Article

Advances in Structural Engineering

Design and performance validation of a 1–13

! The Author(s) 2016
Reprints and permissions:
compact wireless ultrasonic device for sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1369433215624597

localized damage detection ase.sagepub.com

Shuo Chen1,2, Xinjun Dong2, Jin-Yeon Kim2, Shengxing Wu1 and

Yang Wang2

Abstract
Recent years have seen growing adoption of wireless structural sensor nodes that can significantly reduce the cost and installation
effort of a structural health monitoring system. While previous wireless sensor development has mainly focused upon dynamic and
vibration measurements at lower frequency domains, in this study, a new wireless ultrasonic sensing node capable of megahertz exci-
tation and sampling is proposed. In addition to presenting the design of the wireless ultrasonic sensing node, experimental notch test
and fatigue test of a dog-bone specimen are described in this article. The experimental results demonstrate that the ultrasonic charac-
teristics of surface cracks can be identified in both scenarios with the proposed wireless sensor node. Furthermore, a signal process-
ing procedure is proposed to obtain an accurate estimation of the ultrasonic signal amplitude, which can be a key indicator for crack
identification. The procedure involves signal reconstruction with the cardinal sine function and envelope detection using discrete
Hilbert transform.

Keywords
fatigue crack, Hilbert transform, notch test, structural health monitoring, ultrasonic measurement, wireless sensing

Introduction structure without generating excitation signals first

(Dhakal et al., 2013). Acoustic emission technique, for
Throughout the decades of its service life, a civil struc- instance, is a typical passive SHM method used to
ture can experience various adverse environmental detect elastic waves generated in a continuous cracking
conditions (e.g. extreme weather condition, frost-wed- process (Ohtsu, 1996). In contrast, active SHM is dif-
ging, and freeze-thaw cycles) and excessive loads (e.g. ferent by generating excitation signals to intentionally
overloaded trucks), both of which may gradually lead excite the structure and then measure the response.
to structural deterioration. Infrastructure deterioration Therefore, the active SHM methods can be more con-
has become an issue of concern in both developed and trollable compared with the passive ones. Ultrasound
developing countries. According to the American technique is a widely used active SHM method, which
Society of Civil Engineers (ASCE) 2013 Report Card employs a transducer to generate excitation signals
for America’s Infrastructure, one-ninth of the nation’s and detect structural response. The response signal
bridges are classified as structurally deficient. may indicate the presence, extent, intensity, and geo-
Meanwhile, 32% of America’s major roads are in poor metric shape of the damage (Grosse et al., 2004).
or mediocre condition, which costs US motorists Different types of ultrasonic waves can be used for
US$67 billion a year in repairs and operation. With
limited maintenance budget for repair, it is important
to be aware of updated structural conditions in real 1
College of Civil and Transportation Engineering, Hohai University,
time, in order to determine which structures are in the Nanjing, China
2
direst situation. School of Civil and Environmental Engineering, Georgia Institute of
To collect real-time information on infrastructure Technology, Atlanta, GA, USA
conditions, a vast amount of structural health monitor- Corresponding author:
ing (SHM) techniques have been studied. Overall, the Yang Wang, School of Civil and Environmental Engineering, Georgia
approaches can be classified as either passive or active. Institute of Technology, Atlanta, GA 30332, USA.
Passive SHM methods collect sensor signals from the Email: [email protected]
2 Advances in Structural Engineering

SHM, such as Rayleigh wave, longitudinal wave, and In comparison, the wireless ultrasonic device devel-
shear wave (Aindow et al., 1981). Main advantages of oped in this study demonstrates significant improve-
the Rayleigh wave include the following: (1) the wave ment in hardware integration. The development in
propagation suffers less from geometric attenuation this study is based upon a latest generation of wireless
than three-dimensional waves (Pertsch, 2009), (2) sensing platform, named Martlet (Kane et al., 2014).
Rayleigh wave is non-dispersive (Viktorov, 1967), and The highly integrated Martlet motherboard
(3) Rayleigh wave is naturally sensitive to surface or (6:35 cm 3 5:71 cm) contains a microprocessor and a
near-surface defects or material degradation. For these wireless transceiver onboard. The ultrasonic board
reasons, Rayleigh wave is best suited for measuring the developed in this study, which has similar size to the
surface crack depth with a low-power ultrasonic SHM motherboard, easily stacks on top of the motherboard
device (such as the one described in this article). (Dong et al., 2014) and achieves a significantly more
However, one should be aware that a major disadvan- compact configuration than the previous prototype. In
tage is that Rayleigh wave loses its sensitivity when a addition, a two-step data processing is proposed in this
crack is deeper than a few times of the wavelength. study, including signal reconstruction and then envel-
Although cabled ultrasonic SHM technologies are ope extraction. Finally, a fatigue crack test is con-
relatively mature, the technologies are generally more ducted with the newly developed wireless ultrasonic
suitable for short-term inspection rather than long- device for more realistic performance validation.
term monitoring in the field. Commercial ultrasonic This article presents the design and validation of the
equipment is bulky and too expensive to be left for wireless ultrasonic sensing node aimed to provide the
long-term field operations (Monje et al., 2012). For advantages of being cable-free, low-cost, and miniatur-
locations difficult to reach, it is even more unfeasible to ized, yet still being functionally comparable with tradi-
carry ultrasonic equipment for in situ test, due to the tional cable-based ultrasonic equipment. Section
weight and size of the equipment. In order to reduce ‘‘General concept and hardware architecture’’ describes
the cost and improve the equipment portability, SHM the hardware design of the ultrasonic device. In section
systems adopting wireless communication technology ‘‘Data processing,’’ a two-step algorithm (i.e. signal
have been investigated. An exhaustive review on wire- reconstruction and envelope detection) is proposed to
less sensing for SHM is provided by Lynch and Loh obtain close approximation of a received ultrasonic
(2006), which summarizes the development of various signal. Using a notch specimen, section ‘‘Validation of
academic and industrial wireless sensing devices. Many wireless ultrasonic sensing node on a notch specimen’’
field experiments have been performed to validate the illustrates the measurement comparison of the wireless
performance of wireless sensing devices. For instance, ultrasonic device and commercial cabled equipment.
a field test is carried out on the Geumdang Bridge Section ‘‘Fatigue crack monitoring using the wireless
located in Incheon, Korea, by Lynch et al. (2006), and ultrasonic sensing device’’ describes measurement
the result shows that the wireless sensor data have results in a laboratory fatigue experiment with a dog-
comparable quality with cabled data. With growing bone specimen. Section ‘‘Conclusions and discussions’’
interest and development in wireless sensor network, provides a summary of the work.
many other research teams have also made significant
development (Baker et al., 2011; Warneke and Pister,
2004). Wireless sensor devices are becoming more General concept and hardware
energy efficient and the cost has been continuously architecture
reducing (Nakamura et al., 2013).
To combine the advantages of traditional SHM and The wireless ultrasonic sensing node proposed in this
non-destructive evaluation (NDE) systems, a prelimi- study is developed upon the Martlet wireless sensing
nary prototype device for wireless ultrasonic evaluation platform (Kane et al., 2014). Featuring a Texas
has been developed by Pertsch et al. (2011). This device Instruments Piccolo microcontroller as the core pro-
is smaller in size compared with traditional NDE sys- cessor, Martlet is a low-cost wireless sensing platform
tems, but still consists of multiple separate physical for various SHM applications. The clock frequency of
components that are somewhat cumbersome. These an earlier version (TMx320F28069) of the microcon-
components include an evaluation board for the Texas troller can be programmed up to 80 MHz, and a more
Instruments F28335 microprocessor, a connector board recent version (TMs320F28069) can support up to
for wireless transceiver, an excitation signal amplifica- 90 MHz. Using the earlier version of microcontroller,
tion board, and a receiving signal–conditioning board. the onboard analog-to-digital conversion (ADC) mod-
The device is tested on a specimen with artificial notches ule can sample data at a frequency up to 2.67 MHz.
cut by electrical discharge machining (EDM), but no While using the more recent microcontroller version,
fatigue crack test is reported. the sampling rate can achieve 3 MHz. The Martlet
Chen et al. 3

node adopts a 2.4 GHz radio for low-power wireless Figure 2 shows the functional diagram of the ultra-
communication through Institute of Electrical and sonic wing, which constitutes of two modules, an exci-
Electronics Engineers (IEEE) 802.15.4 standard tation module and a receiving module. The excitation
(Cooklev, 2004). The communication range can reach module amplifies five-cycle bursts of 500 kHz pulse-
up to 500 m at line-of-sight, and the maximum trans- width modulation (PWM) signals, which are originally
fer rate can reach 250 kbps. generated by Martlet by certain amplification gain.
The extensible hardware design feature of the When powered by two standard low-cost 9-V batteries,
Martlet motherboard enables various accessory sensor the output of the excitation module occupies the vol-
boards to conveniently stack up through four wing tage range of 29 to + 9 V. Although 69 V range is
connectors and work with the Martlet motherboard. used in this study, when higher voltage power source is
The combination of the extensible design feature with available, for example, by simply connecting two 9-V
onboard 9-channel 12-bit ADC allows the Martlet batteries in series, the range can be easily increased up
node to simultaneously sample analog signals from to 620 V when needed. The burst signals are then fed
multiple sensors through different accessory sensor into the transmitting transducer to generate ultrasonic
boards (termed ‘‘wing’’ boards). This study primarily waves as the excitation source for the transmitting
utilizes the newly developed ultrasonic wing (as shown transducer. The transmitting transducer then launches
in Figure 1) for ultrasonic excitation and receiving, fol- ultrasonic waves along a solid structural surface.
lowing a pitch-catch setup of ultrasonic transducers. Upon propagation, a receiving transducer catches the
response signal, which is then collected by the receiving
module of the ultrasonic wing. Signal-conditioning cir-
cuit in the receiving module shifts, filters, and amplifies
the response signals captured by the receiving transdu-
cer and then transfers the analog signal to the ADC
module of the microcontroller in Martlet
motherboard.

Excitation signal generation

The generation of ultrasonic excitation signal is first
initiated by the PWM module of the microcontroller
in Martlet motherboard and then amplified and shifted
by the excitation module of the ultrasonic wing. The
PWM module can generate square waves between 0
and 3.3 V at 500 kHz, which meets the center fre-
quency of the commercial ultrasonic transducers
adopted for this study (Ultran Group WC50-0.5). The
signal amplification module in the ultrasonic wing
Figure 1. Schematic diagram of the ultrasonic wing stacking aims to boost the PWM signal to a higher level. When
on the Martlet motherboard. two pieces of standard 9 V batteries are used, one

Figure 2. Functional diagram of the ultrasonic wing developed for the Martlet wireless sensing node; the ultrasonic wing contains
an excitation module and a receiving module.
4 Advances in Structural Engineering

providing + 9 V and the other providing 29 V, the transducer has relatively low amplitude. A typical out-
schematic is shown in Figure 3. Because of the high- put of a received signal coming out of the receiving
speed switching of the PWM signal, a metal–oxide– ultrasonic transducer oscillates between 20.02 and
semiconductor field-effect transistor (MOSFET) + 0.02 V. This signal cannot be fed into the Martlet
(Alpha & Omega Semiconductor Inc. AO4612) ampli- ADC module directly, as the ADC module requires an
fier circuit is used to buffer the 0–3.3 V signal gener- input voltage between 0 and 3.3 V. Therefore, the sig-
ated by the PWM module of the Martlet nal captured by the receiving transducer needs to be
motherboard. Because the amplitude of the signal initi- shifted, amplified, and filtered before entering the
ated by Martlet is only 3.3 V and does not meet the ADC module. Figure 4 shows the circuit schematic of
threshold voltage requirement between the MOSFET the receiving module of the Martlet ultrasonic wing.
gate (G) and source (S), a hex inverter (Texas The incoming signal, which oscillates around 0 V and
Instruments 74LS04) is adopted to boost the PWM sig- contains negative voltages, first goes through a high-
nals to 0–5 V. The amplification circuit consists of two pass filter with a virtual ground at + 1.65 V, so that
halves. The positive half buffers the signal from 3.3 to the mean value of the signal is shifted up from 0 to
9 V with a p-channel MOSFET, and the negative half + 1.65 V. The high-pass filter has a cutoff frequency
changes the signal from 3.3 to 29 V by an n-channel set at 1.59 kHz, which removes the low-frequency
MOSFET. As a result, the signal is amplified by both component in the signal, but is not high enough to
positive and negative halves, and the output is fed into affect the ultrasonic signal that has a center frequency
the transmitting ultrasonic transducer. at 500 kHz. The shifted signal then goes into a non-
inverting amplifier with an adjustable gain. The gain
can be easily switched among 10, 20, and 30 dB. The
Signal conditioning in the receiving module last stage of the signal conditioning in the receiving
Upon propagation of the ultrasonic waves along a module is a fourth-order voltage–controlled voltage
solid surface, the signal captured by the receiving source (VCVS) low-pass Bessel filter. The low-pass
cutoff frequency is set at 1 MHz. The filter offers a lin-
ear phase performance, which helps maintain the sig-
nal waveform in the time domain (Zhu et al., 2011).
This is important for identifying the arrival time of the
received signal, which is an important index for ultra-
sonic data interpretation. Finally, the signal enters the
Martlet ADC module and is transmitted back wire-
lessly after digitization.

Data processing
Signal reconstruction with the cardinal sine (sinc)
function
Figure 3. Schematic of the amplification circuit in the The commercial ultrasonic transducers adopted in this
excitation module of the Martlet ultrasonic wing. study are the Ultran Group WC50-0.5 model with

Figure 4. Schematic of the signal-conditioning circuit in the receiving module of the Martlet ultrasonic wing.
Chen et al. 5

500 kHz central frequency, which means that the the original peak. Meanwhile, signal 2 (dot line) shows
received signal is mainly concentrated around a more common scenario with original peak missed,
500 kHz, with a side lobe bandwidth usually less than but shares the same envelope with signal 1. Therefore,
200 kHz. Therefore, according to the Nyquist– a more accurate estimation of the amplitude can be
Shannon sampling theorem (Nyquist, 1928; Shannon, obtained from the envelope of the signal, compared
1949), the sampling frequency used in this study with simple peak-picking. Discrete Hilbert transform
(2 MHz) is sufficient for capturing the received signal. (DHT) is adopted in this article to detect the envelope.
However, it is still possible that the digitization process Proposed by Huang et al. (1998), Hilbert transform
cannot capture the exact peak of every received burst has been widely used as an analysis and processing
response, while the peak amplitude can be important method for nonlinear and non-stationary signals.
for quantitative damage assessment. In order to obtain Using the Cauchy principal value (CPV), Hilbert trans-
accurate waveform of the received signals, signal form of the reconstructed signal is defined as
reconstruction for up-sampling is proposed as the first
step of the two-step data processing. ð‘
+
1 xre ðtÞ
In this study, the cardinal sine function (also known xre ðtÞ ¼ Hfxre ðtÞg¼ CPV dt ð2Þ
p t't
as the sinc function) is adopted for signal reconstruc- '‘
tion (Shannon, 1948). The function enables reconstruc-
tion of band-limited signals (Yaroslavsky, 2007). The Based on the definition, it can be seen that xre ðtÞ
formulation of the reconstructed signal reconstruction, can be obtained by the convolution between xre ðtÞ and
xre ðtÞ, is as follows 1=pt. So equation (2) can be rewritten as

X
n 1
xre ðtÞ ¼ (xre ðtÞ ð3Þ
xre ðtÞ ¼ x½i% & sincðt & fs ' ði ' 1ÞÞ ð1Þ pt
i¼1
where ( is the convolution symbol. If Fourier trans-
where t represents time, fs is the sampling frequency of form is applied on both sides of equation (3), one
the original signal (2 MHz in this study), x½i% is the ith obtains
discrete sample point of the original signal, n is the
n o 1
number of data points, and sincðxÞ ¼ sinðpxÞ=ðpxÞ. F xre ðtÞ ¼ & F f1=tg& F fxre ðtÞg
p ð4Þ
¼ 'j & sgnðvÞ & F fxre ðtÞg
Envelope detection using discrete Hilbert transform
where j is the imaginary unit; F f&g denotes Fourier
In this article, envelope amplitude is adopted as the
transform, and sgnð&Þ is the sign function
indicator for quantitative damage assessment. It is
important to calculate amplitude value as accurately as 8
< 1 v.0 rad=s
possible. However, if a simple peak-picking process is sgnðvÞ ¼ 0 v ¼ 0 rad=s ð5Þ
adopted, the original peak can be easily missed during :
'1 v.0 rad=s
the modulation process for transmission. Figure 5
shows an example for illustration. Signal 1 (dashed Using the Hilbert transform pair of xre ðtÞ and xre ðtÞ,
line) is a modulated signal, which happens to capture the reconstructed analytic signal xare ðtÞ is defined as
(Oppenheim et al., 1989)

xare ðtÞ ¼ xre ðtÞ + j & xre ðtÞ ð6Þ

Applying Fourier transform on equation (6), the fre-

quency spectrum of xare ðtÞ is given as
" # n o
F xare ðtÞ ¼ F fxre ðtÞg+ F j & xre ðtÞ
ð7Þ
¼ ð1 + sgnðvÞÞ & F fxre ðtÞg

As a result, F fxare ðtÞg¼ 2F fxre ðtÞg for v.0,

F fxare ðtÞg¼ F fxre ðtÞg for v ¼ 0, and F fxare ðtÞg¼ 0
for v\0. This means that the spectrum of the analytic
Figure 5. Diagram for two modulated signals sharing a same signal, xare ðtÞ, can be easily obtained based on the cal-
envelope waveform. culation of F fxre ðtÞg. In addition, the time-domain
6 Advances in Structural Engineering

expression of the analytic signal xare ðtÞ can be obtained

by an inverse Fourier transform to the right-hand side
of equation (7). Finally, as shown in the literature, the
envelope signal of the analytic reconstructed signal
xenv
re ðtÞ can be calculated as the magnitude of the analy-
tic signal xare ðtÞ (Dugundji, 1958; Pertsch et al., 2011)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
$ a $ 2
xenv $ $ x2re ðtÞ + xre ðtÞ
re ðtÞ ¼ xre ðtÞ ¼ ð8Þ

Validation of wireless ultrasonic sensing

node on a notch specimen Figure 6. Steel specimen with four notches.

In order to validate the performance of the wireless the transducer on either side of the notch is 12.7 mm.
ultrasonic sensing node, an experiment is first carried Couplant oil is used between the wedge and the speci-
out on a steel specimen with four artificial notches of men surface. In the measurement taken with the wire-
different depths. Then, the same ultrasonic measure- less ultrasonic sensing node (Figure 7(a)), a five-cycle
ment is repeated with traditional commercial equip- ultrasonic tone-burst with a peak-to-peak amplitude of
ment (an arbitrary waveform generator and an 18 V at 500 kHz is generated by the excitation module
oscilloscope) to use its result as a reference. In these and fed to the transmitting transducer. The ultrasonic
experiments, the wedge technique is employed to waves (Pertsch et al., 2011), which are launched in the
launch and detect Rayleigh waves in the steel speci- specimen, propagate through the notch and are then
men. Two commercial narrow band ultrasonic trans- detected by the receiver on the other side of the notch.
ducers (Ultran Group, WC50-0.5) are attached to the The received signal is amplified as much as 30 dB and
transmitting and receiving wedges, respectively. low-pass filtered with a 1 MHz cutoff frequency. The
Different types of ultrasonic waves, such as bulk conditioned signal enters the ADC module of the
wave, Rayleigh wave, and Lamb wave (Aindow et al., Martlet and, finally upon digitization, is wirelessly sent
1981), can be used for SHM. One advantage of to a computer for post-processing. Besides wireless
Rayleigh wave is that it suffers less from geometric measurement, a cabled baseline measurement is per-
attenuation than other bulk waves (Pertsch, 2009) due formed using the same transducer arrangement as in
to its two-dimensional propagation nature. Moreover,
the previous measurements (Figure 7(b)). A same exci-
Rayleigh wave is non-dispersive unlike Lamb waves,
tation signal is generated by an arbitrary waveform
and thus, waveforms from an experiment are relatively
generator (Agilent 33220A) and a response signal is
simple and easy to analyze (Staszewski, 2004). These
captured by an oscilloscope (Tektronix MSO4034) at
advantages make Rayleigh wave suitable to be used
the sampling frequency of 250 MHz.
for low-power ultrasonic SHM applications.

Notch specimen and measurement procedures Measurement results from wireless ultrasonic
sensing node and cabled equipment
The specimen, which is made of steel with dimensions
241:3 mm 3 152:4 mm 3 25:4 mm, is shown in The signal reconstruction algorithm is applied as
Figure 6. Four notches with different depths are cut by described in section ‘‘Signal reconstruction with the
EDM as simulated cracks. The notch depths are 0.51, cardinal sine (sinc) function,’’ and Figure 8(a) and (b)
1.27, 2.29, and 3.05 mm; the width is about 9.3 mm; each shows a sampled tone-burst signal (cross sym-
the opening gap is about 0.58 mm long, and the dis- bols) taken from the undamaged area and the recon-
tance between them is 25 mm. These notches are suffi- structed signal (continuous lines). Figure 8(a) shows a
ciently away from both the specimen boundaries and general scenario when peaks in the original sampled
each other, and thus, ultrasonic measurements can be signal are missing due to the low sampling rate. In con-
performed for each individual crack with little distur- trast, Figure 8(b) illustrates the scenario when the
bances from boundaries and adjacent notches. sampled signal coincidentally captures the peak ampli-
In the ultrasonic measurement, two ultrasonic trans- tude of the original signal. The plots demonstrate that
ducers are placed on each side of the notch in a pitch- signal reconstruction helps restore the peak value of
catch setup. The transducers are carefully aligned so the original signal.
that the generated waves are normally incident on and To validate the reconstruction performance of
transmitted through the crack. The distance between waveform as well, a reconstructed wireless signal
Chen et al. 7

Figure 7. Experimental setup for both cabled and wireless test: (a) photo of the wireless test setup and (b) photo of the cabled
test setup.

(shown in Figure 8(c)) is compared with a signal cap- To reduce noise effect, at each notch location, aver-
tured by cabled equipment (shown in Figure 8(d)), age peak amplitude is obtained from the envelopes of 60
both collected from the undamaged area. Note that a reconstructed wireless signals and obtained from 15
gain of 30 dB is removed from the reconstructed wire- cabled signals. The results are compared in Figure 8(f).
less signal so that its waveform can be compared with This figure shows that the average peak amplitudes of
that of cabled signal. In order to evaluate the wave- the signals taken from the wireless and cabled equip-
form difference between cabled and reconstructed ment are in good agreement. There is a clear trend that
wireless signals, relative error (i.e. root-mean-square the peak amplitude of the transmitted signals decreases
error, RMSE) is calculated with the increasing of the notch depth. This is because
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi more energy of the incident Rayleigh wave is scattered
1X n & '2 and reflected from a deeper notch, and thus, less energy
RMSE ¼ xcabledðiÞ ' xreconstructed
wireless ði Þ ð9Þ is captured by the receiving transducer on the other side.
n i¼1
These results illustrate that the Martlet wireless
where xcabled is a signal captured by cabled equipment ultrasonic sensing node is capable of generating, detect-
from undamaged area at 250 MHz; xreconstructed is the ing, and post-processing signals in 500 kHz range in
wireless
corresponding reconstructed wireless signal of the experiment where ultrasonic signals propagate
250 MHz (with amplification removed for compari- through a notch of a few millimeters depth. Moreover,
son). Using data plotted in Figure 8(c) and (d), the the resulting signals, after the signal reconstruction and
relative error is found to be 7:53 3 10'4 V, which is envelope detection, are almost equivalent to those of
small compared with the amplitude over 0.01 V. cabled ultrasonic experiment. Finally, the detected sig-
Therefore, it can be concluded that the cabled and nals are sensitive to the surface defects in the steel spe-
reconstructed wireless signals are very close, and that cimen and thus can be used to distinguish the notch (or
the reconstruction algorithm can effectively rebuild the crack) depth.
waveform. Furthermore, Figure 8(e) shows an example
of reconstructed tone-burst signal together with the Fatigue crack monitoring using the
envelope curve extracted by DHT, as described in sec-
wireless ultrasonic sensing device
tion ‘‘Envelope detection using discrete Hilbert trans-
form.’’ This particular signal has two peaks with In order to evaluate the possible crack detection capa-
similar magnitudes, but the envelope curve easily iden- bility of the wireless ultrasonic sensing device in a more
tifies the highest peak. realistic fatigue scenario, the device is applied, in situ,
8 Advances in Structural Engineering

Figure 8. Comparison of cabled and wireless measurement results: (a) comparison between original and reconstructed wireless
signals (the original missing peaks), (b) comparison between original and reconstructed wireless signals (the original including peaks),
(c) reconstructed wireless signal with gain removed, (d) cabled signal at 250 MHz, (e) envelope of a sample tone-burst at undamaged
area, and (f) comparison of cabled and wireless peaks.

to a specimen undergoing fatigue test. In this experi- parameters. Thus, a tensile test is carried out on an
ment, a dog-bone specimen is cut from a 6.35 mm Instron 22EMF material test machine with a maximum
thick Aluminum 6061-T6-alloy plate. To facilitate fati- loading capacity of 245 kN. The flat plate tension test
gue crack initiation at the desired location, that is, in specimen is designed according to the American Society
the middle of the specimen, a through-thickness hole for Testing and Materials (ASTM) standard E8/E8M-
with a diameter of 3.81 mm is drilled in the middle and 13a:2013 (2013) as shown in Figure 9(b). Two 120 O
then two small slots are cut on both lateral sides of the strain gauges are attached onto both sides of the speci-
hole. The detailed dimensions of the dog-bone speci- men and the strain data are collected with a data acqui-
men are shown in Figure 9(a). The hatched areas sition module (National Instruments, 9235) as shown in
(6:35 mm 3 31:75 mm) in this figure indicate the loca- Figure 10(a). The tensile loading schedule is designed
tions, where two ultrasonic transducers are placed. The according to the above standard as well; the loading
specimen is then fatigued under a constant-amplitude rate is set as 0.13 kN/s. The test result is shown in
loading condition. Figure 10(b). The yield strength determined from the
curve is 294 MPa, and Young’s modulus is 65.3 GPa.

Material properties determination Fatigue load and test procedure

A prior knowledge of the actual yield strength of the The loading parameters for fatigue test are determined
specimen is critical to determine fatigue-loading according to both the ASTM standard E647-13: 2013
Chen et al. 9

Figure 9. Details in the specimens’ design: (a) detailed dimensions of the dog-bone specimen for fatigue test (mm) and (b) layout of
flat plate tension specimen for yield strength test (mm).

Figure 10. Test to determine Aluminum 6061-T6 yield strength: (a) layout of the tensile test and (b) strain–stress curve for
aluminum 6061-T6 under uniaxial load (average strain).

(2013) and the yield strength result of the Aluminum collecting ultrasonic signals when the specimen is held
6061-T6 material. In a preliminary fatigue test, the at a static load. Different holding loads may cause dif-
maximum and minimum loads are 26.69 and 0 kN, ferent levels of crack opening.
with the R factor equaling 0 (R ¼ Pmin =Pmax ). The setup of the test is shown in Figure 11(b). At
However, due to the limitation of the loading machine, the beginning of the test, the specimen is pre-fatigued
the minimum load cannot be controlled to be precisely to generate an initial crack (in fact, cracks happen on
0 kN, and thus is set to be a small value at 0.44 kN. both sides of the central hole) whose length is mea-
The final loading profile is presented in Figure 11(a), sured using a caliper. The transducers are attached to
and other detailed loading parameters are shown in the front side of the specimen with oil couplant. The
Table 1. The static loading steps are allocated for ultrasonic response signals are collected at each static
10 Advances in Structural Engineering

Table 1. Fatigue test loading procedures.

Test No. Test type Loading description

#1a (pre-crack) Cyclic load 0.44–26.69 kN at 5 Hz; 2000 cycles

#1b (after #1a) Static holding load 13 steps at 0.44–13.34 kN; load increment from first to second holding
step is 0.67 kN, and 1.11 kN for steps afterward
#2a Cyclic load 0.44–26.69 kN at 3 Hz; 1000 cycles
#2b (after #2a) Static holding load 9 steps at 0.44–26.69 kN; load increment from first to second holding
step is 2.9 kN, and 3.34 kN for steps afterward
#3a Cyclic load Same as Test #2a
#3b (after #3a) Static holding load Same as Test #2b
#4a Cyclic load Same as Test #2a
#4b (after #4a) Static holding load Same as Test #2b

Figure 11. Loading parameters and experimental setup of the fatigue test: (a) diagram of load profile to monitor crack initiation
and evolution (showing up to Test #3a for brevity) and (b) photo of the fatigue test with the wireless ultrasonic device implemented.

holding load. Sampling frequency for wireless acquisi- propagation (Jogi et al., 2008), the fatigue crack
tion is 2 MHz. In addition, the received signals are growth rate can indicate the crack stage in the fatigue
post-amplified by 30 dB. test (Logsdon and Liaw, 1986). In the beginning of the
test, a plane-strain condition predominates. However,
loss of constraint is expected, as the crack grows lon-
Crack behavior and ultrasonic signal versus loading ger than a critical length, and the initial steady fatigue
cycles crack propagation mode switches to a fast fracture
The crack lengths are measured throughout the fatigue mode. The fractography result (Figure 12(b)) also
test and after the specimen’s failure at 5873 cycles (on agrees with this explanation as a clear flat-to-slant
the fracture surfaces). Figure 12(a) presents the mea- transition is observed on both sides of the fracture sur-
surement results, where stable crack propagation is face. Since during the crack propagation period the
observed. The crack growth rate is approximately crack growth is stable and slow enough, the presented
7:62 3 10'4 mm=cycle at the beginning and sharply ultrasonic method is suitable for detecting the develop-
increases to 1:4 3 10'2 mm=cycle at the end. Since the ment of small fatigue cracks.
fatigue crack growth rate has a positive correlation At every static holding step, the ultrasonic signals
with the mechanical driving force for crack are captured by the wireless ultrasonic sensing node
Chen et al. 11

Figure 12. Growth behavior and fractography result of the crack: (a) crack growth rate against loading cycles and (b) fractography
result at the fracture surface.

phenomenon called crack closure, which is well recog-

nized in the fatigue-fracture community (Elbert, 1971)
and from the ultrasonic response from a closed crack
(Rokhlin and Kim, 2003). When the specimen is
unloaded, small fatigue cracks initiated in ductile
metals (e.g. Aluminum 6061-T6) tend to close, for rea-
sons including the closing action of the crack tip plasti-
city. At low holding load levels, the tensile load is not
high enough to open the cracks (see the beginning
parts of Test #1b, #2b, and #3b in Figure 13), so the
crack surfaces are still tightly closed and thus allow the
surface Rayleigh wave to transmit through. This has
led to the stable ultrasonic signal amplitudes, that is,
Figure 13. Average amplitude of the envelope curves of the plateau region in the beginning parts of Test #1b,
ultrasonic receiving signals at every holding load stage. #2b, and #3b in Figure 13. As the holding load level
increases further and the cracks open gradually, the
and wirelessly transmitted. The ultrasonic data are amplitude of the transmitted ultrasonic signals starts
processed by the algorithms described in section to decrease with load. After the load gets to some
‘‘Measurement results from wireless ultrasonic sensing higher load, and the crack fully opens, the amplitude
node and cabled equipment.’’ The peak amplitudes of of the transmitted ultrasonic signal saturates at a lower
the transmitted wave signals obtained from envelopes holding load level. This opening–closure behavior is
are plotted in Figure 13. In this figure, it is seen that well illustrated in the ultrasonic signals, especially the
the amplitude decreases significantly with the increase signals collected in Test #3b, where the arrows indicate
of the load at which the ultrasonic signal is taken. This the start and end of the opening.
trend is in good agreement with previous literature A very similar behavior is observed in the reflected
(Hevin et al., 1998). This can be attributed to the fact ultrasonic signals from a surface crack initiated around
that more energy is reflected and thus less energy is a surface defect. However, this trend is different near
transmitted and detected by the receiving transducer the end of the fatigue life (i.e. at 5000 cycles), and the
as the length of the cracks grows. Since the amplitude initial flat stage is missing in the curve. This phenom-
of a received ultrasonic signal correlates directly with enon is attributed to the stress release. When the speci-
the crack length, the change in the ultrasonic ampli- men is on the brink of failure, the stress required to
tude can be used to detect crack growth. open a crack decreases significantly (Rokhlin and
Another trend that can be observed in Figure 13 is Kim, 2003). As demonstrated at the beginning part of
the decreasing signal amplitude with the increase in Test #4b in Figure 13, it is possible that the stress for
static holding load level. This is due to the the crack to open is so small that the plateau region
12 Advances in Structural Engineering

would not appear. The same phenomenon has also used to predict the fatigue life of a structural member.
been described and reported in other investigations Furthermore, upon successful laboratory validation,
(Kim and Rokhlin, 2002; Song and Sih, 2003). At a field deployment of the miniature wireless devices will
given number of fatigue cycles, the crack length desig- be the next step toward enabling unprecedentedly
nates the length of the fully open crack. Therefore, if dense and long-term ultrasonic measurements for criti-
the holding load level is not carefully taken into con- cal structural members in the field.
sideration, the ultrasonically determined crack length
can be significantly underestimated. More detailed
Acknowledgements
quantitative analysis is required in subsequent studies.
The authors also wish to express their gratitude to Dapeng
Zhu, Xiaohua Yi, and Andrew Udell of Georgia Institute of
Conclusion and discussions Technology for their assistance during the design and testing
of the Martlet ultrasonic wing. Any opinions, findings, and
This article presents the design and validation tests of conclusions or recommendations expressed in this publica-
a wireless ultrasonic sensing node, which is developed tion are those of the authors and do not necessarily reflect
based on the Martlet platform for wireless sensing: the view of the sponsors.

1. Powered by two standard 9-V batteries, the

Declaration of Conflicting Interests
ultrasonic wing can generate 69 V PWM sig-
nals at 500 kHz for ultrasonic excitation. The author(s) declared no potential conflicts of interest with
Meanwhile, the device performs signal condi- respect to the research, authorship, and/or publication of this
article.
tioning to receive ultrasonic signal for a higher
signal-to-noise ratio. The conditioned signal
can be sampled at up to megahertz level by the Funding
Martlet sensing node and then wirelessly The author(s) disclosed receipt of the following financial sup-
transmitted. port for the research, authorship, and/or publication of this
2. So as to obtain a more accurate assessment of article: This research is partially sponsored by the National
the peak amplitude of the received ultrasonic Science Foundation (#CMMI-1150700), the Research and
signal, a two-step data processing approach is Innovative Technology Administration of US DOT
introduced (i.e. signal reconstruction and envel- (#DTRT12GUTC12), and the Georgia DOT (#RP12-21).
ope identification using DHT). Furthermore,
experimental results show the proposed References
approach is effective for accurately capturing Aindow A, Dewhurst R, Hutchins D, et al. (1981) Laser-gen-
signal peaks. erated ultrasonic pulses at free metal surfaces. The Journal
3. Validation test for the wireless ultrasonic sen- of the Acoustical Society of America 69(2): 449–455.
sing node is carried out using a notch specimen, ASTM E647-13:2013 (2013) Standard test method for mea-
in order to evaluate the performance of surface surement of fatigue crack growth rates (West Consho-
hocken, PA: ASTM).
defect detection. Experimental results demon-
ASTM E8/E8M-13a:2013 (2013) Standard test methods for
strate that the wireless ultrasonic sensor signal
tension testing of metallic materials (West Conshohocken,
is sensitive to surface defects and is reasonably PA: ASTM).
close to that of a cabled non-destructive testing Baker J, Hasler WL, Camilleri M, et al. (2011) Impairment
(NDT) system. of overall and regional high amplitude colon contractions
4. Performance of the wireless ultrasonic sensing in severe slow transit constipation measured by wireless
node is further validated through fatigue test of motility capsules. Gastroenterology 140(5): S-531.
a dog-bone specimen. The experimental results Cooklev T (2004) Wireless Communication Standards: A
show agreement with those in previous studies Study of IEEE 802.11, 802.15, and 802.16. New York:
and can be explained by crack scattering the- IEEE Standards Information Network Press.
ory. The reliability of the wireless ultrasonic Dhakal DR, Neupane K, Thapa C, et al. (2013) Different
sensing node is validated. techniques of structural health monitoring. Research and
Development (IJCSEIERD) 3(2): 55–66.
Dong X, Chen S, Zhu D, et al. (2014) High-speed heteroge-
This study is undertaken as a first step toward the
neous data acquisition using Martlet—a next-generation
development and testing of the ultrasonic functionality wireless sensing node. In: 6th World conference on struc-
with the Martlet wireless sensing platform. Future tural control and monitoring, Barcelona, Spain, 15–17 July
efforts are needed toward more detailed and quantita- 2014.
tive evaluation of damage. The study will become
more interesting if the ultrasonic measurements can be
Chen et al. 13

Dugundji J (1958) Envelopes and pre-envelopes of real wave- deployed GPS network. Journal of Infrastructure Systems
forms. IRE Transactions on Information Theory 4(1): 20(2): 04014003-1–04014003-10.
53–57. Nyquist H (1928) Certain topics in telegraph transmission
Elbert W (1971) The significance of fatigue crack closure. In: theory. Transactions of the American Institute of Electrical
Rosenfeld MS (ed.) Damage Tolerance in Aircraft Struc- Engineers 47(2): 617–644.
tures, STP 486. West Conshohocken, PA: American Soci- Ohtsu M (1996) The history and development of acoustic
ety for Testing and Materials, pp. 230–242. emission in concrete engineering. Magazine of Concrete
Grosse C, Finck R, Kurz J, et al. (2004) Monitoring tech- Research 48(177): 321–330.
niques based on wireless AE sensors for large structures in Oppenheim AV, Schafer RW and Buck JR (1989) Discrete-
civil engineering. In: 26th European conference on acoustic Time Signal Processing. Upper Saddle River, NJ: Prentice
emission testing, Berlin, Germany, 15–17 September 2004. Hall, Inc.
Hevin G, Abraham O, Pedersen H, et al. (1998) Characteri- Pertsch A, Kim JY, Wang Y, et al. (2011) An intelligent
zation of surface cracks with Rayleigh waves: a numerical stand-alone ultrasonic device for monitoring local struc-
model. NDT & E International 31(4): 289–297. tural damage: implementation and preliminary experi-
Huang NE, Shen Z, Long SR, et al. (1998) The empirical ments. Smart Materials and Structures 20(1): 015022.
mode decomposition and the Hilbert spectrum for non- Pertsch AT (2009) An intelligent stand-alone ultrasonic device
linear and non-stationary time series analysis. Proceedings for monitoring local damage growth in civil structures. MS
of the Royal Society of London: Series A: Mathematical, Thesis, Georgia Institute of Technology, Atlanta, GA.
Physical and Engineering Sciences 454(1971): 903–995. Rokhlin SI and Kim J-Y (2003) In situ ultrasonic measure-
Jogi B, Brahmankar P, Nanda V, et al. (2008) Some studies ment of crack closure. International Journal of Fatigue
on fatigue crack growth rate of aluminum alloy 6061. 25(1): 51–58.
Journal of Materials Processing Technology 201(1): Shannon CE (1948) A mathematical theory of communica-
380–384. tion. Bell Systems Technical Journal 27: 379–423.
Kane M, Zhu D, Hirose M, et al. (2014) Development of an Shannon CE (1949) Communication in the presence of noise.
Extensible Dual-Core Wireless Sensing Node for Cyber- Proceedings of the IRE 37(1): 10–21.
Physical Systems. San Diego, CA: SPIE Smart Structures, Song Z and Sih G (2003) Crack initiation behavior in magne-
NDE. toelectroelastic composite under in-plane deformation.
Kim J-Y and Rokhlin SI (2002) Surface acoustic wave mea- Theoretical and Applied Fracture Mechanics 39(3):
surements of small fatigue cracks initiated from a surface 189–207.
cavity. International Journal of Solids and Structures 39(6): Staszewski WJ (2004) Structural health monitoring using
1487–1504. guided ultrasonic waves. In: Holnicki-Szulc J and Soares
Logsdon W and Liaw PK (1986) Tensile, fracture toughness C (eds) Advances in Smart Technologies in Structural Engi-
and fatigue crack growth rate properties of silicon carbide neering, vol. 1. Berlin; Heidelberg: Springer, pp. 117–162.
whisker and particulate reinforced aluminum metal Viktorov IA (1967) Rayleigh and Lamb waves: Physical The-
matrix composites. Engineering Fracture Mechanics 24(5): ory and Applications. New York: Plenum Press.
737–751. Warneke BA and Pister KS (2004) An ultra-low energy
Lynch JP and Loh KJ (2006) A summary review of wireless microcontroller for smart dust wireless sensor networks.
sensors and sensor networks for structural health moni- In: IEEE international solid-state circuits conference, San
toring. Shock and Vibration Digest 38(2): 91–130. Francisco, CA, 15–19 February.
Lynch JP, Wang Y, Loh KJ, et al. (2006) Performance moni- Yaroslavsky L (2007) Fast discrete sinc-interpolation: A gold
toring of the Geumdang Bridge using a dense network of standard for image resampling. In: Astola J and Yaro-
high-resolution wireless sensors. Smart Materials and slavsky L (eds) Advances in Signal Transforms: Theory
Structures 15(6): 1561. and Applications. Cairo: Hindawi Publishing Corporation,
Monje PM, Casado L, Aranguren G, et al. (2012) Integrated pp. 337–388.
electronic system for ultrasonic structural health monitor- Zhu D, Wang Y and Brownjohn J (2011) Vibration testing
ing. In: 6th European workshop on structural health moni-
of a steel girder bridge using cabled and wireless sensors.
toring, Dresden, Germany, 3–6 July 2012.
Frontiers of Architecture and Civil Engineering in China
Nakamura I, Saeki M, Oguni K, et al. (2013) Energy-saving
5(3): 249–258.
wireless sensor node for relative positioning of densely