5 Jang
5 Jang
Abstract. Structural health monitoring (SHM) of civil infrastructure using wireless smart sensor networks
(WSSNs) has received significant public attention in recent years. The benefits of WSSNs are that they are
low-cost, easy to install, and provide effective data management via on-board computation. This paper reports
on the deployment and evaluation of a state-of-the-art WSSN on the new Jindo Bridge, a cable-stayed bridge
in South Korea with a 344-m main span and two 70-m side spans. The central components of the WSSN
deployment are the Imote2 smart sensor platforms, a custom-designed multimetric sensor boards, base stations,
and software provided by the Illinois Structural Health Monitoring Project (ISHMP) Services Toolsuite. In
total, 70 sensor nodes and two base stations have been deployed to monitor the bridge using an autonomous
SHM application with excessive wind and vibration triggering the system to initiate monitoring. Additionally,
the performance of the system is evaluated in terms of hardware durability, software stability, power consumption
and energy harvesting capabilities. The Jindo Bridge SHM system constitutes the largest deployment of
wireless smart sensors for civil infrastructure monitoring to date. This deployment demonstrates the strong
potential of WSSNs for monitoring of large scale civil infrastructure.
Keywords: structural health monitoring; wireless smart sensor network; cable-stayed bridge; deploy-
ment; evaluation.
1. Introduction
Civil infrastructure systems such as bridges, buildings, pipelines and offshore structures are valuable
national assets that must be maintained to ensure economic prosperity and public safety. Many
bridge structures in modern countries have reached their design life and will need to be replaced or
retrofitted to remain in service. Indeed in the United States, more than 149,000 bridges are
structurally deficient or functionally obsolete according to the 2006 report from the Federal Highway
Administration (FHWA 2009). Thus, the ability to assess the structural condition and possibly
2. Bridge description
The Jindo Bridges are twin cable-stayed bridges connecting Haenam on the mainland with the
Structural health monitoring of a cable-stayed bridge using smart sensor technology 441
Jindo Island (see Fig. 1). The Jindo Island is the third largest island in South Korea, and Haenam,
which is located in the south-west tip of the Korean peninsula. Each of these bridges consists of
three continuous spans, with a 344-m central main span and two 70-m side spans.
The original Jindo Bridge, constructed in 1984 by Hyundai Engineering & Construction Co., Ltd.,
was the first cable-stayed bridge in South Korea. The width of the bridge is 11.7-m, the design
traffic velocity is 60 km/hr, and the design live load is based on AASHTO HS-20-44 (DB-18). The
second Jindo Bridge was constructed in 2006 by Hyundai Engineering & Construction Co., Ltd.,
Daelim Industrial Co., Ltd. and Namhei Co., Ltd. The width of the second bridge is 12.55 m, the
traffic design velocity is 70 km/hr, and the design live load is based on AASHTO HS-20-44 (DB-
24, DL-24). It has a streamlined steel box girder supported by 60 high-strength steel cables connected
to two pylons. The structural drawing of the bridge is shown in Fig. 2.
Both bridges have existing SHM systems based on wired sensors. The first Jindo Bridge has 38
strain gages, four inclinometers, two anemometers, two seismic accelerometers, five uniaxial capacitive
accelerometers, and 15 uniaxial piezoelectric accelerometers. The second Jindo Bridge has 15
thermometers, 15 strain gages, four biaxial inclinometers, two string pots, two laser displacement
meters, 24 Fiber Bragg Grating sensors, 20 uniaxial capacitive accelerometers, two biaxial force
balance type accelerometers, and three triaxial seismic accelerometers. Among two bridges, the
second Jindo Bridge is selected as the test bed for this research, for two primary reasons: (i) the
existing SHM system is quite versatile, including accelerometers and fiber optic sensors and (ii) the
design and construction documents are more complete.
3.1 Hardware
The wireless smart sensor is the key element of the bridge monitoring system. Two hardware
configurations are employed: a gateway node attached to the base station PC, and battery-operated
leaf nodes at remote locations to the base station. Because the wind excitation is the major source
of bridge vibration, two types of sensor boards have been employed: the SHM-A sensor board to
measure vibrations and the SHM-W sensor board to measure signals from an anemometer. To
evaluate the potential for long-term deployment, solar panels and rechargeable batteries have been
installed on selected nodes.
SHM-A board for this bridge monitoring application; however, the sampling rate can be chosen
nearly arbitrarily by designing appropriate filters for the QF4A512. The components of the SHM-A
sensor board are identified in Fig. 3. More details can be found in (Rice et al. 2010).
Two hardware configurations of smart sensor nodes are required for the wireless communication
and sensing: a gateway node for sending commands and receiving wireless data from the network,
and the battery-powered nodes remote to the base station. The gateway node consists of an Imote2
stacked on an IIB2400 interface board connected to the base station PC via a USB/UART port. The
leaf nodes consist of an SHM-A sensor board and Imote2 stacked on the battery board as shown in
Fig. 4. To increase the communication range, both nodes are equipped with an Antenova gigaNova
Titanis 2.4 GHz external antenna (Antenova 2009). The sensor nodes are placed in environmentally
hardened enclosures to endure the harsh environment at the Jindo Bridge site.
small modifications.
The SHM-W board is developed by modifying the SHM-A board to have three external 0-5V
input channels and one acceleration channel. The wind speed (channel 1), horizontal and vertical
wind directions (channels 2 and 3) are measured through analog input interface connectors on the
SHM-W board as shown in Fig. 5(c). Because the SHM-W board also uses the QFA512-based
sensor board, it uses the same software drivers as the SHM-A board; as a result, the wind data is
acquired precisely synchronized with the acceleration data from the SHM-A board. Another
modification for SHM-W board is that it is adjusted so to that the full range of the 0-5V inputs is
utilized, resulting in better resolutions for the wind data.
Fig. 6 Power supply options for Imote2 and battery board modification for recharging option
Fig. 7 Solar panel and rechargeable battery for Jindo Bridge SHM
the another zero-ohm resistor (R3) on the battery board (see Fig. 6).
Considering the range of input power that the Imote2 can support (4.6~10 V, up to 1400 mA), the
SPE-350-6 solar panel from Solarworld (9 V-350 mA) shown in Fig. 7(a) was employed. A
Powerizer lithium-polymer rechargeable battery (Powerizer 2009) was selected, which reaches up to
4.2 V when fully charged. This system has 10,000 mAh capacity (half of that of D-cell batteries),
will allows this system to supply power to the Imote2 for an extended period without recharging.
The Illinois SHM Project developed a suite of services, denoted the ISHMP Services Toolsuite,
that implements key middleware functionality to provide high-quality sensor data and to transfer the
data reliably to the base station via wireless communication across the sensor network, and a library
of numerical algorithms. This open-source software is available at http://shm.cs.uiuc.edu/software.html.
More detailed information regarding this software can be found in Rice et al. (2010).
The toolsuite components are categorized into foundation services, application services, tools and
utilities, and continuous and autonomous monitoring services. The foundation services provide the
fundamental functionalities to measure synchronized sensor data reliably (Mechitov et al. 2004,
446 Shinae Jang et al.
Nagayama and Spencer 2007, Rice et al. 2008). The application services are the numerical algorithms to
implement SHM applications on the Imote2, including modal identification and damage detection
algorithms. The tools and utilities support network maintenance and debugging. This category has
essential services for full-scale monitoring as well as sensor maintenance. The key features of these
services include RemoteSensing, a remote data measurement application; DecentralizedData
Aggretation, data measurement and on-board computation to calculate the correlation functions for
decentralized local groups; imote2comm and autocomm, an automatic terminal tool to interact
between the base station PC and a gateway node; Vbat, a command for checking a leaf node’s
battery level; TestRadio, a tool for assessing radio communication quality, and many others.
The continuous and autonomous monitoring service critical for this deployment is AutoMonitor,
an autonomous SHM network management application, which combines RemoteSensing, ThresholdSentry,
and SnoozeAlarm (Rice and Spencer 2009). SnoozeAlarm is a strategy that allows the network to
sleep most of the time, thus improving energy efficiency and allowing long-term system deployment. To
wake the network for an important event, the ThresholdSentry application defines a specified
number of the leaf nodes as sentry nodes. The sentry nodes wake up at predefined times and
measure a short period of acceleration or wind data. When the measured data exceeds a pre-defined
threshold, the sentry node sends an alarm to the gateway node, which subsequently wakes the entire
network for synchronized data measurement. In this way, AutoMonitor enables the automatic,
continuous monitoring with reduced power consumption.
The AutoMonitor application employs multiple threshold levels for the Jindo Bridge SHM system.
When a single threshold level is used to define the maximum number of events during given period,
if the threshold level is too high, the data measurements would rarely if ever occur. If too small a
threshold level is used, the number of events can be exhausted with small structural responses,
hence possible strong vibrations during remaining period cannot be captured. The multiple threshold
levels, having separate maximum number of events, make it possible to measure both rare but
strong responses and frequent but low-level ambient vibration during a given period, which enables
more effective SHM.
Another feature of the AutoMonitor application is a wind threshold sentry. The wind sentry node
is equipped with SHM-W sensor board with ultrasonic anemometer. Strong winds, such as those
that occur during typhoons, induce large structural responses. Therefore, the wind sentry will trigger
network sensing when the velocity exceeds a threshold. In this deployment, both vibration- and
wind-based sentry nodes have been installed for the AutoMonitor application. In summary, the final
software version is the AutoMonitor application having both data measurement and on-board
computational functionality, a multiple threshold triggering strategy using vibration/wind sentries
pursuing energy efficiency using SnoozeAlarm.
The developed hardware and software framework has been deployed on the Jindo Bridge to
realize the first large-scale, autonomous, WSSN-based SHM system. Due to the harsh environment at
the site, the delicate electrical components of the SHM system have been hardened to prevent
corrosion, overheating, or other damage. Also, many verification and optimization steps have been
required to extend the laboratory-scale SHM system to a full-scale deployment for a long-span bridge
structure. The details of the hardening and optimization process are provided in this section.
Structural health monitoring of a cable-stayed bridge using smart sensor technology 447
The network topology is carefully determined to ensure reliable network sensing for the Jindo
Bridge. The major factors to define the network topology are the size of network, the communication
range, etc. The total length of the bridge is 484 m, the communication range of Imote2 with external
antenna is ~200 m, and the number of sensor nodes in network is 70. Considering the range and the
communication time, the network was divided into two sub-networks: one on the Jindo side and the
other on the Haenam side.
The Jindo sub-network consists of 33 nodes with 22 nodes on the deck, three nodes on the pylon,
and eight nodes on the cables. The Haenam sub-network consists of 37 nodes with 26 nodes on the
deck, three nodes on the pylon, and seven nodes on the cables. Each sub-network is controlled by a
base station located on the concrete piers supporting the steel pylons of the first Jindo Bridge. These
locations were chosen to achieve consistent line-of-site communication with leaf nodes.
An industrial-grade PC AAEON AEC-6905 was adopted as a base station for its fan-less architecture,
protecting it from dust and moisture. The UPS backup APC ES550 protects the base station
components from an unexpected electric surges and outages. The gateway node is consisted of an
Imote2 stacked with an interface board for interactive serial communication and SHM-A sensor
board and 2.4 GHz 5 dBi dipole antenna as shown in Fig. 9. A wired internet line was installed to
the PC with an ADSL internet modem to provide remote access to base station, to control the
WSSN, and to download the measured data.
Software for remote control and data download is installed on the base station. On the top of
basic OS and anti-virus software, the key components of software are Cygwin and the autocomm
executable application for interfacing with the gateway node, the VNC (Virtual Network Computing)
server for remote desktop control, and an FTP (File Transfer Protocol) server to remotely download
the measured data.
magnet, to prevent the vibration due to the vortex-shedding in the gap between the enclosure and
structure surface (see Fig. 12(b)). This approach reduces the installation time and effort under the
deck; however, the magnet-based method is not feasible for the cable sensors due to round surface.
Therefore, the leaf nodes on the cables are mounted using two U-bars and aluminum mounting plate
as shown in Fig. 11(c). Because these nodes are directly exposed to sun light, the stainless steel
cover is employed to protect the PVC enclosure.
Because single-hop communication was used for this deployment, radio communication tests were
conducted on-sight using the TestRadio application to measure the packet reception rates at various
communication distances. During the communication range tests, the gateway node was located at
the Haenam side pylon and the leaf nodes were gradually moved out to the mid-span. The number
of sensors which were unresponsive to the TestRadio request from the gateway node was counted as
shown in Fig. 13. In total, 73 Imote2s were tested and five nodes were found to be unresponsive
and subsequently replaced. At the mid-span at 172 m, 23 Imote2s were found to communicate
effectively. Based on these radio communication tests, sensors location were optimized as shown in
Fig. 14.
The communication parameters have been optimized to operate the large WSSN on a long-span
452 Shinae Jang et al.
send the sensing parameters to all of the nodes in the network. This total wait time is expressed by
T = Sc + n ⋅ Sn (1)
where, Sc is a constant wait time, Sn is a short nodal wait time for each sensor denoted as, and n is the
number of sensors. These two parameters are increased in the updated state considerably because the
communication took a longer time for the long distances between the leaf nodes.
The data measurement parameters include the sampling frequency, the measurement directions,
and the number of data point for each channel. 10,000 data points are measured in three directions
at 10 Hz.
The SnoozeAlarm parameters include the duration of the wake/listen time and the sleep time. The
sleep time can be chosen based on the target excitation. For Jindo Bridge, loading from strong
winds from typhoons having a duration of several hours to a day is significant. For such excitations,
waking the network up within several minutes of a trigger event is quite reasonable. To accommodate
this, the deep sleep interval is set to 15 seconds, and the wakeup time to listen for commands is 750
ms. With this setup, the wake up time for the entire network was in the range of 1~5 minutes based
on the radio communication conditions.
The ThresholdSentry parameters include the number of sentry nodes used in the network. The
number of sentry nodes for this deployment is three for Haenam side and two for Jindo side. The
sentry checks the acceleration level frequently to catch large vibration due to strong winds. The sentries
are set to wake up every 10 minutes and measure the acceleration for 10 seconds. The threshold
values for the vibration sentry are 10 mg and 50 mg, so that large vibration events more than 50 mg
should be recorded. For the wind sentry, the thresholds are 3 m/sec and 8 m/sec to measure both the
usual state of wind-induced vibration (3-4 m/sec) and stronger wind vibration events.
Finally, a Watchdog Timer is used to reset the nodes to ensure the network reliability in the case
of a node hanging due to some unexpected error. The Watchdog update time is the waiting time
before the processor resets the node when the node gets hung up. Using this setup, an indefinite
hang-up of nodes may be avoided. The Watchdog period should be longer than the measurement
duration so that the Watchdog timer does not interrupt the measurement and reset the leaf nodes.
The Watchdog update was set to one minute, and the Watchdog timer period is set to 10 minutes
during the initial state, because the network sensing takes less than 10 minutes. This is updated to
60 minutes and 60 minutes respectively for the updated state to measure longer data with increased
distances of sensor distribution.
The hardware components including smart sensor nodes base stations and the anemometer with
many installation details have shown reliable performance during this deployment. One of the
biggest enablers of the deployment on the Jindo Bridge is the SHM-A sensor board. The resolution
of the sensor board is ~0.3 mg and adequate to measure the bridge vibration in the range of 5~30 mg.
The programmable sampling frequency features are critical to capture major natural frequencies of
the bridge under 1 Hz as opposed to ITS400CA basic sensor board (MEMSIC 2010) support higher
sampling frequency than 280 Hz. Furthermore, the modified SHM-A board with an anemometer, the
454 Shinae Jang et al.
SHM-W sensor board, has reliably measured the wind information; however, the horizontal component
of the wind direction is not measured by channel 2 of the SHM-W board because of a hardware
malfunctioning problem on the anemometer.
The two base stations have been functioning reliably for four months, enabling stable remote
monitoring of the Jindo Bridge and communication with each gateway node. When checking
the condition of the inside of the base station enclosure after four months of operation, it was
confirmed that it has successfully protected the computer from overheating and the harsh
environment.
The enclosures for the leaf nodes have also performed well. The inside of sensor enclosure was
dry and the temperature was acceptable. The magnet-based attachment has proven to be an excellent
solution for the Jindo Bridge because all leaf nodes have been attached firmly for four months.
The AutoMonitor application has shown stable performance after appropriate optimization of the
sensing and radio communication parameters. All software is operating reliably and remotely
desktop using a VNC® server. RemoteSensing and DecentralizedDataAggregation work successfully
with optimized communication parameters.
The ambient vibration data at the Jindo Bridge has been acquired from the WSSN. One sample of
recorded data has shown in Fig. 16, of which levels are ~8 mg, ~40 mg, ~5 mg for the deck, cable
and pylon, respectively.
The power spectral densities (PSD) of the vibration data have been investigated. Fig. 17(a) shows
the PSDs of the deck vibrations at the mid span, quarter span and at the pylon. The PSD magnitude
of deck sensors near pylon is almost zero; however, the other two sensors at the mid span and the
quarter span show significant energy around 0.44, 0.66 and 1.03 Hz, implying the natural frequencies of
the bridge. Fig. 17(b) shows the comparison between the PSD from the existing wired sensor at the
quarter span and the PSD from the deployed wireless sensor at the same location. Though the wired
sensor data was measured in 2007 and the wireless sensor data was measured in 2009, most natural
frequencies agree well with the previous data.
Fig. 18 shows the PSD from six cables sensors and from the pylon sensors the one in the side
and at the top. For the cable sensors, the consistent peaks are shown around 0.44, 0.66, 1.03 Hz
for all sensors and different peaks for various cables in other frequency regions. The natural
Fig. 17 Power spectral density comparison at the quarter span of the Jindo Side
frequencies of the pylon sensor in the side (1) and of the one at the top are different. Detailed
modal analysis and detailed cable tension estimation is described in the companion paper in this
issue (Cho et al. 2010).
The wind speed and direction has been successfully measured using the 3D ultra-sonic anemometer
at the mid span. The outputs of the SHM-W sensor board are raw voltage measurements from the
anemometer. This data have been converted to the wind speed and voltage using the anemometer
sensitivity. The data are synchronized with vibration data measured by SHM-A sensor board. Fig.
19 shows example data. In this sample data, the wind speed was 4-6 m/sec and the direction is zero
degrees to the longitudinal direction of the bridge.
456 Shinae Jang et al.
Fig. 20 Communication time of 23 sensor nodes based on the number of data points collected
Fig. 20 shows the operating time for the RemoteSensing application in terms of various phases for
different requested sample sizes. In total, three axes of accelerations were measured at 50 Hz from
23 sensor nodes. In this specific deployment, the communication time sending data back to the
gateway node has been measured. Although the total communication time depends on various
environmental and communication factors, the communication time shows a linear relationship with
the number of requested data points as shown in Fig. 20. Based on this result, the total communication
time to acquire 30,000 points from 46 sensor nodes of two networks is about 30 minutes.
The battery voltage levels for all nodes in both sides have been recorded for two months as
shown in Fig. 21. The average initial voltage of the three D-cell batteries on each sensor nodes was
4.6 V, and the average on board voltage reading using RemoteVbat command was 4.2 V. The on
board voltage reading is 0.3~0.4 V less than the actual battery voltage because of the diode drop on
the battery board to prevent damage due to incorrect installation of the batteries. From 8/27/09 to 9/
8/09, the parameters stored in FLASH on each Imote2 were uploaded many times to optimize
Structural health monitoring of a cable-stayed bridge using smart sensor technology 457
network performance, which is a power demanding procedure. After 9/8/09, the AutoMonitor application
has run continuously to measure the data. During this period, the power consumption has been
approximately linear. The minimum onboard voltage for sensing is 3.4 V. The power consumption
depends on the frequency of data measurement, data length, and sleeping parameters. Based on
once-a-day measurement with the network parameters mentioned in Table 2, three D-cell batteries
can operate about two months.
In total, eight sensor nodes employ solar panels and rechargeable batteries. Five of the solar
powered sensor nodes are on the cables, two on the pylons, and one on the deck. Fig. 22 shows the
voltage of the solar rechargeable batteries during monitoring. The voltage levels of the rechargeable
Fig. 22 Charging status of solar rechargeable batteries on Jindo Bridge SHM system
458 Shinae Jang et al.
battery have maintained around 4.15 V, which confirms that the charging process of the solar power
system is working well. However, the solar powered node located under the deck (Haenam side,
node 6) shows a continuous decrease in the voltage level. The reason for the decrease is that it only
can receive indirect reflected sunlight, because its orientation is downward. Either a more sensitive
solar panel, reorienting the panel, or another type of energy harvesting system should be considered
for sensors located under the deck in the next deployment.
7. Conclusions
A state-of-the-art SHM system using a WSSN has been successfully deployed on the Jindo Bridge
in South Korea to verify the performance of the system and to serve as a driver for advancement of
smart sensor technology. The Imote2 has been selected as the wireless sensor platform to use along
with custom-designed SHM-A and SHM-W sensor boards. An autonomous structural monitoring system
has been developed employing a threshold detection strategy and an energy-efficient sleeping mode
to extend the network lifetime. Solar powered nodes have been employed to investigate the possibility of
energy harvesting to power the sensor network.
In total, 70 sensor nodes have been installed, divided into two sub-networks to decrease the
communication time and because of the limit of the radio communication range. All sensors are carefully
located based on radio communication capability determined by extensive radio communication
tests. The measured data shows a good agreement with data from the existing wired system, which
verifies that the data quality of the WSSN is reliable. Successful deployment of this WSSN demonstrates
the suitability of the Imote2 smart sensor platform, the SHM-A sensor board, and the ISHMP software
for full-scale, continuous, autonomous SHM.
Acknowledgements
This research is supported in part by the National Science Foundation Grant CMS 06-00433 (Dr.
S.C. Liu, Program Manager), Global Research Network program from the Natural Research Foundation
in Korea (NRF-2008-220-D00117). The support of the Ministry of Land, Transport and Maritime
Affairs in Korea, Daewoo Engineering Co. Ltd. and Hyundai Engineering & Construction Co. Ltd.
is also gratefully acknowledged.
References
Abe, M., Fujino, Y., Yanagihara, M. and Sato, M. (2000), “Monitoring of hakucho suspension bridge by ambient
vibration measurement”, Proceedings of the Nondestructive Evaluation of Highways, Utilities, and Pipelines
IV, Newport Beach, USA, March.
Antenova (2009), Titanis: p/n B4844/B6090, http://www.antenova.com/?id=536.
Çelebi, M., Purvis, R., Hartnagel, B., Gupta, S., Clogston, P., Yen, P., O’Connor, J. and Franke, M. (2004),
“Seismic instrumentation of the Bill Emerson Memorial Mississippi River Bridge at Cape Girardeau (MO): A
cooperative effort”, Proceedings of the 4th International Seismic Highway Conference, Memphis, Tenn, USA.
Crossbow Technology (2009), Imote2-High-performance Wireless Sensor Network Node, http://www.xbow.com/
Products/Product_pdf_files/Wireless_pdf/Imote2_Datasheet.pdf.
Structural health monitoring of a cable-stayed bridge using smart sensor technology 459