al.,
2001b; Zhang and Iwan, 2002). One way to
approach this problem is to model the dynamic
properties of structur
es through wave propa-
gation methods, specifi
cally to predict the
displacement response of a building when it is
subjected to near-field or far-field shaking
from an earthquake. We have computed
propagating waves through the building model
to examine the building’s linear and nonlinear
behavior for damage scenarios. For this, we
calculated the propagating impulse response
function and inter-story drift resulting from the
different scenarios. From the results we are
examining predicted travel-time variations,
torsional, and nonlinear behavior.
We used the commercial engineering
software ETABS to construct a Factor building
model (Fig. 3). The dynami
c simulations in the
linear regime are carried out by entering
ground acceleration excitation at the base in
the form of dynamic linear time histories.
Figure 3: The ETABS model of the Factor
building showing the
primary major struc-
tural elements, and cross sections of major
structural elements.
The Proceedings of the 4
th
China-Japan-US Symposium on St
ructural Control and Monitoring
Oct. 16-17, 2006
5
We used earthquake data in the form of
impulse response functions to construct and
validate our linear building model (Kohler et
al., 2006). Observed displacements for 20
small and moderate, local and regional earth-
quakes were used to co
mpute the impulse re-
sponse functions of the building by decon-
volving the subbasement records. We use the
subbasement as a proxy for the free field,
thereby separating out the source effects and
propagation effects be
tween the source and
subbasement. The subbasement is the second
level below ground and thus has soil-structure
interactions included
its recordings. We use
small to intermediate earthquakes in our
analysis, precluding the possibility of signifi-
cant source effects (e.g., complex source-time
functions, source complexity, directivity, and
non-standard source spectrum) or significant
soil-structure interaction. The impulse re-
sponse functions were then stacked to bring out
wave propagation effects more clearly (Fig. 4,
top).
The simulation results using 2% damping
with a Gaussian curve input are shown in Fig. 4
(bottom). This figure shows the synthetic
propagating wave starting as a Gaussian im-
pulse response function in the subbasement
and extending to the top of the building. The
primary pulse reflects o
ff the top of the build-
ing. The synthetic seismograms also show the
secondary reflections from the bottom of the
10
th
floor, especially in the east-west compo-
nents. Both data and synthetics for the
north-south component
s clearly show a
propagating pulse of S-
wave energy traveling
up and down the building three times during
the 3 s of data.
Figure 4: Stacked impulse response func-
tions for the north-south (left, top) and
east-west (right, top) components, and as-
sociated synthetic waveform results (bot-
tom) using the ETABS modeling for a
Gaussian input with 2% damping.
Mode shapes
We have examined mode shapes both
from real data and from forward modeling
analysis for the Factor building to document
what is actually happening and to examine
what we might expect
from strong shaking
from different types of damage applied to the
building model. Using mode shape changes to
identify damage has been investigated before
(Sohn et al., 2004), but never at the small ob-
servational scale available in the Factor
building due to the density of the array. Inves-
tigation of horizontal displacement within
narrow frequency bands has been conducted to
verify Factor building modes and their shapes.
Fig. 5 shows results of the maximum dis-
placement measurements for the 12/16/04
Santa Monica Bay earthquake (M
L
=3.6). These
curves are typical of
those obtained for small
earthquakes and ambient
vibrations (Kohler et
al., 2005). The figure shows displacement for
the N-S components
(top row) and E-W
components (bottom row) of the first eight
horizontal modes.
The Proceeding of 4
th
China-Japan-US Symposium on Structural Control and Monitoring,
Oct.16-17, 2006
6
It is likely that quantitative methods of
measuring the mode shapes from a damaged
building will eventually be used to determine
whether a significant change has taken place,
what stiffness changes may be responsible for
the changes, and to map the stiffness changes
into damage occurrence.
Figure 5: Mode shapes determined from
observed displacements (filled circles) at the
Factor building during the 12/16/04 Santa
Monica Bay earthquake
(distance=35 km,
M
L
=3.6).
In order to gain a preliminary under-
standing of what changes in mode shape we
would expect from strong shaking, we have
used our building model to show the effect of
changing the moment framed connections on a
specific floor to simple pinned connections.
For this example we created two ETABS
models identical to each other, with the ex-
ception of the connections on the 8
th
floor. We
replaced every moment connection at this floor
with a pinned connection as a proxy for con-
nection failures throughout the floor. This is
somewhat similar to breaking all the welds on
that floor. This is
not 100% realistic; the
damage patterns would likely be more evenly
distributed through the building, but this was a
useful example. It is clear from the dynamic
analysis results using ETABS that both the
frequencies and the mode shapes change dra-
matically (Fig. 6). Though this is an oversim-
plified way to approach the problem, it gives us
an initial idea of the potential use of the mode
shapes to isolate where the source of stiffness
change may have occurred and infer damage
from them.
N-S HORIZONTAL
Mode 1 Mode 2 Mode 3
Undamaged Damaged Difference Undamaged Damaged Difference Undamaged Damaged Difference
E-W HORIZONTAL
Mode 1 Mode 2 Mode 3
Undamaged Damaged Difference Undamaged Damaged Difference Undamaged Damaged Difference
TORSIONAL
Mode 1 Mode 2 Mode 3
Undamaged Damaged Difference Undamaged Damaged Difference Undamaged Damaged Difference
Figure 6: Synthetic mode shapes from a
hypothetical damage pattern that simulates
broken welds on the 8
th
floor. The “Dam-
aged” plots are for the model that simulates
the broken welds. The “Difference” plots
have been scaled by a normalization factor.
CONCLUSIONS
In addition to high spatial density, the
24-bit continuous recording system provides
The Proceedings of the 4
th
China-Japan-US Symposium on St
ructural Control and Monitoring
Oct. 16-17, 2006
7
high-resolution data for a wide variety of
sources including earthquakes of all sizes,
wind excitation, and ambient vibrations. We
have observed dynamic characteristics not
usually observable for long, continuous time
scales and for different
sources of excitation.
Vibration frequencies are known to change
both permanently and temporarily due to
strong shaking but frequency change alone
may not be an accurate or complete measure of
when or where a building has been perma-
nently damaged. The combination of fre-
quency change information coupled with that
provided by wave propagation data could
pinpoint the time and location of damage more
accurately.
The Factor building on the UCLA campus,
heavily instrumented with sensors embedded
throughout its entire hei
ght, presents a unique
opportunity to study the
building response after
removing the free-field response. Earthquake
data are revealing changes in shear wave ve-
locity where there are major changes in stiff-
ness due to material properties and dimension
changes in the structural elements. The wired
data illustrate that lo
w-amplitude data provide
a wealth of information that should be consid-
ered in wireless system design.
Acknowledgments
We appreciate discussions w
ith Erdal Safak and advice on
building the ETABS model from Swami Krishnan. The work
was supported by the National Science Foundation through the
Center for Embedded Networ
ked Sensing under Grant No.
CCF-0120778. Operations and maintenance of the Factor
seismic array was supported by personnel and funds from the
U.S. Geological Survey Advanc
ed National Seismic System
(ANSS) program and the NSF Center for Embedded Net-
worked Sensing (CENS) at UCLA.
References
Kohler, M. D., P. M. Davis,
and E. Safak, 2005, Earthquake
and ambient vibration mon
itoring of the steel-frame
UCLA Factor building,
Earthquake Spectra
,
21
, 715-736.
Kohler, M. D., T. Heaton, and C. Bradford, 2006, Wave
propagation in buildings,
Bull. Seis. Soc. Am.
, submitted.
Lynch, P. J., and K. J. Loh, 2006, A summary review of
wireless sensors and sensor netw
orks for structural health
monitoring,
The Shock and Vibration Digest
,
38
, 91-128.
Safak, E., 1999, Wave-propaga
tion formulation of seismic
response of multistory buildings,
J. Struct. Eng
.,
125
,
426-437.
Skolnik, D., Y. Lei,
E. Yu, and J. W. Wa
llace, 2006, Identifi-
cation, model updating, and response prediction of an
instrumented 15-story steel frame building,
Earthquake
Spectra
, submitted.
Snieder, R., and E. Safak, 2006, Extracting the building re-
sponse using interferometric imaging; theory and appli-
cation to the Millikan Library in Pasadena, California,
Bull. Seis. Soc. Am.
,
96
, 586-598.
Sohn, H., C. R. Farrar, F. M. He
mez, J. J. Czarnecki, D. D.
Shunk, D. W. Stinemates, B.
R. Nadler, 2004, A review of
structural health monitoring literature: 1996-2001,
Los
Alamos National Laboratory Report
,
LA-13976-MS
.
Todorovska, M. I., S. S. Ivanovi
c, and M. D. Trifunac, 2001a,
Wave propagation in a seven-
story reinforced concrete
building, I. Theo
retical models,
Soil Dyn. Earth. Eng., 21
,
211-223.
Todorovska, M. I., S. S. Ivanovi
c, and M. D. Trifunac, 2001b,
Wave propagation in a seven-
story reinforced concrete
building, II. Observed wavenumbers,
Soil Dyn. Earth.
Eng., 21
, 225-236.
Zhang, Y., and W. D. Iwan, 2002, Active interaction control of
tall buildings subjected to
near-field ground motions,
J.
Struct. Eng., 128
, 69-79.