Journal of Intelligent Material Systems and Structures …shm.cs.illinois.edu/Full Scale Papers/13. Whelan et al... · 2013. 8. 21. · E-mail: [email protected] JOURNAL OF INTELLIGENT

http://jim.sagepub.com/Structures

Journal of Intelligent Material Systems and

http://jim.sagepub.com/content/20/7/849The online version of this article can be found at:

DOI: 10.1177/1045389X08098768

2009 20: 849 originally published online 28 November 2008Journal of Intelligent Material Systems and StructuresMatthew J. Whelan and Kerop D. Janoyan

Design of a Robust, High-rate Wireless Sensor Network for Static and Dynamic Structural Monitoring

Published by:

http://www.sagepublications.com

can be found at:Journal of Intelligent Material Systems and StructuresAdditional services and information for

http://jim.sagepub.com/cgi/alertsEmail Alerts:

http://jim.sagepub.com/subscriptionsSubscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

http://jim.sagepub.com/content/20/7/849.refs.htmlCitations:

What is This?

- Nov 28, 2008 OnlineFirst Version of Record

- May 15, 2009Version of Record >>

at UNIV OF ILLINOIS URBANA on June 27, 2013jim.sagepub.comDownloaded from

http://jim.sagepub.com/http://jim.sagepub.com/content/20/7/849http://www.sagepublications.comhttp://jim.sagepub.com/cgi/alertshttp://jim.sagepub.com/subscriptionshttp://www.sagepub.com/journalsReprints.navhttp://www.sagepub.com/journalsPermissions.navhttp://jim.sagepub.com/content/20/7/849.refs.htmlhttp://jim.sagepub.com/content/20/7/849.full.pdfhttp://jim.sagepub.com/content/early/2008/11/28/1045389X08098768.full.pdfhttp://online.sagepub.com/site/sphelp/vorhelp.xhtmlhttp://jim.sagepub.com/

Design of a Robust, High-rate Wireless Sensor Network forStatic and Dynamic Structural Monitoring

MATTHEW J. WHELAN1 AND KEROP D. JANOYAN2,*1Graduate Student, Clarkson University, Dept. of Civil and Environmental Engineering, Potsdam, NY 13699-5712

2Associate Professor, Clarkson University, Dept. of Civil and Environmental Engineering, Potsdam, NY 13699-5710

ABSTRACT: Over recent years, there has been much interest in the use of low-cost wirelesstransceivers for communication of sensor data to alleviate the expense of widely distributedcable-based sensors in structural monitoring systems. However, while the number of uniquewireless sensor platforms has continued to expand rapidly, the lack of success in replicating thenumber of deployed sensors and sampling rates utilized in previous cable-based systems hasled to disillusionment over their use for this application. This article presents a wireless sensingsystem designed for concurrent measurement of both static and dynamic structural responsethrough strain transducers, accelerometers, and temperature sensors. The network protocoldeveloped supports real-time, high-rate data acquisition from large wireless sensor arrays withessentially no data loss. The current network software enables high-rate acquisition of up to40 channels across 20 wireless units on a single peer-to-peer network with system expansionenabled through additional networks operating simultaneously on adjacent communicationchannels. Elements of the system design have been specifically tailored towards addressingcondition assessment of highway bridges through strain-based load ratings as well asvibration-based dynamic analysis. However, the flexible system architecture enables thesystem to serve essentially as an off-the-shelf solution for a wide array of wireless sensingtasks. The wireless sensing units and network performance have been validated throughlaboratory tests as well as dense large-scale field deployments on an in-service highway bridge.

Key Words: structural health monitoring, wireless sensor networks, vibration monitoring,load rating, bridge inspection.

INTRODUCTION

AS a significant portion of the aging network ofhighway bridges have met or exceeded theirintended design lifetime and service limits, highwayadministrations are faced with the challenging task ofallocating limited resources for replacement and rehabi-litation of the structures most critical for repair whilemanaging the remaining end-of-life bridges withoutjeopardizing public safety. As demonstrated in theaftermath of recent bridge collapses over the past severaldecades, current schedule-based visual inspections fallshort of ensuring a safe operational model for highwaybridge management with bridge closures precedingimminent failure. Visual inspections introduce signifi-cant subjectivity and variability as evidenced by Mooreet al. (2001) in a study conducted for the FederalHighway Administration (FHWA). In this study pri-mary members individually inspected by a group ofinspectors were assigned, on average, four to five

different ratings on the scale of 0–9. It was found thatinspectors were hesitant to assign condition ratingsoutside of the mid-range ratings, often lacked a‘formulated, systematic approach’ in assigning conditionratings, and were unlikely to detect localized defects,such as weld crack initiations. Visual inspections simplylack the ability to identify deterioration that isinaccessible or is simply invisible to the inspector.Overloading, settlement, fatigue damage, and lockedbearings can often only be visually identified in the mostextreme cases. Furthermore, the FHWA acknowledgesthat assessment ratings provided in the National BridgeInventory (NBI) by visual inspections do not provideadequate detail for managing maintenance programsand planning rehabilitations (Chase, 2005). Even basicoperational data such as traffic counts, operationalservice demands, and truck weights are unknown,thereby impeding quantitative cost-benefit analysis indetermining allocation of resources.

Structural monitoring through sensor technology tocharacterize deterioration of bridge components, in orderto evaluate safety and advise repair in advance of failure,has long been proposed (Kato and Shimada, 1986).*Author to whom correspondence should be addressed.

E-mail: [email protected]

JOURNAL OF INTELLIGENT MATERIAL SYSTEMS AND STRUCTURES, Vol. 20—May 2009 849

1045-389X/09/07 0849–15 $10.00/0 DOI: 10.1177/1045389X08098768� SAGE Publications 2009

Los Angeles, London, New Delhi and Singapore


http://jim.sagepub.com/

In general, approaches either prescribe installation ofinstrumentation on the bridge for continuous monitoringthroughout the service life of the bridge or enhanceperiodic, schedule-based inspections through the incor-poration of quantitative sensor data. Currently, even thelow-power wireless sensors are limited in terms ofduration of unattended deployment, due to the limitedcapacity of battery power supplies. As a consequence ofthe number of highway bridges in need of in-serviceassessment and the obstacles to continuous monitoring,it is likely that wireless sensors will be foremost used toaccompany inspection routines for periodic conditionassessment, with long-term monitoring reserved only forcritical structures having significant investment in termsof cost and potential for high loss of life.Transmitting data using a wireless transceiver pre-

sents several obstacles to distributed sensing, particu-larly limited and shared transmission bandwidth,coordination of decentralized hardware, and the possi-bility that the data packets will be dropped due to radiofrequency signal corruption. A review of recent wirelesssensor deployments for structural health monitoring ofbridges (Table 1) reveals that the networks havegenerally relied upon one of either two approaches: (1)low sampling rates and/or limited numbers of sensors toachieve real-time transmission, or (2) local data loggingand post-sampling transmission of sensor data. Reducedsampling rates may be acceptable for some bridgeswhere there are many low natural frequencies; howevermoderately stiff and stiff bridges, such as integralabutment and short-span bridges, necessitate highersampling rates as well as large number of sensors tocapture and spatially resolve a sufficient number ofmodes for analysis. For short-term monitoring, datalogging may be an acceptable approach to ease theburden on the transceiver bandwidth limitation; how-ever this architecture eliminates the possibility ofsampling histories beyond several minutes and generallynecessitates a much longer time period to recover thedata across the wireless link. Furthermore, the addi-tional time required for post-transmission of sampled

data mandates a significant increase in the duration oftime that the microcontroller and, in particular, theradio transceiver must be active and drawing powerfrom limited battery resources. For instance, a recentdeployment using a data logging and post-samplingapproach required 9 h to transmit a total of 20MB ofnetwork data following onboard sampling for aneffective average transmission bandwidth of �0.6 kbps(Pakzad et al., 2008). This additional active timeultimately restricted the network deployment to 13data sets before the relatively large power reserve,which consisted of four 6V lantern batteries at eachnode or 180Whr, was exhausted. In contrast, the currentstudy maintained an effective network bandwidth ofover 100 kbps while concurrently handling samplingtasks, thereby marking an improvement in severalorders of magnitude in terms of network bandwidth,effective sampling duration, and, ultimately, powerconsumption.

Other studies have suggested that onboard dataprocessing to alleviate bandwidth limitations andreduce power consumption should be pursued in favorof transmission of complete time histories (Lynch et al.2004). However, without complete time histories theanalysis is restricted to the onboard computationalanalysis, thereby eliminating the possibility of employ-ing several analysis or damage detection algorithms tothe data. This approach also prohibits the developmentof a database of sensor measurements for complemen-tary data mining, i.e., for extraction of operationalinformation related to traffic counts, stress cycles, andservice demands. In short, utilizing wireless sensornetworks as an alternative to cable-based instrumenta-tion systems should not be accompanied by excessiveconcessions in terms of performance and dataextraction. The system described in this article hasachieved the higher sampling rates required whilemaintaining reliable communication of time historieswithin a large, dense array of sensors, effectivelyreplicating previous cable-based structural health mon-itoring test programs.

Table 1. Survey of wireless bridge monitoring field deployments.

Deployment Network description Data delivery No. of sensors Sampling rate

Pakzad et al.(2008)

64 nodes log data from two channels over asampling time of 1600s. Data is streamed aftersampling resulting in a significantly more timeconsuming stage (9 hours total).

Post-sample deliveryof logged data

128 Accel. 50 Hz

Paek et al.(2006)

Five local networks of 4 nodes each with a single-boardcomputer base station connected to anIEEE802.11b wireless radio.

Real-time 20 Tri-axis Accel.(4 per network)

20 Hz

Lynch et al.(2006)

A wireless sensor network on a concrete boxgirder bridge alongside a wired system.

Real-time 14 Accel. 70 Hz

Current study A dense, multi-sensor wireless network on asingle-span concrete deck on steel girder bridge.

Real-time 40 Channels,mixed Accel/Strain

128 Hz

850 M. J. WHELAN AND K. D. JANOYAN



HARDWARE DESIGN

The Wireless Sensor Solution (WSS) developed withinthe Laboratory for Intelligent Infrastructure andTransportation Technologies (LIITT) at ClarksonUniversity was designed as a universal platform forhigh-rate, large-scale monitoring of structural response(Figure 1). The sensor network is composed of an arrayof distributed sensing nodes that interface with sensors,condition analog signals, then convert them to digitalformat and transmit the readings to a base coordinator.The base coordinator features the same wirelesstransceiver hardware as the remote nodes; however, itsfunction is to control bi-directional wireless commu-nications between the host computer and the distributedsensing units. The base coordinator is connected to thehost computer across a Universal Serial Bus (USB)connection as a virtual COM device. This physicalhardware interface is advantageous as it enables eithernetwork control from a CPU local to the measurementsite or remote access across an internet connectionthrough TCP/IP protocol using a network-enabled USBhub. The primary hardware issues addressed in design ofthe wireless sensing units were appropriate signalconditioning for the range of responses typical ofthe spectrum of highway bridge designs and span-lengths, minimized power consumption for batteryresource conservation, and high-throughput networkcommunications.

Wireless Sensor Network Platform

A wireless sensor node is comprised of a traditionalsensor, appropriate signal conditioning hardware, and atransceiver platform for onboard processing, control,and communications. The advent of low-cost radio-frequency chip transceivers has led to the developmentof a significant number of commercial wireless sensornetwork platforms with various microcontroller andtransceiver chip combinations, each with certain

advantages and disadvantages relative to the sensorapplication (Lynch and Loh, 2006). Processor speed,onboard memory, analog-to-digital converter (ADC)specifications, digital I/O port access, power consump-tion, communication range, transceiver data through-put, and host communication bus throughput must allbe considered in the selection of an optimal wirelesssensor network platform for any application.

The developed wireless sensor node (Figure 1)incorporates the Tmote Sky wireless sensor networkplatform developed by researchers at the University ofCalifornia at Berkeley and marketed by the MoteIVCorporation. This platform integrates an ultra-lowpower microcontroller and chip transceiver on a singleprinted circuit board with a USB interface to the hostcomputer for microcontroller programming and com-munication. The onboard Chipcon CC2420 2.4GHztransceiver offers an effective data rate of 250 kbps,enabling real-time packet transfer from high-samplingrate deployments. The transceiver is a spread spectrummodem, which provides substantial resilience to theinterference relative to narrow-band communicationmodems; this is vital for reliable communication in theincreasingly noisy 2.4GHz frequency band. The printedcircuit invert-F antenna on the Tmote Sky enables anapproximate communication range of 50m indoors and125m outdoors, while an external antenna can be usedto extend the range beyond 500m (Whelan et al., 2008).As previously mentioned, low power consumption isimperative for long-duration wireless deployments as thesystem life is dictated by the power supply resources.The CC2420 chip transceiver features one of the lowestcurrent consumption specifications of the IEEE802.15.4family of modems; the receiving state consumes18.8mA, the transmission state consumes 17.4mA at 0dBm output power, and three ultra-low idle modesreduce consumption to as low as51 mA. The Tmote Skyhardware is compliant with US and Canadian radiofrequency regulations and is certified by the FederalCommunications Commission (FCC) and IndustryCanada for unlicensed use in either country.

The Texas Instruments MSP430F1611 ultra-low-power microcontroller provides the computational coreof the Tmote Sky platform. This 16-bit microcontrollerhas 48 kB of flash memory for embedded code storage aswell as 10 kB of RAM. When running at 1MHz at asupply voltage of 3V, the microcontroller consumes anominal 500 mA of current; low-power modes can reducethe consumption to 51 mA. An integrated 12-bitsuccessive approximation register (SAR) ADC provideseight external channels and greater than 200 kspsmaximum conversion rate. Conversions are triggeredby a timer sourced from a clock oscillator for accurate,hardware-timed sampling rates. Two universal synchro-nous/asynchronous serial communication buses areavailable for four-wire serial peripheral interface (SPI),Figure 1. WSS node with accelerometer and strain transducer.

Wireless Sensor Network for Static and Dynamic Structural Monitoring 851



two-wire serial (I2C), or universal asynchronous recei-ver/transmitter bus protocol. Dual communicationbuses permit the microcontroller to transmit databetween the radio and the host computer at the basecoordinator without sharing communication lines, so asto maintain high throughput. Other notable peripheralsinclude a two-channel digital-to-analog converter(DAC), a three-channel direct memory access (DMA)controller for high-speed data transfers, a hardwaremultiplier for efficient computations and onboard dataprocessing, two 16-bit timers with interrupt capability,and a watchdog timer for automated system recovery inthe event the software hangs.

Sensor Interface and Signal Conditioning

The WSS hardware features a low-power signalconditioning board that improves the quality of theanalog sensor signals relative to the ADC range andsampling parameters prior to digital conversion. Theconditioning interfaces were designed to be optimizedfor measurement of vibrations resulting from bothambient and forced excitation as well as acquisition ofstrain transducer outputs during typical load ratings.However, whenever possible, integrated circuits with awide range of reprogrammable features were selected tomaintain the flexibility for additional sensing applica-tions outside of bridge monitoring. The sensor nodes aremulti-functional in that they accommodate acquisitionof up to two signal-conditioned single-ended voltagesignals, a differential analog sensor signal, and up tothree resistive or diode-based sensors, such as thermis-tors or thermodiodes for temperature measurement.

SINGLE-ENDED ANALOG SIGNALCONDITIONINGTo facilitate high-resolution acquisition of distributed

acceleration measurements for modal analysis of struc-tures, a custom signal conditioning sub-circuit providesanalog low-pass filtering, digital offset correction, anddigitally programmable gain for up to two single-endedanalog signals. A 3V voltage reference sources ultra-lownoise, stable power to the sensors, and filter operationalamplifiers. Providing a regulated supply to the sensorsreduces output noise as well as maintaining thesensitivity to enable conversion from voltage to accel-eration using a single calibration constant. The analogfilters enforce a Butterworth frequency response with a100Hz frequency bandwidth (-3 dB) and are provided toprevent aliasing of higher frequency signal components.Each filter is a fifth-order Sallen-Key circuit designfeaturing dual second-order sections on the signalconditioning board with the real-pole provided at theexternal accelerometer. Placement of the real-pole filterwith the accompanying buffer amplifier enables greaternoise immunity through low-impedance output at the

signal source and permits the connection of otherMEMS accelerometers, which generally have differentinternal resistance on the signal output path. The low-noise, low-power Linear Technologies LT6915 program-mable gain amplifier (PGA) was selected to maximizethe resolution of the conversion specific to the signalinput range or on-site vibration amplitude. In-networkcommands enable remote programming of 14 gainsettings available in binary multiples. Independentnon-volatile programmable voltage references setthrough the use of the microcontroller 12-bit DAC areused to digitally correct signal offsets, such as thegravitational offset of each accelerometer, prior toamplification. Signal offset nulling adjusts the signalinput range such that it is balanced in both the positiveand negative directions, and permits the use of highergain amplification without driving the signal out ofrange. An embedded software algorithm performs thisadjustment automatically to enable rapid configurationof the sensors with a single network command. ThePGA output is biased with a 1.25V reference, which isthe mid-span of the ADC conversion range that is setwith an external 2.5V reference. Hardware shutdown ofthe PGA, voltage reference supply, filter sections, andsensor excitation conserves limited battery resourcesduring periods of inactivity.

DIFFERENTIAL ANALOG SIGNALCONDITIONING

An independent signal conditioning interface isprovided for the acquisition of differential sensorsignals, such as Wheatstone-bridge resistive sensorslike strain transducers, load cells, pressure sensors, anddisplacement sensors. To perform a strain-based loadrating, a large array of strain transducers are requiredfor measurement of the induced strains and strainprofiles to known loads, for calculation of neutral axislocations, distribution factors, end fixity, and impactfactors to derive an overall rated load capacity of thebridge. An application-specific integrated circuit (ASIC)was incorporated into the design, which condenses thesignal conditioning and acquisition hardware for differ-ential signal and full-bridge resistive sensors into a singleintegrated circuit. The ZMD31050 AdvancedDifferential Sensor Signal Conditioner features 13stages of programmable gain of up to 420V/V, digitallyprogrammable analog offset nulling, and a 15-bitinternal ADC with adjustable input range.Additionally, an input channel is provided for tempera-ture measurement using a thermistor in a half-bridgeconfiguration. An internal calibration microcontrollerintroduces a digital conditioning algorithm for up tothird-order correction of sensor nonlinearity as well asproviding temperature compensation using the externalthermistor. Temperature compensation is a criticalcorrection for long duration strain measurements using




strain transducers, as the temperature-induced expan-sion of bridge elements often differs with temperature-induced transducer expansion, thereby resulting in astrain output in the absence of applied deck load that isan incorrect measure of temperature-induced strain.Another advantageous feature for long-term monitoringis the novel error detection hardware embedded in theASIC for automatically sensing damaged sensors orbroken cabling; two comparators monitor the inputvoltages, and the output register is set to an error code inthe event of a broken wire. The ASIC communicateswith the MSP430 microcontroller across the I2Ccommunication bus, which is used to program thesignal conditioning hardware and acquire readings fromthe ADC. A conditioned analog output is also wired tothe microcontroller ADC to enable higher rate acquisi-tions concurrently with the single-ended sensor circuits.Power conservation is addressed similarly to the single-ended signal conditioning circuitry, with digitally con-trolled hardware shutdown of the 3V voltage regulatorused to supply low-noise, regulated power to thetransducer and ASIC.

WSS SOFTWARE DESIGN

Software was designed specifically for the wirelesssensor nodes to utilize the advantageous hardwarefeatures selected and incorporated in the design. Theintegration of advanced hardware peripherals withoptimized software algorithms is directly responsiblefor the large-scale high-rate wireless sensor networkoperation achieved in field testing that has exceeded theperformance characteristics of concurrent wirelessstructural health monitoring platforms. Software forthe WSS system encompasses both embedded softwarefor the remote and central coordinator nodes, as well asPC software for bi-directional communication with thecentral coordinator nodes, real-time display, and datalogging.

Embedded Software

The embedded software applications for both theremote and coordinator nodes were written in C code,compiled, and programmed under the loose frameworkof the TinyOS-1.x open-source operating system. Whilethe TinyOS project and the accompanying assemblercode were utilized for programming the motes, thesoftware modules and interfaces were generally found tobe insufficient for meeting the requirements of large-scale, high-rate structural health monitoring and weretherefore seldom used. Extensive development of low-level software to incorporate the advanced hardwareperipherals resulted in significantly increased datathroughput, with a greater number of nodes per network

while maintaining reliable transmission. For instance,the introduction of direct hardware interrupt-drivenhandling of certain microcontroller events, in contrast tofirst-in-first-out (FIFO) handling of software events,allowed for prioritization of certain critically timedtasks. For instance, hardware interrupt handling of thecompletion of analog-to-digital conversions permittedthe alleviation of software delays resulting in memoryoverflow of the ADC buffer and the associated temporalsampling jitter. Software development also enabled theintroduction of digital signal processing of the sensormeasurements, specifically through the use of the micro-controller hardware multiplier for digital low-passfiltering.

The embedded software was designed to balanceefficiency and throughput with flexibility, so that thesame code could be utilized for an array of monitoringtasks and sensor network configurations. A set of in-network radio messages, identified by a single-byte ID,are handled by the embedded software to performconfiguration and status return tasks as well as initiatesampling routines. The software allows in-networkselection of node independent sensor channels, samplingrates, and monitoring durations. Configuration of thesignal conditioning circuitry is also handled by theembedded software in response to node independentmessages from the coordinator. The PGA gain settingfor each single-ended channel is user-selected throughthe PC host software. Upon reception of the configura-tion message for the single-ended signal conditioningcircuit, the embedded software measures the offset ofeach channel, nulls the offsets to the mid-span of theADC input range using programmable voltage refer-ences, and establishes the signal gains using a digitaloutput to a 4-bit binary counter interfaced with theparallel inputs of the PGA. Automatic offset correctionand programmable gain facilitate rapid sensor deploy-ment and alleviate the burden of physical user inter-action and knowledge of the underlying hardware.Configuration of the differential signal conditioner forthe strain transducer is handled similarly, though theuser is provided full access to the memory registers ofthe ASIC through the PC host software. Followingprogramming of the ASIC, the signal offset of theconditioned differential signal is measured and anoptional iterative scheme is applied through theembedded software, to adjust the extended analogoffset compensation register until the signal is withinachievable tolerance of the mid-span of the outputrange. Commands from the central coordinator nodealso trigger queries of configuration settings, establishsub-circuit power supply operation, and initiate low-power sleep periods. The query response is displayedon the PC host software to indicate the status of thewireless nodes, battery voltage, configuration settings,temperature readings, and received signal strength




indication (RSSI) for providing a measure of the RFtransmission power.

SAMPLING ARCHITECTUREAn over-sampling approach is implemented within

the embedded software during the acquisition ofmeasurements from the MSP430 12-bit ADC, inorder to increase the effective resolution of theconversion as well as to virtually eliminate signalattenuation in the measured bandwidth. Sensor datais over-sampled and then passed through a digital low-pass filter prior to down-sampling to the desiredeffective data rate that is transmitted across the radio.The digital filter is implemented by the embeddedsoftware with the microcontroller hardware multiplier;coefficients and decimation ratio can be remotelyprogrammed to override defaults. The use of over-sampling reduces the effect of ADC quantization noiseand is generally accepted to provide an additional bitof effective resolution for each power of four rate ofover-sampling.A default digital low-pass filter and decimation rate

stored in the microcontroller memory have beenestablished to maximize the non-attenuated bandwidthof the measured signal while producing sufficientattenuation of frequency components passed by theanalog low-pass filter (Figure 2). A 56th-order finite-impulse response (FIR) digital filter of the Equi-Rippledesign has been implemented with a specified maximumripple of 0.01 dB in the pass-band and a minimumattenuation of 52 dB in the cut-off. At a sampling rate of128Hz, the pass-band encompasses 0–50Hz and the cut-off frequency of the filter is 78Hz. This design allowsfrequency components from 64–78Hz to alias into the50–64Hz bandwidth in order to maximize the non-attenuated measurement bandwidth. The cut-off fre-quency of the analog filter was deliberately set higherthan the anticipated measurement bandwidth due to therelatively poor roll-off characteristics of the analogfilter. The filtering approach taken also affords greaterflexibility of the measurement bandwidth relative to ananalog-only filtering design as the digital filter andsampling rate can be reprogrammed to enable any

desired measurement bandwidth up to the 100Hzcut-off of the analog hardware filter.

RADIO TRANSMISSION PROTOCOLIn order for wireless sensors to serve as an effective

alternative to cable-based instrumentation for bridgemonitoring tasks, the network capabilities must at leastpermit replication of the number of sensors as well as thesampling rates utilized in typical cable-based deploy-ments. The use of a bi-directional, adaptive, andcoordinated radio transmission protocol to governnetwork communications is essential for reliable datareception in a large sensor array operating with highdata throughput. The protocol developed, displayedschematically in Figure 3, has been found to support upto 40 channels of sensor data with per-channel samplingrates of up to 128 sps in a single network. Furthermore,data completion rates in field tests have validatedthat the protocol yields virtually 100% packet delivery.This system-level performance signals that the use ofwireless sensor networks for structural health monitor-ing has emerged as a currently technologically feasibleapproach.

Scheduling of packet transmissions was found to beessential to prevent significant packet loss from colli-sions when maintaining concurrent communicationamong a high volume of deployed sensor nodes. Theprotocol implemented assigns a sequential time offsetbetween transmissions based on the local address of thenode. However, sampling initiation is not offset amongstthe nodes in the network; all nodes initiate samplingsimultaneously as triggered by a single command fromthe central coordinator node. Transmission scheduling isintegrated into the ADC sampling routine so that theprotocol is adaptive to sampling frequency. The nodesutilize the FIFO transmission buffer in the CC2420transceiver to temporarily hold the packet to betransmitted until the time window scheduled fortransmission occurs. To additionally guard against lossof data due to packet collision, the clear channelassessment (CCA) feature of the chip transceiver isutilized, whereby the packet is only transmitted if themeasured RSSI is below the specified threshold valueand no IEEE802.15.4 data is being transmitted acrossthe current channel.

Despite local transceiver scheduling, there is noguarantee that transmissions will be received by thecentral coordinator and without bit errors. However,providing nodes with an indication of successful packetreception through bi-directional communication allowsthe embedded software to determine whether to schedulethe packet for retransmission. All data packets requestan automatic acknowledgement packet from the centralcoordinator, which is returned only in the event thatthe packet is received and passes a cyclic-redundancycheck (CRC) for bit errors. The transmission schedule

0 100 200 300 400 500 600−160−140−120−100

−80−60−40−20

020

Atte

nuat

ion

(dB

)

Analog filterDigital filterComposite

Frequency (Hz)

Measurement Bandwidth 50 HzSamplingfrequency

Figure 2. Anti-aliasing filter response.




developed provides an additional time slot for retrans-mission of packets failing to receive an acknowledge-ment of host reception. In the event that theretransmission also fails to receive an acknowledgement,the complete data packet is transferred to a transmissionqueue for retransmission during available radio accesstime. These instants of availability occur during the localnode-scheduled time for retransmission of packetsfailing to receive acknowledgement on the first attempt,if the previous packet was successfully acknowledged, orat the conclusion of the data sampling. During systemvalidation, the transmission queue rarely containedmore than a few packets and any packets transmittedat the conclusion of sampling were generally from thefinal seconds of the sampling duration. A watchdogtimer is also implemented during the sampling and radiotransmission routine, to recover the system in case theembedded software become unresponsive during thisperiod of high overhead for the microcontroller.

TIME SYNCHRONIZATIONSince all nodes are independent hardware devices,

they each operate with their own clocks that will have

unique offset and drift characteristics that affect therelative timing of tasks among all nodes in the network.As system identification algorithms often require syn-chronous sampling of sensor signals and the networkradio protocol developed relies on coordination of theindividual nodes, simultaneous initiation and timesynchronization of the remote WSS nodes wasaddressed in the software design. The CC2420 transcei-ver allows packets to be addressed either to specificnodes or to be broadcast to all nodes in the network.The use of a generic message to initiate sampling acrossthe entire network with a broadcast packet enables theinitiation of sampling across all nodes at nearly the sameinstant. Deviations in sampling initiation time arise fromthe RF propagation duration and deviations in proces-sor clock frequencies when handling the samplinginitiation packet and preparing the hardware forsampling. Given that electromagnetic waves propagateat the speed of light (Griffiths, 2006), the time ofreception deviation between nodes with a 100mdifference in their distance to the central coordinatorwill be on the order of nanoseconds and thereforeintroduce virtually no phase error for the sampling

New data packet ready

Access channel and transmit if free

Local address offset reached

Listen for acknowledgement

Wait for next offset multiple

Send packet to queue



Success

Success

Channel busy Channel free / packet sent

New packet ready?

Send packet to queue

No

Yes New packet ready?

Yes

No

New packet ready?

Transmit from queue (if queue is not empty)

No

No ACK ACK

ACK

Transmit from queue (if queue is not empty) Success

ACK

No ACK

End of sampling duration

Figure 3. Radio transmission protocol flow chart.




periods used, which are on the order of milliseconds.Since the main clock is sourced by a 4MHz digitallycontrolled oscillator (DCO), phase differences in theclocks affecting the interrupt handling of the samplinginitiation packet from the chip transceiver are also onthe order of nanoseconds. To ease any time delayintroduced through handling the sampling initiationroutine, the signal conditioning and ADC/Timer registerconfiguration is handled by a separate message. Sincethis reduces the initiation of sampling to only a fewinstructions, any variability in DCO frequency acrossthe network should result in negligible offset in samplinginitiation. Consequently, the predominant source of anyinitial phase error in the sampling timers is derived fromany phase difference among the sample-and-hold sourceclock for the ADC. The 32 kHz crystal oscillator sourcecurrently utilized contributes a maximum 30 ms differ-ence in sampling initiation.Maintaining a stable, accurate timer source for the

ADC is imperative in order to produce accuratemeasurement of modal frequencies and extract modeshapes, as well as to prevent clock drift from affectingthe radio transmission protocol. Main clocks integratedin microcontrollers, such as the internal DCO of theMSP430F1611, generally exhibit poor stability in regardto supply voltage and temperature. The MSP430F1611is specified with a nominal supply voltage drift of 10%per volt with a temperature drift of þ/�0.1% perdegree Celsius. While the WSS nodes are equipped withvoltage regulators supplying the microcontroller vol-tage, which eliminates supply voltage-induced frequencydeviation, the initial accuracy of the DCO is generallyunacceptable for maintaining time synchronizationacross the network over any length of time. As analternative, a 32.768 kHz crystal oscillator is used asthe sample-and-hold source clock to trigger ADCconversions.The 32.768 kHz crystal oscillator incorporated into

the Tmote Sky design has a frequency tolerance ofþ/�20 ppm at 258C. Assuming a worst-case scenario inwhich the maximum difference in crystal frequencyamong sequential nodes in radio transmission protocolis 40 ppm, or 1.3Hz, the transmission protocol willmaintain the prescribed transmission schedule over aminimum sampling duration of 3min and 15 s for aneffective sampling rate of 128 Sps. Laboratory investiga-tion of long-term performance of the radio transmissionprotocol has confirmed that packet transmission fromselect nodes in a twenty-node network is generallyimpaired after 4 or 5min of continuous sampling whenutilizing the 32.768 kHz crystal to trigger ADC samples.Fortunately, 3min sampling intervals provide more thansufficient frequency resolution for modal analysis andare consistent with previous cable-based studies (Wenzeland Pichler, 2005). In order to enable longer durationsampling and address issues associated with additional

temperature-induced clock drift, a packaged tempera-ture-compensated real-time clock can be introduced tothe hardware at the expense of slightly increased powerconsumption. Alternatively, the number of sensor nodesper network can be reduced to increase the continuoussampling duration.

CENTRAL COORDINATOR SOFTWAREThe embedded software operating on the central

coordinator node provides the means for bridging thewireless sensor network to a host computer. Interrupt-driven routines provide a transparent interface betweenthe CC2420 transmission and reception buffers and thevirtual serial COM port operating at 262,144 baud overa USB connection. Priority is given to transfer ofreceived wireless packets to the host computer in orderto prevent overflow of the 128-byte reception buffer inthe transceiver. In the event of reception overflow, themicrocontroller transfers any complete packets remain-ing in the buffer and then flushes it to resume normaltransceiver operation.

To enable deployment of a high-density wirelesssensor network, frequency division multiple access(FDMA) has been implemented to enable multiple20-node networks to operate simultaneously on differentfrequency channels. During programming, nodes areassigned a group identification and default communica-tion frequency. In-network commands can reassign agroup to another frequency channel in the event ofinterference from other 2.4GHz devices. TheIEEE802.15.4 standard specifies 16 channels within thelegal bandwidth in the US from 2405MHz to2480MHz, thereby enabling concurrent deployment ofup to 320 sensor nodes across 16 networks. To maintaintime synchronization among multiple networks, acommon external switch supplies an interrupt signal toall central coordinators for simultaneous transmission ofthe sampling initiation command.

PC Host Software

A high-level LabVIEW host PC application has beenwritten to operate in conjunction with the remotetransceivers to control and coordinate the wirelessnetworks. User-friendly interfaces permit in-networkadjustment of sampling parameters and sub-circuitconfigurations of the sensor nodes through the extensivein-network command library, supported by the devel-oped embedded software and documented previously inthis article. The PC host software enables independentconfiguration of nodes in the network, as configurationmessages are addressed to specific local addresses.Additionally, the host software permits the configura-tions to be specified with on-screen toggles andmenu rings to reduce the in-network programming ofthese individual configurations to triggering by a single




on-screen button, rather than requiring the user tosequentially send commands to each node. This facil-itates the rapid deployment of functionally diverse nodesperforming only the tasks specific to the currentmonitoring program and sensor layout. An extensivedigital filter design sub-section of the host softwareallows the operator to design a custom filter, thecoefficients of which can be wirelessly transmitted tothe nodes for implementation in the sampling routine.The response to network status queries are reported on-screen to allow indication of the network status, state ofthe component power supplies, regulated mote supplyvoltage for indication of low battery resources, tem-perature readings, and received signal strength. Thestatus query response also includes the programmedoffset correction of the accelerometer signal for indica-tion of the sensor orientation relative to the gravita-tional field. Upon initiation of sensor sampling, the hostsoftware receives incoming data packets, displaysthe measurements in real-time on a waveform chart,and logs the readings to individual spreadsheet filesfor post-processing.

WSS PROTOTYPE

Prototype sensor nodes were assembled in-houseusing printed circuit boards commercially manufacturedfrom the board layout design (Figure 4). The printedcircuit boards and battery supply are housed in aweatherproof enclosure to ruggedize the nodes fordeployment on the exterior of structures. Sensorconnectors are watertight IP68 compliant PCB-mountconnectors with o-rings to maintain the integrity of theenclosure. An external switch with an o-ring seal isprovided to easily cycle power to the node withoutopening the enclosure. The use of the inverted-Fantenna printed on the Tmote Sky circuit boardeliminates the need to mount an external antennathrough the weatherproof enclosure. The exteriordimensions of the wireless sensor node are 12 cm�6 cm� 6 cm.

Power Consumption

Power consumption was weighted heavily in thedesign of the circuitry and selection of integrated circuitssince the service life and/or duty cycle of the wirelesssensor unit will be dependent on the capacity of thebatteries or dictated by the generally low powerdelivered by power harvesting alternatives. Digitallycontrolled shutdown of devices through internal low-power modes or hardware switching of supply voltagelines is consequently a critical consideration in compo-nent selection and circuit design for wireless sensornodes. The approximate power consumption of a WSSsensor node based on device specifications and verifiedwith laboratory measurements is itemized in Table 2.Active power draw refers to the power consumptionwhile the node is providing excitation to sensors,actively conditioning the outputs, converting theanalog signals to digital format, and transmitting thedata across the radio. During periods of idle node

Figure 4. Prototype signal conditioning board and assembled wireless sensor node.

Table 2. Approximate power consumption of wirelesssensor node (*-revision).

Activepower

Lowpower

Lowpower*

CC2420 transceiver 59.4 mW 0.07 mW 0.07mWMSP430F1611 microcontroller 2.5 mW 10mW 10 mWNon-utilized Tmote sky devices 36 mW 36mW 36 mWMain voltage regulator 0.56 mW 0.56 mW –ADC voltage reference 0.25 mW 0.25 mW –LTC6915 PGA 2.9 mW

(�2)3.3mW

(�2)3.3 mW(�2)

Analog filters 1.55 mW(�2)

– –

Voltage references(single-ended)

7.7 mW 7.6 mW 79 mW

LIS2L02AL accelerometer 2.4 mW – –ZMD31050 ASIC 7.5 mW – –Voltage regulator (Differential) 0.13 mW 0.01 mW 0.01mWBDI strain transducer (350 �) 25.7 mW – –Total: acceleration monitoring 79.4 mW 8.5 mW 132 mWTotal: strain monitoring 106.3 mW 8.5 mW 132 mWTotal: strain and acceleration 115.3 mW 8.5 mW 132 mW




activity, the hardware can be placed in a low-powersleep mode to conserve power resources. In this state,either the voltage supply to the circuitry is switched offor the integrated circuit is set to function in a low-powerstate, if available. Since the PGA, programmable offsetvoltage references, and differential signal conditioningASIC all feature non-volatile memory, the power to thesensors and signal conditioning can be cycled withouthaving to reconfigure the signal conditioning settings.Furthermore, this allows the sensors and signal con-ditioning to be powered only during active sampling, toenable optimized power conservation during installa-tion, in-network configuration, and network diagnosticperiods.In a long-term wireless sensor network deployment, a

duty-cycle approach to sampling is generally preferredto reduce power consumption for extending the life ofthe battery resources and maintaining a manageabledatabase of measurements. In order to maintain theproper regulation of the 3V sensor excitation to ensurethe integrity of the sensitivity for accurate measure-ments, the sensor nodes require a power supplyproviding at least 3.1V. In the current prototype, threeAA batteries with 2650mAh capacity are used to powerthe wireless nodes, thereby providing a supply voltage of4.2V at full charge and utilizing about 90% of thebattery resources before discharging below the opera-tional voltage required by the nodes. Consequently, the�2385mAh effective capacity results in a continuousmonitoring service life, as dictated by the batteryresources, of �68 h, or 2.8 days. Conversely, if a duty-cycle approach is implemented in which the sensor nodeonly participates in active sampling for 1min each hour,or 24min per day, the service life is extended to 32 days.Consequently, an effective long-term structural healthmonitoring deployment will balance the requirements ofthe monitoring approach with the battery resources andacceptable service life of the instrumentation.Several issues in power conservation were identified

after the development of the prototype sensor nodes.First, the circuit does not provide a means to disruptpower to the voltage references used to offset the single-ended measurements, which each require 3.63mW.Since these devices are non-volatile, supply could bedisrupted to reduce power consumption duringinactivity. Additional devices, such as the ADC refer-ence and the PGA reference, could be isolated fromthe supply during inactivity for an additionalreduction of 0.54mW. Additionally, in deploymentsusing low-voltage power supplies, such as batteries,the main voltage regulator can be removed to reduceidle consumption by an additional 0.56mW.Implementation of these revisions in the board designwill reduce idle power consumption to 132 mW andsignificantly increase the long-term duty-cycled servicelife of the nodes. For active sampling at a rate of 1min

per hour, the estimated service life would be extendedto 160 days.

PERFORMANCE VALIDATION

Thorough laboratory validation tests, as well as fielddeployments, have been undertaken to evaluate theperformance of the developed wireless bridge monitor-ing hardware platform as well as the sampling routineand radio transmission protocol. The scope of labora-tory testing included spectral verification of acceler-ometer sensors, time synchronization verification ofindependent nodes, investigation of data success rates,and multi-axis modal analysis of a laboratory-scalebridge model. A large-scale high-rate field deploymenton a single-span bridge served to further verify thesystem performance in a real-world setting.

Laboratory Testing

A Polytec scanning vibrometer was used to verify theacceleration spectra recorded by the wireless sensornodes due to sinusoidal and sine-sweep inputs from asmall shaker (Figure 5). A sequence of small amplitudetests was performed over a range of excitation frequen-cies across the measurement bandwidth. Comparisonsbetween the spectra obtained by the vibrometer and thewireless accelerometer yield excellent correlation interms of both the signal amplitude and frequency.Consequent to this comparison with state-of-the-artinstrumentation benchmark measurements, the transferfunction of the signal conditioning on the wireless nodeas well as the accuracy of the sampling clock arereasonably validated.

Field Deployment

A large-scale network consisting of 40 channels ofsensor measurements acquired through 20 remotewireless transceiver nodes was deployed on an integralabutment, single-span bridge in St. Lawrence County,NY (Figure 6). The bridge is a 17.07m (56 ft) spanreinforced concrete deck on four steel girders spaced2.74m (9 ft) center-to-center. Both quasi-static, similarto load-rating protocol, and dynamic monitoring of thebridge was conducted using a total of 29 accelerometersand 11 strain transducers. LIS2L02AL MEMSAccelerometers were selected for low-noise, low-powervibration monitoring and were potted in an externalsensor housing for direct placement on the structure.BDI Intelliducers, manufactured by Bridge DiagnosticsIncorporated, were utilized for strain monitoringas these sensors feature a 3 inch gauge length, can beredeployed, and are currently used by several trans-portation agencies to perform in-service load ratings.




Only ambient loading from vehicular traffic wasprovided for structural excitation. Each channel ofsensor data was over-sampled at 512Hz, passed throughthe default 56th-order digital low-pass filter, anddecimated to an effective sampling rate of 128Hzfor real-time transmission to the host computer. Thisnetwork configuration and sampling rate resulted in atransmission overhead in the range of 97–126 kbpsdepending on the initial packet success and retransmis-sion rates.

The average packet success rate across all of thesensor nodes over 10 186 second test cycles was 99.91%,with 92% of the nodes reporting 100% packet deliverysuccess. The minimum packet success rate over thesetests was 98.0% (Figure 7). The small loss of data hasbeen attributed to a software bug identified in theportion of the code responsible for transmitting anypackets remaining in the transmission queue aftercompletion of sampling. It is anticipated that correctionof the software section responsible for transmitting

10 Hz sine wave Sine sweep0.14

0.12

0.1

0.08

0.06

0.04

0.02

08 9 10

Frequency (Hz) Frequency (Hz)

Acc

eler

atio

n (m

/s2 )

Acc

eler

atio

n (m

/s2 )

Wireless nodeLaser vibrometer

Wireless nodeLaser vibrometer

11 12 00

0.02

0.03

0.04

0.05

0.06

0.07

0.01

10 20 30 40 50 60 70

Figure 5. Acceleration spectra comparison with laser vibrometer.

Accelerometer

Wirelesssensor node

Strain transducer

(a)

(c)

(b)

0.00m

0.00m 3.35m 6.86m 8.53m 10.21m 13.72m 17.07m

2.13m 3.96mA1/A21

A5 S5S6

S10

S9S8

S7

A7 A8 A9 A10

A15

A19

A11 A12A6

S3 S4S2

A2/A22

A13/A24 A14/A25 A16/A26

A20/A29A18/A28A17/A27

A3

S1

S11

G4

G3

G2

G1

A4/A235.79m 7.62m 9.45m 11.28m 13.11m 14.94m 17.07m

A# - AccelerometerS# - Strain

G# - GirderTransducer

View of wright road bridge Typical sensor node installation

Accelerometer and strain instrumentation plan

Figure 6. Wright road bridge field deployment.




packets remaining in the queue will result in completedata sets; however at the current level of packet successover the sampling time, system identification analysissuffered from no noticeable or adverse distortion.

This degree of transmission reliability at the high-datathroughput rate attained in this testing reveals thatwireless sensor networks are currently capable ofperforming large-scale structural health monitoringtasks with real-time transmission.

Integral abutment bridges are unique in that theprimary longitudinal members, or girders, are castintegrally with the abutments rather than supported bybearings. Consequently, bridges with this design tend toexhibit stiffer behavior than designs of similar spanlength and, as a result, experience lower amplitudevibrations. Therefore, field deployment of an instru-mentation system designed to measure the responsefrom the wide spectrum of bridge designs can advanta-geously be tested on a single-span integral abutmentbridge to provide a baseline ‘worst-case’ scenario from asignal-to-noise perspective. Throughout the deployment,local accelerations on the bridge generally producedpeaks ranging from only around 2mg to up to 10mg,though were well-captured in the sensor time historiesand frequency spectra (Figure 8).

200

180

160

140

120

100

80

60

40

20

0100 99.8 99.6 99.4 99.2 99

Data success rate

Num

ber

of ti

me

hist

orie

s

Num

ber

of ti

me

hist

orie

s

98.8 98.6 98.4 98.2 98

99.80

1

2

3

4

5

6

99.6 99.4 99.2 99

Data success rate

98.8 98.6 98.4 98.2 98

Figure 7. Histogram of packet success rates over field deploymenttesting.

Acc

eler

atio

n (m

g)

8

6

4

2

0

0 20

10−1

10−2

10−3

10−410 20 30

Frequency (Hz)

mg/

√⎯⎯Hz

40 50 60

40 60 80 100 120

Time (s)

140 160 180 200

−2

−4

−6

−8

Figure 8. Typical time history and average normalized power spectral density from 20 vertical accelerometers over a 186 s sampling duration.




A significant percentage of damage detection algo-rithms proposed for the analysis of the dynamicresponse of civil structures rely on derivation of in-service mode shapes. To produce accurate mode shapes,not only do the modal frequencies need to be wellrepresented in the frequency spectra, but the nodes mustpreserve reasonable time synchronization in order tomaintain phase relationships. Furthermore, the phaserelationship is hardest to maintain for high-frequencymodes; the period of the signal, and hence the timesynchronization tolerance, is relative to the frequency.In this study, mode shapes were derived from the 20vertical acceleration measurements using the classicalpeak peaking method employed through Fourier analy-sis as well as stochastic sub-space identification (SSI) toreveal the 4th and 5th modes which were not excited wellby traffic loading (Figure 9). These mode shapes areconsistent with those of a plate with parallel fixed ends,and correlate well in terms of frequency and shape withfinite element analysis from a model constructed fromas-built drawings. To the authors’ knowledge, deriva-tion of nine experimental mode shapes with frequenciesranging up to 47Hz is the highest order mode

development from a real-time wireless sensor network.This analysis is afforded only through sufficient timesynchronization coupled with the high nodal densityfacilitated by the radio transmission protocol.

Deployment of strain transducers alongside theaccelerometers demonstrated the versatility of thesensor nodes and capability of the system to be deployedas an alternative to cable-based systems currently usedfor load ratings. The layout of the strain transducers(Figure 6(c)) provides typical placement to measurebridge properties of interest, such as neutral axislocations, section modulus, and distribution factors.Despite that the applied loads from the passengervehicle are significantly lower than typically imposedduring a load rating, strain profiles resulting frompassenger vehicle traffic were well represented by the 15-bit ADC provided in the differential signal conditioningASIC. The development of bending strain in the girdersduring a crawl-speed pass of a large sports utility vehiclewas well captured (Figure 10). The localized tensionspike recorded at the top of the girder at the mid-spanoccurs when the vehicle wheel is directly overhead thesensor location. Strain profiles were found to be

10.80.60.40.2

00 10 20 30 40 50 60

0

10

20

30

0 10 20 30 40 500

10

20

30

1

0.5

0

−0.5−1

−1

0

1

0 10 20 30 40 500

10

20

30

−1

0

1

0 10 20 30 40 50 600

10

20

30

1

0.5

0

−0.5

−10 10 20 30 40 50 60

0

10

20

30

1

0.5

0

−0.5

−10 10 20 30 40 50 60

0

10

20

30

1

0.5

0

−0.5

−10 10 20 30 40 50 60

0

10

20

30

1

0.5

0

−0.5−1

0 10 20 30 40 50 600

10

20

30

1

0.5

0

−0.5−1

0 10 20 30 40 50 600

10

20

30

Mode 1 – 9.5 Hz

Mode 4 – 22.4 Hz

Mode 7 – 31.6 Hz Mode 8 – 33.8 Hz Mode 9 – 46.9 Hz

Mode 5 – 25.7 HzMode 6 – 30.0 Hz

Mode 2 – 11.1 Hz Mode 3 – 17.7 Hz

Figure 9. Mode shapes derived from experimental measurements.




consistent with vehicle loading patterns, compositeaction of the deck and girders was verified, and thecalculated neutral axis locations were found to correlatewell with theoretical calculations.

CONCLUSION AND DISCUSSION

A WSS for high-rate real-time sampling from large,distributed sensor arrays has been developed in theLIITT at Clarkson University. The system is multi-functional in that it provides a low-power platform forthe concurrent deployment of accelerometers, straintransducers, and temperature sensors. The hybridsensing capabilities of these nodes satisfies the immedi-ate requirements for economic, low-maintenance loadratings and short-term dynamic vibration measure-ments, in addition to providing the hardware function-ality for development of a long-term continuous bridgemonitoring system.The wireless sensor network presented in this article is

unique in that it has enabled the simultaneous acquisi-tion of up to 40 channels of sensor data across 20 sensornodes at high sampling rates in real-time without thedata loss typically associated with wireless sensornetworks. Extensive laboratory development has pro-duced a robust network transmission protocol capableof sustaining a large number of nodes with high datathroughput in real-time. Field deployments have verifiedthe ability of the system to capture natural frequenciesand accurately construct mode shapes from even arelatively stiff highway bridge, as well as record thelocalized strain profiles induced by vehicular traffic.The ability of this wireless sensor network to replicatethe performance of cable-based deployments, in termsof the number of sensors and sampling rates, and abilityto produce concise analysis results, signals a break-through in wireless structural health monitoring. Suchemerging technology appears to be presently capableof performing the bridge monitoring tasks that havebeen highly proposed and promised, though seldom

demonstrated, since the advent of low-cost wirelesssensing technology.

ACKNOWLEDGMENTS

This research has been funded by the New YorkState Energy Research and Development Authority(NYSERDA), in collaboration with the St LawrenceHighway Department, and the New York StateDepartment of Transportation (NYSDOT). The authorswould also like to acknowledge Dr Ratneshwar Jha andgraduate students Michael Gangone, Dan Nyanjom,Michael Fuchs, and Kevin Cross for their assistancein the study and field deployment.

REFERENCES

Chase, S.B. 2005. ‘‘The Role of Sensing and Measurement inAchieving FHWA’s Strategic Vision for HighwayInfrastructure,’’ In: Ansari, F. (ed.), Sensing Issues in CivilStructural Health Monitoring, pp. 23–32, Springer, Netherlands.

Griffiths, D.J. 2006. Introduction to Electrodynamics, 3rd edn, p. 576,Prentice-Hall, New Delhi, India.

Kata, M. and Shimada, S. 1986. ‘‘Vibration of PC Bridgeduring Failure Process,’’ Journal of Structural Engineering,112(7):1692–1703.

Lynch, J.P. and Loh, K.J. 2006. ‘‘A Summary Review of WirelessSensors and Sensor Networks for Structural HealthMonitoring,’’ The Shock and Vibration Digest, 38(2): 91–128.

Lynch, J.P., Sundararajan, A., Law, K.H., Kiremidjian, A.S. andCarryer, E. 2004. ‘‘Embedding Damage Detection Algorithms ina Wireless Sensing Unit for Operational Power Efficiency,’’Smart Materials and Structures, 13(4):800–810.

Lynch, J.P., Wang, Y., Loh, K.J., Yi, J.-H. and Yun, C.-B. 2006.‘‘Performance Monitoring of the Geumdang Bridge using aDense Network of High-resolution Wireless Sensors,’’ SmartMaterials and Structures, 15(6):1561–1575.

Moore, M., Phares, B., Graybeal, B., Rolander, D. and Washer, G.2001. Reliability of Visual Inspection for Highway Bridges,Volume 1: Final Report. Federal Highway Administration.FHWA-RD-01-020.

Paek, J., Jang, O.G.K.-Y, Nishimura, D., Govindan, R., Caffrey, J.,Wahbeh, M. and Masri, S. 2006. ‘‘A ProgrammableWireless Sensing System for Structural Monitoring,’’ In: 4th

7

(b) Recorded strain at abutments

6

5

4

3

2

1

0

1

0

−1

−2

−3

−4

−5

−6

−7104 106 108 110 112

Time (s)

Localized tensionspikes

Mic

rost

rain

(m

m/m

m)

Mic

rost

rain

(m

m/m

m)

114 116 118 104 106 108 110 112

Time (s)

114 116 118

Top of griderBottom of grider

(a) Recorded strain at midspan

Figure 10. Representative strain transducer response.




World Conference on Structural Control and Monitoring,San Diego, CA.

Pakzad, S.N., Fenves, G.L., Kim, S. and Culler, D.E. 2008. ‘‘Designand Implementation of Scalable Wireless Sensor Networkfor Structural Monitoring,’’ Journal of Infrastructure Systems,14:89–101.

Wenzel, H. and Pichler, D. 2005. Ambient Vibration Monitoring,John Wiley & Sons Ltd, West Sussex, England.

Whelan, M.J., Fuchs, M.P. and Janoyan, K.D. 2008. ‘‘LargeScale Remote Sensing for Environmental Monitoringof Infrastructure,’’ Journal of Environmental Monitoring,10:812–816.




Documents

Journal of Intelligent Material Systems and Structures …shm.cs.illinois.edu/Full Scale Papers/13. Whelan et al... · 2013. 8. 21. · E-mail: [email protected] JOURNAL OF INTELLIGENT