sensors

Article
Noncontact Damage Topography Reconstruction by
Wavenumber Domain Analysis Based on Air-Coupled
Ultrasound and Full-Field Laser Vibrometer
Hui Zhang, Dongmei Liang , Xiaobo Rui * and Zhuochen Wang

State Key Laboratory of Precision Measurement Technology and Instrument, Tianjin University,
Tianjin 300072, China; hzhang@tju.edu.cn (H.Z.); dm_liang@tju.edu.cn (D.L.); zc_wang@tju.edu.cn (Z.W.)
* Correspondence: ruixiaobo@tju.edu.cn

Abstract: Noncontact ultrasonic detection technology is an effective method to detect damage in time.
This paper proposes a noncontact damage detection system based on air-coupled ultrasound and
full-field laser vibrometer, which realizes the excitation of relatively single-mode guided waves and
the wavefield automatic detection. The system performance is verified through experiments, and the
experimental wavenumber is consistent with the theoretical dispersion characteristics of the Lamb
wave in the A0 mode. Based on this system, the topography reconstruction algorithms, including the
Wavenumber Filtering Algorithm and Spatial Wavenumber Algorithm, were tested and compared
with the aluminum alloy plate and the carbon fiber reinforced polymer plate. The results show that,
based on the air-coupled ultrasound and full-field laser vibrometer detection system, the Spatial
Wavenumber Algorithm has better imaging error and contrast, and the damage edge detection is
smoother.



Keywords: damage evaluation; air-coupled ultrasound; Lamb wave; laser vibrometer; wavenumber

 domain analysis
Citation: Zhang, H.; Liang, D.; Rui,
X.; Wang, Z. Noncontact Damage
Topography Reconstruction by
Wavenumber Domain Analysis Based 1. Introduction
on Air-Coupled Ultrasound and Ultrasonic technology is widely used in structural damage assessment due to its
Full-Field Laser Vibrometer. Sensors
advantages of nondestructive and fast testing [1–3]. It can evaluate the health status of the
2021, 21, 609. https://doi.org/
structure during the whole life cycle of production, use and maintenance. Ultrasound has
10.3390/s21020609
a wide frequency range and various waveforms, which provide abundant possibilities for
nondestructive testing [4,5]. In recent years, noncontact ultrasonic nondestructive testing
Received: 2 December 2020
technology has received widespread attention, because of its advantages of no coupling,
Accepted: 14 January 2021
Published: 17 January 2021
no surface contact, and fast scanning [6,7].
Electro-Magnetic Acoustic Transducer (EMAT) technology, laser ultrasound and air-
Publisher’s Note: MDPI stays neu-
coupled ultrasound are currently three widely used noncontact ultrasonic detection meth-
tral with regard to jurisdictional clai- ods. EMAT technology is suitable for thickness measurement and defect characterization
ms in published maps and institutio- of various metal pipes or plates, and can be applied under high-temperature conditions [8].
nal affiliations. The low energy conversion efficiency, the short distance between the coil and the tested
structure limit the application and development of related technologies, and the material
to be tested is required to have good conductivity or ferromagnetism [9].
Laser ultrasound technology uses pulsed lasers to achieve high-frequency ultrasonic
Copyright: © 2021 by the authors. Li-
excitation, based on thermo-elastic and thermo-erosion effects. However, the conversion
censee MDPI, Basel, Switzerland.
efficiency of photoacoustic energy is not high, and the surface of the material will be
This article is an open access article
damaged when the laser energy is too high. In laser ultrasound, laser Doppler vibrometers
distributed under the terms and con-
are often used as testing equipment, and the Doppler frequency shift phenomenon is used
ditions of the Creative Commons At-
to measure the in-plane or out-of-plane surface vibration displacement and velocity [10].
tribution (CC BY) license (https://
The laser-related technology can perform spatial distributed detection, has a high spatial
creativecommons.org/licenses/by/
4.0/).
resolution, and can detect large-size or complex curved surfaces [11,12].

Sensors 2021, 21, 609. https://doi.org/10.3390/s21020609 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 609 2 of 18

Air-coupled ultrasound technology uses air as the coupling medium between the
transducer and the structure, which can generate bulk waves, surface waves and also Lamb
waves. Among them, the Lamb waves in a thin-plate structure can be propagated for a
long distance with low attenuation energy [13]. Air-coupled Lamb wave detection methods
have been extensively studied, including visualization of impact, delamination, debonding,
and surface damage imaging of metal and composite plates. The main limitation of air-
coupled ultrasonic transducers is the enormous acoustic impedance mismatch between the
piezoelectric material and air. Researchers are also coming up with new ways to overcome
the problem [14,15]. Each detection technology has its unique advantages and limitations.
Hybrid detection systems have the potential to combine multiple advantages and to break
through the bottleneck of existing technologies, such as laser ultrasound combined with
electromagnetic ultrasound, piezoelectric ultrasound and laser ultrasound, etc., [16].
For structural damage detection, the damage topography reconstruction of plate-
shaped structures is the most concerned. Typical algorithms include reconstruction algo-
rithm of probabilistic inspection (RAPI) [17], single-transmitter and multireceiver method
(STMR) [18], wavenumber domain imaging [19], etc. Among them, RAPI and STMR are
mainly based on the difference in the energy passability of damage to achieve location and
topography reconstruction. The wavenumber domain imaging method analyzes the full
wavefield data of the range to be measured by sampling in the time–space domain, and the
signal information is more abundant and accurate [20,21].
Juarez et al. came up with a methodology based on multifrequency local wavenumber
analysis for quantitative assessment of multi-ply delamination damage, including the
delamination size and ply depth in carbon fiber reinforced polymer (CFRP) composite
specimens [22]. Tian et al. presented a Spatial Wavenumber Algorithm through global
inspection by phased array beamforming and local damage evaluation via wavenumber
analysis [23]. The detected delamination location, size and shape agree well with those of
an ultrasonic C-scan, and the total inspection time was reduced greatly. Jeon et al. imple-
mented two signal processing methods, including Local Wavenumber Mapping (LWM)
and Acoustic Wavenumber Spectroscopy (AWS), to determine dominant wavenumber
components of the measured wavefield [24]. In this way, the structural delamination
and debonding damage were clearly and accurately identified. Kudela et al. aimed at
enhancing damage visualization in thin-walled structures, through a combined system of
the PZT transducer and the laser Doppler vibrometer [25]. They proposed a Wavenumber
Filtering Algorithm for the estimation of the length and orientation of the crack. After
that, Kudela et al. used wavenumber adaptive image filtering and signal processing to
visualize and assess the impact-induced damage location and size [26]. The results showed
that damage caused by the impact of 10 J or higher could be successfully detected using
the presented approach. Radzienski et al. developed a detection technique, where guided
waves were excited by PZT transducer, and measured by scanning laser vibrometer [27].
By means of mapping irregularities of propagating guided waves, damage characteristics
in various types of composite plates were obtained. The above-mentioned methods usually
use the traditional contact PZT combined with a vibrometer to complete the analysis in
the wavenumber domain, which lacks the flexibility of detection. So a hybrid technology
using air-coupled transmitters and laser receivers was proposed. Harb et al. confirmed that
using this system and implementing the simple Snell’s law method is highly sensitive and
effective in characterizing the dispersion curves of Lamb waves in composite structures as
well as its angular dependency [28]. In addition, they illustrated the feasibility of using
the hybrid system and the guided wave zero-lag cross-correlation imaging condition in
characterizing the barely visible impact damages in thin composite structures [29].
This paper uses a noncontact combined detection scheme, including air-coupled
ultrasound and laser vibrometer. The air-coupled ultrasonic technology can flexibly excite
the Lamb wave of a specific mode to achieve long-distance and high-sensitivity detection.
Laser vibrometer technology can obtain the vibration of the structure surface and realize
the full wavefield data detection with high spatial resolution. The above system is applied
 This paper uses a noncontact combined detection scheme, including air-coupled ul-
trasound and laser vibrometer. The air-coupled ultrasonic technology can flexibly excite
the Lamb wave of a specific mode to achieve long-distance and high-sensitivity detection.
Sensors 2021, 21, 609
Laser vibrometer technology can obtain the vibration of the structure surface and realize 3 of 18
the full wavefield data detection with high spatial resolution. The above system is applied
to the damage topography reconstruction for metal plates and composite material plates
for
to testing.
the damageThe topography
Spatial Wavenumber Algorithm
reconstruction for metalandplates
the Wavenumber
and compositeFiltering
materialAlgo-
plates
rithm are compared by the noncontact system. The system has the characteristics
for testing. The Spatial Wavenumber Algorithm and the Wavenumber Filtering Algorithm of flexi-
ble
areoperation
compared and
byfast
the scanning,
noncontact and can be The
system. combined
systemwithhas athe
variety of algorithms,
characteristics and
of flexible
has the application potential for rapid scanning of large-scale equipment structures.
operation and fast scanning, and can be combined with a variety of algorithms, and has
the application potential for rapid scanning of large-scale equipment structures.
2. Air-Coupled Ultrasonic and Full-Field Laser Vibrometer Detection System
2. Air-Coupled
2.1. Ultrasonic
System Composition and Full-Field Laser Vibrometer Detection System
and Function
2.1. System Composition and Function
The noncontact detection system proposed in this paper is shown in Figure 1, which
mainlyThe noncontact
includes detection system
an air-coupled proposed
ultrasonic in thisand
transmitter paper is shown laser
a full-field in Figure 1, which
vibrometer.
mainly includes an air-coupled ultrasonic transmitter and a full-field laser vibrometer.
The air-coupled ultrasonic transmitter is responsible for exciting the different modal ul- The
air-coupled ultrasonic transmitter is responsible for exciting the different
trasonic waves by adjusting the incident angle. Laser vibration measurement technology modal ultrasonic
iswaves by adjusting
responsible the incident
for realizing angle. Laser
the distributed vibration
point scanning measurement
measurement technology
with highisspatial
respon-
sible for realizing the distributed point scanning measurement with high spatial
resolution. The two are applied to the excitation and the receiving parts, which can give resolution.
The two are applied to the excitation and the receiving parts, which can give full play
full play to their advantages to conduct the guided wave excitation and the full-wave field
to their advantages to conduct the guided wave excitation and the full-wave field data
data acquisition. The sheet structural damage can be characterized by the wave field
acquisition. The sheet structural damage can be characterized by the wave field changes by
changes by different algorithms.
different algorithms.

Air-coupled θ
Ultrasonic Full-field Laser
Transmitter Vibrometer
Scanning Area
Figure
Figure1.1.The
Theair-coupled
air-coupledultrasonic
ultrasonicand
andfull-field
full-fieldlaser
laservibrometer
vibrometerdetection
detectionsystem
systemdiagram.
diagram.

Specifically,ininthe
Specifically, thesubsequent
subsequentexperimental
experimentalsystem,
system,the theultrasonic
ultrasonictransmitting
transmittingcom- com-
ponent is shown in Figure 2a,
ponent is shown in Figure 2a, including including a signal generator (Tektronix AFG3152C),
generator (Tektronix AFG3152C), a volt- a voltage
amplifier
age amplifier(Tegam
(TegamModel 2340)2340)
Model and anandair-coupled ultrasonic
an air-coupled transducer
ultrasonic (Ultran(Ultran
transducer Group
NCG200-D25).
Group NCG200-D25). The transducer is a circular
The transducer noncontact
is a circular transducer
noncontact with a center
transducer with frequency
a center
of 200 kHz
frequency of and a bandwidth
200 kHz of 40% ofofthe
and a bandwidth 40%center
of thefrequency. It is fixedItby
center frequency. is afixed
clamping
by a
device and
clamping emitsand
device ultrasonic waves to the
emits ultrasonic wavessurface of surface
to the the thinof plate. The incident
the thin plate. The angle of the
incident
transducer is easy to accurately adjust. According to Snell’s law of refraction,
angle of the transducer is easy to accurately adjust. According to Snell’s law of refraction, the ultrasonic
Lamb
the wave with
ultrasonic Lamba wave
relatively
withsingle mode single
a relatively is excited
mode byisa excited
specificbyangle. The out-of-plane
a specific angle. The
displacement of the A0 mode wave is more obvious than the in-plane
out-of-plane displacement of the A0 mode wave is more obvious than the in-plane dis- displacement and
is easy to actuate, thus the A0 mode Lamb wave is usually considered
placement and is easy to actuate, thus the A0 mode Lamb wave is usually considered as as the preferred
mode
the for damage
preferred mode detection.
for damageBecause
detection.of the dispersion
Because of the characteristics of the Lambofwave
dispersion characteristics the
in the plate, the higher the frequency is, the more complex the
Lamb wave in the plate, the higher the frequency is, the more complex the lamb wavelamb wave modes will be.
In order to reduce the number of wave modes as much as possible
modes will be. In order to reduce the number of wave modes as much as possible to obtainto obtain the relatively
pure
the guidedpure
relatively wave, a lower
guided frequency
wave, a lower offrequency
200 kHz isofselected
200 kHzasisthe excitation
selected as thefrequency
excitationof
the probe. The signal generator outputs a 1 ms, 4 Vpp, 5 cycles, 200
frequency of the probe. The signal generator outputs a 1 ms, 4 Vpp, 5 cycles, 200 kHz sinekHz sine pulse train,
and 50 times amplified to realize the air-coupled ultrasonic excitation.
pulse train, and 50 times amplified to realize the air-coupled ultrasonic excitation.
The signal receiving component of the system is shown in Figure 2b, which is per-
formed by a full-field laser Doppler vibrometer (PSV-500-V). The system consists of one
laser head, a data acquisition system with velocity and displacement decoders as well as
a signal generator, and a data management system. The He-Ne laser of the system has a
wavelength of 633 nm, the velocity resolution is 0.01 μm/s (1 Hz), a minimum displace-
ment resolution can be up to pico-meter level, and the frequency measurements range
 greatly reducing the complexity of the experiment operation and improving the spatial
resolution of the signal. In the experiment, the time-domain measurement mode is used,
the sampling frequency is 1.25 MHz, the range of the band-pass filter is 100~300 kHz, and
the trigger acquisition signal is an external input analog signal from the signal generator.
In order to ensure the high precision of laser vibration measurement and improve the
Sensors 2021, 21, 609 4 of 18
signal-to-noise ratio, each point is measured 32 times and the average is taken as the result.

Signal Generator Voltage Amplifier

Transducer

200 Vpp

200 kHz Clamping Device

(a)

Scanning
Area

Scanning Point Laser Vibrometer

(b)
Figure
Figure 2.
2. The
The specific
specific composition
composition of
of the
the system
system in
in the
the experiment:
experiment: (a)
(a) the
the ultrasonic
ultrasonic transmitting
transmitting component;
component; (b)
(b) The
The
signal
signal receiving
receiving component.
component.

2.2. Excitation
The signal andreceiving
Receivingcomponent
Performanceof
Testing of the System
the system is shown in Figure 2b, which is per-
formed by a full-field
2.2.1. Aluminum Alloy Platelaser Doppler vibrometer (PSV-500-V). The system consists of one
laser head, a data acquisition system with velocity and displacement decoders as well as
An aluminum alloy plate with a density of 2700 kg/m3, Young’s modulus of 71.7 GPa,
a signal generator, and a data management system. The He-Ne laser of the system has a
a Poisson’s ratio of 0.33, and a thickness of 2 mm is used for the system test. Lamb waves
wavelength of 633 nm, the velocity resolution is 0.01 µm/s (1 Hz), a minimum displacement
in thin plates have dispersion phenomena, that is, the propagation speed of sound waves
resolution can be up to pico-meter level, and the frequency measurements range from
along the surface is a function of frequency or wavelength. The phase velocities and group
DC to 25 MHz. The Doppler frequency shift effect is used to realize the measurement of
velocities of different wave modes at different frequencies can be obtained according to
the surface velocity or displacement, with high spatial resolution, scanning speed, and
the Rayleigh-Lamb wave equation. It can be determined that the phase velocity of the A0
measurement accuracy. The excitation position of the air-coupled sensor is fixed, and the
laser vibrometer is synchronously triggered to receive the response of a point every excita-
tion time. In this way, the Lamb wave field of the area to be measured is collected point
by point, so as to detect the Lamb wave propagation at any point at any time. This laser
scanning measurement method can replace the method with a large number of sensors,
greatly reducing the complexity of the experiment operation and improving the spatial
resolution of the signal. In the experiment, the time-domain measurement mode is used,
the sampling frequency is 1.25 MHz, the range of the band-pass filter is 100~300 kHz, and
the trigger acquisition signal is an external input analog signal from the signal generator.
In order to ensure the high precision of laser vibration measurement and improve the
signal-to-noise ratio, each point is measured 32 times and the average is taken as the result.
Sensors 2021, 21, 609 5 of 18

2.2. Excitation and Receiving Performance Testing of the System

2.2.1. Aluminum Alloy Plate
An aluminum alloy plate with a density of 2700 kg/m3 , Young’s modulus of 71.7 GPa,
a Poisson’s ratio of 0.33, and a thickness of 2 mm is used for the system test. Lamb waves
in thin plates have dispersion phenomena, that is, the propagation speed of sound waves
Sensors
Sensors2021,
2021,21,
21,xxFOR
FORPEER
PEERREVIEW
REVIEW 55ofof1818
along the surface is a function of frequency or wavelength. The phase velocities and group
velocities of different wave modes at different frequencies can be obtained according to
the Rayleigh-Lamb wave equation. It can be determined that the phase velocity of the A0
mode
modeLamb
mode Lambwave
Lamb waveat
wave ataafrequency
at frequencyofof200 200kHz
kHzisis1744
1744 m/s,
m/s,thereby
1744m/s, therebydetermining
thereby determiningthat
determining thatthe
that the
optimal ◦ 0
optimalincident
optimal incidentangle
incident angleof
angle ofthe
of theA0
the A0mode
A0 modeLamb
mode Lambwave
Lamb waveexcitation
wave excitationisisaboutabout11°14′.
11 14 .
11°14′.
The
Thedriving
The drivingability
driving abilityofofaaapure
purefundamental
pure fundamental
fundamental A0A0
A0 mode
mode
mode atatat
the selected
the
the selected
selected incident
incident
incident angle by
angle
angle by
by
thethe
the air-coupled
air-coupled
air-coupled ultrasonic
ultrasonic
ultrasonic transducer,
transducer,
transducer, andand
and the the ability
theability
ability to acquire
totoacquire
acquire the the wavefield
thewavefield
wavefield thein
ininthe the
plate
plate
plate
bybythe
theby the full-field
full-field
full-field laser laser
laservibrometervibrometer
vibrometer are are tested.
aretested.
tested.The The distance
Thedistance
distance between
between between
the the air-coupled
theair-coupled
air-coupled trans-
trans-
transducer
ducer and the plate surface is about 40 mm. The wavefield u(t,
ducer and the plate surface is about 40 mm. The wavefield u(t, x) of the Lamb wave inthe
and the plate surface is about 40 mm. The wavefield u(t, x) of x)
the of
Lambthe Lamb
wave wave
in the
in
platetheisis
plate plate is obtained
obtained
obtained by by scanning
bylinear
linear linear scanning
scanning ofofthe of theDoppler
thelaser
laser laser Doppler
Doppler vibrometer.
vibrometer.
vibrometer. The The scanning
Thescanning
scanning range
range
range
isisabout
aboutis about
30 30 mm~60
30mm~60
mm~60 mm
mmfrom mm the
from from the excitation
theexcitation
excitation point,
point,
point, withwith
with an an interval
aninterval
interval ofof11of 1 mm,
mm,
mm, as as
asshownshown
shown inin
in Figure
Figure
Figure 3.3. 3.

Tested
TestedPlate
Plate
1mm
1mm

xx(mm)
(mm)
Excitation
ExcitationPoint
Point 30
30 Scanning
ScanningPoint
Point 60
60

Figure
Figure3.
Figure 3.3.Schematic
Schematicdiagram
Schematic diagramof
diagram oflinear
of linearscanning
linear scanningof
scanning oflaser
of laservibrometer.
laser vibrometer.
vibrometer.

The
Thetime-domain
The time-domain waveform
time-domainwaveformwaveformdata data from
datafrom the
fromthethe 40
40 40mm
mm mm are
areareselected
selected
selected tototoanalyze,asas
analyze,
analyze, asshown
shown
shown in
in Figure
in Figure
Figure 4. It
4. It4.cancan
It can be seen
be seen
be seen that
thatthat the signal
the signal
the signal noise
noisenoise in
in one one acquisition
in acquisition
one acquisition is large,
is large,
is large, as
as in asin Figure
in Figure
Figure 4a.
4a.
4a. The
The
Therandom
random noise
random noise
noise isisreduced
is reduced afterafter
reduced averaging
averaging
after 32
32times
32 times
averaging ofofdata,
of data,
times as inas
data, ininFigure
Figure
as 4b.
4b.After
4b. After
Figure band-
band-pass
After band-
pass
passfiltering,
filtering, the
the noise
filtering, thenoise isisfurther
is further
noise reduced,
reduced,
further as in as
reduced, asininFigure
Figure 4c. 4c.
Figure 4c.

(a)
(a) (b)
(b) (c)
(c)
Figure
Figure4.4.The
Figure The wave
Thewave
waveatatat
4040
40
mmmmin in
mm inaluminum
aluminum
aluminum plate:
plate:
plate: (a)
(a)Single
(a) SingleSingleacquisition
acquisition
acquisition signal;
signal;signal;(b)
(b)Average
(b) AverageAverage ofof32
32acquisitions;
(c) After (c)
acquisitions;
of 32 acquisitions; (c)After
After
filtering.
filtering.
filtering.
Based on the above data, the wavefield at different moments of the one-dimensional
Based
straight lineon
Based inthe
on theabove
the abovedata,
aluminum the
thewavefield
data,plate atatdifferent
can be obtained,
wavefield moments
as shown
different ofofthe
in Figure
moments 5a.one-dimensional
the The air-coupled
one-dimensional
straight
Lamb
straight line
wave ininthe
linespreadsthealuminum
out in theplate
aluminum platecan
range be
beobtained,
of about
can 50 µs inas
obtained, shown
the
as ininFigure
time domain.
shown Figure 5a.
The The
Theair-cou-
5a.wave moves
air-cou-
pled
pledLamb
forward,Lamb wave
and spreads
spreadsout
the propagation
wave ininthe
outspeed range
theof ofofabout
the wave
range is the50
about μs
μsinin
same
50 the
thetime
during the domain.
time whole
domain. The
Thewave
process. The
wave
time-domain
moves
movesforward,
forward, wavefield
and
andthe data are transformed
thepropagation
propagation speedofofinto
speed the the
thewave
wave frequency-wavenumber
isisthe
thesame
sameduring
duringthe domain
thewhole
wholepro- by
pro-
two-dimensional
cess. The time-domain Fourier transform
wavefield for
data characterization,
are transformed and
into
cess. The time-domain wavefield data are transformed into the frequency-wavenumber compared
the with the
frequency-wavenumber theoretical
spectrum,
domain
domainby byas shown in FigureFourier
two-dimensional
two-dimensional 5b. It can
Fourier be seen for
transform
transform that the center frequency
forcharacterization,
characterization, and of the received
andcompared
compared with
with
lamb
the wave
thetheoretical is about
theoreticalspectrum, 200
spectrum,askHz, and
asshown the center
shownininFigure wavenumber
Figure5b. 5b.ItItcan
canbe is
beseen about
seenthat 0.72
thatthe rad/mm.
thecenter Almost
centerfrequency
frequency
only
ofofthe
thethe A0 mode
received
received lamb Lamb
lamb wave
wavewave is excited,
isisabout
about 200
200kHz,which
kHz, and
andis the
consistent
thecenter with
centerwavenumber the target.
wavenumber The
isisaboutabove
about 0.72
0.72
experiments
rad/mm.
rad/mm.Almost confirmed
Almost only that
onlythe
theA0 themode
A0 system
mode Lamb can excite
Lamb waveaisis
wave relatively
excited,
excited,whichpure
which A0isismode Lambwith
consistent
consistent wavethe
with in
the
the aluminum
target. The above plate by exciting
experiments at the
confirmed theoretically
that the optimal
system angle.
can
target. The above experiments confirmed that the system can excite a relatively pure A0 exciteHowever,
a the
relatively ability
pure A0to
acquire
mode
modeLamb the wavefield
Lamb wave
waveininthedata has also been
thealuminum
aluminum plate
plateverified.
by
byexciting
excitingatatthe thetheoretically
theoreticallyoptimal optimalangle.
angle.
However,
However,the theability
abilitytotoacquire
acquirethe thewavefield
wavefielddata datahas hasalso
alsobeenbeenverified.
verified.
Sensors 2021, 21, x FOR PEER REVIEW 6 of 18
Sensors 2021, 21, 609 6 of 18

Sensors 2021, 21, x FOR PEER REVIEW 6 of 18

(a) (b)
Figure 5.
Figure 5. Test
Test results
resultsofof22mm
mmthick
thickaluminum plate:
aluminum (a)(a)
plate: Lamb wave
Lamb one-dimensional
wave timetime
one-dimensional domain wave
domain field;field;
wave (b) Fre-
(b)
quency-wavenumber spectrum.
Frequency-wavenumber spectrum.
(a) (b)
Figure 5. Test results of 2 mm2.2.2.
thick Carbon
aluminumFiber
plate:Reinforced
(a) Lamb wave one-dimensional time
Polymer(CFRP) Platedomain wave field; (b) Fre-
quency-wavenumber spectrum. 2.2.2. Carbon Fiber Reinforced Polymer(CFRP) Plate
For the CFRP plate, the characteristic parameters are shown in Table 1, which is a 16-
layer For the CFRP
composite
2.2.2. Carbon Fiber plate,
plate the
layered
Reinforced characteristic
in [+45/−45/0/0]
Polymer(CFRP) parameters
Plate 2s, as shown are
inshown
Figure in6. Table 1, which is a
16-layer composite plate layered in [+45/ − 45/0/0] 2s
For the CFRP plate, the characteristic parameters are shown , as shown
in Table 1, which is6.a 16-
in Figure
Table
layer 1. CFRPplate
composite material parameters.
layered in [+45/−45/0/0]2s, as shown in Figure 6.
Table 1. CFRP material parameters.
Density Elastic Modulus
Table 1. CFRP material parameters.
Shear Modulus
Poisson’s Ratio
kg/m3
Density GPa
Elastic Modulus Shear GPa
Modulus
Poisson’s Ratio
kg/m3
Density Elastic Modulus
E1 GPa E2 E3 ShearGModulus
12
GPa
G13 GPoisson’s
23 v12Ratio v13 v23
1600
kg/m 3 GPa GPa
E1 172 E2 11.6 E311.6 G12 7.8 G7.8
13 G
3.9
23 v
0.36
12 v
0.36
13 v23
0.55
1600 E1 E11.6
2 E3 11.6 G12 7.8G13 G23 v3.9
12 v0.36
13 v230.36
1600 172 7.8 0.55
172 11.6 11.6 7.8 7.8 3.9 0.36 0.36 0.55

Figure
Figure 6. Schematic
6. Schematic diagram
diagram of CFRP
of CFRP plate scheme.
plate laying laying scheme.
Figure 6. Schematic diagram of CFRP plate laying scheme.
For multilayer composite materials, due to its heterogeneity, inherent anisotropy of the
material and multilayer structure, the propagation of waves in composite materials is very
Sensors2021,
Sensors 2021,21,
21,xxFOR
FORPEER
PEERREVIEW
REVIEW 77of
of18
18

Sensors 2021, 21, 609 7 of 18

For multilayer
For multilayer composite
composite materials,
materials, due due to to its
its heterogeneity,
heterogeneity, inherent
inherent anisotropy
anisotropy of of
the material and multilayer structure, the propagation of waves
the material and multilayer structure, the propagation of waves in composite materials is in composite materials is
very
very complicated.
complicated.
complicated. Inpaper,
In
In this this paper,
this paper, the dispersion
the
the dispersiondispersion relationship
relationship
relationship between between
between the symmetric
the
the symmetric symmetric
wave wavewave
mode
mode
mode and
and the
the antisymmetric
antisymmetric wave
wave mode
mode in
in the
the thin
thin plate
plate is
is obtained
obtained
and the antisymmetric wave mode in the thin plate is obtained from the three-dimensional from
from the
the three-di-
three-di-
mensional
mensional elastic
elastic
elastic theory, theory,
andtheory, andititisisto
and
it is extended extended
extended
composite tocomposite
to composite
laminates with laminates
laminates withstacking
with
arbitrary arbitrary
arbitrary stacking
stacking
order. The
order.
order.
phase TheThe phase velocity
phaseofvelocity
velocity the A0 mode of the
of the A0 A0
Lamb mode
mode waveLamb
Lambat 200 wave
wave kHz at 200 kHz
at frequency frequency
200 kHz frequency
is 1598 m/s, is 1598
is 1598 m/s,
andm/s,
the
and
and the optimal
the
optimal optimal incident
incident incident
angle angle
of angle
the A0of of the A0
the
mode A0 mode
Lamb modewave Lamb
Lamb wave
wave
at this at this
at
frequency thisis about 12◦is
frequency
frequency is 0about
17about
. The
12°17′.
12°17′. The experiment
The
experiment experiment
process is the process
process
same is asis the
the same
that same asaluminum
of theas that of
that of the
the aluminum
aluminum
plate, plate,
and theplate,
results and
and the
arethe results
as results
shown
are
are as shown
as shown
in Figure 7. inin Figure
Figure 7. 7.

(a)
(a) (b)
(b) (c)
(c)
Figure7.
Figure
Figure 7.The
7. The wave
Thewave at
waveat 40
at40 mm
40mm in
mmin CFRP
inCFRP plate:
CFRPplate: (a)
plate:(a) Single
(a)Single acquisition
Singleacquisition signal;
acquisitionsignal; (b)
signal;(b) Average
(b)Average of32
Averageof
of 32acquisitions;
32 acquisitions; (c)After
acquisitions;(c)
(c) Afterfiltering.
After filtering.
filtering.

The one-dimensional
The one-dimensionallinear
one-dimensional lineartime-domain
linear time-domain
time-domain wavefield
wavefield
wavefield andand and frequency-wavenumber
frequency-wavenumber
frequency-wavenumber spec-
spectrum
spectrum
trum in the in the
inCFRP CFRP
the CFRP plate
plate plate are shown
are shown
are shown in Figure
in Figure
in Figure 8. The 8. 8. The Lamb
The wave
Lamb wave
Lambpropagates propagates
wave propagates straight
straight straight
forward
forward
forward and the
and the slope
and the slope remains
remains
slope remains
the same,the the same, indicating
indicating
same, indicating
that the wave that the
that the wave propagation
propagation
wave propagation
speed is the speed
speed isis
same
the same throughout
throughout
the same throughout
the process. theAccording
the process. According
process. According to the
to the comparison
to the comparison
comparison
between the between the theoretical
theoretical
between the theoretical
frequency-
frequency-wavenumber
wavenumber spectrum and
frequency-wavenumber spectrum and the
the experimental
spectrum and the experimental
experimental
wavenumberwavenumberwavenumber spectrum,
found thatititthe
spectrum, it isspectrum, isis
found
found that
centerthat thecenter
frequency
the center frequency
of this mode Lamb
frequency ofthis
of this mode
wave
mode Lamb200
is about
Lamb wave
wave isisabout
kHz, about
and the200
200 kHz,wavenumber
center
kHz, andthe
and thecenter
center
is
about 0.78
wavenumber rad/mm.
is about The experiments
0.78 rad/mm. have
The proved
experiments that through
have
wavenumber is about 0.78 rad/mm. The experiments have proved that through excitation excitation
proved that at
throughthe theoretical
excitation
optimal
at
at the angle of incidence,
the theoretical
theoretical optimal angle
optimal a relatively
angle of pure A0
of incidence,
incidence, mode Lamb
aa relatively
relatively wave
pure
pure can
A0 mode
A0 modealso be obtained
Lamb
Lamb wave can
wave in
can
the composite
also
also be obtained
be obtained plate.
inthe
in the composite
composite plate.
plate.

(a)
(a) (b)
(b)
Figure8.
Figure
Figure 8.Test
8. Testresults
Test resultsof
results of222mm
of mmthick
mm thickCFRP
thick CFRPplate:
CFRP plate:(a)
plate: (a) Lamb
(a)Lamb wave
Lambwave one-dimensional
waveone-dimensional time
one-dimensionaltime domain
timedomain wave
domainwave field;
wavefield; (b)
field;(b) Frequency-
(b)Frequency-
Frequency-
wavenumberspectrum.
wavenumber
wavenumber spectrum.
spectrum.

3. Topography Reconstruction Algorithms

3.1. Wavenumber Filtering Algorithm (WFA)
WFA processes the full wavefield information through filtering in the wavenumber do-
main, and then the filtered time domain wavefield is obtained through the inverse Fourier
transform, and the damage situation is reflected by the time domain energy signal [20].
Sensors 2021, 21, 609 8 of 18

(1) Perform a two-dimensional Fourier transform on the acquired wavefield data

v(x,y,t) of each point, and transform it from the spatial domain (x,y) to the wavenumber
domain (kx ,ky ):
+∞ Z
Z +∞
V (k x , k y , t) = v( x, y, t)e− j(k x x+ky y) dxdy, (1)
−∞ −∞

where v(x,y,t) is the out-of-plane velocity, and V(kx ,ky ,t) is the wavefield data in wavenum-
ber domain.
(2) The wavenumber domain pattern of propagating guided waves is defined in the
wavenumber domain according to the selected number of wavefield images. Calculate
the average wavenumber spectrum Vavg (kx ,ky ) of the wavefield data, which represents the
propagation of waves in this period of time in the wavenumber domain, which can be
defined as:
1
M∑
Vavg (k x , k y ) = V ( k x , k y , t ), (2)
t

where M is the total number of time points of wavefield data.

(3) According to the distribution of the average wavenumber spectrum Vavg (kx ,ky ), the
threshold filter function M(kx ,ky ) is calculated as follows:

0 Vavg (k x , k y ) > threshold
M(k x , k y ) = , (3)
1 otherwise

The purpose of the filter function is to remove the mainstream health signals in the
wavefield and retain the unhealthy signals caused by damage. Since the wavenumber
intensity of the mainstream health signal in the wavenumber domain is larger, and the
wavenumber intensity of the damage signal is smaller, the wavenumber with the larger
intensity value is filtered out, and the smaller one is retained, thereby completing the
threshold filtering. The optimal value of the threshold in the function is not known a priori.
In the experiment, according to the cumulative distribution of the average wavenumber
spectrum Vavg (kx ,ky ), the wavenumber intensity value with a larger amplitude is selected
as the candidate threshold. While ensuring the damage detection effect, reduce noise
interference as much as possible. Select the candidate threshold that can make 5% of the
element values in the filter function matrix 0 as the final experimental threshold to realize
the automatic setting of the threshold.
(4) The generated threshold filter function is multiplied by the wavefield image at
each moment in the wavenumber domain to achieve filtering. It is important to note that
the same filter function is applied to each wavefield image:

V ( k x , k y , t ) = V ( k x , k y , t ) M ( k x , k y ), (4)

(5) Subsequently, the wavefield data filtered by each threshold are subjected to a
two-dimensional inverse Fourier transform to obtain a wavefield image in the time–space
domain after filtering:
−1

v( x, y, t) = F2D V (k x , k y , t) , (5)
(6) The proposed algorithm given by Equations (1)–(5) eliminates the main compo-
nents of guided wave, mainly including incident wave. All structural features and damage
locations may be easily identified. Repeat the above process for all wavefield data and
merge all filtered data with root mean square (RMS) as:
v
u N
u1
RMSx,y = t ∑ v( x, y, t)2 , (6)
N t =1
Sensors 2021, 21, 609 9 of 18

Through the filtered Root Mean Square energy field, the structural features and
damage locations can be easily identified.

3.2. Spatial Wavenumber Algorithm (SWA)

SWA directly reflects the damage situation through the change of wavenumber [18].
The three-dimensional Fourier Transform is performed on the time-domain full-field data
combining the spatial window function to obtain spatial wavenumber information at
different frequencies. The specific frequency range is selected according to the excitation
signal bandwidth, and average the wavenumber signals in the entire frequency band.
The average spatial wavenumber diagram is used to identify the damage. This method
synthesizes the changes of the wavenumber at each frequency and effectively contains
most of the damage information.
(1) Perform a short-space three-dimensional Fourier transform on the wavefield data
v(x,y,t) at different spatial positions to obtain the frequency-wavenumber spectrum. The
short-space three-dimensional Fourier transform is a direct extension of short-time Fourier
transform to multidimensional problems, that is, the time–space wave field is decomposed
into small segments in two-dimensional space before the Fourier transform. To do this, the
wavefield data are multiplied by a window function that is nonzero only for a short period
in space, but is constant throughout the time dimension. Then three-dimensional Fourier
transform is applied to the generated wavefield segments. The specific transform form is
as follows:
Z∞ Z∞
S ( x0 , y0 , k x , k y , f ) = v( x, y, t)W ( x − x0 , y − y0 )e− j(2π f t+k x x+ky y) dtdxdy, (7)
−∞ −∞

where (x0 ,y0 ) is the reserved spatial position point, and W(x-x0 ,y-y0 ) is the spatial window
function based on the two-dimensional Hanning window:
( √
2π x 2 + y2 p
2 2 D
W ( x, y) = 0.5 × [1 + cos( D )] px + y ≤ 2D , (8)
0 2 2
x +y > 2

where D is the length of the spatial window. By sliding the window function along the
spatial dimension, the frequency-wavenumber spectrum can be obtained.
(2) According to the space-frequency-wavenumber spectrum obtained, find the maxi-
mum wavenumber vector of each spatial location point under each frequency component:

k∗ ( x0 , y0 , f ) = argmax S( x0 , y0 , k x , k y , f ) , (9)
k

(3) According to the frequency band range of the excitation signal, the average spatial
maximum wavenumber value is calculated as:

1 N ∗
N i∑
k ∗ ( x0 , y0 ) = |k ( x0 , y0 , f i )|, (10)
=1

where f i is a frequency during the selected range. Hence the average spatial wavenumber
value represents the wavenumber at each location. The wavenumber diagram of the entire
space is obtained, and displaying the damage imaging situation through the direct change
of the wavenumber.

4. Results and Discussion

4.1. Experiment Platform
The experimental site and the damage information of the tested plates are shown in
Figure 9.
 of the wavenumber.

4. Results and Discussion

4.1. Experiment Platform
Sensors 2021, 21, 609 The experimental site and the damage information of the tested plates are shown
10 ofin
18
Figure 9.

150mm 200mm 350mm

Aluminum Plate

150mm
Signal Generator Laser Vibrometer

10mm
150mm
10mm

Depth1.0mm Depth1.5mm
(b)
Voltage Amplifier 150mm 200mm 350mm
Scanning Area
CFRP Plate

150mm
Transducer

2mm
150mm
20mm

Depth1.0mm Depth1.5mm
(a) (c)
Figure
Figure9.9.Experiment
Experimentplatform:
platform:(a)(a)
Experimental setup
Experimental forfor
setup thethe
detection andand
detection estimation of damage
estimation on the
of damage on CFRP plate;
the CFRP (b)
plate;
The damage information of the tested aluminum plate; (c) The damage information of the tested CFRP plate.
(b) The damage information of the tested aluminum plate; (c) The damage information of the tested CFRP plate.

The
Thealuminum
aluminumplate plateand
andCFRP
CFRPplate
platearearetested
testedinin the
the experiment.
experiment. The Thedamages
damagesare are
processed
processedon onthe
thesurface
surfaceofofplates
platesbyby
cutting
cutting grooves. TheThe
grooves. cutting depth
cutting is 1 is
depth mm and and
1 mm 1.5
mm respectively.
1.5 mm TheThe
respectively. preset damage
preset damage size
sizeofofaluminum
aluminumplate plate is
is 10 mm × × 10
10 mm ××11mm mm
and
and1010mm mm× 10
× mm10 mm × 1.5×mm,
1.5 and
mm,the andpreset
the damage size of composite
preset damage plate is 20plate
size of composite mm ×is
220mmmm × 1×mm2 mmand× 201 mmmmand× 2 mm
20 mm × mm.
× 1.5 2 mmThe × 1.5
scanning
mm. The range of the aluminum
scanning range of theplates
alu-
isminum
40 mm plates
× 40 mm, and
is 40 mmthe×scanning
40 mm,range
and the of the composite
scanning plates
range is 60composite
of the mm × 20 mm. Theis
plates
60 mm ×center
scanning 20 mm. point
Theisscanning
coincident withpoint
center the damage centerwith
is coincident point,thethe sampling
damage point
center in-
point,
the sampling
terval is 1 mm,point
and the interval is 1 mm,
sampling and the
frequency sampling
is 1.25 MHz,frequency is 1.25 MHz,
the total sampling timethe
for total
one
samplingpoint
scanning time is
for500
one scanning point is 500 µs.
μs.

4.2.Topography
4.2. TopographyReconstruction
ReconstructionProcess
Process
Taking the
Taking the CFRP
CFRP plate
plate with
with 20 mm ×
20 mm mm××1 1mm
× 22mm mmdamage
damageas
asan
anexample,
example,the
thedata
data
processing processes of the WFA and SWA are shown
processing processes of the WFA and SWA are shown below. below.
(1) WFA
(1) WFA
The
Thetime–space
time–spacethree-dimensional
three-dimensionalwavefield
wavefielddata
dataare
aresubjected
subjectedtotoaatwo-dimensional
two-dimensional
Fouriertransform
Fourier transformtoto obtain
obtain thethe wavenumber
wavenumber distribution
distribution diagram
diagram at different
at different times.times.
Fig-
Figure
ure 10a,b
10a,b are are
the the time–space
time–space domain
domain andand wavenumber
wavenumber domain dataatatt t==136
domaindata 136μs. Sum
µs. Sum
andaverage
and averageallallwavenumber
wavenumberdomain domaindatadatato
toobtain
obtain the
the average
average wavenumber
wavenumber spectrum
spectrum
as shown
as shown inin Figure
Figure 10c.
10c. Based
Basedonon the
the average
average wavenumber
wavenumber spectrum
spectrum data,
data, select
select the
the
threshold that
threshold that can make 5% of the data in the filter function zero, and get the threshold
make 5% of the data in the filter function zero, and get the threshold filter
function as Figure 10d. Finally, the threshold filter function is applied to the wavenumber
data at each time point, and a two-dimensional inverse Fourier transform is performed.
Figure 10e,f are the wavenumber data and the time–space data after filtering at t = 136 µs.
The filtered space-time domain data are combined with the root mean square energy
function to obtain the final damage imaging.
 filter function as Figure 10d. Finally, the threshold filter function is applied to the wave-
number data at each time point, and a two-dimensional inverse Fourier transform is per-
formed. Figure 10e,f are the wavenumber data and the time–space data after filtering at t
Sensors 2021, 21, 609 = 136 μs. The filtered space-time domain data are combined with the root mean square 11 of 18
energy function to obtain the final damage imaging.

(a) (b)

(c) (d)

(e) (f)
Figure 10. WFA process: (a) space-time domain data; (b) wavenumber domain spectrum; (c) average wavenumber spec-
Figure 10. WFA process: (a) space-time domain data; (b) wavenumber domain spectrum; (c) average wavenumber spectrum;
trum; (d) the threshold filter function; (e) wavenumber domain data after filtering (f) time–space domain data after filter-
(d) the threshold filter function; (e) wavenumber domain data after filtering (f) time–space domain data after filtering.
ing.

(2) SWA
(2) SWA
The spatial window
The windowfunction
functionisis
used to to
used cutcut
thethe
time–space domain
time–space domainwavefield. Figure
wavefield. 11a,b
Figure
are the
11a,b areoverall wavefield
the overall at t =at
wavefield 136
t =µs and
136 μsthe
andwindowed
the windowedwavefield with with
wavefield the (0,0)
the coor-
(0,0)
dinate point
coordinate as the
point center.
as the Subsequently,
center. Subsequently, the
thedata
dataare
aresubjected
subjectedtoto a three-dimensional
three-dimensional
Fourier transform
Fourier transform toto obtain
obtain wavenumber
wavenumber distribution
distributiondiagrams
diagramsat at various
various frequencies.
frequencies.
Figure 11c,d
Figure 11c,d are
are wavenumber
wavenumber distribution
distributiondiagrams
diagramsat at200
200kHz
kHzand
and 220
220 kHz.
kHz. The
Themaxi-
maxi-
mumvalue
mum valueof ofthe
thewavenumber
wavenumberat ateach
eachfrequency
frequency isis averaged
averaged as
as the
the actual
actual wave
wave value
value of
of
the current coordinate point, thereby the wave number diagram of each position in the
space is obtained, which can be used to characterize the damage.
Sensors 2021, 21, x FOR PEER REVIEW 12 of 18

Sensors 2021, 21, 609 the current coordinate point, thereby the wave number diagram of each position 12
in ofthe
18
space is obtained, which can be used to characterize the damage.

(a) (b)

(c) (d)
Figure
Figure 11.
11. SWA
SWA process:
process: (a)
(a) space-time
space-time domain
domain data;
data; (b)
(b) the
the windowed
windowed space-time
space-time domain
domain data;
data; (c)
(c) wavenumber
wavenumber domain
domain
spectrum at 200 kHz; (d) wavenumber domain spectrum at 220
spectrum at 200 kHz; (d) wavenumber domain spectrum at 220 kHz. kHz.

4.3. Topography Reconstruction Result

4.3. Topography
4.3.1. AluminumReconstruction
Plate Damage Result
Imaging
4.3.1. Aluminum Plate Damage Imaging
Based on the above two imaging algorithms, the damage of the aluminum plate is
imaged,Based on the in
as shown above two
Figure 12.imaging algorithms,
The topography thedamage
of the damageisofwell
the presented,
aluminum and platethe
is
imaged, as shown in Figure 12. The topography of the damage is well presented,
size is consistent with the actual preset size of 10 mm × 10 mm, and the detection size of 1 and the
size deep
mm is consistent
damagewith the actual
is larger than preset
1.5 mm size of 10
deep mm × 10
damage. Formm, and the
damage edge detection sizethe
detection, of
1 mm deep damage is larger than 1.5 mm deep damage. For damage edge
SWA imaging results have smoother edges. Comparing the results of damage imaging at detection, the
SWAdifferent
two imagingdepths,
resultsthe
have smoother
energy valueedges. Comparing
and wave value ofthe
theresults
damage ofat
damage
1.5 mmimaging
depth are at
two different depths, the energy value and wave value of the damage at 1.5 mm depth
both higher than 1 mm, and the imaging effect is better. This is due to the thinning of the
are both higher than 1 mm, and the imaging effect is better. This is due to the thinning of
plate thickness in the damaged area, forming an acoustic black hole structure, which
the plate thickness in the damaged area, forming an acoustic black hole structure, which
causes energy to concentrate, the wave propagation speed becomes slower, and the wave
causes energy to concentrate, the wave propagation speed becomes slower, and the wave
number becomes larger.
number becomes larger.
Sensors 2021, 21, x FOR PEER REVIEW 13 of 18
Sensors 2021, 21, 609 13 of 18

(a) (b)

(c) (d)
Figure
Figure 12.
12. Damage Imagingresults
Damage Imaging resultsofofthe
the aluminum
aluminum plate:
plate: (a) (a) Imaging
Imaging of damage
of damage at 1depth
at 1 mm mm depth using
using the the(b)
WFA; WFA; (b)
Imaging
Imaging of damage at 1.5 mm depth using the WFA; (c) Imaging of damage at 1 mm depth using the SWA; (d)
of damage at 1.5 mm depth using the WFA; (c) Imaging of damage at 1 mm depth using the SWA; (d) Imaging of damage at Imaging
of damage at 1.5 mm depth using the SWA.
1.5 mm depth using the SWA.

4.3.2. CFRP Plate Damage Imaging

4.3.2. CFRP Plate Damage Imaging
Similarly, the damage of the CFRP plate is imaged, as shown in Figure 13. The above
Similarly, the damage of the CFRP plate is imaged, as shown in Figure 13. The above
detection scheme is still effective for composite materials and elongated damage, and the
detection scheme is still effective for composite materials and elongated damage, and the
effect is similar to that of the aluminum plates. Damage topography reconstruction algo-
effect is similar to that of the aluminum plates. Damage topography reconstruction algo-
rithms based on WFA and SWA are applied to the detection of different damage condi-
rithms based on WFA and SWA are applied to the detection of different damage conditions
tions of different plates. Experimental studies have proved that both methods can better
of different plates. Experimental studies have proved that both methods can better present
present the damage
the damage topography
topography and areand are suitable
suitable for damage
for damage topography
topography reconstruction
reconstruction of the
of the air-coupled
air-coupled ultrasonic
ultrasonic and
and laser laser vibrometer
vibrometer hybrid noncontact
hybrid noncontact detectiondetection system
system proposed
proposed in this
in this paper. Thepaper. The
specific specificquality
imaging imaging quality evaluation
evaluation willout
will be carried be carried outsection.
in the next in the
next section.
 Sensors
Sensors 2021,
2021, 21, 21,
609x FOR PEER REVIEW 14 of
1418of 18

(a)

(b)

(c)

(d)

Figure 13. Damage Imaging results of the CERP plate: (a) Imaging of damage at 1 mm depth using the WFA; (b) Imaging of
damage at 1.5 mm depth using the WFA; (c) Imaging of damage at 1 mm depth using the SWA; (d) Imaging of damage at
1.5 mm depth using the SWA.
Sensors 2021, 21, 609 15 of 18

4.4. Quality Assessment of Damage Imaging Results

In order to specifically evaluate the damage imaging quality, two evaluation indicators,
imaging error and image contrast, are used to evaluate the damage imaging effect.
For the two square damages in the aluminum plate, the imaging error δ1 is represented
by the average of the length measurement errors in the x and y directions. For the damage
of two rectangular cracks in the CFRP plate, since the real length in the y direction is too
small, only the length measurement error in the x direction is chosen to represent the
imaging error δ2 .  
| Mx − L x | | My − L y |
Lx + Ly
δ1 = , (11)
2
| Mx − L x |
δ2 = , (12)
Lx
In the above formula:
1 N
N i∑
Mx = | MLi − MRi |, (13)
=1

1 K
K j∑
My = MUj − MDj , (14)
=1

Among them, Lx and Ly represent the true value of the length in the x and y directions
respectively, and Mx and My represent the measured value of the length in the x and y
directions respectively. N and K are the number of samples for length measurement in the
x and y directions respectively. MLi and MRi represent the starting point and end point of
measuring the length in the x direction in each sample, and MUj and MDj represent the
starting point and end point of measuring the length in the y direction in each sample.
In addition to the imaging error defined above, in order to more objectively evaluate
the imaging quality of the two algorithms, the Weber contrast is used to evaluate the
imaging quality, which is specifically defined as:

I − Ib
C= , (15)
Ib

Among them, I and Ib respectively represent the average brightness value of the target
in the image and the average brightness value of the background. For Weber contrast, the
larger the value, the better the visual contrast between the target and the background, that
is, the better the imaging.
Table 2 compares the imaging effects of aluminum plates. First, comparing the P1 and
P2 damage detection results under the same algorithm, it can be seen that in WFA, the
P2 damage imaging error is reduced by 12.5% from P1, and the contrast is increased by
0.08. In SWA, the P2 damage imaging error is reduced by 9.5%, and the contrast increased
by 0.44. The P2 damage is more serious, the more damage information is carried by the
collected signal, the more accurate the judgment of the damage, the more accurate the
length detection in the x and y directions of the damage imaging, and the higher the image
contrast. Second, comparing the imaging results of different detection algorithms for the
same damage, the SWA is slightly better than the WFA. The damage imaging error in the
aluminum plate is reduced by 4.5% and 1.5%, and the contrast is increased by 0.13 and
0.49, respectively.
Sensors 2021, 21, 609 16 of 18

Table 2. Evaluation of damage imaging results of aluminum plates.

Damage- x True y True x Measured y Measured Imaging Weber

Method
Depth Length Length Length Length Error Contrast
WFA 12.6 mm 11.7 mm 21.5% 0.36
P1-1.0 mm
SWA 12.1 mm 11.3 mm 17.0% 0.49
10 mm 10 mm
WFA 11.3 mm 9.5 mm 9.0% 0.44
P2-1.5 mm
SWA 11.2 mm 10.3 mm 7.5% 0.93

Similarly, the detection effect in the CFRP plate was evaluated, as shown in Table 3,
and similar results were obtained. The detection effect of P4 damage is better than that
of P3. In the WFA, the imaging error is reduced by 1.5%, and the contrast is increased by
0.49. In the SWA, the imaging error is reduced by 4.0%, and the contrast is increased by
1.72. The damage imaging error of the SWA is reduced by 1% and 3.5% compared with the
WFA, and the image contrast of the SWA is better, increasing by 0.91 and 2.14 respectively.

Table 3. Evaluation of damage imaging results of CFRP plates.

Damage- x True y True x Measured Imaging Weber

Method
Depth Length Length Length Error Contrast
WFA 21.8 mm 9.0% 0.47
P3-1.0 mm
SWA 21.6 mm 8.0% 1.38
20 mm 2 mm
WFA 18.5 mm 7.5% 0.96
P4-1.5 mm
SWA 19.2 mm 4.0% 3.10

In summary, for the noncontact ultrasonic detection system with air-coupled excitation
and laser vibrometer reception proposed in this paper, the SWA is better than the WFA,
especially the damage edge detection is smoother and the image contrast is higher. The
main reasons are as follows:
(1) WFA uses filtered time-domain sound field data for imaging, but the time-domain
data have energy loss, which affects the imaging effect. The SWA is not affected
by energy attenuation characteristics. It reflects the damage through the change of
wavenumber information and has nothing to do with energy.
(2) The filter function in the WFA has a great influence on the imaging effect, and the
selection of the threshold is the core element that determines the imaging quality.
When sound waves encounter damage, they will undergo changes such as refraction
and reflection, which virtually increases the difficulty of threshold selection. The SWA
has stronger stability, does not need to filter the spatial wavenumber, and reduces the
cumbersomeness of selecting the filter threshold.
(3) SWA selects a specific frequency range according to the bandwidth of the excitation
signal, and averages the signals in the entire frequency band to generate a spatial
wavenumber diagram, which basically covers most of the effective data signals,
making the imaging results more accurate.
However, although the SWA improves the imaging effect, it also sacrifices a certain
detection efficiency. The use of short-space-time three-dimensional Fourier transform
makes its calculation time long. In contrast, the WFA does not consider the frequency
domain information, and only needs to perform the conversion between the spatial domain
and the wavenumber domain. The algorithm calculation time is shorter and the efficiency
is higher.
Based on the above test results, air-coupled ultrasound and full-field laser vibrome-
ter detection system has the characteristics of fast scanning, high spatial resolution and
complete wave field. It is a structural damage detection system with good development
prospects and can be applied to a variety of different algorithms.
Sensors 2021, 21, 609 17 of 18

5. Conclusions
The air-coupled ultrasound and full-field laser vibrometer detection system is pro-
posed in this paper. The air-coupled oblique incident ultrasonic excitation method realizes
the generation of single-mode ultrasonic Lamb waves in the aluminum plate and composite
plate. The laser vibrometer is used to measure the surface displacement to achieve rapid
signal acquisition with high spatial resolution. The above system enables the wavenumber
domain analysis methods to be realized in a completely noncontact manner, giving them a
broader application scenario. In this paper, aiming at the damaged structure in aluminum
and composite thin plates, the damaged topography reconstruction in the plate structure
is realized.
Based on the noncontact system, this paper uses multidimensional Fourier transform
to analyze the spatial energy, frequency, and wavenumber of the wavefield data in the time–
space domain. Through the Wavenumber Filtering Algorithm and Spatial Wavenumber
Algorithm, the visual imaging of the damaged area is realized, and completely present
the specific size and shape of the damage. Through two evaluation indicators of imaging
error and image contrast, the imaging results of different damages on different plates are
evaluated, and it is found that the overall detection effect of the SWA is better than that
of the WFA. The imaging error reduction amplitude is between 1.0% and 4.5%, and the
image contrast of the imaging effect has been effectively improved between 0.1 and 2.2,
and the damaged edge of the former is smoother. The advantages and disadvantages of
the two imaging algorithms are analyzed, the WFA is greatly affected by the attenuation of
signal energy propagation and the threshold selection is complicated. The SWA intuitively
reflects the damage situation through the change of the wavenumber, the detection is more
stable and the effect is better. However, the SWA process time is long, and the detection
efficiency is relatively lower.
Through the above verification and analysis, the noncontact damage detection system
based on air-coupled ultrasound and full-field laser vibrometer combines the advantages
of two detection methods and is a completely nondestructive structural damage detection
system with good development prospects.

Author Contributions: Conceptualization, H.Z. and X.R.; methodology, H.Z. and D.L.; validation,
D.L., X.R. and Z.W.; formal analysis, D.L.; investigation, D.L. and Z.W.; resources, H.Z. and Z.W.;
data curation, H.Z. and D.L.; writing—original draft preparation, H.Z. and X.R.; writing—review
and editing, H.Z., D.L. and X.R.; visualization, D.L. and Z.W.; supervision, H.Z. and X.R.; project
administration, H.Z. and X.R.; funding acquisition, H.Z. All authors have read and agreed to the
published version of the manuscript.
Funding: This work was funded by the National Natural Science Foundation of China (61771337),
Tianjin Key R&D Program (No. 19YFSLQY00080).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Yu, X.; Zuo, P.; Xiao, J.; Fan, Z. Detection of damage in welded joints using high order feature guided ultrasonic waves. Mech.
Syst. Signal. Pr. 2019, 126, 176–192. [CrossRef]
2. Fu, H.; Sharif-Khodaei, Z.; Aliabadi, M.H.F. An energy-efficient cyber-physical system for wireless on-board aircraft structural
health monitoring. Mech. Syst. Signal. Pr. 2019, 128, 352–368. [CrossRef]
3. Rautela, M.; Gopalakrishnan, S. Ultrasonic guided wave based structural damage detection and localization using model assisted
convolutional and recurrent neural networks. Expert Syst. Appl. 2020, 114189. [CrossRef]
4. Yan, W.J.; Chronopoulos, D.; Papadimitriou, C.; Cantero-Chinchilla, S.; Zhu, G.S. Bayesian inference for damage identification
based on analytical probabilistic model of scattering coefficient estimators and ultrafast wave scattering simulation scheme. J.
Sound Vib. 2020, 468, 115083. [CrossRef]
5. Dafydd, I.; Sharif Khodaei, Z. Analysis of barely visible impact damage severity with ultrasonic guided Lamb waves. Struct.
Health. Monit. 2020, 19, 1104–1122. [CrossRef]
6. Bustamante, L.; Jeyaprakash, N.; Yang, C.H. Evaluation of Defect Detection in Aluminium, CFRP and Epoxy Resin Plates Using
Non-contact Air-Coupled Ultrasonic Waves. Int. J. Precis. Eng. Manuf. 2020, 21, 1843–1856. [CrossRef]
Sensors 2021, 21, 609 18 of 18

7. Meng, W.R.; Zhou, Z.G. Non-contact Detection of Debonding Defects in a Steel-Lead Adhesive Structure Using Laser Ultrasonic
Technology. Lasers Eng. 2018, 41, 373–392.
8. Lopez, B.; Syed, A. Advances in EMAT Technology and Applications. Mater. Eval. 2020, 78, 1017–1024. [CrossRef]
9. Xiang, L.; Greenshields, D.; Dixon, S.; Edwards, R.S. Phased electromagnetic acoustic transducer array for Rayleigh wave surface
defect detection. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 2020, 67, 1403–1411. [CrossRef] [PubMed]
10. Kudela, P.; Wandowski, T.; Malinowski, P.; Ostachowicz, W. Application of scanning laser doppler vibrometry for delamination
detection in composite structures. Opt. Laser. Eng. 2016, 99, 46–57. [CrossRef]
11. Zhang, Y.; Yue, G.; Qi, L.; Rui, X. Research on continuous leakage location of stiffened structure based on frequency energy ratio
mapping method. J. Phys. D Appl. Phys. 2020, 53, 165301. [CrossRef]
12. Yang, L.J.; Li, Y.; Sun, J.J.; Zou, Y. Reflection and Transmission of Laser Ultrasonic Waves on Surface Defects. Laser Optoelectron.
Prog. 2019, 56, 041203. [CrossRef]
13. Chang, J.J.; Wan, T.L. Probabilistic damage imaging of composite plates by air coupled Lamb waves. IEEE Far East NDT New
Technol. Appl. Forum. 2019, 148–152. [CrossRef]
14. Wang, X.; Wu, H.; Zhang, X.; Zhang, D.; Gong, X.; Zhang, D. Investigation of a multi-element focused air-coupled transducer.
AIP Adv. 2018, 8, 095010. [CrossRef]
15. Wang, X.; Gong, X.; Li, C.; Wu, R.; Cao, X. Low insertion loss air-coupled ultrasonic transducer with parallel laminated
piezoelectric structure. AIP Adv. 2020, 10, 105331. [CrossRef]
16. Hettler, J.; Tabatabaeipour, M.; Delrue, S.; Van Den Abeele, K. Linear and nonlinear guided wave imaging of impact damage in
CFRP using a probabilistic approach. Materials 2016, 9, 901. [CrossRef]
17. Zhang, H.; Liu, Y.; Rui, X.; Wang, Z.; Yu, L. Boundary-enhanced reconstruction algorithm for probabilistic inspection of composite
plate defects with air-coupled Lamb waves. Smart. Mater. Struct. 2020, 29, 035019. [CrossRef]
18. Wang, Z.; Qiao, P. Backward wave separation method in a single transmitter and multi-receiver sensor array for improved
damage identification of two-dimensional structures. Int. J. Damage Mech. 2017, 26, 229–250. [CrossRef]
19. Michaels, T.E.; Michaels, J.E.; Ruzzene, M. Frequency–wavenumber domain analysis of guided wavefields. Ultrasonics 2011, 51,
452–466. [CrossRef]
20. Flynn, E.B.; Chong, S.Y.; Jarmer, G.J.; Lee, J.R. Structural imaging through local wavenumber estimation of guided waves. NDT E
Int. 2013, 59, 1–10. [CrossRef]
21. Tian, Z.; Yu, L.; Leckey, C. Delamination detection and quantification on laminated composite structures with Lamb waves and
wavenumber analysis. J. Intel. Mat. Syst. Str. 2015, 26, 1723–1738. [CrossRef]
22. Juarez, P.D.; Leckey, C.A. Multi-frequency local wavenumber analysis and ply correlation of delamination damage. Ultrasonics
2015, 62, 56–65. [CrossRef] [PubMed]
23. Tian, Z.; Yu, L.; Leckey, C. Rapid guided wave delamination detection and quantification in composites using global-local sensing.
Smart. Mater. Struct. 2016, 25, 085042. [CrossRef]
24. Jeon, J.Y.; Gang, S.; Park, G.; Flynn, E.; Kang, T.; Woo Han, S. Damage detection on composite structures with standing wave
excitation and wavenumber analysis. Adv. Compos. Mater. 2017, 26 (Suppl. S1), 53–65. [CrossRef]
25. Kudela, P.; Radzieński, M.; Ostachowicz, W. Identification of cracks in thin-walled structures by means of wavenumber filtering.
Mech. Syst. Signal. Pr. 2015, 50, 456–466. [CrossRef]
26. Kudela, P.; Radzienski, M.; Ostachowicz, W. Impact induced damage assessment by means of Lamb wave image processing.
Mech. Syst. Signal. Pr. 2018, 102, 23–36. [CrossRef]
27. Radzieński, M.; Kudela, P.; Marzani, A.; De Marchi, L.; Ostachowicz, W. Damage identification in various types of composite
plates using guided waves excited by a piezoelectric transducer and measured by a laser vibrometer. Sensors 2019, 19, 1958.
[CrossRef]
28. Harb, M.; Yuan, F. Non-contact ultrasonic technique for Lamb wave characterization in composite plates. Ultrasonics 2016, 64,
162–169. [CrossRef]
29. Harb, M.S.; Yuan, F.G. Barely visible impact damage imaging using non-contact air-coupled transducer/laser Doppler vibrometer
system. Struct. Health Monit. 2017, 16, 663–673. [CrossRef]