You are on page 1of 15

Original Article

Structural Health Monitoring


1–15

Multimodal structural health Ó The Author(s) 2017


Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
monitoring based on active and passive DOI: 10.1177/1475921717699375
journals.sagepub.com/home/shm
sensing

Amir Nasrollahi1, Wen Deng2, Zhaoyun Ma3 and Piervincenzo Rizzo1

Abstract
We present a structural health monitoring system based on the simultaneous use of passive and active sensing. The pas-
sive approach is based on acoustic emission, whereas the active approach uses the electromechanical impedance and the
guided ultrasonic wave methods. As all these methods can be deployed with the use of wafer-type piezoelectric transdu-
cers bonded or embedded to the structure of interest, this article describes a unified structural health monitoring sys-
tem where acoustic emission, electromechanical impedance, and guided ultrasonic wave are integrated in the same
hardware/software unit. We assess the feasibility of this multimodal monitoring in a large flat aluminum plate instrumen-
ted with six transducers. Acoustic emission events are simulated by exciting a tone burst or by using the conventional
pencil lead break test, and the detected signals are processed with a source localization algorithm to identify the position
of the source. For the active sensing, damage is simulated by adding a small mass to the plate: the raw waveforms are
processed with a delay-and-sum algorithm to create an image of the plate, whereas the electrical admittance of each
transducer is analyzed using the statistical index of the root-mean-square deviation. The results presented in this article
show that the proposed system is robust, mitigates the weaknesses of each method considered individually, and can be
developed further to address the challenges associated with the structural health monitoring of complex structures.

Keywords
Acoustic emission, electromechanical impedance method, guided ultrasonic waves, structural health monitoring, damage
detection

Introduction sensors different than the commercial AE transducers


including wafer-type lead zirconate titanate (PZT).37,38
A wide body of literature proved that structural health GUW-based SHM methods rely on the active gener-
monitoring (SHM) methods based on acoustic emission ation and detection of ultrasonic signals such as tone
(AE),1–9 electromechanical impedance (EMI),10–18 and bursts, chirps, or continuous waves. GUWs are very
guided ultrasonic waves (GUWs)19–31 are able to assess attractive as they enable the inspection of large areas
the health of simple waveguides and a certain number from a single or a few positions. When the method is
of complex structures.
AE method exploits the propagation of transient
elastic waves generated by the rapid release of energy 1
Laboratory for Nondestructive Evaluation and Structural Health
from a localized or multiple sources within the material Monitoring Studies, Department of Civil and Environmental Engineering,
being monitored. The energy propagates as a stress University of Pittsburgh, Pittsburgh, PA, USA
2
School of Automation, Northwestern Polytechnical University, Xi’an,
wave in the structure and is detected by one or more
China
AE sensors. Typically, commercial dedicated systems 3
Department of Mechanical Engineering, University of South Carolina
are used to run AE. These systems perform, for exam-
ple, parametric analysis by quantifying features, such Corresponding author:
as counts, amplitude, rise time, and energy, and these Piervincenzo Rizzo, Laboratory for Nondestructive Evaluation and
Structural Health Monitoring Studies, Department of Civil and
parameters are correlated to the source that has gener- Environmental Engineering, University of Pittsburgh, 3700 O’Hara Street,
ated the transient wave.32,33 Some researchers proposed Pittsburgh, PA 15261, USA.
alternative platforms,34,35 whereas others36–39 used Email: pir3@pitt.edu
2 Structural Health Monitoring

targeted for permanent monitoring, PZT patches are interface coded in LabVIEW. Two piezoelectric trans-
used.39–42 The EMI technique exploits instead the rela- ducers were used; the first PZT transmitted a constant
tionship between the electrical impedance of a PZT and amplitude 30-cycle sine wave across a range of frequen-
the mechanical impedance of the host structure to cies and also enabled EMI measurements. At the same
which the PZT is bonded or embedded.10,43–48 time, the digitizer sampled the signal from the second
It is known that AE, EMI, and GUW have advan- PZT at predefined actuation frequencies, giving rise to
tages and limitations. The method of the AE assumes the pitch–catch configuration typical of many guided
that all kinds of defect emit sounds (or ultrasounds) wave testing. The system was successfully tested on a
when they occur or when they grow. However, there plate and on a pipe, but each measurement required
are still some disputes about the ability to monitor/ several minutes and therefore was not feasible for the
detect corrosion, about the repeatability of the AE sig- field applications. An et al.56,62 built an integrated
nals, and the ability to use this technique in noisy envir- impedance and guided wave system made of an arbi-
onments. Another concern about AE is that past trary waveform generator (AWG), a two-channel digiti-
history or damage, that is, damage that has occurred zer, and two multiplexers. The multiplexers allowed for
before implementing an AE monitoring system, is not the selection of GUW generation or EMI sensing. An
‘‘seen’’ unless the damage grows. Methods based on the algorithm was developed to compensate any effect
propagation of GUWs may be insensitive to the pres- associated with temperature variation; moreover, they
ence of cracks or other kinds of defect oriented along combined the two indices relative to the two methods
the direction of the wave propagation. The ability to into a single damage index. Gulizzi et al.66 improved
detect defects is related to the wavelength of the waves. the approach presented in43 by (1) designing a more
The smaller the wavelength, the smaller is the defect efficient hardware, (2) changing the monitoring scheme,
potentially detectable; however, this comes at the and (3) using an advanced signal processing to perform
expense of the propagation distance. Moreover, in com- damage detection. For the actuation, a chirp signal
plex structures, GUWs are scattered by bends, elbows, x(c)[n], defined as
edges, or fasteners (bolts, rivets, welds, etc.) leading to   
possible false positives/negatives. Finally, the EMI tech- (c) 2p f2  f1
x ½n = Aw½n sin n n + f1 , n = 0, . . . , N  1
nique is very effective at probing a region that is close fs 2N
(in some cases only a few inches) to the probing ele- ð1Þ
ment, that is, it is a local approach. The EMI method is
also sensitive to many variables including the geometric was used. In equation (1), w[n] is a window function, fs
and electrical properties of the PZT. is the sampling frequency, A is the amplitude, and f1
To boost the advantages and null the shortcomings and f2 are the start and end frequencies. N/fs is the
of the each method, we present, in this article, a multi- duration of the chirp signal, where N is related to the
modal SHM system where the three methods are driven number of cycles (Nc) of the signal itself N = 2Nc fs/( f1
by a single hardware/software. + f2). Two PZTs were used to monitor an aluminum
Owing to the widespread use of AE, EMI, and plate. The SHM paradigm exploited the broadband
GUW considered individually, an exhaustive review characteristics of chirp signals to obtain, from a single
of the past literature exceeds the scope of this article. measurement, the response associated with pitch–catch
However, it is interesting to report those researchers and pulse-echo configurations as well as the electrome-
who have succeeded at integrating two techniques at chanical response of the actuator. One advantage of
least. Some authors49–58 used the impedance and the using a chirp over a tone burst is the availability of a
wave propagation methods separately utilizing differ- large number of tone bursts with a single measurement
ent and disjointed hardware, software, and sensing. and the ability to select a virtually infinite number of
Park et al.53,54 utilized two PZTs to monitor rail- frequency ranges for the EMI analysis.
roads. Zagrai et al.55 integrated the acoustoelastic In the study presented in this article, we transformed
effect and the EMI to assess bolted structures, and the monitoring scheme introduced in66 by: (1) adding
for the EMI measurement, a conventional impedance the passive method of AE; (2) using an array of six
analyzer was used. Other authors executed pulse-echo PZTs instead of two; (3) processing the ultrasonic data
measurements by adding hardware55 or by placing to map the area of the structure bounded by the array;
two transducers close to each other acting as trans- (4) implementing a source localization algorithm in
mitter and receiver.59–61 LabVIEW; and (5) quantifying the repeatability of the
In previous works,43,56,62–66 the GUW-based and the EMI method. The long-term vision is a system that
EMI-based methods were employed together. Zhu and runs the passive AE mode, in which n transducers listen
Rizzo43 designed a low-cost circuit coupled to a to any burst emitted by the onset of new damage or the
National Instruments (NI)-PXI driven by an user growth of existing ones. When a burst is detected, the
Nasrollahi et al. 3

Table 1. Distance from each damage location to each PZT.

Damage Distance (mm)


P1 (762, 874) P2 (914, 610) P3 (762, 345) P4 (457, 345) P5 (305, 610) P6 (457, 874)

d1 305 305 305 305 305 305


d2 255 152 255 400 445 400
d3 405 405 330 220 210 315
d4 152 405 550 550 405 152
d5 210 210 510 687 687 510
d6 560 625 520 280 40 345

PZT: lead zirconate titanate.


The green cells denote the closest PZT to the damage, and the orange cells refer to the farthest one. The numbers in parenthesis represent the x, y
coordinates of the PZTs. All numbers are expressed in mm.

unit processes the AE data and then runs the active


sensing. Here, the EMI probes the region nearby the
actuators, whereas the GUWs create the image of the
plate in search of defects in the far field.
The article is structured as follows: section
‘‘Materials and test setup’’ describes the experimental
setup. Section ‘‘Data processing’’ explains the metho-
dology used for the signal processing. Section
‘‘Experimental results’’ presents the results. Section
‘‘Conclusion’’ ends the article with some concluding
remarks.

Materials and test setup


We tested a 1219 mm 3 1219 mm 3 2.54 mm alumi-
num plate with an array of six 10 mm 3 10 mm 3
2 mm PSI-5A4E transducers labeled P1, ..., P6 and
bonded to the plate at the location indicated in
Table 1. Each PZT was located 305 mm away from the
centroid of the plate. To avoid permanent damage on Figure 1. Experimental setup. Location of the PZTs and
simulated damage. The coordinates are expressed in mm.
the specimen, the passive monitoring was proven by
simulating the occurrence of an event with ultrasonic
bursts emitted by a commercial ultrasonic transducer
(Olympus V103) connected to a function generator and calibrate the sensitivity of the AE transducers deployed
a linear amplifier EP-104 and with the conventional in real structures.
pencil lead break (PLB) test. The ultrasonic bursts were For the active sensing, damage was mimicked by
used to compare the information provided by our inte- gluing two 25.4 mm 3 25.4 mm 3 1.27 mm aluminum
grated unit with the information provided by the para- strips on the same side of the plate. The strips were
metric analysis of a commercial AE hardware/software. sequentially detached and glued at six different loca-
The comparison included the location of the event and tions (d1, ..., d6) in order to assess the reliability of the
the quantitative evaluation of AE parameters such as active monitoring scheme. Figure 2 displays the case
number of counts, number of events, and AE energy. relative to the strip at position d6. The Euclidean dis-
For the sake of space and owing to the scope of this tance between the simulated damage and the PZTs is
article, the findings of this comparison are not pre- presented in Table 1. The shortest and the longest dis-
sented here. The use of the PLB is instead a conven- tances are highlighted for convenience. The PZTs were
tional approach to mimic real AE events and to driven by a NI-PXI unit running under LabVIEW. The
4 Structural Health Monitoring

presented later in section ‘‘Sound source localization


(AE approach),’’ two graphical user interfaces (GUIs)
were created. The software included a digital bandpass
filter, 50–550 kHz, to remove unwanted noise. The trig-
ger of the digitizer was arbitrarily set to Channel 1 and
the threshold was equal to 10 mV, which is a common
value adopted in AE commercial systems used in a
laboratory setting. Each channel was pre-triggered by
40% of the data sample length to avoid any truncation
of the real signals.

Figure 2. Photo of the test specimen when the damage was


Hardware design: active sensing (EMI and GUW)
located at position d6, very close to PZT 5. The EMI and the guided wave methods were unified
using the NI-PXI unit, and the auxiliary circuit is
PXI included an AWG NI PXI-5412, an eight-channel shown in Figure 3(a). The overall scheme of the active
digitizer NI PXI-5105, and a controller NI PXIe-8108. sensing mechanism is presented in Figure 3(b), which
refers to the case where P2 served for the EMI sensing
and for the excitation of the guided waves, and the
Hardware design: passive sensing (AE) remaining transducers acted as GUW sensors. The
In the passive AE approach, six PZTs sensed the Lamb AWG produced an user-defined chirp signal x0(t) that
waves x1(t) to x6(t) triggered by any AE event. A con- was 8 ms long, 2000 cycles, 24 Vpp, frequency range
trol panel was coded in LabVIEW to digitize and pro- 90–200 kHz, and duration N/fs equal to 100 ms. The
cess the detected signals in order to locate the event. chirp reached the circuit and the Channel 0 of the digi-
With the purpose of implementing the algorithm tizer where the signal was sampled at 2 MHz rate and

Figure 3. (a) Auxiliary circuit used to perform the active sensing approaches and (b) schematics of the hardware used to perform
ultrasonic monitoring and EMI measurements. In the figure, r1 and r2 represent the electrical resistances of the cables and Zin is the
impedance of the digitizer.
Nasrollahi et al. 5

shows the experimental waveforms detected by sensor


P1 when a 300-kHz tone burst was emitted at location
d3, and Figure 4(b) shows the corresponding Hilbert
transform.
These transforms were used to determine the arrival
of the first arrival73 that was then used to calculate the
difference Dtij of the first signal that hit PZTs i and j.
From Dtij and the known group velocity of the S0
mode, the difference Ddij = Dtij 3 cg of the path dis-
tances between the source and the receivers i and j was
computed. The dominant frequency of the first arrival
was determined with the fast Fourier transform (FFT)
and used to retrieve the appropriate value of cg from a
library of Lamb wave dispersion curves integrated in
the unit. These dispersion curves were created using the
SAFE (semi-analytical finite element) approach.74,75
With the value of Ddij, the two coefficients
 
Ddij di  dj 
aij = = ð2aÞ
2 2
Figure 4. (a) Acoustic emissions detected by the array when a
300-kHz tone-burst event occurred at location d3 and and
(b) corresponding Hilbert transforms. Owing to the position of
d3 with respect to the sensors, the S0 mode and any other qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
trailing waves hit the transducers P4 and P5 first. bij = c2ij  a2ij ð2bÞ

were calculated. In equation (2), di and dj are the


0.5 ms pre-trigger. In the circuit, x0(t) passed through a unknown sources to PZTs i and j distance, respectively,
resistor Rs = 1967 O in order to run the EMI measure- and cij is the known half distance between the two
ments and then was split into the xa(t) signal to excite PZTs. Finally, these coefficients were used to find the
the transmitter and the xi(t) signal sent to the channel equation of the hyperbola
of the digitizer associated with the actuator. The resis-
tor enabled the EMI measurements because it distribu- x20 y20
ted the input voltage between itself and the PZT.  =1 ð2cÞ
a2ij b2ij

Data processing For each pair of PZTs, two branches of the hyper-
bola exist. For illustrative purposes, Figure 5 presents
Sound source localization (AE approach) the branches associated with sensor pairs (P1, P3) and
Typical commercial AE systems embed algorithms to (P2, P5). The four intersections identify the possible
perform source localization. When the structure being location of the AE event. To identify the true location,
monitored is a plate-like structure, these localization the area within the array was virtually divided into four
techniques are sometimes referred to as planar localiza- quadrants. Upon the occurrence of an AE event, the
tion,67 and their accuracy is strongly dependent on the two smallest arrival times ti and tj identified the quad-
knowledge of the group velocity (cg) of the Lamb rant affected by the event and selected automatically
waves. Typically, only the S0 and A0 Lamb wave the pairs of transducers to compute the hyperbolas. If
modes are detected because of the frequency–thickness quadrant Q1 or Q4 was identified, the sensor pairs
range associated with the AE events. (P1, P3) and (P2, P5) were selected; otherwise, the pairs
In this study, we used the hyperbola method68–71 (P4, P6) and (P2, P5) were considered. Note that the
that is sometimes referred to as ‘‘Classical source locali- pair (P2, P5) was always chosen because both sensors
zation method.’’72 lie on the boundary of two quadrants. Combining the
In this study, the position of the AE event was identi- information about the quadrant and the intersection of
fied by considering the arrival of the S0 mode because it the hyperbolas, the location of the AE event was deter-
is non-dispersive in the frequency range of the expected mined. The example of Figure 5 illustrates the case of
bursts and faster than the dispersive A0. Figure 4(a) an event occurring in third quadrant.
6 Structural Health Monitoring
( h i
Aw½m sin 2pf mfs m = 0, 1, . . . , N  1
x(tb)
0 ½m = ð5Þ
0 m = N, . . . , M  1

where w[m] is the modulating window function, f is the


central frequency, A is the amplitude, and N is equal to
Ncfs/f where Nc is the number of tone-burst cycles.
Then, we calculated the FFT Xi(tb)[k] of the response
signals to the tone-burst signal

Xi(tb) ½k = Hi0(c) ½kX0(tb) ½k, i = 1, 2, . . . , NT ð6Þ


This FFT was then used to calculate the response
xi(tb)[m] to the tone burst with the inverse FFT of equa-
tion (6), that is
 (tb)
x(b)
i ½m = IFFT Xi ½k , i = 1, 2, . . . , NT ð7Þ

The sensor PZTj, where j = 1, 2, ..., NT j 6¼ i (i is


the actuator), captured the guided waves propagating
Figure 5. AE localization algorithm. along the plate. The response of PZT as if a tone burst
was excited by the transducer PZTi was obtained using
Waveform reconstruction algorithm the transfer function Hi0[k] associated with the chirp.
The FFT Xsj[k] of xsj[k] at the sensor’s electrodes was
For the active sensing part, we used the approach pro-
related to the FFT Xj[k] of the output signal xj[k] using
posed by Moll and Fritzen76 and Michaels et al.77 to
the relationship as follows
extract the narrowband components of the detected
GUWs and to derive the electromechanical response of r2 + jZjin ½k
the actuators within a certain frequency range. The Xsj ½k = Xj ½k, j = 1, 2, . . . , NT j j 6¼ i ð8Þ
jZjin ½k
parameters Nc, f1, f2, and fs were set through one of the
user-friendly panels. Then, the FFT of the transmitted If r2 \\ Zin
and detected chirps was computed as follows
Xsj(tb) ½k’Xj(tb) ½k = Hj0(c) ½kX0(tb) ½k, j = 1, 2, . . . , NT j j 6¼ i
X0 ½k = FFT fx0 ½mg ð3aÞ
ð9Þ
Xi ½k = FFT fxi ½mg, i = 1, . . . , NT ð3bÞ
and
where m = 0, 1, ..., M 2 1 (M is the length of the sig- n o
nal, M = fsttotal), NT is the number of the transducers, Xsj(tb) ½m = IFFT Xsj(tb) ½k , j = 1, 2, . . . , NT j j 6¼ i
that is, 6 in this study, and subscript 0 refers to the chirp
generated by the AWG. The frequency response func- ð10Þ
tion (FRF) Hi0[k] was then calculated as the ratio of the Using the above algorithm, the FRFs were used in
FFT of each signal to the FFT X0[k], that is the EMI processing. In addition, the pitch–catch
responses of the PZTs were obtained and applied to
Xi(c) ½k damage localization algorithm based on a delay-and-
Hi0(c) ½k = , i = 1, . . . , NT ð4Þ
X0(c) ½k sum algorithm.
The superscript (c) indicates that the FRFs were
obtained by measuring the responses associated with EMI analysis
the excitation of the chirp. The overall system, includ- In general, the conductance Zi,PZT of the ith transducer
ing the circuit and the waveguide, could be assumed is related to its FRF response through the following
linear and time invariant. As such, the FRFs enabled relationship66
to find the response of the sensors to any input, pro-
vided that the frequency content of such input falls Zin ½kZi, PZT ½k
between the frequency range f1–f2 of the chirp. Hi0(c) ½k = ,
ðRs + r1 Þðr2 + Zin ½kÞ + Zi, PZT ½kðRs + r1 + r2 + Zin ½kÞ
To reconstruct the tone-burst (tb) response from the
i = 1, 2, . . . , NT
sensed chirps, an input tone burst was defined first as
ð11Þ
Nasrollahi et al. 7

Table 2. The results of commercial transducers and the corresponding errors.

Location d1 d2 d3 d4
x (mm) y (mm) x (mm) y (mm) x (mm) y (mm) x (mm) y (mm)

Detected 607 610 756 609 512 558 610 853


Damage 610 610 755 610 510 560 610 874
Error (mm) 3 1 3 21
Error ratio (%) 0.49 0.16 0.49 3.44

The numbers for each detected damage are the average of nine measurements.

Solving equation (11) for Zi,PZT, we obtained

Hi0(c) ½kZin ½kRS


Zi, PZT ½k = , i = 1, 2, . . . , NT
Zin ½k  Hi0(c) ½kðRS + Zin ½kÞ
ð12Þ

The resistances r1 and r2 were neglected because


they were much smaller than RS. Once the Zi,PZT was
determined for both the pristine and faulty condi-
tions, the root-mean-square deviation (RMSD)78,79
defined as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP
u
G ½i  G ½i 2
n
u p d
ui = 1
RMSD(%) = u P n
2 3100 ð13Þ
t
Gp ½i
i=1

was calculated. In equation (13), Gp[i] and Gd[i] repre-


sent the conductance of the PZT in the pristine and
the undamaged states, respectively, and n is the num-
ber of the sampled points in the considered frequency
range. The analysis was conducted in the frequency
range of 91–94 kHz as this range was sufficiently far
from the PZT resonances and had many peaks that
serve as indicators of the vibration modes of the host
structure.66 Figure 6. Delay-and-sum algorithm: (a) general principle of an
ultrasonic wave scattered by damage and (b) virtual grid
overlapped with the test specimen.
Delay-and-sum imaging algorithm
The delay-and-sum technique80–85 was applied to the 2. Calculate the expected arrival time (t = t1 + t2)
reconstructed tone bursts in order to create a tomo- associated with each point (x, y) of the grid using
graphic image of the plate. The technique is schema-
tized in Figure 6(a). The presence of damage scatters tði, j, x, yÞ
the waveform emitted by PZT i. The scattered signal is qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

2
2ffi
received by PZT j at the time required for the wave to ðx  xi Þ2 + ðy  yi Þ2 + x  xj + y  yj
travel from the transmitter to the scatterer (t1) and from = + toff
cg
the scatterer to the receiver (t2). The following steps
ð14Þ
were implemented:
where (xi, yi) and (xj, yj) are the coordinates of the ith
1. Create a virtual grid of points on the plate actuator and the jth receiver, respectively, and toff is the
(Figure 6(b)) located at coordinate (x, y). offset time.
8 Structural Health Monitoring

3. Scale the amplitude of the received signal to remove


the effect of signal attenuation due to the path
distance
0 11=2
S ði, j, mÞ B d0 C
S^ði, j, mÞ = @
2
2 1=2 A
Aij
xi  xj + xi  xj
ð15Þ
where S(i, j, m) is the signal transmitted by PZT ith
and received by sensor jth, d0 is the unit distance (e.g.
1 m in this study), and Aij is the maximum amplitude
of S(i, j, m).

4. Compute the difference of the scaled signals Figure 7. Acoustic emissions detected by the array under the
between the pristine (p) and the damaged state pencil lead break test.

^ j, m)  S^p (i, j, m)
U (i, j, m) = S(i, ð16Þ
results of the localization algorithm for each simu-
5. Calculate the Hilbert transform of the signal differ- lated event. The small relative error between the true
ence and the corresponding modulus and the estimated locations of the event proves the
effectiveness of the algorithm under the case of nar-
rowband signals. The error is the difference in the
C(i, j, m) = Hilbert ðU (i, j, m)Þ = U (i, j, m) + ð1Þ0:5 V (i, j, m) Euclidean distance of the true and the estimated dam-
ð17Þ age from the origin of the reference system, located at
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the bottom left corner of the plate. The error ratio is
D(i, j, m) = j C(i, j, m) j = ðU (i, j, m)Þ2 + ðV (i, j, m)Þ2 instead the ratio between the error and the farthest
ð18Þ distance (610 mm) between two PZTs of the array.
The results relative to the PLB test are instead pre-
where V(i, j, m) is the harmonic conjugate of U (i, j, m). sented in Figures 7 and 8. Figure 7 displays the wave-
forms received when the event was simulated at point
6. Calculate the image intensity value E(x, y) for each d3. The waveforms mainly contain the A0 mode; how-
point of the grid ever, the amplitude of the S0 mode was sufficient to be
considered for the analysis. The wave hits PZT 5 first
and PZT 2 last. Figure 8 shows the estimated location
X X
1 NT1 NT
of the PLB. The hollow circles represent the estimated
E(x, y) = Dði, j, mt Þ ð19Þ
NP i = 1 j = i + 1 locations, while the green asterisks represent the true
locations. Despite the challenge associated with the low
where NP=NT(NT 2 1)/2 is the number of the trans- amplitude of the symmetric mode and the broadband
ducer pairs and mt=fst(i, j, x, y). The image value is nature of the excitation, the algorithm was able to
expected to be large when the grid point overlaps with detect the location of the event very precisely. To quan-
the location of damage, that is, when the wave is scat- tify the relative inaccuracies, the results of Figure 8 are
tered by damage as part of its energy reaches the also presented in Table 3 where the coordinates of the
sensor. true and the estimated damage are presented along
with the error and the error ratio. The fact that the
Experimental results maximum error is only 2.13% demonstrates the accu-
racy of the passive system.
AE monitoring
One Olympus V103 ultrasonic transducer was fixed at
the locations d1–d4 shown in Figure 1. At each loca-
EMI measurements
tion, a five-cycle sine wave from 100 to 500 kHz at In the literature, the use and the analysis of multiple
50 kHz step was excited. Thus, nine signals were PZTs were rarely considered. To the best of the
detected at each location. Table 2 summarizes the authors’ knowledge, typical EMI experimental
Nasrollahi et al. 9

Table 3. The results of PLB and the corresponding errors.

Location d1 d2 d3 d4
x (mm) y (mm) x (mm) y (mm) x (mm) y (mm) x (mm) y (mm)

Detected 611 609 756 602 512 548 598 868


Damage 610 610 755 610 510 560 610 874
Error (mm) 1 8 12 13
Error ratio (%) 0.16 1.31 1.96 2.13

PLB test: pencil lead break test.

monitoring is conducted using one PZT. When two or


more PZTs were used, there was no direct cross-
comparison among the values of the conductance and
susceptance of the PZTs. Moreover, damage detection
was performed by comparing the conductance and
eventually the susceptance of each PZT with their cor-
responding baseline values. In this study, we began by
comparing the responses of the six PZTs under free
boundary conditions and after bonding them to the
pristine plate. The results are presented in Figure 9.
The signatures of the free PZTs (Figure 9(a)) show that
the transducers were approximately identical in terms
of amplitude and resonant frequency. As expected, the
amplitudes at resonance were significantly smaller
when the PZTs were bonded to the plate (Figure 9(b)).
As discussed earlier, the quantitative analysis of the
conductance was restricted to the 91- to 94-kHz fre-
quency range. To quantify the differences in this range,
Figure 8. Results of the AE localization under pencil lead break the RMSD of the conductance with respect to PZT 5
test. was computed and is presented in Figure 10(a). PZT 5
was selected as its distance to P6 and P1 was identical

Figure 9. Conductance as a function of the excitation frequency of the six transducers under (a) free conditions and (b) glued to
the pristine plate.
10 Structural Health Monitoring

Figure 10. (a) RMSD of the conductance of the transducers prior and after bonding them to the pristine plate. The conductance
associated with PZT 5 was considered as reference. (b) RMSD of the conductance for all transducers and for all damage scenarios.
In this case, for each PZT, the baseline data are those collected under pristine condition and measured by each transducer.

to its distance to transducers P4 and P3, respectively. Ultrasonic imaging


Moreover, its location with respect to the plate edges Figure 11 presents the waveforms reconstructed when
was identical to the location of PZT 2. The plot sug- P1 was probing the pristine plate. The waveforms refer
gests that the differences among the PZTs were mainly to the reconstruction of a five-cycle 100- and 170-kHz
due to manufacturing. In fact, if the PZTs were identi- tone bursts. The electromagnetic crosstalk (emc),
cal, the RMSD under free conditions should have
almost identical for all PZTs, is followed by the arrival
returned a value equal to zero. Once bonded to the
of the S0 and A0 modes, respectively. The wave packets
plate, these differences remain similar.
traveling along the line of sight between the transmitter
The comparative analysis suggests that some differ-
and each receiver are labeled along with some of the
ences may exist among the PZTs of an array. As such,
waves reflected from the edges.
the absolute value of the conductance may not be effec-
The reconstructed waveforms were used in the
tive at revealing the presence of damage. To prove it,
delay-and-sum algorithm described earlier. The results
Figure 10(b) shows the RMSD associated with each
relative to the localization of damage 2 are presented in
transducer and for all six damage scenarios. The index
Figure 12 for six different reconstructed tone bursts.
was computed by comparing the conductance of the
The figure demonstrates that the frequency to be con-
damaged case to the conductance taken under pristine
sidered should be close to the resonant frequency of the
condition with the same PZT. The bar relative to the
PZTs which Figure 9 proved to be ;170 kHz. Below
undamaged case refers to two consecutive measure-
150 kHz, the damage was not localized partially
ments; ideally, this value should have been zero.
because the propagating wavelength is larger than the
Overall, the values of the statistical index are smaller
size of added mass, and therefore the signal-to-noise
than those presented in Figure 10(a) but they are all
ratio (SNR) of the scattered signal is too small, and
above 0.3% which is the RMSD relative to the pristine
mainly because the SNR of the transmitted and propa-
condition. This means that the EMI method could
gating signals was small due to the resonant character-
detect the presence of the added mass, provided each
istics of the PZTs, as shown in Figure 11. For example,
PZT was compared to its own baseline value. Notably,
at 150 kHz, the wavelength of the S0 mode is approxi-
for each transducer, the highest RMSD corresponds
mately 33 mm. It should be noted that the SNR is also
with the closest damage scenario, labeled as C in the
small because the resonance of the PZTs is close to
figure, and vice versa. When the added mass was only
170 kHz, as shown in Figures 9 and 11. For this rea-
40 mm from PZT 5, the RMSD was the largest.
son, any frequency higher than the resonant frequency
Overall, the data presented in Figure 10(b) seem to indi-
of the PZT is also detrimental for the detection of
cate that the RMSD is inversely proportional to the
damage.
Euclidean distance of the added mass to the sensors.
Nasrollahi et al. 11

In the future, the chirp signal should be tailored to


include only those frequencies of interests for the mea-
surements of the sensors electrical impedance and the
Lamb waves with the highest SNR.
Finally, Figure 13 shows the ultrasonic images rela-
tive to damage scenario 6 and 170 kHz frequency. The
images demonstrate that when the damage was very
close to a transducer, the delay-and-sum algorithm
failed to detect it. When the damage is close to a PZT
and far from the other sensors, the amplitude of the
scattered wave attenuates below the noise floor before
reaching the sensors or it is masked by the amplitude
of the direct transmitter–receiver signal. Moreover, the
presence of the PZT itself masks any small defect
located nearby. This circumstance proves that the inte-
gration of EMI and GUW methods enables the locali-
zation of damage regardless of the damage position
with respect to the transducers.

Conclusion
This article presented an integrated SHM system based
on an array of wafer transducers and smart data acqui-
sition system able to run passive sensing based on AE,
and active sensing based on EMI and propagation and
detection of GUWs. The system was tested on a large
Figure 11. Reconstructed pitch–catch signals for the actuation aluminum plate subjected to simulated damage by glu-
frequency of (a) 100 kHz and (b) 170 kHz. Note the vertical ing a small mass on the surface of the specimen (active
scale. The latter signals are 10 times larger. sensing) and by exciting narrowband and broadband

Figure 12. Tomographic images of the plate when the added mass was located at d2 using the reconstructed tone burst centered
at (a) 90 kHz, (b) 110 kHz, (c) 130 kHz, (d) 150 kHz, (e) 170 kHz, and (f) 190 kHz.
12 Structural Health Monitoring

Moreover, the hardware/software can be improved to


take into account the fact that the different methods
may have different frequency ranges. In the experiment
presented in this study, the data associated with the
EMI and the GUW were analyzed at different ranges.
In the future, the use of the chirp to drive the active
modes may be refined in order to reduce the computa-
tion time associated with the EMI and GUW
approaches. For the AE setup, instead the frequency
range is strongly dependent on the geometry and the
material of the structure under investigation, and there-
fore, the multimodal system would need to be adapted
to the specific application. We do not exclude that
more sophisticated AE localization algorithms should
Figure 13. Tomographic images of the plate when the added be implemented into our integrated system when more
mass was located at d6 using the reconstructed tone burst complex geometries are to be monitored.
centered at 170 kHz.
Acknowledgements
signals (passive sensing). We found that the system can The authors thank Mr Vincenzo Gulizzi at the University of
efficiently integrate the three sensing approaches and Palermo, Italy for many fruitful discussions. W.D. conducted
can effectively locate the simulated AE events and this research as a visiting scholar at the University of
Pittsburgh under a sponsorship from the China Scholarship
detect the presence of the added mass, which was as
Council. Z.M. conduct this research at the University of
small as less than 1% of the area of the plate. This arti- Pittsburgh as part of her M.S.
cle demonstrated that the integrated system could be
used synergistically using the array to passively detect
any burst generated by damage onset or growth (AE Declaration of conflicting interests
approach) and then using the array to actively measure The author(s) declared no potential conflicts of interest with
the electrical impedance of the transducers and to sense respect to the research, authorship, and/or publication of this
the guided waves sensed by the array. Using the three article.
methods altogether, the structure can be monitored in a
fraction of the time that would occur to run the three Funding
methods individually using separate instruments. The author(s) disclosed receipt of the following financial sup-
Moreover, the integrated system maintains, if not port for the research, authorship, and/or publication of this
enhances, the advantages and overcomes the limitations article: This study was supported by the American Society for
that each method possesses when considered individu- Nondestructive Testing through a graduate research fellow-
ally. As such, we expect that the proposed multimodal ship awarded to the A.N., the first author.
system can be applied to anisotropic materials as well
as can detect defects located outside the area within the References
array of transducers. For the latter, the image process-
1. McCrory JP, Al-Jumaili SK, Crivelli D, et al. Damage
ing or the passive localization algorithms may need to
classification in carbon fibre composites using acoustic
be modified to improve the outcomes. emission: a comparison of three techniques. Compos Part
We want to emphasize that the GUWs can also be B: Eng 2015; 68: 424–430.
processed in a pulse-echo mode, like a radar system, in 2. Al-Jumaili SK, Pearson MR, Holford KM, et al. Acous-
which the same PZT excites the tone burst and detects tic emission source location in complex structures using
any echo reflected from a defect, and in a pitch–catch full automatic delta T mapping technique. Mech Syst Sig-
mode where two PZTs communicate with each other nal Pr 2016; 72: 513–524.
such that one PZT transmits the signal and the other 3. Crivelli D, Guagliano M, Eaton M, et al. Localisation
receives the same signal and a comparison is made and identification of fatigue matrix cracking and delami-
between current situation and some baseline data; using nation in a carbon fibre panel by acoustic emission. Com-
imaging algorithms as done in this article. As such, the pos Part B: Eng 2015; 74: 1–12.
4. Clark G and Knott J. Acoustic emission and ductile
multimodal integrated system can exploit up to five dif-
crack growth in pressure-vessel steels. Met Sci 2013; 11:
ferent methodologies. 531–536.
Future studies shall consider more complex struc- 5. Barile C, Casavola C, Pappalettera G, et al. Analysis of
tures and test the effect of temperature and moisture. crack propagation in stainless steel by comparing
Nasrollahi et al. 13

acoustic emissions and infrared thermography data. Eng 23. Alleyne DN and Cawley P. The interaction of Lamb
Fail Anal 2016; 69: 35–42. waves with defects. IEEE T Ultrason Ferr 1992; 39:
6. Agletdinov E, Pomponi E, Merson D, et al. A novel 381–397.
Bayesian approach to acoustic emission data analysis. 24. Rose JL. Ultrasonic waves in solid media. Cambridge:
Ultrasonics 2016; 72: 89–94. Cambridge University Press, 2004.
7. Babu MN, Mukhopadhyay C, Sasikala G, et al. Study of 25. Niethammer M, Jacobs LJ, Qu J, et al. Time-frequency
fatigue crack growth in RAFM steel using acoustic emis- representations of Lamb waves. J Acoust Soc Am 2001;
sion technique. J Constr Steel Res 2016; 126: 107–116. 109: 1841–1847.
8. Prem PR and Murthy AR. Acoustic emission and flexural 26. Giurgiutiu V. Tuned Lamb wave excitation and detection
behaviour of RC beams strengthened with UHPC over- with piezoelectric wafer active sensors for structural
lay. Construct Build Mater 2016; 123: 481–492. health monitoring. J Intel Mat Syst Str 2005; 16:
9. Ohtsu M. Elastic wave methods for NDE in concrete 291–305.
based on generalized theory of acoustic emission. Con- 27. Rizzo P and di Scalea FL. Ultrasonic inspection of multi-
struct Build Mater 2016; 122: 845–854. wire steel strands with the aid of the wavelet transform.
10. Park G, Sohn H, Farrar CR, et al. Overview of piezoelec- Smart Mater Struct 2005; 14: 685.
tric impedance-based health monitoring and path for- 28. Su Z, Ye L and Lu Y. Guided Lamb waves for identifica-
ward. Shock Vib Digest 2003; 35: 451–463. tion of damage in composite structures: a review. J Sound
11. Peairs DM, Park G and Inman DJ. Improving accessibil- Vib 2006; 295: 753–780.
ity of the impedance-based structural health monitoring 29. Raghavan A and Cesnik CE. Review of guided-wave
method. J Intel Mat Syst Str 2004; 15: 129–139. structural health monitoring. Shock Vib Digest 2007; 39:
12. Bhalla S and Soh C. Electro-mechanical impedance tech- 91–116.
nique. In: Soh CK, Yang Y and Bhalla S (eds) Smart 30. Rizzo P, Han J-G and Ni X-L. Structural health monitor-
materials in structural health monitoring, control and bio- ing of immersed structures by means of guided ultrasonic
mechanics. Berlin, Heidelberg: Springer, 2012, pp. 17–51. waves. J Intel Mat Syst Str 2010; 21: 1397–1407.
13. Gulizzi V, Rizzo P and Milazzo A. On the repeatability 31. Rizzo P and Lanza di Scalea F. Effect of frequency on
of electromechanical impedance for monitoring of the acoustoelastic response of steel bars. Exp Techniques
bonded joints. AIAA J 2015; 53: 3479–3483. 2003; 27: 40–43.
14. Na S, Tawie R and Lee H-K. Electromechanical impe- 32. Rizzo P, Spada A, Degala S, et al. Acoustic emission
dance method of fiber-reinforced plastic adhesive joints monitoring of chemically bonded anchors. J Nondestruct
in corrosive environment using a reusable piezoelectric Eval 2010; 29: 49–61.
device. J Intel Mat Syst Str 2012; 23: 737–747. 33. Degala S, Rizzo P, Ramanathan K, et al. Acoustic emis-
15. Gulizzi V, Rizzo P and Milazzo A. Electromechanical sion monitoring of CFRP reinforced concrete slabs. Con-
impedance method for the health monitoring of bonded struct Build Mater 2009; 23: 2016–2026.
joints: numerical modeling and experimental validation. 34. Wang C, Au YJ, Li L, et al. Guided wave mode disper-
Struct Durab Health Monit 2014; 10: 19–54. sion of transient acoustic emission on copper pipes—its
16. Lim YY and Soh CK. Electro-mechanical impedance visualisation and application to source location. Mech
(EMI)-based incipient crack monitoring and critical crack Syst Signal Pr 2016; 70: 881–890.
identification of beam structures. Res Nondestruct Eval 35. Murasawa G, Takahashi R, Morimoto T, et al. Inhomo-
2014; 25: 82–98. geneous deformation twinning measurement using digital
17. Selva P, Cherrier O, Budinger V, et al. Smart monitoring image correlation and acoustic emission. Exp Mech 2015;
of aeronautical composites plates based on electromecha- 55: 65–76.
nical impedance measurements and artificial neural net- 36. LédecziÁ, Hay T, Volgyesi P, et al. Wireless acoustic
works. Eng Struct 2013; 56: 794–804. emission sensor network for structural monitoring. IEEE
18. Talakokula V, Bhalla S and Gupta A. Corrosion assess- Sens J 2009; 9: 1370–1377.
ment of reinforced concrete structures based on equivalent 37. Maji A and Shah S. Process zone and acoustic-emission
structural parameters using electro-mechanical impedance measurements in concrete. Exp Mech 1988; 28: 27–33.
technique. J Intel Mat Syst Str 2014; 25: 484–500. 38. Perelli A, De Marchi L, Marzani A, et al. Acoustic emis-
19. Adams D. Health monitoring of structural materials and sion localization in plates with dispersion and reverbera-
components: methods with applications. Hoboken, NJ: tions using sparse PZT sensors in passive mode. Smart
John Wiley & Sons, 2007. Mater Struct 2012; 21: 025010.
20. Giurgiutiu V. Structural health monitoring: with piezoelec- 39. Giurgiutiu V, Bao J and Zhao W. Piezoelectric wafer
tric wafer active sensors. New York: Academic Press, 2007. active sensor embedded ultrasonics in beams and plates.
21. Balageas D, Fritzen C-P and Güemes A. Structural health Exp Mech 2003; 43: 428–449.
monitoring. New Port Beach, CA, USA: John Wiley & 40. Zhu X, Hao H and Fan K. Detection of delamination
Sons, 2010. between steel bars and concrete using embedded piezo-
22. Farrar CR and Worden K. Structural health monitoring: electric actuators/sensors. J Civ Struct Health Monit
a machine learning perspective. New York: John Wiley & 2013; 3: 105–115.
Sons, 2012.
14 Structural Health Monitoring

41. Zhu X and Rizzo P. Sensor array for the health monitor- monitoring, San Diego, California, USA, 6 March 2011,
ing of truss structures by means of guided ultrasonic p. 79811Q. Bellingham, WA: International Society for
waves. J Civ Struct Health Monit 2014; 4: 221–234. Optics and Photonics.
42. Giurgiutiu V and Santoni-Bottai G. Structural health 57. Cuc A, Giurgiutiu V, Joshi S, et al. Structural health
monitoring of composite structures with piezoelectric- monitoring with piezoelectric wafer active sensors for
wafer active sensors. AIAA J 2011; 49: 565–581. space applications. AIAA J 2007; 45: 2838–2850.
43. Zhu X and Rizzo P. A unified approach for the structural 58. Sharif Khodaei Z, Ghajari M, Aliabadi M, et al. SMART
health monitoring of waveguides. Struct Health Monit platform for structural health monitoring of sensorised
2012; 11: 629–642. stiffened composite panels. Key Eng Mater 2013;
44. Wang Q and Wang C. Optimal placement and size of 525–526: 581–584.
piezoelectric patches on beams from the controllability 59. Yang M and Qiao P. Modeling and experimental detec-
perspective. Smart Mater Struct 2000; 9: 558. tion of damage in various materials using the pulse-echo
45. Baptista FG, Vieira Filho J and Inman DJ. Sizing PZT method and piezoelectric sensors/actuators. Smart Mater
transducers in impedance-based structural health moni- Struct 2005; 14: 1083.
toring. IEEE Sens J 2011; 11: 1405–1414. 60. Raghavan A and Cesnik CE. Guided-wave signal pro-
46. Bhalla S and Soh CK. Electromechanical impedance cessing using chirplet matching pursuits and mode corre-
modeling for adhesively bonded piezo-transducers. J Intel lation for structural health monitoring. Smart Mater
Mat Syst Str 2004; 15: 955–972. Struct 2007; 16: 355.
47. Park G and Inman DJ. Structural health monitoring 61. Shen Y and Giurgiutiu V. WaveFormRevealer: an
using piezoelectric impedance measurements. P Roy Soc analytical framework and predictive tool for the simula-
A: Math Phy 2007; 365: 373–392. tion of multi-modal guided wave propagation and inter-
48. Koo K-Y, Park S, Lee J-J, et al. Automated impedance- action with damage. Struct Health Monit 2014; 13:
based structural health monitoring incorporating effec- 491–511.
tive frequency shift for compensating temperature effects. 62. An Y-K, Giurgiutiu V and Sohn H. Integrated impedance
J Intel Mat Syst Str 2008; 20: 367–377. and guided wave based damage detection. Mech Syst Sig-
49. Li Y, Jiang Z, Chonan S, Feng G and Wen B. Impe- nal Pr 2012; 28: 50–62.
dance-based technique and wave propagation measure- 63. An YK, Kim MK and Sohn H. Airplane hot spot moni-
ment for non-destructive evaluation. Proceedings of toring using integrated impedance and guided wave mea-
international conference on vibration engineering. Dalian, surements. Struct Control Hlth 2012; 19: 592–604.
China: Northeastern University Press, 1998, pp. 476–481. 64. Park H-J, Sohn H, Yun C-B, et al. Wireless guided wave
50. Kabeya K, Jiang Z and Cudney H. Structural health and impedance measurement using laser and piezoelectric
monitoring by impedance and wave propagation mea- transducers. Smart Mater Struct 2012; 21: 035029.
surement. In: Proceedings of the international motion and 65. Providakis CP, Stefanaki KD, Voutetaki ME, et al.
vibration control, Zurich, 25–28 August 1998. Damage detection in concrete structures using a simulta-
51. Jiang Z, Kabeya K and Chonan S. Longitudinal wave neously activated multi-mode PZT active sensing system:
propagation measuring technique for structural health numerical modelling. Struct Infrastruct E 2014; 10:
monitoring. In: Proceedings of the 1999 symposium on 1451–1468.
smart structures and materials, Newport Beach, CA, 1 66. Gulizzi V, Rizzo P, Milazzo A, et al. An integrated struc-
March 1999, pp. 343–350. Bellingham, WA: International tural health monitoring system based on electromechani-
Society for Optics and Photonics. cal impedance and guided ultrasonic waves. J Civ Struct
52. Giurgiutiu V, Zagrai A and Bao JJ. Piezoelectric wafer Health Monit 2015; 5: 337–352.
embedded active sensors for aging aircraft structural 67. Kurz HJ, Köppel S, Linzer ML, et al. Source localization.
health monitoring. Struct Health Monit 2002; 1: 41–61. In: Grosse C and Ohtsu M (eds) Acoustic emission testing:
53. Park S, Lee J-J, Yun C-B, et al. A built-in active sensing basics for research—applications in civil engineering. Ber-
system-based structural health monitoring technique lin, Heidelberg: Springer, 2008, pp. 101–147.
using statistical pattern recognition. J Mech Sci Technol 68. Kalafat S and Sause MG. Acoustic emission source loca-
2007; 21: 896–902. lization by artificial neural networks. Struct Health Monit
54. Park S, Inman DJ, Lee J-J, et al. Piezoelectric sensor- 2015; 14: 633–647.
based health monitoring of railroad tracks using a two- 69. Fendzi C, Mechbal N, Rebillat M, et al. A general Baye-
step support vector machine classifier. J Infrastruct Syst sian framework for ellipse-based and hyperbola-based
2008; 14: 80–88. damage localization in anisotropic composite plates. J
55. Zagrai A, Doyle D, Gigineishvili V, et al. Piezoelectric Intel Mat Syst Str 2016; 27: 350–374.
wafer active sensor structural health monitoring of space 70. Rossi P, Robert JL, Gervais JP, et al. Acoustic emission
structures. J Intel Mat Syst Str 2010; 21: 921–940. applied to study crack propagation in concrete. Mater
56. An Y-K and Sohn H. Integrated impedance and guided Struct 1989; 22: 374–384.
wave based damage detection under temperature varia- 71. Dunn F, Hartmann W, Campbell D, Fletcher N and
tion. In: Proceedings of the SPIE smart structures and Rossing T. Springer handbook of acoustics. New York:
materials + nondestructive evaluation and health Springer Science+Business Media LLC, 2015.
Nasrollahi et al. 15

72. Kalafat S and Sause MG. Localization of acoustic emis- review and future issues. J Intel Mat Syst Str 2010; 21:
sion sources in fiber composites using artificial neural net- 41–59.
works. 31st Conference of the European working group on 80. Michaels JE. Detection localization and characterization
acoustic Eeission. DGZIP, 2014, pp. 1–8. of damage in plates with an in situ array of spatially dis-
73. Sedlak P, Hirose Y and Enoki M. Acoustic emission loca- tributed ultrasonic sensors. Smart Mater Struct 2008; 17:
lization in thin multi-layer plates using first-arrival deter- 035035.
mination. Mech Syst Signal Pr 2013; 36: 636–649. 81. Flynn EB, Todd MD, Wilcox PD, et al. Maximum-likeli-
74. Marzani A, Viola E, Bartoli I, et al. A semi-analytical hood estimation of damage location in guided-wave
finite element formulation for modeling stress wave pro- structural health monitoring. P Roy Soc A: Math Phy.
pagation in axisymmetric damped waveguides. J Sound Epub ahead of print 6 April 2011. DOI: 10.1098/
Vib 2008; 318: 488–505. rspa.2011.0095.
75. Sale M, Rizzo P and Marzani A. Semi-analytical formu- 82. Salmanpour M, Sharif Khodaei Z and Aliabadi M.
lation for the guided waves-based reconstruction of elas- Transducer placement optimisation scheme for a delay
tic moduli. Mech Syst Signal Pr 2011; 25: 2241–2256. and sum damage detection algorithm. Struct Control
76. Moll J and Fritzen C. Advanced aspects of mode-selec- Hlth. Epub ahead of print 21 June 2016. DOI: 10.1002/
tive excitation of ultrasonic guided waves. 24th Confer- stc.1898.
ence on noise and vibration engineering, Leuven, Belgium, 83. Michaels JE. 9-sparse array imaging with guided waves
2010, pp. 969–984. ISMA. under variable environmental conditions. In: Yuan F-G
77. Michaels JE, Lee SJ, Croxford AJ, et al. Chirp excitation (ed.) Structural health monitoring (SHM) in aerospace
of ultrasonic guided waves. Ultrasonics 2013; 53: 265–270. structures. Duxford: Woodhead Publishing, 2016, pp.
78. Giurgiutiu V and Rogers CA. Recent advancements in 255–284.
the electromechanical (E/M) impedance method for 84. Michaels JE, Croxford AJ and Wilcox PD. Imaging algo-
structural health monitoring and NDE. In: Proceedings rithms for locating damage via in situ ultrasonic sensors.
of the 5th annual international symposium on smart struc- In: Proceedings of the IEEE sensors applications sympo-
tures and materials, San Diego, CA, 1 March 1998, sium, 2008 (SAS 2008), Atlanta, GA, 12–14 February
pp. 536–547. Bellingham, WA: International Society for 2008, pp. 63–67. New York: IEEE.
Optics and Photonics. 85. Liu G, Xiao Y, Zhang H, et al. Baseline signal recon-
79. Annamdas VGM and Soh CK. Application of electrome- struction for temperature compensation in Lamb wave-
chanical impedance technique for engineering structures: based damage detection. Sensors 2016; 16: 1273.

You might also like