(
2
2
)
2
_
1
(1)
where the positive exponential relates to symmetric waves, the negative exponential relates
to antisymmetric waves,
2
=
2
c
2
l
2
,
2
=
2
c
2
t
2
, =
c
p
and , c
l
, c
t
, c
p
and h are the wave angular frequency, longitudinal velocity, transverse velocity,
phase velocity, and plate half thickness, respectively. For isotropic materials,
c
2
l
=
+ 2
and c
2
t
=
0
a
e
it
_
_
S
J
1
(
S
a)
N
S
(
S
)
D
S
(
S
)
H
(2)
1
(
S
r) +
A
J
1
(
A
a)
N
A
(
A
)
D
A
(
A
)
H
(2)
1
(
A
r)
_
_
(4)
4
where
D
S
= (
2
2
)
2
cos(h) sin(h) + 4
2
sin(h) cos(h)
D
A
= (
2
2
)
2
sin(h) cos(h) + 4
2
cos(h) sin(h)
N
S
= (
2
+
2
) cos(h) cos(h)
N
A
= (
2
+
2
) sin(h) sin(h)
and
0
, a, r, J
1
(.) and H
(2)
1
(.) are the maximum shear stress at the end of the actuator, the
actuator diameter, radial distance from the actuator, the rst order Bessel function of the
rst kind and the rst order Hankel function of the second kind, respectively. Expanding
upon the displacement results as found by Raghavan and Cesnik, the corresponding surface
strain wave solution is obtained from
r
(r, t) =
u
r
(r, t)
r
= i
0
a
e
it
_
_
S
J
1
(
S
a)
N
S
(
S
)
D
S
(
S
)
H
(2)
1
(
S
r)
r
+
A
J
1
(
A
a)
N
A
(
A
)
D
A
(
A
)
H
(2)
1
(
A
r)
r
_
_
. (5)
Applying the derivative for a rst order Hankel function of the second kind as dened by
H
(2)
1
(r)
r
=
1
2
_
H
(2)
0
(r) H
(2)
2
(r)
_
, (6)
the closed form surface strain wave solution for disc actuators is
r
(r, t) = i
0
a
2
e
it
_
_
S
J
1
(
S
a)
N
S
(
S
)
D
S
(
S
)
_
H
(2)
0
(
S
r) H
(2)
2
(
S
r)
_
+
A
J
1
(
A
a)
N
A
(
A
)
D
A
(
A
)
_
H
(2)
0
(
A
r) H
(2)
2
(
A
r)
_
_
_
. (7)
Based on the displacement and strain relations dened by Equations (4) and (7), re-
spectively, the modal eects of actuation frequency are investigated. Normalizing the dis-
placement and strain with respect to their peak values, the mode tuning for the composite
plate/transducers used in the following experiments (carbon-epoxy, AS4/3501-6, [0/90]
2s
and
10mm actuators) with respect to A
0
and S
0
Lamb waves are shown in Figure 3. Such that
the actuation frequency is tuned properly, a single mode excitation can be simulated (as is
further discussed later in the paper).
TIME REVERSAL METHOD
Current structural health monitoring techniques are often based on the comparison of current
structural responses with a baseline healthy response case. Variations from the baseline case
alerts the presence of damage. Unfortunately, operational and environmental conditions cause
variations in the healthy response and, therefore, it is generally necessary to have baseline
5
(a) (b)
Figure 3: Theoretical mode tuning for 1mm thick AS4/3501-6 [0/90]
2s
: (a) normalized dis-
placement and (b) normalized strain
data for all possible operational and environmental conditions. For most structures, there
exists a complex array of possible functional conditions, making it necessary to perform front-
end data mining for all of the structural response characteristics. Furthermore, the necessary
data storage increases the cost/decreases the eciency of these diagnostic techniques. The
concept of using a Time Reversal Methods (TRM) has recently been proposed as a baseline-
free diagnostics technique in attempts to address these problems. Currently, the TRM has
been successfully implemented to detect the presence and location of damage and has the
potential to determine extent and type of damage.
To implement the TRM method for damage diagnosis, a transducer array is setup and
used in a pitch-catch arrangement (that is, a signal is actuated from one transducer and
recorded at another). Each transducer must work both as a sensor and actuator, so typically
piezoelectric (PZT) transducers are used. The time reversal process is as follows (and shown
in Figure 4):
1. A Lamb wave tone burst is sent from transducer A and recorded at transducer B
2. The received signal at B is reversed in time [i.e., V
B
(t) V
B
(t)]
3. The time reversed signal is sent from B back to A where it is recorded
4. The received signal at A is time reversed and compared to the original signal actuated
Since the TRM is based on the linear reciprocity of the system, for a linear elastic structure,
reemission of the time reversed signal from B focuses the waves such that the signal received at
A is qualitatively identical to the original signal emitted. When nonlinearities are introduced
into the signal path, the linear reciprocity of the system breaks down and the time reversal
fails. In application to damage diagnostics, dierences between the original signal and the
nal signal signify the existence of damage within the signal path.
Theoretically, the TRM for plate diagnostics was rst validated by Wang (2003) and
Park (2007) for low frequency A
0
Lamb waves (actuated by disc transducers) and based on
Mindlin plate theory. Although, Mindlin plate theory improves upon classical plate theory
by incorporating shear deformation, it is still a simplied version of the three dimensional
elasticity equations for a plate and only approximates the propagation of low frequency A
0
6
Lamb waves. An enhanced investigation of the TRM was conducted by Xu and Giurgiutiu
(2007) using the exact solutions to the three dimensional elastic equations, resulting in an
exact validation of A
0
and S
0
Lamb waves (actuated by rectangular transducers) in TRM
applications.
B
A
1) Actuate signal
2) Read signal
3) Time reverse
5) Read signal
4) Actuate signal
6) Compare signals
Figure 4: Time reversal processes
Modied Time Reversal
In application, the time reversal method needs a complex actuator/sensor array. Although
PZT transducers alleviate some of the problems associated with actuation and sensing at the
same position, other complexities still exist. In particular, the hardware (power amplier,
data acquisition) necessary to support both actuation and sensing at all sensor locations is
quite high. Therefore, a Modied Time Reversal Method (MTRM) is developed.
The MTRM is as shown in Figure 5. Step by step, the procedure is as follows (where the
modied steps are denoted in bold font):
1. A Lamb wave tone burst is sent from transducer A and recorded at transducer B
2. The received signal at B is reversed in time [i.e., V
B
(t) V
B
(t)]
3. The time reversed signal is sent from A to B where it is recorded
4. The received signal at B is time reversed and compared to the original signal
actuated
Although the MTRM is similar to the original TRM, only one actuator is necessary for the
process. Over a large structure, instead of actuating and sensing between every transducer,
only a few actuators are necessary within a grid of sensors.
The MTRM can be validated by analyzing the process in Fourier space. For the trans-
ducer arrangement in Figure 5, the signal received at B due to an input signal at A can be
represented by
V
1
() =
I()G(), (8)
7
B
A
2) Read signal
3) Time reverse
1) Actuate signal
4) Actuate signal
5) Read signal
6) Compare signals
Figure 5: Modied time reversal process
where denotes the Fourier transform of the function, I is the input signal at A, and G is
the structural transfer function (and is dened in a later section). In Fourier space, the time
reversal of the signal at B is dened by
V ()
TR
(), (9)
where
denotes the complex conjugate of the function. Therefore, the time reversed signal
at B (step 2) can be represented by
V
tr
1
() =
I
()G
(). (10)
In the modied time reversal method proposed here, the time reversed signal is actuated from
A to B (step 3), resulting in
V
2
() =
V
tr
1
()G() =
I
()G
()G(). (11)
Time reversing the nal signal
V
2
(step 4),
V
tr
2
() =
I()G
()G() =
I |G()|
2
(12)
and transforming the signal back into the time domain,
V
tr
2
(t) =
1
2
_
I() |G()|
2
e
it
d. (13)
Such that |G()|
2
is independent of (as discussed later), the signal at the end of the time
reversal process is directly related to the originally actuated signal as found by
V
tr
2
(t) = CI(t), (14)
where C = |G()|
2
/2. This is the same result as found for the original TRM (Park et. al.,
2004 and Xu and Giurgiutiu, 2007).
8
Structural transfer function
The structural transfer function for Lamb wave propagation in a plate, as dened in Equation
(8), is rewritten as
G() =
V ()
I()
. (15)
Based on the coupling between a PZT sensor (with thickness t
a
and Youngs Modulus E
a
)
ideally bonded to a plate (with thickness t
p
and Youngs Modulus E
p
), the electric potential
experienced by the sensor is dened by
V (t) =
( +)t
a
d
13
(t) (16)
where is the plate surface strain at the center of the sensor, is related to the plate
displacement, strain and stress eld (for low frequencies = 4), d
13
is the piezoelectric
coupling coecient relating electric potential across the transducer thickness to in-plane radial
strain, and = E
p
t
p
/E
a
t
a
. For K t
a
( +)/d
13
,
G() =
F {K(t)}
I()
. (17)
For a unit input actuation I(t) = e
i
0
t
, the resulting strain response at a distance r from
the actuator is as found by Equation (5). Such that only the A
0
+ S
0
modes are excited,
substitution of Equation (5) into Equation (17) results in
G() =
_
S()
H
(2)
1
(
S
r)
r
+A()
H
(2)
1
(
A
r)
r
_
F
_
e
i
0
t
_
F {e
i
0
t
}
(18)
and reduces to
G() = S(
S
0
)
H
(2)
1
(
S
0
r)
r
+A(
A
0
)
H
(2)
1
(
A
0
r)
r
(19)
where
S() = iK
0
a
J
1
(a)
N
S
()
D
S
()
A() = iK
0
a
J
1
(a)
N
A
()
D
A
()
.
Such that r >> 3/4 (ADD FIGURE TO SHOW RELATION?), the rst order Hankel
function can be approximated to be
H
(2)
1
(r)
_
2
r
e
i(r
3
4
)
, (20)
resulting in the approximate derivative
H
(2)
1
(r)
r
T()e
ir
(21)
9
where
T() =
_
2
r
e
i
3
4
_
1
2r
+i
_
. (22)
The time reversal transfer function is therefore approximated to be
|G()|
2
T(
S
)S(
S
)
2
+
T(
A
)A(
A
)
2
+S(
S
)A
(
A
)T(
S
)T
(
A
)e
ir(
A
S
)
+
S
(
S
)A(
A
)T
(
S
)T(
A
)e
ir(
S
A
)
. (23)
Examining the eects of the time reversal transfer function on Equation (13) it can be
noted that, such that the original signal actuated is a single mode (A
0
or S
0
) and a single
frequency (i.e., G becomes independent of driving frequency), then the results of Equation
(14) are valid. However, for a multiple mode excitation S
0
+A
0
, the exponentials in the last
two terms in Equation (23) cause side-bands, symmetric about a central wave, to appear as
dened by the time shifting principle of the Fourier transform. Therefore, only single mode
excitations for use in the time reversal method are valid. A single mode excitation is typically
accomplished by tuning the excitation frequency such that one mode is dominate over the
other. Referring to Figure 3(b), it can be seen that the A
0
mode is dominant for frequencies
around 20kHz, 190kHz etc and that S
0
is dominant for frequencies around 130kHz, 310kHz
etc. In the following experiments, a 20kHz signal was used to simulate an A
0
single mode
excitation.
Since it is not possible to excite a single frequency, a narrow-band tone burst is typically
used to simulate a single frequency excitation. A tone burst consists of a windowed sine wave
of nite energy. There are various forms of windowing functions that are used to smooth the
signal such as, Gaussian, Hamming or Hanning windows. Typically, a Hanning window is
used and is dened by
h(t) =
_
0.5[1 cos(2t/)] for t [0, ]
0 otherwise
(24)
where is the period of the window. The resulting tone burst is represented by
x(t) = h(t) sin(2ft) (25)
where f is the frequency of the signal in Hz (Xu and Giurgiutiu, 2007). The respective
bandwidth of a Hanning windowed tone burst is as dened by
b =
4f
n
(26)
where n is the number of cycles in the tone burst (n = /f) (Raghavan and Cesnik, 2004).
Since the frequency of the actuation signal is chosen such that one mode is dominant, the
bandwidth of the signal is decreased by increasing the number of cycles in the tone burst.
However, although an extremely large number of cycles results in a very narrow bandwidth,
the corresponding signal can become exceedingly complex (especially for short signal paths).
Therefore, a 9.5 count tone burst was used in experiments such that the bandwidth was
narrow enough, but the resulting signals were not too complex.
10
Damage Index
The detection of damage using the time reversal method is accomplished by comparing the
input actuation signal with the signal due to the time reversal process. Deviations from the
actuated signal signify the existence of damage which is typically quantied using a Damage
Index (DI). However, dierent methods of determining and interpreting these DI values have
been used by researchers.
Sohn and Park (2003, 2007) used a damage index dened by
DI
SP
= 1
_
_
_
t
1
t
0
I(t)V (t)dt
_
2
_
t
1
t
0
I(t)
2
dt
_
t
1
t
0
V (t)
2
dt
(27)
where I(t) is the original signal, V (t) is the signal due to the time reversal process, and t
0
and t
1
dene the time interval over which the signals are compared. Based on this denition,
DI
SP
= 0 represents identical signals whereas DI
SP
= 1 represents very dissimilar signals.
Furthermore, it can be shown that for I(t) = V (t), DI
SP
= 0 for all . Note however that
is a generic constant which does not relate to C. Therefore, this damage index focuses on
shifts in frequency and changes in the overall shape of the time reversed signal but does not
capture dierences in signal amplitudes.
Contrary to Sohn and Parks quantication of dissimilarity, Giurgiutiu (2008) uses sim-
ilarity based on an L
2
error norm. So as to have the same basis as DI
SP
, we redene the
resulting damage index to be
DI
G
= VI
2
=
N
[V I]
2
N
(I)
2
(28)
where V and I are discrete data sets of the time reversed signal and the actuated signal,
respectively, and N is the number of data points. Since the amplitude of the two signals V
and I are typically of dierent orders of magnitude as dened by C in Equation (14), each
signal is normalized with respect to itself before determining the similarity. This damage
index also only focuses on shifts in frequency and changes in the overall shape of the time
reversed signal and not dierences in signal amplitudes since the amplitude information of
each signal is lost in the normalization process.
EXPERIMENTAL SETUP
An experimental set-up using a quasi isotropic carbon/epoxy plate with surface bonded piezo-
electric disc sensors was used to implement the MTR method. An eight ply carbon/epoxy
plate, in a symmetric [0/90]
2s
layup, was fabricated using AS4/35016 pre-preg and a vacuum
bag/oven curing technique. A piezoelectric (PZT) actuator/sensor network was integrated
onto the composite plate as shown in Figure 6(a) where each transducer was axed using
epoxy. The transducers have dimensions of 10mm diameter and 0.5mm thickness and the
plate was tested in a cantilever set-up. Two impact damages were introduced into the plate
as denoted by A and B in Figure 6(a). Damage A was constant for all tests where as the
damage at B was incrementally increased. For a controlled increase damage at B, a 175 g
11
steel ball was dropped down a tube, onto B, from a xed height of 4 ft. Based on the mode
tuning curve in Figure 3, a single frequency excitation of an A
0
Lamb wave was simulated
using a 20kHz, 9.5 cycle tone burst with amplitude of 100V . Do to a faulty sensor at location
4, only the signal paths relating to transducers 1, 2 and 3 were used in the analysis.
The experimental setup is as shown in Figure 6(b). A National Instruments PCI 6229
data acquisition card and a BNC-2110 board were used to actuate and record Lamb waves.
A QuickPack R power amplier, made by Active Control Experts, was used to amplify the
actuation signal and 3 Dual Mode Charge Ampliers, made by Kistler, were used to amplify
the signals read from the PZT sensors.
1
2 3
4
A B 9.5
5
(a)
Sensor Steel ball
Cantilevered plate
(b)
Figure 6: Experimental setup: (a) transducer arrangement and (b) experimental test equip-
ment
RESULTS AND DISCUSSIONS
Modied time reversal
To experimentally validate the use of the modied time reversal, a comparison between the
results generated by the TRM and MTRM was made. For each signal path, both the TRM
and MTRM were conducted and the results were compared. Figure 7 shows the results for
signal path 1-2. Qualitatively, it can be noted that the two signals are close to identical.
Quantitatively, the similarity between the two signals was determined as dened by
Similarity = 1 V
trm
V
mtrm
2
(29)
where V
trm
and V
mtrm
are the time reversal responses of the TRM and MTRM, respectively.
Results are as shown in Table 1. For all three signal paths, the results from the MTRM method
are very similar to the results as found by the TR method. For those areas that are dissimilar,
it is mostly due to the noise in the signal and could be mitigated using basic denoising
techniques. Finally, it can be shown that the TRM/MTRM results are not dependent on
initial actuator (i.e., the results are the same for initial actuation at 1 for path 1-2 as the results
for initial actuation at 2 for path 1-2). This realization cuts down on the plate diagnostic
procedure from (N 1)(N 2) TR/MTR evaluations down to N(N 1)/2 TRM/MTRM
procedures, where N is the number of transducers.
12
Figure 7: TR/MTR comparison
Table 1: Comparison between TR and MTR
Path Similarity(%)
1-2 89.18
1-3 92.18
2-3 98.42
Damage severity
A set of test cases, with incremental damage levels, were examined to determine the applica-
bility of the MTR for evaluation of damage severity. For each test case, the MTRM process
for all signal path permutations was performed. The test cases analyzed consisted of a base-
line case and three damage levels (DL) referenced to as DL
base
, DL
1
, DL
2
and DL
3
and
corresponding to 0, 4, 8 and 12 impacts at B, respectively . In evaluation of the results, both
damage indexes (Eqns. 27 and 28) were used.
Figure 8: Baseline DI calculations
With respect to the baseline results, as shown in Figure 8, path 1-3 exhibits the highest
13
(a) (b)
(c)
Figure 9: Damage index evaluation and comparison for increasing damage at B: (a) path
1-2, (b) path 1-3 and (c) path 2-3
DI values for both damage indexing systems. Due to the initial damage at A and healthy
conditions for paths 1-2 and 2-3, this is as expected. In analyzing the eects of increased
damage on the DI values, the DI results at each damage level were compared to the baseline
results (by DI
i
/DI
base
where i references the damage level case) and are as shown in Figure 9.
For both damage indexing systems, it is shown that the damage index for path 1-2 increases
incrementally as the damage level is increased at B whereas the damage indices for paths 1-3
and 2-3 remain close to the baseline values. It is important to note that baseline data is not
necessary to determine the presence of damage but it rather used as reference to show the
eects of damage on the indexing methods.
CONCLUSIONS
This paper has presented a time reversal method to detect the presence of damage within a
thin composite plate. A cross-ply carbon-epoxy composite plate with 8 layers was fabricated
and four PZT actuator/sensor were bonded to the plate surface. A modied time reversal
method was developed to enhance the applicability of the process by decreasing the necessary
hardware to monitor the health of the structure. Two damage indexing systems, as developed
by Park et al (2003 and 2007) and Giurgiutiu (2008), were used to evaluate the eects of
severity of damage. It was found that both damage indexing evaluations increased as severity
of damage increased, verifying the ability of the modied time reversal method to determine
14
not only the presence of damage, but also the severity.
This evaluation of damage severity is still in the preliminary stages of research and needs
further validation. Firstly, a larger array of damage types is necessary to investigate the
applicability of using the time reversal method for all damage cases. In particular, composite
delaminations are of interest in that they can not be seen by the naked eye. Furthermore,
a more in depth correlation between the actual severity of damage within the plate and the
indicated damage severity as indicated by the time reversal method is necessary. In this
study, there was little know about the actual extent of damage within the plate although
it was assumed that each impact increased the damage at the impact location. Use of
currently accepted non destructive testing methods, such as use of a C-scan or x-ray imaging,
to determine that actual state of damage within the structure would allow for a better com-
parison of the accuracy of the the time reversals ability to detect damage severity. Finally, it
also found that each indexing system exhibited dierent characteristics as the damage level
was increased. Further study of the damage indexing systems and their ability to identify
increases in damage severity is necessary to determine which method is best to be used in the
time reversal method.
REFERENCES
[1] Banerjee, S., Ricci, F., Monaco, E., Lecce, L. and Mal, A. 2007. Autonomous Impact Damage Monitoring
in a Stiened Composite Panel, Journal of Intelligent Material Systems and Structures, 18.
[2] Boller, C. 2000. Next Generation Structural Health Monitoring and its Integration into Aircraft Design,
International Journal of Systems Science, 31(11):13331349.
[3] Butenas, G. and Kazys, R. 2006. Investigation of Time-reversal Approach for Detection and Character-
ization of Ultrasonic Defects in Dispersive Plates, Ultragarsas, 61(4).
[4] Giurgiutiu, V. 2005. Tuned Lamb Wave Excitation and Detection with Piezoelectric Wafer Active Sensors
for Structural Health Monitoring, Journal of Intelligent Material Systems and Structures, 16:291305.
[5] Giurgiutiu, V. 2008. Structural Health Monitoring with Piezoelectic Wafer Active Sensors, Elsevier.
[6] Ing, R. and Fink, M. 1998. Time-reversed Lamb Waves, IEEE transactions on Ultrasonics, Ferro-
electrics, and Frequency Control, 45(4):10321043.
[7] Kaczmarek, H. 2003. Lamb Wave Interaction with Impact-induced Damage in Aircraft Composite: Use
of the A0 Mode Excited by Air-coupled Transducer, Journal of Composite Materials, 37(3).
[8] Kessler, S. S. 2002. Piezoelectric-Based In-Situ Damage Detection of Composite Materials for Structural
Health Monitoring Systems, PhD thesis at Massachusetts Institute of Technology.
[9] Kim, Y.-H., Kim, D.-H., Han, J.-H. and Kim, C.-G. 2007. Damage Assessment in Layered Composites
using Spectral Analysis and Lamb Wave, ScienceDirect:Composites Part B: Engineering, 38:800809.
[10] Lestari, W. and Qiao, P. 2005. Damage Identication for Carbon/Epoxy Laminated Composite Struc-
tures Based on Wave Propagation Analysis, 46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural
Dynamics & Materials Conference, .
[11] Park, H. W., Sohn, H., Law, K. H. and Farrar, C. R. 2007. Time Reversal Active Sensing for Health
Monitoring of a Composite Plate, Journal of Sound and Vibration, 302:5066.
[12] Raghavan, A. and Cesnik, C. E. S. 2004. Modeling of piezoelectric-based Lamb wave generation and
sensing for structural health monitoring Society of Photo-Optical Instrumentation Engineers (SPIE)
Conference Series, Bellingham, WA.
[13] Raghavan, A. and Cesnik, C. E. S. 2007. Review of Guided-wave Structural Health Monitoring, The
Shock and Vibration Digest, 39(2):91114.
[14] Sohn, H., Park, G., Wait, J. R., Limback, N. P. and Farrar, C. R. 2004. Wavelet-based Active Sensing
for Delamination Detection in Composite Structures, Smart Materials and Structures, 13:153160.
[15] Sohn, H., Park, H. W., Law, K. H. and Farrar, C. R. 2007a. Combination of a Time Reversal Process
and a Consecutive Outlier Analysis for Baseline-free Damage Diagnosis, Journal of Intelligent Material
Systems and Structures, 18.
15
[16] Toyama, N., Okabe, T. and Takeda, N. 2003. Lamb Wave Evaluation and Localization of Transverse
Cracks in Cross-ply Laminates, Journal of Material Science, 38:17651771.
[17] Tua, P. S., Quek, S. T. and Wang, Q. 2004. Detection of Cracks in Plates using Piezo-actuated Lamb
Waves, Smart Materials and Structures, 13:643660.
[18] Wang, C. H., Rose, J. T. and Chang, F.-K. 2003. A Computerized Time-reversal Method for Struc-
tural Health Monitoring, Nondestuctive Evaluation and Health Monitoring of Aerospace Materials and
Composites II, San Diego, California.
[19] Xu, B. and Giurgiutiu, V. 2007. Single Mode Tuning Eects on Lamb Wave Time Reversal with Piezo-
electric Wafer Active Sensors for Structural Health Monitoring, Journal of Nondestructive Evaluation,
26:123134.
16