Post on 11-Oct-2020
Northumbria Research Link
Citation: Gao, Shang, Dai, Xuewu, Liu, Zheng and Tian, Guiyun (2016) High-Performance Wireless Piezoelectric Sensor Network for Distributed Structural Health Monitoring. International Journal of Distributed Sensor Networks, 2016. p. 3846804. ISSN 1550-1329
Published by: Hindawi
URL: http://dx.doi.org/10.1155/2016/3846804 <http://dx.doi.org/10.1155/2016/3846804>
This version was downloaded from Northumbria Research Link: http://nrl.northumbria.ac.uk/26632/
Northumbria University has developed Northumbria Research Link (NRL) to enable users to access the University’s research output. Copyright © and moral rights for items on NRL are retained by the individual author(s) and/or other copyright owners. Single copies of full items can be reproduced, displayed or performed, and given to third parties in any format or medium for personal research or study, educational, or not-for-profit purposes without prior permission or charge, provided the authors, title and full bibliographic details are given, as well as a hyperlink and/or URL to the original metadata page. The content must not be changed in any way. Full items must not be sold commercially in any format or medium without formal permission of the copyright holder. The full policy is available online: http://nrl.northumbria.ac.uk/pol i cies.html
This document may differ from the final, published version of the research and has been made available online in accordance with publisher policies. To read and/or cite from the published version of the research, please visit the publisher’s website (a subscription may be required.)
Research ArticleHigh-Performance Wireless Piezoelectric Sensor Network forDistributed Structural Health Monitoring
Shang Gao,1 Xuewu Dai,2 Zheng Liu,3 and Guiyun Tian3
1State Key Lab of Mechanics and Control of Mechanical Structures, Nanjing University of Aeronautical and Astronautics,Nanjing 210016, China2Faculty of Engineering and Environment, Northumbria University, Newcastle upon Tyne NE1 8ST, UK3School of Electrical, Electronic and Engineering, University of Newcastle, Newcastle upon Tyne NE1 7RU, UK
Correspondence should be addressed to Xuewu Dai; xuewu.dai@northumbria.ac.uk
Received 18 September 2015; Revised 21 January 2016; Accepted 10 February 2016
Academic Editor: Alvaro Marco
Copyright © 2016 Shang Gao et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper presents the development of a newly designed wireless piezoelectric (PZT) sensor platform for distributed large-scalestructure health monitoring, where real-time data acquisition with high sampling rate up to 12.5Msps (sample per second) anddistributed lamb-wave data processing are implemented. In the proposed wireless PZT network, a set of PZT transducers aredeployed at the surface of the structure, a lamb-wave is excited, and its propagation characteristics within the structure are inspectedto identify damages. The developed wireless node platform features a digital signal processor (DSP) of TMS320F28335 and animproved IEEE 802.15.4 wireless data transducer RF233 with up to 2Mbps data rate. Each node supports up to 8 PZT transducers,one of whichworks as the actuator generating the lamb-wave at an arbitrary frequency, while the responding vibrations at other PZTsensors are sensed simultaneously. In addition to hardware, embedded signal processing and distributed data processing algorithmare designed as the intelligent “brain” of the proposed wireless monitoring network to extract features of the PZT signals, so thatthe data transmitted over the wireless link can be reduced significantly.
1. Introduction
Structural health monitoring (SHM) is a system to monitorthe integrity of civil structures (e.g., bridges and aircraftwings) and ensure their performance and safety, which hasbecome an attractive research topic in the disciplinary fieldof mechanical, civil, and electronic engineering. One ofthe main targets of SHM is the online damage detection,which not only reduces costs by minimizing maintenanceand inspection cycles, but also prevents catastrophic failuresat earlier stage. This is particularly useful for developingself-monitoring structures, into which “smart” materials areintegrated. As a nondestructive evaluation (NDE) method,the well-known lamb-wave-based damage detection has beenwidely used in SHM [1, 2], which utilizes the features ofpiezoelectric (PZT) materials and shows great promises foronline SHM.
In lamb-wave-based approach, there has been recentinterest in the use of PZT transducers, because of their
simplicity, robustness, and potentially low cost. In sucha PZT-based SHM system, a set of PZT transducers aredeployed at the surface of the structure, and one ormore PZTtransducers work as exciters to induce lamb-waves into thestructures. Since the propagation of lamb-waves are affectedby the structure’s degradation, defects, and damage (e.g.,cracks), the characteristics of the lamb-waves propagatingfrom the exciter to these receiving PZT sensors need to beclosely monitored and carefully inspected to identify theoccurrence of defects and damages within the structure.
Traditional wired SHM systems require long deploymenttime and significant cost for cable installation.With thematu-rity of wireless communication techniques, one of the recentchallenges in the structural engineering community is theemerging wireless SHM system, which provides a promisingsolution for rapid, accurate, and low-cost structural monitor-ing [3–5]. A large number of researches have been focusedon the lamb-wave method for damage localization or impactlocalization for structural health monitoring [6, 7]. However,
Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2016, Article ID 3846804, 16 pageshttp://dx.doi.org/10.1155/2016/3846804
2 International Journal of Distributed Sensor Networks
the majority of these researches are wired SHM systemsrequiring intensive cabling work, which make it unsuitablefor distributed deployment in large structural health mon-itoring. Martens et al. [8] investigated a platform for PZTsensor based on TMS320F28335 chip with high-resolutionPWM and multichannel ADC with 4MHz sampling rate.This platform is a wired system and, without wirelessmodule,it is not suitable for forming a large PZTnetwork forwide areamonitoring. Some researches [9–13] develop some compres-sive sensing method on lamb-wave. They verified lamb-wavecan be reconstructed well by compressive sensing method.However, the compressive sensing method was not easilyembedded into wireless node so that distributed processingfor PZT network is not able to be achieved.
On the other hand, conventional design of wireless sensornode is not suitable for active sensing in SHM, wherethe lamb-waves to be sensed are high-frequency ultrasonicsignals and contain frequency components up to a fewhundred Hz. As a result, high sampling rate is requiredand a huge amount of data will be collected during theprocedure of lamb-wave interrogation. Depending on thestructure’s material and shape, the frequency of lamb-wavemay range from 10 kHz to 1MHz [1, 2, 8–10, 14–16] and thespeed of lamb-wave is usually in the range 5000m/s (S0mode). The speed of A0 mode is lowered to 3000m/s. Inorder to achieve a resolution of 0.1–2.5mm in TOF-baseddamage localization, the sampling rate will be from 2MHz to10MHz. In literatures, various sampling frequencies are useddepending on the lamb-wave frequency, wave speed, andresolution requirement of damage detection, for example,1MHz in [9, 10], 1.8MHz in [6], 4MHz in [8], and 10MHzin [7, 17]. In our proposed system, we fully made use of thebuild-in ADC with up to 12.5MHz sampling rate to provideflexibility to various applications and requirements. Thesampling frequency is adjustable by configuring the samplingclock and/or the number of PZT channels.
In addition, due to the complexity of lamb-wave prop-agation, the damage detection algorithms in SHM usu-ally are computation intensive and require considerabledata processing capabilities. However, the existing wirelesscommunication protocol (i.e., IEEE 802.15.4) and wirelesshardware motes (e.g., Mica2, MicaZ, and TelosB) in wirelesssensor networks are designed for low data rate and lowcomputation applications, which make it is impossible totransmit all the lamb-wave data to a central server thatcarries out centralized data processing and structure damageidentification. To achieve a practical wireless PZT sensor andactuator network for structure health monitoring, the wire-less senor nodesmust be armedwith some kind of distributeddata processing and compressive sensing or downsamplingalgorithm capabilities, such that the amount of data to betransmitted over the wireless link can be reduced.
In literatures, some PZT sensor and actuator nodes havebeen proposed for structural health monitoring. A wirelesssensor node is proposed in [14, 17], where field programmablegate array (FPGA) was used for PZT active diagnosis. How-ever, these wireless sensor nodes have very limited on-boarddata processing capability for acquired signal.The data trans-mission is the real-time. Furthermore, the authors in [15, 16]
design the wireless PZT sensor and actuator node basedon TMS320C2811, TMS320C6713, and TMS320F2812, respec-tively. However, there are no distributed data processing anddistributed wireless network based on nodes investigated inthese researches. Dong et al. [18] discussed a Martlet nodewith TMS320F28069 chip which can support 3MHz sam-pling rate for MEMS accelerometer. Conventional wirelesssensor node without high-performance core chip is not ableto execute the complicated data processing algorithm.
Furthermore, lamb-wave based damage localization isbased on the time-space principle that propagation distance(in the space domain) is proportional to the propagationdelays (in the time domain) of lamb-waves. To achieve ahigh accuracy of localization and high resolution of damageimage, a precise time synchronization is critical to ensure thatthe data acquisition and propagation delay calculation areprecisely synchronized.
To address the challenges of big data in lamb-wave inter-rogation, the bottleneck of limited wireless communicationbandwidth, and precise time synchronization, a wireless PZTsensor and actuator platform is proposed and developed.The newly designed wireless node features a TMS320F28335digital signal processor (DSP) and an improved IEEE 802.15.4wireless data transducer with up to 2Mbps data rate. Eachnode supports up to 8 PZT transducers, one of which worksas the actuator that generates lamb-wave actuation signalat an arbitrary frequency to drive the external PZT sensorwhile the responding vibrations at other PZT sensors aresensed simultaneously. In addition to hardware, embeddedsignal processing and distributed data processing algorithmare also designed as the intelligent “brain” of the proposedwireless monitoring. As a result, the amount of data to betransmitted over the wireless link is reduced significantly.These features enable the developed PZT sensor-actuatornode to be deployed easily and suitable for wide area SHM.
This paper is organized as follows. Section 2 presentsthe network architecture and function diagram of the pro-posed wireless PZT sensor and actuator network in SHM,followed by the sensor and actuator hardware developmentin Section 3. The software development is presented inSection 4. Some results are discussed in Section 5 for thepurpose of demonstration. And Section 6 is the summary ofconclusions and future work.
2. Network Architecture
Similar to themost PZT-based SHMsystems, the architectureand function diagramof thewireless PZT sensor and actuatornetwork is as illustrated in Figure 1.
The proposed WSN consists of the following types ofcomponents.
(1) PZT Transducer. The PZT transducer either convertsmechanical to electric signals, or vice versa. The PZT trans-ducers have two work modes: they can work as either aPZT actuator to excite an elastic lamb-wave according tothe electrical signal applied on the PZT crystal, or a PZTsensor to transform the responding elastic lamb-waves intoan electrical signal.
International Journal of Distributed Sensor Networks 3
Automatedanalysis
Data processing
Manage network
Time synchronization
Damage detection
Base station
Structure
Get and set RFconfiguration
Start DAC
Wave actuation
Start ADC
Start DMA
Data sensing
Data processing
Data transmission
Tim
e syn
chro
niza
tion
PZTtransducer
Sensornode
Sensornode
Sensornode
Sensornode
Figure 1: The architecture and function diagram of the wireless PZT sensor and actuator network.
(2) Wireless PZT Node. Each node is able to connect totwo PZT transducers at least. The node generates excitationsignal for driving PZT transducer and acquires data fromPZT transducers. PZT sensor and actuator node have embed-ded distributed lamb-wave data processing for informationextraction and downsampling algorithm for reducing dataamount of wireless transmission. All wireless PZT nodesform a network and send lamb-wave signal to the base stationaccording to TDMA or CSMA protocol. And wireless PZTnodes in one structure plate should comply with time syn-chronization for ensuring the damage localization accuracy.
(3) Base Station. This is the sink node which can supportbig data transmission access, multipoint operations, andtime synchronization mechanism for all PZT sensor andactuator nodes. Moreover, this node can implement signalreconstruction for lamb-wave and damage localization.
Groups of PZT sensor nodes are arranged into severalstructure plates and densely deployed around specific areas ofinterest and report PZT lamb parameters to the monitoringand human-computer interaction (HCI) system at a remotestation. Depending on the application requirements and thesize of the area to be monitored, the topology of the networkcan be a single cluster or multiclusters. Although these PZTtransducers are connected to the nearby wireless PZT nodesin their regions via a set of short wires, the featured differenceof the proposed system is that the longer distance lamb-wave analog signal cables in wired SHM are replaced by thedigital wireless data links between PZT nodes and the remote
monitoring system. Therefore, the massive signal cables andcostly cable installation are avoided, which is the key benefitof the proposed wireless PZT sensor and actuator network.
While setting up the wireless communication network,time synchronization is also carried out to ensure that allwireless PZT nodes are well synchronized and a commonsense of time has been achieved. Once the network is set upand wireless PZT nodes are synchronized, the base stationfirst initializes the lamb-wave interrogation by specifying oneor more PZT transducers as the actuators (with appropriateconfigurations, such as lamb-wave central frequency andamplitudes, according to the requirements from the remotemonitoring and HCI center). The base station starts thelamb-wave interrogation procedure by broadcasting a startcommand to include the start time to all wireless PZT nodes.After sending start command, the base station waits forreceiving data from all wireless PZT nodes. Upon receivingthe SC from base station, all wireless PZT nodes switch fromlistening state to a preworking state. Once RF configurationis finished, all wireless PZT nodes start digital-to-analogconverter (DAC) function for generating PWM wave exci-tation. Then, all wireless PZT nodes input PWM excitationsignal to PZT transducers and then configure function ofanalog-to-digital converter (ADC) and direct memory access(DMA). Once sensing event is detected, wireless PZT nodesstart to execute data acquisition of lamb-wave signal. TheDMA controller controls lamb-wave data to be sent fromADC register to data queue. The DSP controller in wirelessPZT node acquires PZT transducer’s data from data queue
4 International Journal of Distributed Sensor Networks
Radio board
Base board
Sensor array
Microcontroller (ATSAMR21G18)(i) RF233 wireless transceiver(ii) IEEE 802.15.4 system
Distributed data processing (i) Wavelet transform
(ii) Time delay calculation (iii) Compressive sensing
DMA controller
RF module
DSP moduleinterface
Antenna
Time synchronization
SRAM controller
ADC 12-bit up to
RF moduleinterface
Time synchronization
DACmodule
Amplification
Conditioning board
Actuation
Switc
h
16
16
Pow
er(+5
V)
6.25Msps
Figure 2: The hardware schematic representation of sensing wireless PZT node.
and executes data processing. Finally, all wireless PZT nodestransfer processed data to the base station in the light ofTDMA or CSMA protocol.
3. Hardware Architecture
3.1. Hardware Schematic. The block diagram of the proposedwireless PZT node is shown in Figure 2. The wireless PZTnode consists of three components, namely, conditioningboard, DSP base board, and radio board.
The conditioning board is an analog signal processingboard that has two tasks: (1) lamb-wave execution: this isto amplify the 3.3 V lamb-waveforms generated by the DSPto an appropriate level, so that a required lamb-wave can beinduced at the PZT transducer (actuator mode); (2) lamb-wave detection: this is to amplify and filter the weak and noisylamb-wave signal detected by the PZT transducer (sensormode) to 3.3 V level for analog-to-digital conversion at theDSP. The excitation designed is on the basis of our previouswork at our lab [2] and the excitation circuit for wirelessactive sensing node is presented in [16]. The lamb-wavedetection circuit consists of a group of AD8608 operationalamplifiers (Op AMPs) configured in charge amplificationmode. Unlike the voltage preamplifiers that suffer fromdistance/attenuation effects, charge amplifier is selected tomaintain the signal sensitivity regardless of distance from thepassive sensor to the preamplifier. Therefore, it is suitablefor the long lengths of sensor input cable. The AD8608 ischosen because of its low bias current (1 pA maximum),
low noise (12 nV/maximum), and low offset voltage (65 𝜇Vmaximum). The output of the U2A OPAMP is 0.1 V to 3.2 Vwhichmatches the input range of the ADC (0V to 3.2 V) with100mV headroom to maintain linearity (Figure 3).
The switching circuit is presented in [1] at our lab. Inactive sensing, the PZT transducer may switch between twooperation modes: excitation mode and receiving mode. Aswitching circuit is needed to connect the excitation circuitsand detection circuits to different PZT transducers. Theswitching module consists of two sets of low crosstalk singlepole double throw (SPDT) relays that are controlled DSP. Ata given time, only one PZT is connected to the excitationamplifier as the lamb-wave generator; other PZT transducersare connected to charge amplifiers as the signal detector. Theconnections change in round-robin scheduling.
In base board, one distinct feature of the microcon-troller TMS320F28335 is the capability of high-speed dataacquisition. The direct memory access (DMA) module onTMS320F28335 allows thewireless PZTnode to collect sensordata at 10MHz sampling rate up to by 12-bit ADC with6.25Msps.The core of the base board is distributed data pro-cessing unit which is able to execute wavelet or Hilbert trans-form, time delay calculation for lamb-wave, and downsam-pling algorithms. The radio board contains ATSAMR21G18microcontroller, DSP module interface, and RF module.
3.2. Base Board. A typical smart wireless sensor node haslimited memory and limited computational ability and bat-tery power. These conditions should be taken into account
International Journal of Distributed Sensor Networks 5
270
POS
DNP
V01
V02U2A
U1A
AD8608
AD8608
NEG
GND
GND
2
34
5
12
12
4
52
3
R2
R3
R2
R5
R4
1K
1K
1nF
+1.25
10K
3.3V
3.3V
C2
−
+
−
+
Figure 3: The schematic diagram of lamb-wave detection circuit.
TMS320F28335 chip
P8 I/O interface
Crystal oscillator
P9 analogInterface connector
SPISIMOSPISOMISPICLKSPISTA
ADCINA0ADCINB0
GND0 ADCLOGND3GND1
7.5
cm
13.5 cm
(a)
Crystal oscillator
PB03PB02PB22PB23
GND0
ATSAMR21G18Antenna
6 cm
6cm
+5V
(b)
Figure 4: Prototype of the proposed wireless PZT actuator/sensor node, which consisted of two parts: (a) DSP base board; (b) IEEE 802.15.4radio board.
when the downsampling algorithm is embedded into wirelesssensor board. The base board should be chosen according tothe following aspects: (1) it should have SPI or I2C interfacefor communication with RF board; (2) it should support highdata transmission rate for large amount of lamb-wave or othercomplex signals; (3) it should support high enough crystalfrequency for working; (4) it should support certain memorystorage for data cache. For low-cost low-power measurementdevices, the programmable DSP could be an efficient digital,programmable, and real-time platform.
The base board, as detailed in Figure 4(a), is a four-layer PCB roughly with the size of 13.5 cm by 7.5 cm. ATexas Instrument microcontroller TMS320F28335 (real-timedigital signal processor with controller features of Delfino-family) in base board, running up to 150MHz clock fre-quency, is adopted in the wireless sensing node to execute
high sampling data acquisition and on-board data processing.Additional important features of this DSP are on-boardDMA controller, 16-bit enhanced pulse-width modulatorsPWM (HR-EPWM), 12-bit multichannel sampling/hold cir-cuit, internal flash memory of the program, floating pointhardware arithmetic, and so forth. The PWM works atclock frequency of 100MHz and can be used as precisedigital-to-analog converters (DAC) or trigger the analog-to-digital conversion (ADC). The DSP chip also includes 12-bit resolution multiplexed ADC with a conversion rate ofup to 10MHz or more and has 8 simultaneous sample-and-holds. So, this chip has high-performance digital-to-analogand analog-to-digital conversion without extra components.In this board, the P8 I/O interface and P9 analog interfaceconnector are used for communicating with RF board andacquiring sensor data, respectively.TheADCL0 pin should be
6 International Journal of Distributed Sensor Networks
User inputs
Send commandsWait for command
Send commandsand wait for data
Receive feature datafrom surrounding
sensor nodes
RF initialization
Wait for command
Measure and store data locally
Board initialization
Local data processingand extract signal feature
Send feature data
Data request?
Transmit entire measurements
Data request?
Request data andtransmit entire
measurements to PC
Require data?
Receive entiremeasurements
PZT node Base station
Receive evaluationfrom surrounding
sensor nodes
Computer
Evaluate damage/defect existence and send
report to PCRequire damage?
YesYes
Yes
Yes
Figure 5: Functional blocks of the wireless PZT network.
connected to GND pin to make sure that the ADCmodule inDSP adopts internal reference voltage.
3.3. Radio Board. The radio board, as detailed in Figure 4(b),is employed in this research. The Atmel SMART SAM R21board with low-power microcontroller ATSAMR21G18 ischosen. It adopts a 32-bit ARM Cortex-M0+ Processor andan integrated ultra-low-power RF233 as 2.4GHz ISM bandtransceiver with a maximum data rate of 2Mbps. This deviceis available in 48-pin packages with up to 256 kb Flash, 32 kbof SRAM, and is operating at a maximum frequency of48MHz.
Sensor data is sent from base board using SPI bus (PB03,PB22, PB02, and PB23) to the radio board, respectively. Theradio board is a two-layer PCB and the size of the board is6 cm× 6 cm roughly, as shown in Figure 4(b).The two verticalheader pins are shown on both the right side and bottomside. The 4-pin horizontal headers interface with verticalheaders (SPICLK, SPISIMO, SPISOMI, and SPISTA) on thebase board to enable external data to access from base board.In transmitting state, the radio board consumes 35.5mW ofpower (26mA of power at 1.8 V) but less than 2 𝜇Wof power(1.1 𝜇W of power at 1.8 V) in power down mode.
Another task of the radiomodule is time synchronization.It is well known that time synchronization plays a keyrole in all Time-of-Flight- (TOF-) based structural healthmonitoring systems. The accuracy of arrival time calculationand the resolution of damage localization are degraded by theerrors in time synchronization. Recognizing the importanceof time synchronization, we take dual-processor architecturein our hardware system, in which a second processor (i.e.,the ATSANR21G18 at radio module) is separated from thecomputation-intensive data processing and reserved for onlywireless communication and time synchronization. There-fore the time accuracy of interrupt-based time synchroniza-tion and time-stamping is not compromised by other tasks(e.g., ADC interrupts, delays for intensive data processing).
At a lower computation burden, the interrupt response isfaster and lower jitters and delays can be guaranteed. Thus abetter time synchronization accuracy can be achieved. In thispaper, a simple time synchronization scheme is presented,which is suitable for a single-hop network only. In his scheme(as illustrated in Figure 5), the PZT sensor nodes start withan idle state (no sampling nor data processing) after initial-ization and keep waiting for the “START” command fromthe base station. Once it receives the “START” command,all the nodes start sampling immediately. However, it is notfit for the multihop network, where we need some advancedtechniques (such as packet-exchange time synchronization inour previous work) to keep the node’s clock synchronizedall the time. The proposed dual-processor architecture hasthe capability for precise time synchronization by packet-exchange (e.g., the IEEE 1588 PTP-like time synchroniza-tion) and our future work is to implement our previoustime synchronization work [19, 20] into this hardwareplatform.
4. Software Architecture of System
4.1. Operation Overview of PZT Sensor Network Operation.Figure 5 illustrates a procedure within a three-stage wirelesssensor network for autonomous damage or defect detection.The entire wireless sensor network may consist of PZT nodeand base station connected to PC. Several PZT nodes and abase station form a group to cover a geometrical area. Thediagnosis command is initiated from the PC through basestation. After informing all wireless PZT nodes, the base sta-tion broadcasts an initial command to wireless PZT node andtriggers the signal excitation of sensor nodes. Each wirelessPZT node immediately starts data acquisition after receivingthe initial command and records the measurements. Themeasurements are subtracted from the prestored sensor datafor healthy structure to obtain the scattered wave signal formthe damage.
International Journal of Distributed Sensor Networks 7
MAC header Payload CRC
PANID Destination address Source addressSequence numberFrame header
Network header
Source address Destination address TimestampSequence numberFrame control(1 byte) (1 byte)
(1 byte)
(2 bytes)(2 bytes)(2 bytes)
(2 bytes) (2 bytes) (2 bytes)(2 bytes)
Figure 6: The format of wireless packet.
Conventional wireless sensor nodes normally utilize cen-tralized architecture. In this architecture, analog-to-digitalconverters (ADC) and memory chips are directly connectedto I/O ports of microcontroller. The microcontroller mustaccess the peripheral chips sequentially and clock multipleclock cycles that are required to complete an operationwhich involves several peripheral chips. Most peripheralchips should be active while waiting for trigger signal frommicrocontroller. For such architecture using conventionaldesign, high sampling rate is almost impossible.
4.2. Wireless Command and Data Communication. The net-work header, MAC header, and application payload areencapsulated inside the standard IEEE 802.15.4 data framepayload, as shown in Figure 6.
The general format is composed of a IEEE 802.15.4MAC header, network header, application payload, optionalmessage integrity code (MIC), and a check sum (CRC).Whenthe RF board gets the acquired data from base board by SPIinterface, the acquired data will be filled into the payloadformat. Once the payload has been filled to the full, wirelesspacket will be sent to the base station.Themaximum payloadis 112 bytes. The 2-byte timestamp format in network headerstores the timestamp of wireless PZT nodes or base station,as shown in Figure 6.
4.3. Data Acquisition and Processing. In Figure 7(a), theflowchart demonstrates typical data sampling cycle of con-ventional design, such as prototype of wireless nodes, TelosBmotes, Imote2, and Mica motes. These conventional moteshave an internal 12-bit ADC in microcontroller. The micro-controller sends clock signal and control signals to the ADCto trigger sampling cycle conversion.The internal buffer is setup to get data from the ADC. The reading takes 12 periodicoperations and an ADC clock signal is generated in eachperiod for filling the buffer bit by bit. After the data has beenread out entirely, the microcontroller provides a clock signaland control signals to the on-board flash. Before writing thedata into the flash bit by bit, the microcontroller takes someinstructions to send address to the flash. The flash savingoperation also takes 12 periodic operations, each of whichinvolves several instructions and a few clock cycles. Then,the sampling cycle ends and another sampling cycle may
start again. This architecture clearly reveals the inefficiencyof wireless sensor node designed with an ordinary microcon-troller.
Figure 7(b) presents an improved sampling design usingTMS320F28335 chip. The chip internally has some con-trollers, including first-in-first-out (FIFO) DMA controllerfor sampling data input and output, SRAM controller, and aclock generator.TheDMA algorithm has been widely used toallow hardware to access memory independently. To enablea wireless sensor node for high-speed applications, somealgorithms such as semi-DMAapproach are applied to extendthe traditional architecture of wireless sensor node. In ourdesign, with the DMA controller, the main microcontrollerTMS320F28335 chip can be released from the task of datatransfer. The sampling data transfer and data acquisition canbe more efficient when DMA is adopted. The 12-bit samplingdata is acquired by the ADC and saved into the internal DMAbuffer in FIFO input mode by DMA controller. Meantime,the DMA controller and address controller control the datato access the SRAM in FIFO output mode. The samplingdata is moved from the internal DMA buffer to SRAM at16-bit length which is the I/O width of SRAM. Each clockcycle have 8 continuous sampling operations and one writingoperation for writing data for SRAM. As a result, samplingdata are encapsulated according to bit alignment in internalDMA buffer and the DMA, ADC, and the SRAM can be fullyutilized without compromise.
The process diagram illustrates the main stages of dataprocessing for lamb-wave signal, as shown in Figure 8. Theprocess begins with the data record of lamb-wave signal afterexcitation. The first two stages are operated in wireless PZTnode. At first stage, data about the response signal with S0,A0, and other modes are collected. Then, the acquired rawdata is sent to base station if nondownsampling method isadopted or is downsampled first and then is sent to basestation. This stage including downsampling method aimsat reducing amounts of lamb-wave signal’s transmission forsaving energy consumption and raising the efficiency of datatransmission. The third stage in the process is operated inbase station. At base station, the envelope is reconstructedfrom the raw data or downsampled data, and then TOFestimation is calculated by propagation delay. Finally, TheTOF estimation is used for damage localization.
8 International Journal of Distributed Sensor Networks
ADC clock
Sampling
Read ADC
Flash clock
Increase flash addr
Write flash
End
Finished?
Internalbuffer
Finished?
Flash
12-bitADC
Yes
No
Yes
No
Write SPI to RF
1bit
1bit
(a)
FIFO inputbuffer
DMAcontroller
FIFO outputbuffer
SRAM controller
12-bit ADC
InternalDMA buffer
Clock generator
SRAM
addr
ADC clock
SRAM clock
DMA clock
Write SPI to RF
Finished?
Yes
End
No
12bits
16bits
16bits
(b)
Figure 7:Data sampling and collection process comparison between (a) the conventional wireless sensor nodewith low sampling rate (usuallyup to a few kHz) and (b) the proposed wireless PZT sensor/actuator node with burst sampling rate up to 12.5MHz.
Apply excitation and record plate
lamb-waves
1 2
Raw data transmissionto base station
Reconstructenvelope signal
and calculate TOF
3 4
Faultlocalization
Aluminum plate
Lamb-wave
Wireless PZT node
(a)
Apply excitation and record plate
Lamb-waves
1 2
Reconstruct envelope signal
and calculate TOF
3
Fault localization
Aluminum plate
Lamb-wave
Wireless PZT node
Downsampling raw data and transfer data to
base station
4
(b)
Figure 8: Data processing for lamb-wave signal with (a) nondownsampling algorithm (b) and downsampling algorithm.
International Journal of Distributed Sensor Networks 9
Signalgenerator
Poweramplifier
Chargeamplifier PC
Basestation
Wireless piezoelectricnodeStructure
Piezoelectricsensor
Figure 9: Lab test platform for the wireless PZT sensor network.
5. Proof-of-Concept Tests andPerformance Evaluation
In order to evaluate the performance of the developedwireless PZT sensor nodes, a series of proof-of-concept labtests have been carried out and the results are presented inthis section. Two key performance indices of the proposedwireless PZT sensor and actuator network are (1) distributedsignal processing for envelope detection and propagationdelay estimation; (2) downsampling algorithm for reducingthe amount of data to match the wireless communicationbandwidth, while retaining the structure health informationas much as possible.
The lab test platform is shown in Figure 9. As a proof-of-concept lab test, the platform consists of a computer workingas a graphic user interface to the end-user, a wireless basestation that connects to the PC with the wireless network,and the designed wireless sensor node (highlighted by ared square). As our focus is to verify the distributed signalprocessing and wireless transmission, the conditioning boardis replaced by a power amplifier and a charge amplifier inthis lab test. Two PZT sensors are attached to the surface ofan aluminum plate bar. Left one is the actuator connectedto a power amplifier that amplify the five-cycle sinusoidaltone burst signal to generate a lamb-wave. The PZT at rightend of the aluminum bar is the PZT sensor to receive thepropagated lamb-waves. The lamb-wave signal detected bythe PZT sensor is first conditioned by a charge amplifier tothe range of 0–3.3 volts.Then the amplified signal connects totheDSP board andRF board for digital-to-analog conversion,local signal processing, and wireless transmission. Once thebase station receives the transmitted data, it simply forwardsthe data to the PC via a USB cable.
In our lab tests, the waveforms generated from a lamb-wave propagation simulator are used for performance eval-uation [21]. These waveforms are injected into the signal
generator to mimic the PZT transducer. The output signalsfrom the signal generator are connected to theDSPbaseboardthat processes the signal locally and sends the extractedfeature to base station through the wireless link. Two differ-ent envelopes are detected by using Hilbert transform andwavelet transform, as shown in Figures 10 and 11, respectively.
The signals used in this proof-of-concept test are for a pairof PZT actuator and sensor on an aluminum plate of 2mmthickness, where the PZT sensors are separated from the PZTactuator at 100mm distance. Figures 10(a) and 11(a) depictthe hamming-windowed 5-peak tone burst excitation signal(blue line) of carrier frequency 100 kHz and its hammingwindow (red line). The excitation signal has 5 cycles ofthe sinusoidal signal and last for 25 𝜇s. For the purpose ofevaluation, the excitation frequency was tuned to 100 kHz fora good separation of the fundamental symmetric mode S0wave and the fundamental antisymmetric mode A0 wave.
At the PZT sensor node, the reception lamb-wave after100mm propagation was sampled for 400 𝜇s by the build-in ADC of 28335DSP at a sampling rate of 12.5Msps withthe resolution of 12 bits. The sampled data is of 10 k bytes(5 k words) and stored in the DSP RAM. Figures 10(b) and11(b) show the sampled data (the blue line labeled with rawsignal) and the clear separation of the S0 and A0 mode.The dispersion of the A0 mode for such a wave at 100 kHzfrequency after 100mm propagation can also be clearly seen.
5.1. Basic TOF Extraction by Hilbert Transform. The funda-mental principle of lamb-wave structure healthmonitoring isthe fault localization by time-of-flight of the received waves.With the given wave propagation speed, once the TOF 𝜏 isfound; the location of the reflector damage can be determinedaccording to
𝑑 = V𝑔 × 𝜏, (1)
10 International Journal of Distributed Sensor Networks
0 100 200 300 400
0
0.5
1
Exci
tatio
n am
plitu
de (V
)
Hilbert transform envelope
−1
−0.5
Excitation signal x[n]
Time (𝜇s)
(a) Excitation signal (DAC output)
0 100 200 300 400
0
0.5
1
1.5
Rece
ived
sign
al am
plitu
de (V
)
Hilbert transform envelope
Propagation
−1
−0.5
Received signal y[n]
S0A0
delay 𝜏
Time (𝜇s)
(b) Receiving signal at the PZT sensor
0 50 100 150 200 250 300 350 400
0
20
40
60
80
Hilbert transform envelope
Cros
s-co
rrel
atio
n co
ef
𝜏
−20
−40
−60
−80
Cross-correlation c[n]
Peak time Peak time
Time (𝜇s)
at 122.8 𝜇s at 246𝜇s
(c) Cross-correlation coefficients of excitation signal 𝑥[𝑛] and receivedsignal 𝑦[𝑛]
Figure 10: Lamb-wave detection. (a) Excitation signal 𝑥[𝑛] at the PZT excitation; (b) received signal 𝑦[𝑛] and its envelope at the PZT sensor.(c) The cross-correlation function of (𝑥[𝑛], 𝑦[𝑛]) and its envelope. The envelope is detected by using the Hilbert transform.
where V𝑔 is the group velocity of thewave and𝑑 is the distanceof the reflector from the sensor. In the TOF estimation, themain concerns are the envelope of the lamb-waves and theenvelope of the cross-correlation function, rather than theraw 100 kHz waves. The excitation signal 𝑥[𝑛] (Figures 10(a)and 11(a), blue line) works as the baseline signal to calculatethe cross-correlation function with respect to the receivedsignal 𝑦[𝑛] (Figures 10(b) and 11(b), blue line):
𝑐 [𝑛] = 𝐸{∑
𝑖
𝑥 [𝑖] 𝑦 [𝑖 − 𝑛]} . (2)
The cross-correlation coefficients are further processedby the Hilbert transform to derive its envelope for TOFestimation. The calculated cross-correlation function 𝑐[𝑛]and its envelope are shown in Figures 10(c) and 11(c).
5.2. TOF Extraction with Shannon Wavelet Transform.Wavelet transformation method is able to analyze low andhigh frequencies at the same time, even respecting theuncertainty principle [22].Therefore, it is chosen as amethod
to analyze lamb-wave. The Continuous Wavelet Transform(CWT) is a linear transform that correlates the harmonicwaveform 𝑢(𝑥, 𝑡) with basic functions that are simply dilata-tions and translations of a mother wavelet 𝑤(𝑡), by thecontinuous convolution of the signal and the scaled or shiftedwavelet:
WT (𝑥, 𝑎, 𝑏) = 1
√𝑎∫
+∞
−∞
𝑢 (𝑥, 𝑡) 𝜓∗(𝑡 − 𝑏
𝑎)𝑑𝑡, (3)
where 𝜓∗(𝑡) presents the complex conjugate of the motherwavelet 𝜓(𝑡), 𝑎 is the dilatation or scale parameter definingthe support width of the wavelet, and 𝑏 is the translationparameter localizing the wavelet in the time domain. Thekernel function of the Continuous Wavelet Transform is
𝜓𝑎,𝑏 (𝑡) =1
√𝑎𝜓(𝑡 − 𝑏
𝑎) (4)
International Journal of Distributed Sensor Networks 11
0 100 200 300 400
1
2
3Ex
cita
tion
ampl
itude
(V)
Envelope of Shannon waveletReal value of Shannon wavelet
Excitation signal x[n]
0
−1
−2
−3
Time (𝜇s)
(a) Excitation signal (DAC output)
0 100 200 300 400
3
5
Rece
ived
sign
al am
plitu
de (V
)
Envelope of Shannon waveletReal value of Shannon wavelet
S0 A0Propagation
delay 𝜏
Received signal y[n]
−3
−5
Time (𝜇s)
(b) Receiving signal at the PZT sensor
0 100 200 300 400
Cros
s-co
rrel
atio
n co
ef
0
200
400
−200
−400
𝜏 Peak time Peak time
Envelope of Shannon waveletReal value of Shannon wavelet
Cross-correlation c[n]
Time (𝜇s)
at 122.8 𝜇s at 247.3 𝜇s
(c) Cross-correlation coefficients of excitation signal 𝑥[𝑛] and receivedsignal 𝑦[𝑛]
Figure 11: Lamb-wave detection. (a) Excitation signal 𝑥[𝑛] at the PZT excitation; (b) received signal 𝑦[𝑛] and its envelope at the PZT sensor;(c) the cross-correlation function of (𝑥[𝑛], 𝑦[𝑛]) and its envelope. The envelope is detected by using the wavelet transform.
which is generated by shifting and scaling a mother wavelet𝜓(𝑡). Using the definition of the Fourier Transform, assuming(𝑡 − 𝑏)/𝑎 = 𝜏, we have
∧
𝜓𝑎,𝑏 (𝜔) = ∫
+∞
−∞
𝜓𝑎,𝑏 (𝑡) 𝑒−𝑗𝜔𝑡𝑑𝑡
= √𝑎𝑒−𝑗𝜔𝑏
∫
+∞
−∞
𝜓 (𝜏) 𝑒−𝑗𝜔𝑎𝜏
𝑑𝜏
= √𝑎
∧
𝜓 (𝑎𝜔) 𝑒−𝑗𝜔𝑏.
(5)
In this study, the Shannon wavelet is employed as motherwavelet for separate amplitude and phase. The Shannonwavelet is expressed by
𝜓 (𝑡) = √𝑓𝑏 sin 𝑐 (𝑓𝑏𝑡) 𝑒𝑗𝜔𝑐𝑡 (6)
and its Fourier transform is
∧
𝜓 (𝜔) =
{{
{{
{
√2𝜋𝑎
𝜔𝑏
𝑒−𝑖𝜔𝑏,𝜔𝑐
𝑎−𝜔𝑏
2𝑎< 𝐹𝑏 ≤
𝜔𝑐
𝑎+𝜔𝑏
2𝑎
0, others,(7)
where𝐹𝑏 is the shape control parameter (wavelet bandwidth).The function 𝜓𝑎,𝑏(𝑡) using Shannon as mother is then cen-tered at 𝜔𝑎/𝑎 and frequency band is [𝜔𝑐/𝑎 − 𝜔𝑏/2𝑎, 𝜔𝑐/𝑎 +𝜔𝑏/2𝑎].
From the envelope of the cross-correlation function, thefirst peak of 𝑐[𝑛] appears at 122.8 𝜇s, which verifies theobservation of the S0 wave propagation delay in Figures 10(b)and 11(b). The second peaks appear at 246.0 𝜇s and 247.3𝜇sin Figures 10(b) and 11(b), respectively, which indicates thepropagation delay of A0wave. However, due to the significantdispersion of A0wave, the calculated propagation delay of A0is not at the maximum of 𝑦[𝑛]. Once the propagation delay iscalculated, the delay will be sent to the wireless base stationand processed further by advanced data fusion and cross-checking algorithms to localize and visualize the faults.
5.3. Downsampling and Data Recovery. In order to improvethe accuracy of the fault location estimation carried out atthe base station, single TOF may not be informative enoughfor the data fusion algorithm. The envelopes are also neededin most data fusion processes. However, given the samplingrate of 12.5MHz, 12-bit resolution, and monitoring duration
12 International Journal of Distributed Sensor Networks
0 100 200 300 400
0
0.5
1N
orm
aliz
ed am
plitu
deDownsampled envelope signal
−1
−0.5
Lamb-wave (raw)Original envelopeDownsampled envelope
Time (𝜇s)
(a) Downsampled and reconstructed envelope signal using the Hilberttransform
0 100 200 300 400
012345
Sign
al am
plitu
de (V
)
Lamb-wave (raw)Envelope of Shannon waveletDownsampled envelopeReal value of Shannon wavelet
−1
−2
−3
−4
−5
Time (𝜇s)
(b) Downsampled and reconstructed envelope signal using theShannon wavelet transform
Figure 12: Downsampled and reconstructed envelope signal. The downsampling ratio is 1 : 20.
of 400 𝜇s, the amount of collected data is about 10 k Bytesfor just one single PZT sensor channel. The data has to besegmented into 92 packets due to the 112 Bytes’ limits ofwireless packet size and it will take about 1 second under idealconditions (collision-free) to complete the data transmissionat 125 kbps IEEE 802.15.4 standard data rate. When thenumber of wireless nodes increases and the number of PZTsensor channels per node increases, the time for wireless datatransmission will increase dramatically. Further concern willbe the power consumption demanded for such huge datatransmission.
Therefore, the amount of data to be transmitted wirelesslyhas to be reduced to a great extent to enable practicalwireless data transmission. A downsampling algorithm isdemonstrated here to prove the concept.
The downsampling ratio is one of the indicators to mea-sure the degree of data downsampling whose definition is thedownsampling ratio between the original signal data quantityand downsampled data amount written as the follows:
DR =𝑁o𝑁do
, (8)
where𝑁o,𝑁do denote the signal data quantity and downsam-pled data amount, respectively.The larger theDR is, the betterthe downsampling performance will be with smaller trafficload on the network.
The envelope detected in previous stage is further pro-cessed by a downsampling process to reduce the amount ofdata to be transmitted over the wireless link. At data sink,the envelope is reconstructed from the downsampled data.The results are shown in Figure 12, where the downsampledenvelope (blue circle) fits the original envelope (red line) verywell.
The envelope signal before downsampling consists of4000 sample points and is of 8 k Bytes. At the downsamplingratio of 20 and by removing the noise data before the arrivalof the S0 wave, the number of samples is reduced from 4000to 156, which makes it possible for WSN to collect the data.As a result, sending an envelope signal to the base stationis accomplished by transmitting two packets only, which isa significant saving in terms of communication costs andpower consumption.
Another benefit of the proposed downsampling approachis the noise filtering. Due to the use of FIR filter in thedownsampling algorithm, the reconstructed signal from thedownsampled points is smoother than the original envelope(as shown in the zoomed-in insect in Figure 13). It can beseen that the S0 peak value in downsampled envelope signalafter zooming in Figures 13(a) and 13(b) is 107.26𝜇s and107.54 𝜇s, respectively. The two values are almost the same,which proves the downsampling sensing is feasible in wirelessPZT network using two signal methods. As a result, it willcontribute to the improvement of signal-noise-ration and theaccuracy of fault localization.
5.4. Performance Analysis and Discussion. The reconstructedenvelope signal is reconstructed from the downsamplingdata by interpolating values into downsampling data incubic interpolation algorithm. For instance, as the pointnumber of downsampling data is reducing to 200 from4000 points in original envelope signal, 3800 points can beinterpolated into downsampling data by interp1 function toreconstruct the envelope signal. The time instant of peakvalue in interpolating fitting envelope signal and originalenvelope signal are calculated by max function in MATLABtool.
International Journal of Distributed Sensor Networks 13
90 100 110 120
0
0.1
0.2
0.3
0.4
0.5
0.6N
orm
aliz
ed am
plitu
deDownsampled envelope signal
Lamb-wave (raw)Original envelopeDownsampled envelope
105 110
0.5
0.55
0.6
Downsampled envelope signal(zoomed in at around S0 peak)
−0.1
−0.2
Time (𝜇s)Time (𝜇s)
(a) Downsampled envelope signal under Hilbert transform
80 90 100 110 120 130
0
0.5
1
1.5
2
Sign
al am
plitu
de (V
)
Lamb-wave (raw)Envelope of Shannon waveletReal value of Shannon waveletDownsampled envelope
100 105 110 1151.35
1.41.45
1.51.55
1.61.65
1.71.75
Sign
al am
plitu
de (V
)
Downsampled envelope signal
−2
−1.5
−1
−0.5
(zoomed in at around S0 peak)
Time (𝜇s) Time (𝜇s)
(b) Downsampled envelope signal under Shannon wavelet transform
Figure 13: Downsampled envelope signal.
Reconstruction Error Evaluation. Reconstruction error is onbehalf of the similarity degree of the reconstructed signal andthe original one. In general, the formula is
𝜉 =
√[∑𝑛
𝑖=1(∧
𝑥 [𝑖] − 𝑥 [𝑖])
2
]
√(∑𝑛
𝑖=1𝑥 [𝑖]2)
,(9)
where∧
𝑥 [𝑖], 𝑥[𝑖] separately indicated the reconstructed signaland the original one. The smaller the reconstruction error is,the higher the data recovery accuracy of the reconstructionalgorithm is.
As shown in Figure 14, the absolute error of the recon-structed signal with the original signal is within [−0.04, 0.08],[−0.06, 0.12], [−0.025, 0.02], and [−0.05, 0.05], respectively.
14 International Journal of Distributed Sensor Networks
0 50 100 150 200 250 300 350 400
0
0.02
0.04
0.06
0.08
Reconstruction error
Abso
lute
sign
al er
ror (
V)
−0.02
−0.04
Time (𝜇s)
(a) DR = 60, 𝜉 = 0.006
0 50 100 150 200 250 300 350 400
00.020.040.060.08
0.10.12
Abso
lute
sign
al er
ror (
V)
Reconstruction error
−0.02
−0.04
−0.06
Time (𝜇s)
(b) DR = 90, 𝜉 = 0.0102
0
0.005
0.01
0.015
0.02
Abso
lute
sign
al er
ror (
V)
0 50 100 150 200 250 300 350 400
Reconstruction error
−0.005
−0.01
−0.015
−0.02
−0.025
Time (𝜇s)
(c) DR = 60, 𝜉 = 0.0132
Abso
lute
sign
al er
ror (
V)
00.010.020.030.040.05
0 50 100 150 200 250 300 350 400
Reconstruction error
−0.01
−0.02
−0.03
−0.04
−0.05
Time (𝜇s)
(d) DR = 90, 𝜉 = 0.0299
Figure 14: Absolute reconstruction error under Shannon wavelet transform ((a) and (b)) and Hilbert transform ((c) and (d)).
It can be concluded that the absolute errors of all samplingpoints are distributed in [−0.06, 0.12] which is low errorfor signal reconstruction. The relative error 𝜉 of the recon-structed signal is higher as the DR increases. In a word,downsampling method is able to achieve a high CR andaccuracy of the signal reconstruction for the original lamb-wave signal. Also the Shannon wavelet transform has lessrelative error thanHilbert transform for reconstructed signal,which indicates that the Shannon wavelet transform is betterthan Hilbert transform for data reconstruction of lamb-wavesignal.
Time Delay Error Estimation. In the application of damagedetection, one objective of signal processing is to estimatethe delays in time-of-flight of S0 signal.Therefore, time delayerrors of downsampling are evaluated and compared here. Inthis paper, the time delay error is defined as the differencebetween the time instant of peak value in reconstructedenvelope of S0 signal and the time instant of peak value inoriginal envelope of S0 signal, as shown in Figure 15. In thispaper, time delay error is adopted to evaluate the performance
of signal processing algorithm. The percentage of time delayerror is defined as
𝜉 =𝑡𝑏 − 𝑡𝑎
𝑡𝑏
× 100%, (10)
where 𝑡𝑏 and 𝑡𝑎 are the peak instance of the envelope signalsbefore and after downsampling, respectively.
As shown in Table 1, the higher the downsampling com-pression rate is, the higher the time delay error is, which isas we expected. The Shannon wavelet transform and Hilberttransform have similar performance at the lab tests. At highercompression ratio, Shannon wavelet transform is slightlybetter than Hilbert transform. We believe this is due to thefact that our lamb-wave signals at the lab tests are clean. Inpractice, when the signal-noise-ratio is poorer, it is expectedthat Shannonwill be better thanHilbert.This verificationwillbe our future work. The resolution of damage localizationin wireless PZT network will benefit from lower time delayerrors.
International Journal of Distributed Sensor Networks 15
Table 1: Time delay error in different downsampling ratio andtransform methods.
Downsampling ratioTime delay error (%)
Hilberttransform Shannon wavelet transform
20 0.27 0.2740 0.28 0.2850 0.28 0.2860 0.28 0.2870 2.50 2.4980 3.43 3.4390 4.32 4.29
Sign
al am
plitu
de (V
)
100 105 110 115
1.2
1.3
1.4
1.5
1.6
1.7
Fitting curve signalEnvelope signal
tb ta
Time (𝜇s)
Figure 15: S0 time delay error in S0 mode between reconstructedsignal and envelope original signal.
6. Conclusions and Future Works
This paper presents next-generation, low-cost, wireless PZTnode and network for structural healthmonitoring.The nodeprovides a powerful wireless platform which is able to per-form high-frequency precise data acquisition and distributedlocal data processing. A series of proof-of-concept tests havebeen done and the results of both the envelope detectionand downsampling algorithms are presented, from which theperformance of the proposed wireless PZT sensor networkis verified. For the future work, more advanced distributeddata processing algorithm (such as wavelet denoising) willbe deployed for practical experiments. The comparison ofthe performance between the proposed wireless networkand wired network will be also investigated to evaluatewhether the proposed wireless system match the wiredsystem. Another issue to be addressed in such a distributedwireless PZT sensor network for structure health monitoringis the time synchronization among these wireless nodes.
Competing Interests
The authors declare that there is no conflict of interestsregarding the publications of this paper.
Acknowledgments
This work is supported by European Commission projectHealth Monitoring of Offshore Wind Farms (HEMOW)under Grant FP7-PEOPLE-2010-IRSES-GA-269202 andEPSRC project Novel Sensing Network for IntelligentMonitoring (NEWTON) under Grant EP/J012343/1.
References
[1] L. Qiu and S. Yuan, “On development of a multi-channel PZTarray scanning system and its evaluating application on UAVwing box,” Sensors and Actuators A: Physical, vol. 151, no. 2, pp.220–230, 2009.
[2] Q. Lei, Y. Shenfang, W. Qiang, S. Yajie, and Y. Weiwei, “Designand experiment of PZT network-based structural health moni-toring scanning system,” Chinese Journal of Aeronautics, vol. 22,no. 5, pp. 505–512, 2009.
[3] X. Liu, J. Cao, W.-Z. Song, P. Guo, and Z. He, “Distributedsensing for high-quality structural health monitoring usingWSNs,” IEEE Transactions on Parallel and Distributed Systems,vol. 26, no. 3, pp. 738–747, 2015.
[4] X. Liu, J. Cao, M. Bhuiyan, S. Y. Lai, H. Wu, and G. Wang,“Fault tolerant WSN-based structural health monitoring,” inProceedings of the IEEE/IFIP 41st International Conference onDependable Systems and Networks (DSN ’11), pp. 37–48, HongKong, June 2011.
[5] M. Z. Bhuiyan, G. Wang, J. Cao, and J. Wu, “Deploying wire-less sensor networks with fault-tolerance for structural healthmonitoring,” Institute of Electrical and Electronics Engineers.Transactions on Computers, vol. 64, no. 2, pp. 382–395, 2015.
[6] X. Zhao, H. Gao, G. Zhang et al., “Active health monitoring ofan aircraft wing with embedded piezoelectric sensor/actuatornetwork: I. Defect detection, localization and growth monitor-ing,” Smart Materials and Structures, vol. 16, no. 4, article 032,pp. 1208–1217, 2007.
[7] X. Zhao, T. Qian, G. Mei et al., “Active health monitoring of anaircraft wing with an embedded piezoelectric sensor/actuatornetwork: II. Wireless approaches,” Smart Materials and Struc-tures, vol. 16, no. 4, article 1218, 2007.
[8] O. Martens, T. Saar, and M. Reidla, “TMS320F28335-basedpiezosensor monitor-node,” in Proceedings of the 4th EuropeanDSP Education and Research Conference (EDERC ’10), pp. 62–65, IEEE, December 2010.
[9] A. Perelli, T. Ianni, A. Marzani, L. De Marchi, and G. Masetti,“Model-based compressive sensing for damage localizationin lamb wave inspection,” IEEE Transactions on Ultrasonics,Ferroelectrics, and Frequency Control, vol. 60, no. 10, pp. 2089–2097, 2013.
[10] A. Perelli, L. De Marchi, A. Marzani, and N. Speciale, “Acousticemission localization in plates with dispersion and reverber-ations using sparse PZT sensors in passive mode,” SmartMaterials and Structures, vol. 21, no. 2, Article ID 025010, 2012.
[11] W. Wang, D. Lu, X. Zhou, B. Zhang, and J. Mu, “Statisticalwavelet-based anomaly detection in big data with compressivesensing,” Proceeding of EURASIP Journal onWireless Communi-cations and Networking, vol. 2013, no. 1, p. 269, 2013.
[12] Y. Bao, Y. Yu, H. Li et al., “Compressive sensing-based lost datarecovery of fast-moving wireless sensing for structural healthmonitoring,” Structural Control and Health Monitoring, vol. 22,pp. 433–448, 2015.
16 International Journal of Distributed Sensor Networks
[13] S. Ji, Y. Sun, and J. Shen, “A method of data recovery based oncompressive sensing in wireless structural health monitoring,”Mathematical Problems in Engineering, vol. 2014, Article ID546478, 9 pages, 2014.
[14] L. Liu and F. G. Yuan, “Wireless sensors with dual-controllerarchitecture for active diagnosis in structural health monitor-ing,” Smart Materials and Structures, vol. 17, no. 2, Article ID025016, 2008.
[15] D.Musiani, K. Lin, andT. S. Rosing, “Active sensing platform forwireless structural health monitoring,” in Proceedings of the 6thInternational Symposium on Information Processing in SensorNetworks (IPSN ’07), pp. 390–399, ACM, Cambridge, Mass,USA, April 2007.
[16] D. Musiani, Design of an active sensing platform for wire-less structural health monitoring [Ph.D. thesis], University ofBologna, Bologna, Italy, 2006.
[17] L. Liu and F. G. Yuan, “Active damage localization for plate-likestructures using wireless sensors and a distributed algorithm,”Smart Materials and Structures, vol. 17, no. 5, Article ID 055022,2008.
[18] X. Dong, D. Zhu, Y. Wang et al., “Design and validation ofacceleration measurement using the Martlet wireless sensingsystem,” in Proceedings of ASMEConference on SmartMaterials,Adaptive Structures and Intelligent Systems, American Society ofMechanical Engineers, Newport, RI, USA, September 2014.
[19] Y. Huang, T. Li, X. Dai, H. Wang, and Y. Yang, “TS2: a realisticIEEE1588 time-synchronization simulator for mobile wirelesssensor networks,” SIMULATION: Transactions ofThe Society forModeling and Simulation International, vol. 91, no. 2, pp. 164–180, 2015.
[20] B. Lv, Y. Huang, T. Li et al., “Simulation and performanceanalysis of the IEEE1588 PTP with kalman filtering in multi-hop wireless sensor networks,” Journal of Networks, vol. 9, no.12, pp. 3445–3453, 2014.
[21] Y. Shen and V. Giurgiutiu, “Simulation of interaction betweenlamb waves and cracks for structural health monitoring withpiezoelectric wafer active sensors,” in Proceedings of the ASMEConference on Smart Materials, Adaptive Structures and Intelli-gent Systems, pp. 615–623, Stone Moutain, Ga, USA, September2012.
[22] A. Teolis, Computational Signal Processing with Wavelets,Birkhaauser, Boston, Mass, USA, 1998.
International Journal of
AerospaceEngineeringHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporation http://www.hindawi.com
Journal ofEngineeringVolume 2014
Submit your manuscripts athttp://www.hindawi.com
VLSI Design
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Modelling & Simulation in EngineeringHindawi Publishing Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
DistributedSensor Networks
International Journal of