of 9
E
Structural Health Monitoring of Buildings Using
Smartphone Sensors
by Qingkai Kong, Richard M. Allen, Monica D. Kohler, Thomas H.
Heaton, and Julian Bunn
ABSTRACT
This article presents the results of a shaker test of the Millikan
Library in Pasadena, California, using sensors inside smart-
phones to demonstrate their potential usage as a way to mon-
itor health states of buildings. This approach to structural
health monitoring could allow many more commercial and
residential buildings to be monitored because it removes the
cost prohibitive nature of traditional seismic arrays and the
complexity of deploying the instruments. Recordings from
the smartphones during the shaking show high correlation
with those from a reference sensor in the building, illustrating
that the phones can capture the shaking even when not fully
coupled to the floor. The fundamental translational frequen-
cies for the east
west and north
south directions and the
torsional frequencies of the building can be extracted from
single phone recordings. As we compare the displacement de-
rived from the phone recording by double integration to that
from the reference sensor, both phase and amplitude match
well. Signal-to-noise ratio is improved further by stacking re-
cords from multiple phones. These test results demonstrate the
ability to extract the fundamental translational and torsional
frequencies, and absolute displacements from upper levels of
buildings shaken by small local earthquakes. This work builds
on the ongoing MyShake project
a global smartphone seismic
network.
Electronic Supplement:
Time and frequency domain response of
records.
INTRODUCTION
Structures such as buildings, bridges, roads, and dams are an
essential part of modern society. Even though they are designed
to be used in different conditions, extreme unpredictable
events, that is, earthquakes, hurricanes, as well as deterioration
due to aging, can cause serious concerns about the safety and
functionality of the structures. Therefore, there is a need to
monitor the health states of the structures throughout their
lifetime.
Structural health monitoring (SHM) is a process that
involves, first, observing a structural or mechanical system
over time using periodically spaced measurements; second,
extracting the damage-sensitive features from these measure-
ments; and third, statistically analyzing these features to deter-
mine the current state of system health (
Farrar and Worden,
2007
). Different types of sensors are deployed in the structures
either permanently or temporarily to extract measurements
such as acceleration, velocity, displacement, deformation, stress,
and temperature (
Moreno-Gomez
et al.
, 2017
). SHM is expen-
sive both in terms of the hardware and the human effort; it
is implemented in only a few large-scale structures, and must
currently be deployed and maintained by practicing structural
engineering professionals. Therefore, it is virtually impossible
to conduct continuous long-term monitoring of the state of
health of most buildings due to the cost of both hardware and
human efforts.
The emergence of wireless sensors, low-cost microelectro
mechanical systems (MEMS) sensors, and sensor networks has
started to provide lower cost solutions to replace traditional
tethered monitoring systems (
Xu
et al.
, 2004
;
Paek
et al.
,
2005
;
Lynch and Loh, 2006
;
Kim
et al.
, 2007
;
Kohler
et al.
,
2013
;
Yin
et al.
, 2016
). But to use them in a nationwide effort
or even at the city level would require a large-scale effort for
engineers to deploy and maintain these systems. In addition,
many building owners are reluctant to install sensors in their
buildings as they are fearful of legal issues or simply choose not
to prioritize their building performance problems.
MyShake aims to build a global smartphone seismic net-
work by utilizing the power of crowdsourcing. It turns an
everyday handheld device into a portable seismometer by mon-
itoring data from the accelerometer in a smartphone to detect
earthquakes (
Kong
et al.
, 2015
,
Kong, Allen, Schreier,
et al.
,
2016
). After the release of the MyShake app to the public
on 12 February 2016, it has been downloaded by more than
270,000 users globally. Today, about 10,000 active phones con-
tribute data to the system each day for monitoring earthquakes.
The results from the collected data are promising (
Kong,
Allen, and Schreier, 2016
); MyShake can record
M
5 earth-
quakes up to about 200 km from the smartphone, and it
594 Seismological Research Letters Volume 89, Number 2A March/April 2018
doi: 10.1785/0220170111
can record small magnitude earthquakes (
M
2.5) at closer
distances.
The quality of the data recorded by MyShake, and the ease
of building and scaling up with this network, led us to inves-
tigate whether we might expand the use of MyShake data to
the area of SHM. If private smartphone sensors can be used as
a mechanism to collect data on the health state of buildings,
then smartphones could overcome the substantial challenge
of deploying the sensors manually in buildings, and provide
a way to monitor the buildings at very low cost of hardware
and maintenance. Because the smartphones may also be located
throughout a building, this approach could provide a dense in-
building network to monitor the structural health state of the
building floor by floor.
Multiple groups studied the feasibility of using a low-cost
sensor network to conduct SHM (
Cochran
et al.
, 2009
;
Clayton
et al.
, 2011
,
2015
;
Kohler
et al.
, 2013
,
2016
;
Yin
et al.
,
2016
). These studies use specially designed low-cost sensor
boxes that can be deployed in a building either by professional
engineers or community volunteers. These sensors cost tens of
dollars to a few hundred dollars each, but the more significant
cost is associated with the fact that someone must manually
deploy the sensors in buildings. This makes it challenging to
scale up these networks due to the human labor, permitting,
and permission efforts required.
In our study, smartphones provide an opportunity to re-
place this manual deployment process with a straightforward
software download and installation onto the user
s phone. In a
Figure 1.
(a) The Millikan Library building viewed from the northeast. The two dark colored panels on the near-side of the building
comprise the east shear wall (modified from
Bradford
etal.
, 2004
). The inset figure (modified from
Clinton
etal.
(2006)
shows the plan of the
building, where the star is the location of the test phones and the dot is the location of the Episensor station CI MIK. (b) The shaker located
on the roof, used to generate oscillation of the building. The two exposed buckets contain lead masses that spin in opposite directions to
generate a sinusoidal horizontal force. (c) The 25 smartphones used in the test, all placed on the floor of the ninth (top level) floor. The
duration of the shaking in the north
south (N-S) direction is 1:29
3:02 p.m., and in the east
west (E-W) direction is 3:38
5:03 p.m., local
time. The color version of this figure is available only in the electronic edition.
Seismological Research Letters Volume 89, Number 2A March/April 2018 595
typical smartphone, a built-in three-axis accelerometer mea-
sures the movement of the phone in three dimensions (one
vertical and two orthogonal horizontal components if the
phone is oriented with one side parallel to the ground).
Previous shake table tests of the accelerometers inside smart-
phones have shown that typical high-end accelerometers (used
in iPhones and high-end Android phones) are capable of
recording motion in the 0.2
20 Hz frequency range and
10
2000 m
g
amplitude range (
g
is gravitational acceleration)
(
D
Alessandro and D
Anna, 2013
;
Reilly
et al.
, 2013
;
Dashti
et al.
, 2014
;
Kong, Allen, Schreier,
et al.
, 2016
). Engineers ex-
plored the use of smartphone sensors to monitor the health
states of large-scale structures such as bridges.
Yu
et al.
(2015)
conducted a series of tests to show that using smartphones to
carry out health monitoring of bridges is feasible.
Ozer
et al.
(2015)
and
Feng
et al.
(2016)
show that using smartphones
attached to a pedestrian bridge, they can infer modal param-
eters from the recordings of the phones and layout of a struc-
ture for a crowdsourcing platform.
Here we explore the use of private smartphones to mon-
itor building health. This is the first step in investigating
whether the existing global MyShake network of smartphones
as well as similar crowdsourcing smartphone efforts (
Faulkner
et al.
, 2014
;
Finazzi, 2016
) could be harnessed for this purpose,
in addition to monitoring earthquake activity. We show results
from a shaker test of the Millikan Library building on the
Caltech campus in which we placed the phones on the floor
of the ninth (top level) floor. We determine if we can extract
the fundamental frequencies of the building, and estimate the
displacement of the floor due to motions similar to that from a
small, nearby earthquake. Personal smartphones are, of course,
in motion with their owners for portions of the day as the
owner walks, commutes, etc. Our test is similar to the results
expected for stationary phones resting on stands at night, or
when placed on a desk or left in a bag on the floor. The results
presented here illustrate the potential of using MyShake-en-
abled personal smartphones to record building shaking result-
ing from nearby earthquakes and using that data to extract the
building characteristics. We also present a method to deter-
mine the orientation of the smartphone if its orientation is
not known, but prior information about the building charac-
teristics is available.
BACKGROUND OF MILLIKAN LIBRARY
The shaker test was conducted at the Millikan Library building
on the campus of the California Institute of Technology
(Caltech) (Fig.
1a
). Millikan Library is a nine-story, reinforced
concrete building,
44 m
tall, and 21 m by 23 m in plan. The
building has concrete moment frames in both the east
west
(E-W) and north
south (N-S) directions. Shear walls on the
east and west sides of the building provide most of the stiffness
in the N-S direction, and shear walls in the central core provide
added stiffness in both directions (
Bradford
et al.
, 2004
).
The Millikan Library building is instrumented with a
permanent, dense array of uniaxial strong-motion sensors
Table 1
List of Smartphones Used in the Test
Brand/Model
N
Accelerometer Type
Vendor
Maximum Range (
m
=
s
2
)
Resolution (
m
=
s
2
)
Samsung S3
1
MPL
InvenSense
19.6
3
:
83
×
10
2
Samsung S4*
2
K330
STMicroelectronics
19.6
5
:
99
×
10
4
Samsung S5*
1
MPU6500
InvenSense
19.6
5
:
99
×
10
4
Samsung S6*
1
MPU6500
InvenSense
19.6
5
:
99
×
10
4
Samsung Note2
2
LSM330DLC
STMicroelectronics
19.6
9
:
58
×
10
3
Samsung Note4
1
LCM20610
InvenSense
39.2
1
:
20
×
10
3
Samsung Note5*
1
K6DS3TR
STMicroelectronics
39.2
1
:
20
×
10
3
Samsung-Exhibit-II
1
BMA222
Bosch
19.6
1
:
53
×
10
1
Nexus 5
2
MPU6515
InvenSense
39.2
1
:
20
×
10
3
LG-G2*
1
LGE
STMicroelectronics
39.2
1
:
20
×
10
3
LG-G-Stylo
1
LGE
Bosch
156.9
9
:
58
×
10
3
LG-Leon*
1
LGE
Bosch
156.9
9
:
58
×
10
3
LG-G4
1
LGE
Bosch
156.9
9
:
58
×
10
3
HTC-One-M9
1
N/A
HTC Corp
39.2
1
:
00
×
10
2
MotoX
2
LIS3DH
STMicroelectronics
156.9
4
:
79
×
10
3
HuaWei Prism
3
BMA150
Bosch
39.2
1
:
53
×
10
1
Sony Xperia
2
MPL
InvenSense
19.6
3
:
83
×
10
2
HTC Amaze
1
Panasonic
Panasonic
19.6
1
:
20
×
10
2
Brand and model of the phones are shown;
N
is the number phones used in the test. Max Range and Resolution show the range
of the amplitude and the smallest measurable value that the sensor can measure. The phones flagged with * indicate that the
phone is used in the seven-phone stack. The resolution values are from the sensor specifications.
596 Seismological Research Letters Volume 89, Number 2A March/April 2018
(Kinemetrics FBA-11s in
1
g
and
2
g
) with 36
channels throughout the building. On each
floor, there are three horizontal accelerometers;
in addition, three vertical accelerometers are in-
stalled in the basement. A three-axis Episensor is
installed on the ninth floor with a 24-bit data
logger (
Bradford
et al.
,2004
) (see the location
in Fig.
1a
). In addition, the building is instru-
mented with 10 Community Seismic Network
accelerometers distributed on each floor (
Kohler
et al.
,2013
). Previous studies have shown
changes in the modal parameters of the building
through time as a result of large-amplitude
ground shaking (
Clinton
et al.
,2006
). These
studies illustrate the importance of understand-
ing and documenting the dynamic properties of
different classes of structures (i.e., steel-frame
versus reinforced-concrete buildings) within the
linear response regime before a nonlinear re-
sponse might occur.
METHOD
A Kinemetrics model VG-1 synchronized vibra-
tion generator (shaker) was installed on the roof
of Millikan Library in 1972 (Fig.
1b
). The
shaker has two buckets that rotate elliptically
in opposite directions around a center spindle.
These buckets can be loaded with different con-
figurations of lead weights, and depending
on the alignment of the buckets, the shaker can
apply a sinusoidal force in any horizontal direc-
tion (
Bradford
et al.
, 2004
).
In our tests, we applied forces to the build-
ing first in the N-S direction, and then in the
E-W direction at discrete frequencies. The
applied frequencies of oscillation spanned
0.2
2.45 Hz and were varied gradually over
the course of
2
:
5 hrs
. The rate of change of
frequency varied over the course of the sweep.
The sweep progressed at 0.05-Hz intervals with frequencies
held constant for 60 s most of the time, but near the modal
frequencies, the constant-frequency interval was extended to
600 s. The extended runtime was used for the building
s
fundamental E-W-mode frequency of 1.2 Hz, the fundamental
N-S-mode frequency of 1.7 Hz, and the fundamental torsional-
mode frequency of 2.4 Hz. Additional details of the test runs
are provided in
Table S1 (available in the electronic supple-
ment to this article).
Smartphones were placed in the northwest corner of the
ninth floor (top level) of Millikan Library (Fig.
1a,c
) with their
x
axis approximately aligned in the E-W direction. For this test,
the phones are placed at the most advantageous location, as the
displacements on the corner of the ninth floor will be greater
than at many other locations in the structure. Twenty-five dif-
ferent models of Android phones were tested (see Table
1
for
details) as the accelerometers in Android phones are of various
qualities (
Kong, Allen, Schreier,
et al.
, 2016
). The phones have
flat response in the 0.1
12.5 Hz frequency range, and we use a
sampling rate of 25 samples per second for these tests. The
resolution listed in Table
1
is from the phone specifications
and shows the best case when the phone is insulated from envi-
ronmental vibrations.
RESULTS
Acceleration waveforms were recorded by the smartphones and
compared to waveforms from a three-component Episensor,
which is permanently installed on the ninth floor, to validate
the smartphone motion. The Episensor has a 24-bit Q980
data logger, and its data are continuously telemetered to the
Figure 2.
Waveform comparisons between the Episensor (CI MIK) and a Sam-
sung Galaxy S4 phone for the two horizontal components. (a) N-S component;
(b) E-W component. Frequency labels indicate when the test run is at or near
the fundamental or torsional frequencies of the building. The time for each test
run can be found in
Table S1 (available in the electronic supplement to this
article). The amplitude and phase alignment is generally good (see
Fig. S1
which expands the time window of 4820
4850 s). The color version of this figure
is available only in the electronic edition.
Seismological Research Letters Volume 89, Number 2A March/April 2018 597
Southern California Seismic Network as station
MIK. We will refer to it as station CI MIK in
the following figures. Comparison of the accel-
eration waveforms between the Episensor and
one of the Samsung Galaxy S4 phones is shown
in Figure
2
. The recording from the phone gen-
erally has a larger amplitude during peak shaking
and a higher noise level that is above the lower
amplitude shaking compared with the recording
from CI MIK. Between times 2000
8000 s, the
shaking was applied in the N-S direction, and
from 9000
15,000 s it was applied in the E-W
direction (Fig.
2
). During each time interval,
shaking was applied at a range of frequencies
including the fundamental translational mode
in that direction and the fundamental torsional
mode that excites motion in both directions due
to the rotational nature of torsion about a ver-
tical axis. It is clear that the recordings from the
phone and the Episensor from N-S and E-W
shaking show good correlation, though the am-
plitude of the phone signal is greater than the
Episensor. A shorter time window of the com-
parison is shown in
Figure S1 illustrating the phase matching.
The signals during the torsional motion (shaking frequency
around 2.35 Hz), however, do not correlate in the same way,
due to the different locations of the sensors on the ninth floor.
At the location of the Episensor, the torsional motion is mainly
in the N-S direction; the smartphones located at the northwest
corner of the building, however, experienced motion in both the
N-S and E-W directions, causing the amplitude difference. This
can be seen in the time range from 13,000 to 14,000 s (Fig.
2
).
Just one smartphone records building response to the
shaking well, especially at the fundamental frequencies and tor-
sional frequency. From the spectrum (Fig.
3
), we clearly see the
peak at the fundamental frequency and torsional frequency,
and can thus extract them from a single phone. Because we have
25 smartphones at the same location we can also stack them to
improve the signal-to-noise ratio. This assumes that noise re-
corded by different phones is truly random, and stacking across
different phones will cancel the noise but not the coherent sig-
nals caused by the building
s response to shaking.
Because the phones are not synchronized with each other
(each phone has its own network time), we use cross correla-
tion to find the best alignment of the recordings from different
phones. We calculated the cross correlation of the entire
N-S-component time series between different phone record-
ings and a base phone recording by shifting them within 120 s
windows to find the maximum correlation coefficient. To get
the best stack, we select the phones with different thresholds of
the correlation coefficients. Correlation-coefficient thresholds
of 0.6, 0.5, 0.4, and 0.3, resulted in stacks with 7, 10, 14, and 16
phones, respectively. The quality of the waveforms recorded is
also variable due to the different accelerometers in the phones;
thus, stacking more phones does not always improve the result-
ing waveform. By comparing the waveforms and spectra (more
on this below), we conclude that stacking seven phones gives us
the best results (the correlation coefficient is larger than 0.6).
These seven phones are flagged with an asterisk in Table
1
, and
we see that most of them have the best accelerometers based on
specification resolution. Figure
4
shows the comparison of the
waveforms between the Episensor (CI MIK) and the stack of
seven phones. We see that the stack amplitudes are a better
match to those observed on CI MIK. The noise levels during
low-amplitude shaking are lower, and the observed amplitudes
during peak shaking are more similar.
The main goal of our shaker test is to extract the funda-
mental frequencies of the building from the phones and to
compare them for accuracy with those of the Episensor. To
calculate the spectrum of the building
s shaking, we found that
a multitaper spectrum analysis obtains better results than a di-
rect fast Fourier transform. In the multitaper analysis, the time-
series data to be analyzed is multiplied by a series of orthogonal
tapers, and then Fourier transformed and squared to obtain the
estimate of the power spectrum density (PSD). The orthogonal
tapers will generate many independent estimates of the PSD
instead of only one, and an average of them will suppress the
random variance in the estimation (
Prieto
et al.
, 2009
). We
select the N-S and E-W components individually, and apply
the multitaper analysis. Figure
3
shows the amplitude spectrum
for the recordings of the N-S component from the Episensor, a
single phone, and the stacked phone time series. Overall, we
observe that the single phone spectrum has a noise level around
10
×
10
5
and the stacked phone time series has a noise level of
about
10
×
10
6
, which make the peaks around 1.25
1.5 Hz
observable. However, the fundamental frequency in the N-S di-
rection is clearly visible in all cases at 1.7 Hz. The fundamental
torsional-mode frequency at 2.35 Hz is also distinguishable in all
cases. The E-W-component spectrum produces similar results
Figure 3.
Spectrum comparisons in the N-S direction where the fundamental
mode of the building is at 1.7 Hz and the fundamental torsion mode is at 2.35 Hz.
Blue is spectrum for the Episensor (CI MIK), red is spectrum for a single phone
(Samsung Galaxy S4), and the green spectrum is from the 7-phone stack. The mo-
dal peaks are clearly visible in all cases. The color version of this figure is available
only in the electronic edition.
598 Seismological Research Letters Volume 89, Number 2A March/April 2018
and the peak of the fundamental frequency is
clearly visible (see
Fig. S2 for the spectrum
comparison for the E-W component).
Peak and relative displacement amplitudes
at a given floor of the building are also important
to the civil engineering community to quantify
localized deformation of the building (e.g., inter-
story drift). In Figure
5
, we show comparisons
of the displacement time series between the
Episensor, a single phone, and stacked phone
recordings. Overall, the result from seven stacked
phone time series shows better agreement with
the Episensor in both phase and amplitude.
These are obtained through double integration
of the acceleration recordings by first removing
the mean and trend in the record, and then ap-
plying a 0.5 Hz high-pass filter. They show good
agreement with a peak displacement of about
0.05 cm for this frequency range.
ESTIMATION OF THE ORIENTATION
OF THE PHONES
Sometimes, we have the reverse problem in
which we know the building
s fundamental
frequency but do not know the phone
s orien-
tation. If a building
s modal frequencies and
corresponding-mode shapes are already known,
then it is possible to deduce the orientation of
an arbitrarily rotated smartphone that is record-
ing a known mode: rotate the phone until
the resulting filtered motion is aligned with the
mode shape. Analogous methods have been ap-
plied to find the orientation of three-compo-
nent seismic sensors in various conditions
(
Duennebier
et al.
, 1987
;
Ekstrom and Busby,
2008
;
Stachnik
et al.
, 2012
). In our case, appli-
cation of this method will only work for the
translational model shapes, and when the phone produces a
high signal-to-noise ratio recording, most likely from the free
oscillations of the building following earthquake shaking. Also,
the elevation of the phone is a factor as phones on the higher
floors will likely have higher signal-to-noise ratio recordings.
Because we know that the N-S fundamental frequency of
the Millikan Library building is 1.7 Hz, we should be able to
estimate which component is N-S by applying a narrowband
filter to the phone
s two horizontal-component records. Fig-
ure
6
shows that the component shown in the top panel is
likely close to the N-S direction; this is before any corrective
rotation is applied. We observe that there is a small signal on
the bottom-panel component as well, likely due to the fact that
the phone is not perfectly oriented N-S during the tests.
To find the most accurate orientation of the phone during
the tests, we can rotate the two horizontal components until an
angle is found that minimizes the signal on one component.
Figure
7
shows the result of rotating the phone 1.5°; the energy
on the E-Wcomponent is minimized. Of course, this is easy for
our test, because we shake the building in one direction at a
time. During an earthquake, the later parts of the motion that
are dominated by free vibrations should be usable in the same
way to orient the phones. This will allow more meaningful
analysis of earlier parts of the record.
Although a building
s mode may not already be known, in
some cases it is possible to estimate modal properties from the
known geometry of the building. In particular, many buildings
are approximately rectangular and the orientation of the build-
ing
s normal modes is approximately aligned with the natural
axes of the building. If structural design information is avail-
able to determine the most compliant direction of building
deformation, then one can assume that the lowest-frequency
normal mode of a building is also aligned with this direction.
Even if the axis of the lowest frequency mode is unknown, it is
often safe to assume that it is one of the two natural axes of the
building. Knowing which of the two axes is the correct one can
Figure 4.
Waveform comparisons for two horizontal components between the
stack of seven phone recordings and the Episensor (CI MIK). (a) N-S component;
(b) E-W component. Frequency labels indicate when the test run is at or near the
fundamental or torsional frequencies of the building. The time for each test run can
be found in
Table S1. The color version of this figure is available only in the
electronic edition.
Seismological Research Letters Volume 89, Number 2A March/April 2018 599
be determined by observing the building
s response with just
one record of approximately known orientation (it is most
beneficial to have at least one sensor with known orientation).
In some cases, tall buildings may have additional modes
(overtones) at frequencies that are odd-integer multiples of
the fundamental-mode frequency (
Lee
et al.
, 2003
). The
orientations of higher modes can also be used to orient the
phones. Finally, there are cases in which shear waves travel ver-
tically up a building (typical wave velocity of
150 m
=
s
). In
general, the polarization of vertically propagating shear waves
is approximately constant over the building height. If the
polarization of the incident wave is known,
then the orientation of the phones can be de-
termined by finding the appropriate rotation to
reproduce the incident ground motion. Exam-
ples of how this is applied can be found in
Cheng
et al.
(2015)
.
DISCUSSION AND CONCLUSIONS
SHM is important for keeping track of the
changes in the state of buildings, not only of
large-scale structures but also of everyday resi-
dential buildings for the purpose of damage de-
tection. Harnessing the smartphones of
individual private users could open the door
to monitoring many more structures including
residential buildings in the future.
When compared with the current wireless
sensor network monitoring systems, using con-
sumer smartphones as a way to conduct SHM
has the following benefits:
ability to monitor millions of buildings
within short periods of time;
almost no cost for hardware and installa-
tion labor;
low cost for long-term maintenance; and
complementary data to the current moni-
toring system.
The first point is the most important ben-
efit of using smartphones; millions of buildings
could be monitored with just the download of
an application. This will greatly improve our
current monitoring ability at the city scale
or even nation scale to reduce the earthquake
risks. Also, a smartphone monitoring system is
complementary to existing SHM systems by pro-
viding more data in the same building for vali-
dation and to fill in spatial sampling gaps.
This article is only a starting point for the
concept of crowdsourcing SHM. There are still
many challenges, including the following:
1. Determining accurate location, height
(floor), and orientation of the phone is
one of the biggest technical issues to solve.
There are several commercial solutions combining differ-
ent sensors in the phones to get an estimate, but the
results need to be tested. Another potential solution is
to ask users to input their location and floor number after
the earthquake into a questionnaire.
2. This test placed the phones in an ideal location, that is, on
the top floor at the corner of the building with the phones
lying on the floor. In reality, only a few phones may satisfy
these requirements. Testing of phones on different floors,
in different locations, and on different surfaces (desk,
couch, etc.) is also necessary.
Figure 5.
Displacement time-series comparisons (high-pass filtered at 0.5 Hz).
These recordings are extracted from the period of largest-amplitude response to
the shaking that occurred in the N-S direction, between 4820 and 4850 s (see Fig.
2a
for reference). (a) The Episensor (CI MIK) compared with a single phone (Samsung
Galaxy S4); (b) the Episensor compared with the seven-phone stack (see
Fig. S1
for the accelerations in the same time range). The color version of this figure is
available only in the electronic edition.
600 Seismological Research Letters Volume 89, Number 2A March/April 2018
3. Tests in different building types will also provide more
insight into the types of information we can extract from
the phone records for different buildings.
4. Making use of ambient vibrations of the building and
nearby small earthquakes to extract a building
s character-
istic parameters will expand the capability of SHM from the
phones. More work is needed to determine if ambient noise
recordings could be recovered from phones and the lower
limits of earthquake shaking that can pro-
vide useful recordings.
These challenges illustrate that much re-
search is needed before we can have a fully op-
erational crowdsourced SHM system. However,
the initial tests shown here illustrate the prom-
ise of using smartphones for SHM of buildings,
and provide the basis for this future develop-
ment of MyShake functionality.
DATA AND RESOURCES
Data for the shaker tests are available via request
to rallen@berkeley.edu.
ACKNOWLEDGMENTS
This project is funded by Deutsche Telecom
Silicon Valley Innovation Center, and by the
Gordon and Betty Moore Foundation through
Grant GBMF5230 to UC Berkeley. In the analy-
sis of the data, ObsPy (
Beyreuther
et al.
, 2010
;
Megies
et al.
,2011
) and mtspec package were
used (doi:
10.5281/zenodo.321789
); we thank
the authors. We also thank Lucy Yin, Anthony
Massari, Kenny Buyco, and other students and
staff for setting up the shaker tests. The authors
thank the four anonymous reviewers; their com-
ments greatly improved this article.
REFERENCES
Beyreuther, M., R. Barsch, L. Krischer, T. Megies, Y.
Behr, and J. Wassermann (2010). ObsPy: A py-
thon toolbox for seismology,
Seismol. Res. Lett.
81,
no. 3, 530
533, doi:
10.1785/gssrl.81.3.530
.
Bradford, S. C., J. F. Clinton, J. Favela, and T. H.
Heaton (2004).
Results of Millikan Library Forced
Vibration Testing
, Technical report, California In-
stitute of Technology, available at
http://resolver.
caltech.edu/CaltechEERL:EERL
2004
03
(last accessed
December 2017).
Cheng, M. H., M. D. Kohler, and T. H. Heaton (2015).
Prediction of wave propagation in buildings using
data from a single seismometer,
Bull. Seismol.
Soc. Am.
105,
no. 1, 107
119, doi:
10.1785/
0120140037
.
Clayton, R. W., T. Heaton, M. Chandy, A. Krause, M.
Kohler, J. Bunn, R. Guy, M. Olson, M. Faulkner,
M. H. Cheng,
et al.
(2011). Community seismic network,
Ann. Geo-
phys.
54,
no. 6, 738
747, doi:
10.4401/ag-5269
.
Clayton, R. W., T. Heaton, M. Kohler, M. Chandy, R. Guy, and J. Bunn
(2015). Community seismic network: A dense array to sense earth-
quake strong motion,
Seismol. Res. Lett.
86,
no. 5, 1354
1363, doi:
10.1785/0220150094
.
Clinton, J. F., S. C. Bradford, T. H. Heaton, and J. Favela (2006).
The observed wander of the natural frequencies in a structure,
Bull. Seismol. Soc. Am.
96,
no. 1, 237
257, doi:
10.1785/
0120050052
.
Figure 6.
Horizontal-component recordings after application of a narrow-band
filter 1.69
1.71 Hz. Energy can be seen on both components. The upper panel is the
component oriented roughly N-S, and bottom panel is the component oriented
roughly E-W. The phone used here is the same Samsung Galaxy S4 as the one shown
in Figure
2
. The color version of this figure is available only in the electronic edition.
Figure 7.
Result of rotating the horizontal components of the narrow-band fil-
tered signal (Fig.
6
) anticlockwise 1.5° to minimize energy on one component. The
upper panel shows the axis aligned to N-S after having rotated the components;
the lower panel is the axis aligned to E-W. The color version of this figure is avail-
able only in the electronic edition.
Seismological Research Letters Volume 89, Number 2A March/April 2018 601
Cochran, E., J. Lawrence, C. Christensen, and R. S. Jakka (2009).
The Quake-Catcher Network: Citizen science expanding seismic
horizons,
Seismol. Res. Lett.
80,
no. 1, 26
30, doi:
10.1785/
gssrl.80.1.26
.
D
Alessandro, A., and G. D
Anna (2013). Suitability of low-cost three-
axis MEMS accelerometers in strong-motion seismology: Tests on
the LIS331DLH (iPhone) accelerometer,
Bull. Seismol. Soc. Am.
103,
no. 5, 2906
2913, doi:
10.1785/0120120287
.
Dashti, S., J. D. Bray, J. Reilly, S. Glaser, A. Bayen, and E. Mari
(2014). Evaluating the reliability of phones as seismic monitoring
instruments,
Earthq. Spectra
30,
no. 2, 721
742, doi:
10.1193/
091711EQS229M
.
Duennebier, F. K., P. N. Anderson, and G. J. Fryer (1987). Azimuth
determination of and from horizontal ocean bottom seismic
sensors,
J. Geophys. Res.
92,
no. B5, 3567
3572, doi:
10.1029/
JB092iB05p03567
.
Ekstrom, G., and R. W. Busby (2008). Measurements of seismometer
orientation at USArray transportable array and backbone stations,
Seismol. Res. Lett.
79,
no.4,554
561, doi:
10.1785/gssrl.79.4.554
.
Farrar, C. R., and K. Worden (2007). An introduction to structural
health monitoring,
Phil. Trans. Roy. Soc. A
365,
no. 1851, 303
315, doi:
10.1098/rsta.2006.1928
.
Faulkner, M., R. Clayton, T. Heaton, K. M. Chandy, M. Kohler, J. Bunn,
R. Guy, A. Liu, M. Olson, M. H. Cheng,
et al.
(2014). Community
sense and response systems: Your phone as quake detector,
Comm.
ACM
57,
no. 7, 66
75, doi:
10.1145/2622633
.
Feng, G., Z. Li, B. Xu, X. Shan, L. Zhang, and J. Zhu (2016). Coseismic
deformation of the 2015
M
w
6.4 Pishan, China, earthquake esti-
mated from Sentinel-1A and ALOS2 data,
Seismol. Res. Lett.
87,
no. 4, doi:
10.1785/0220150264
.
Finazzi, F. (2016). The earthquake network project: Toward a crowd-
sourced smartphone-based earthquake early warning system,
Bull.
Seismol. Soc. Am.
106,
no. 3, doi:
10.1785/0120150354
.
Kim, S., S. Pakzad, D. E. Culler, J. Demmel, G. Fenves, S. Glaser, and
M. Turon (2007). Health monitoring of civil infrastructures
using wireless sensor networks,
Proc. 6th Int. Conf. Information
Processing in Sensor Networks
, 254
263, doi:
10.1109/
IPSN.2007.4379685
.
Kohler, M. D., T. H. Heaton, and M.-H. Cheng (2013). The community
seismic network and quake-catcher network: Enabling structural
health monitoring through instrumentation by community partic-
ipants,
Proc. of SPIE - Int. Soc. Opt. Eng.
8692,
86923X, doi:
10.1117/12.2010306
.
Kohler, M. D., A. Massari, T. H. Heaton, H. Kanamori, E. Hauksson,
R. Guy, R. W. Clayton, J. Bunn, and K. M. Chandy (2016). Down-
town Los Angeles 52-story high-rise and free-field response to an oil
refinery explosion,
Earthq. Spectra
32,
no. 3, 1793
1820, doi:
10.1193/062315EQS101M
.
Kong, Q., R. M. Allen, and L. Schreier (2016). MyShake: Initial obser-
vations from a global smartphone seismic network,
Geophys. Res.
Lett.
43,
no. 18, doi:
10.1002/2016GL070955
.
Kong, Q., R. M. Allen, L. Schreier, and Y.-W. Kwon (2016). MyShake: A
smartphone seismic network for earthquake early warning and be-
yond,
Sci. Adv.
2,
no. 2, e1501055, doi:
10.1126/sciadv.1501055
.
Kong, Q., Y.-W. Kwony, L. Schreierz, S. Allen, R. Allen, and J. Strauss
(2015). Smartphone-based networks for earthquake detection,
15th
International Conf. on Innovations for Community Services (I4CS)
,
IEEE, 1
8.
Lee, W. H. K., H. Kanamori, P. C. Jennings, and C. Kisslinger (Editors)
(2003).
International Handbook of Earthquake and Engineering
Seismology. Part B
, Elsevier, Amsterdam, The Netherlands.
Lynch, J. P., and K. J. Loh (2006). A summary review of wireless sensors
and sensor networks for structural health monitoring,
Shock Vib.
Dig.
38,
no. 2, 91
128.
Megies, T., M. Beyreuther, R. Barsch, L. Krischer, and J. Wassermann
(2011). ObsPy
What can it do for data centers and observatories?,
Ann. Geophys.
54,
no. 1, 47
58, doi:
10.4401/ag-4838
.
Moreno-Gomez, A., C. A. Perez-Ramirez, A. Dominguez-Gonzalez,
M. Valtierra-Rodriguez, O. Chavez-Alegria, and J. P. Amezquita-
Sanchez (2017). Sensors used in structural health monitoring,
Arch.
Comput. Methods Eng.
1
18, doi:
10.1007/s11831-017-9217-4
.
Ozer, E., M. Q. Feng, and D. Feng (2015). Citizen sensors for SHM:
Towards a crowdsourcing platform,
Sensors
15,
no. 6, 14,591
14,614, doi:
10.3390/s150614591
.
Paek, J., K. Chintalapudi, J. Caffrey, R. Govindan, and S. Masri (2005). A
wireless sensor network for structural health monitoring: Perfor-
mance and experience,
Embed. Netw. Sens.
1
10, doi:
10.1109/
EMNETS.2005.1469093
.
Prieto, G. A., R. L. Parker, and F. L. Vernon (2009). A Fortran 90 library
for multitaper spectrum analysis,
Comput. Geosci.
35,
no. 8, 1701
1710, doi:
10.1016/j.cageo.2008.06.007
.
Reilly, J., S. Dashti, M. Ervasti, J. D. Bray, S. D. Glaser, and A. M. Bayen
(2013). Mobile phones as seismologic sensors: Automating data ex-
traction for the iShake system,
IEEE Trans. Automat. Sci. Eng.
10,
no. 2, 242
251, doi:
10.1109/TASE.2013.2245121
.
Stachnik, J. C., A. F. Sheehan, D. W. Zietlow, Z. Yang, J. Collins, and
A. Ferris (2012). Determination of New Zealand ocean bottom
seismometer orientation via Rayleigh-wave polarization,
Seismol.
Res. Lett.
83,
no. 4, 704
713, doi:
10.1785/0220110128
.
Xu, N., S. Rangwala, K. K. Chintalapudi, A. Broad, R. Govindan, and D.
Estrin (2004). A wireless sensor network for structural monitoring,
Proc. 2nd International Conf. Embedded Networked Sensor System
,
ACM, 3
5 November 2004, 13
24.
Yin, R.-C., Y.-M. Wu, and T.-Y. Hsu (2016). Application of the low-cost
MEMS-type seismometer for structural health monitoring: A pre-
study,
2016 IEEE International Instrumentation and Measurement
Technology Conf. Proc. IEEE
,1
5.
Yu, Y., R. Han, X. Zhao, X. Mao, W. Hu, D. Jiao, M. Li, and J. Ou
(2015). Initial validation of mobile-structural health monitoring
method using smartphones,
Int. J. Distrib. Sens. Networks
11,
no. 2, Article ID 274391, doi:
10.1155/2015/274391
.
Qingkai Kong
Berkeley Seismology Laboratory
University of California, Berkeley
289 McCone Hall
Berkeley, California 94720 U.S.A.
kongqk@berkeley.edu
Richard M. Allen
Berkeley Seismology Laboratory
University of California, Berkeley
279 McCone Hall
Berkeley, California 94720 U.S.A.
Monica D. Kohler
Thomas H. Heaton
Civil and Mechanical Engineering
Caltech
MC: 104-44
Pasadena, California 91125 U.S.A.
Julian Bunn
109 Powell-Booth Computing Center
Caltech
MC: 158-79
Pasadena, California 91125 U.S.A.
Published Online 10 January 2018
602 Seismological Research Letters Volume 89, Number 2A March/April 2018