Sei sulla pagina 1di 32

DK342X_book.

fm copy Page 213 Sunday, May 8, 2005 2:25 PM

10 Comparison of
Hilbert-Huang, Wavelet,
and Fourier Transforms
for Selected Applications
Ser-Tong Quek, Puat-Siong Tua, and Quan Wang

CONTENTS

10.1 Introduction ..................................................................................................214


10.2 Hilbert-Huang Transform.............................................................................215
10.3 Applications of HHT, WT, and FT to Experimental Data..........................216
10.3.1 Locating Crack in Aluminum Beam Using HHT ...........................216
10.3.2 Detection of Edges in Aluminum Plate...........................................219
10.3.3 Determination of Modal Frequencies of Aluminum Beam ............222
10.4 Concluding Remarks....................................................................................236
Acknowledgments..................................................................................................239
References..............................................................................................................243

ABSTRACT

This paper illustrates the suitability of three different signal-processing techniques,


namely, the Hilbert-Huang transform (HHT), the wavelet transform (WT), and the fast
Fourier transform (FFT), under different practical situations in relation to structural
health monitoring. Three experimentally obtained signals are analyzed. First, in the
time-of-flight analysis of flexural wave propagation in an aluminum beam, HHT is
found to be a more direct method compared to WT, as no knowledge of the actuation
frequency is required. However in the case of WT, if prior knowledge of the actuation
frequency is utilized combined with a refined scale search, WT produces more accurate
results. In another experiment involving the time-of-flight analysis of acoustic Lamb
wave propagation in an aluminum plate, piezoelectric actuators and sensors (PZTs) are
used to excite and receive direct and reflected waves at a specific frequency. HHT and

213

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 214 Sunday, May 8, 2005 2:25 PM

214 The Hilbert-Huang Transform in Engineering

WT give similar results in terms of precision and accuracy. Last, a modified Fourier
transform (FT) is investigated based on the concept of HHT and applied to the signals
collected from the free vibration of an aluminum beam under three different support
conditions. The proposed method is able to reveal the presence of higher vibration
modes than can be revealed by conventional FT of the original signal. The results
produced are close to that of the more common WT. It is believed that supporters of
FT will be more receptive to this modified HHT method.

Keywords Hilbert-Huang transform, wavelet transform, fast Fourier transform,


structural health monitoring, flexural wave, Lamb wave, free vibration

10.1 INTRODUCTION
Signal processing has gradually become an indispensable tool in the field of engi-
neering for extraction of important information from raw signals. In civil engineer-
ing, many practical and robust nondestructive evaluation (NDE) techniques devel-
oped for assessment of structural performance involve two key components, namely:
(1) data acquisition and (2) signal processing and interpretation. Although the use
of vibration measurements is a simpler and less costly method with respect to
instrumentation system in comparison with infrared thermography, ground-penetrat-
ing radar, acoustic emission monitoring, and eddy current detection, a key factor for
identifying damage precisely lies in having an appropriate data analysis method.
The most well known conventional signal processing technique is the Fourier
transform (FT). Fourier spectral analysis provides a general method for examining
the global energy–frequency distributions of a given signal. FT has high resolution
in the frequency domain but loses the resolution in the time domain. Despite this,
FT has been employed in health monitoring. For example, Crema et al. [1] used a
fast Fourier transform (FFT) analyzer to perform modal analysis to detect damage
in composite material structures. A broadband impulse input was applied, and the
damage that was induced by loads well above the working load caused a small
decrease in the eigen-frequencies for several modes. However, it can be problematic
to locate damage by using FT-based modal analysis techniques.
A few methods are available for processing signals of nonlinear and nonstation-
ary systems, of which the short-time Fourier transform (STFT) method and wavelet
transform (WT) method are widely adopted. Cook and Berthelot [2] used the
time-dependent energy spectrum from STFT to distinguish backscattered signals
from a single crack against those from a series of closely spaced cracks on steel
surface. Hurlebaus et al. [3] used the time–frequency spectrum obtained by perform-
ing STFT on the Lamb waves generated by laser to obtain the group velocity–fre-
quency domain. The notches are located from the autocorrelation plot of a series of
group velocity spectra at different assumed propagation distances.
Wavelet analysis is another adjustable window signal-processing technique to
handle nonstationary signals. Although WT appears similar to STFT, the basic wave
components are not limited to sinusoidal functions. Rather, classes of functions have
been proposed, from the simplest Haar wavelet to the more complicated higher order
Debauchies wavelet functions, to address various problems of time–frequency

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 215 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 215

resolution. WT has been widely used in the field of damage detection, such as in
modal-based detection (Staszewski [4]). An early work using WT to obtain a dis-
persive relation of group velocity of flexural waves in beam was presented by
Kishimoto et al. [5]. Lee and Liew [6] studied the sensitivity of WT in space domain
for identifying the damage location of a beam for small damage not easily detected
via conventional eigenvalue analysis. Quek et al. [7], on the other hand, used wavelet
coefficients over the time domain to locate cracks in beams via the time-of-flight
analysis of the wave propagation along the beam. However, WT has its shortcomings.
Experience shows that it can produce many spurious harmonics under different scales
that makes the analysis difficult or sometimes meaningless.
The most recently introduced signal-processing technique for analyzing nonlin-
ear and nonstationary time series data is the Hilbert-Huang transform (HHT) [8].
Empirical mode decomposition (EMD) of the data is first performed to segregate
the signals into narrow band components, and the Hilbert transform (HT) is then
applied on each mode. Its ability to analyze nonlinear and nonstationary time series
data [8–10] has attracted many applications. For example, Zhang et al. [11] adopted
HHT to assess structural damage from vibration recordings. Shim et al. [12]
exploited HHT’s ability to exhibit nonlinear and nonstationary behavior to locate
damages on bridges by using frequency sweep and controlled excitations. Quek et
al. [13] also employed HHT to analyze structural dynamic responses to detect and
locate anomalies in beams and plates. In addition, HHT has been used to analyze
experimental data collected by oscilloscope to investigate the dominance of the
lowest antisymmetric (A0) and symmetric (S0) modes of Lamb wave vary across the
frequency range adopted for excitation and good agreement with, which agrees well
with theoretical response prediction in terms of their relative dominance [14].
Individual signal-processing techniques have their own strength and weaknesses.
Suitability of a particular technique may, therefore, be problem dependent, especially
with respect to structural health monitoring problems. This paper compares the three
signal-processing techniques, namely, HHT, WT, and FFT, for signals obtained
experimentally. The first example concerns the time-of-flight analysis of flexural
wave propagation in an aluminum beam, where WT and HHT are compared. Next,
we compare the analysis from the two techniques on a two-dimensional problem
involving the propagation of acoustic Lamb waves in an aluminum plate. Last, we
apply a modified FT based on the EMD concept of HHT to the free vibration signals
of an aluminum beam under three different support conditions, we compare these
results with results from using conventional direct FFT, HHT, and WT to obtain the
modal frequencies.

10.2 HILBERT-HUANG TRANSFORM


Hilbert transform was first developed to process nonstationary narrow-band signals [15].
To apply it to signals in general, Huang et al. [8] proposed using the empirical mode
decomposition technique to decompose any given signal into a set of narrow-band
signals. Each component (which may be nonstationary) of the set is termed an intrinsic
mode function (IMF) of the original signal, on the IMF where HT can be carried out.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 216 Sunday, May 8, 2005 2:25 PM

216 The Hilbert-Huang Transform in Engineering

The distinct features of a narrow-band signal, and hence an IMF, are that (a)
over its entire length, the number of extrema and the number of zero-crossings either
are equal or differ at most by one; and (b) at any point, the mean value of the
envelope of the signal defined by the local maxima and the envelope defined by the
local minima is zero. This forms the basis on which the EMD technique was
developed. For the EMD to be applicable, it is assumed that (a) the signal has at
least one maximum and one minimum; (b) the characteristic time scale is defined
by the time lapse between the extrema; and (c) if the data are totally devoid of
extrema but contain only inflection points, then the signal can be differentiated one
or more times to reveal the extrema. Hilbert spectral analysis can then be performed
on each IMF. MATLAB 6.1 is adopted for the implementation of the HHT in this
study.

10.3 APPLICATIONS OF HHT, WT, AND FT


TO EXPERIMENTAL DATA
An accurate signal-processing technique to obtain the flight times is crucial to locate
damages in structures. Both WT and HHT are potentially applicable. The advantages
of HHT for damage detection in beams using flight times from narrow-band non-
stationary signals have been discussed by Quek et al. [13]. Waves in plates are even
more complex due to the dispersive nature of Lamb waves; hence, a technique that
provides good representation of localized events in both the frequency and energy
is vital for determining the location of the damage precisely. Another practical
problem of interest is the determination of higher frequencies from free vibration
data. In this section we examine a proposal to combine FT with HHT, using an
aluminum beam under three different boundary conditions as illustration.

10.3.1 LOCATING CRACK IN ALUMINUM BEAM USING HHT


Quek et al. [7] presented a wavelet analysis on experimental data to locate a crack
in an aluminum beam based on simple wave propagation considerations. The beam
considered has geometrical and material properties given in Table 10.1. Three sets

TABLE 10.1
Geometrical and Material Properties
of Aluminum Beam/Plate
Beam Plate

Dimensions (mm) 650 × 32 × 6 600 × 600 × 2


Young’s modulus, E (GPa) 73.1 102
Shear modulus, G (GPa) — 26
Mass density, ρ (kg/m3) 2790 2700

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 217 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 217

FIGURE 10.1 Schematic view of wave propagation for impact at 200 mm position.

of data with different boundary conditions are analyzed, namely, fixed-fixed, simply
supported, and cantilever conditions. The basic concept is shown in Figure 10.1, in
a schematic view of wave propagation due to an impact. The piezoelectric sensor
captures the wave signal, which contains timings of the direct, damage-reflected,
and boundary-reflected waves. By estimating these timings with a signal-processing
technique, the damage location can be deduced. In view of the dispersive nature of
the wave, only a wave at a particular frequency, with a corresponding wave propa-
gation velocity, is used so that accurate results can be obtained.
HHT is applied in this study on the same three sets of signals. One of the signals
obtained (and shown in Figure 10.2a) was decomposed via EMD into its IMF
components, shown in Figure 10.2b. It can be observed that the first IMF component
has higher amplitude than the other IMFs. This IMF component is then subjected
to HT, and the corresponding frequency–time and energy–time distributions plotted
in Figure 10.2c and d, respectively. From the energy–time distribution, the timings
of the energy peaks in the signal, which are due to the propagating wave passing
the sensor, are obtained. The frequencies corresponding to the energy peaks are also
noted so as to establish that they do not deviate significantly from each other. It
should be pointed out that not all IMFs are used, as the first IMF is distinct from
the others and contains only the frequency of interest, whereas other IMFs do not.
The results for the estimated damage position are summarized in Table 10.2 and
compared with those obtained with Gabor wavelet analysis with coarse (discrete
WT, scale = 12) and fine (“continuous” WT or CWT, scale = 12.2) resolution. The
wavelet results have been discussed in detail by Quek et al. [7], including the 3D
wavelet plot, and will not be repeated here. It can be observed that the HHT method
is able to identify the damage position with reasonable accuracy; the results are
better than the discrete wavelet results but worse than the continuous wavelet results.
However, using CWT for this application requires determining an appropriate scale
to process the final results. From past experience, this requires a fair amount of skill,
as the selection can be rather subjective and may lead to varied end results (see, for

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 218 Sunday, May 8, 2005 2:25 PM

218 The Hilbert-Huang Transform in Engineering

10 (b) 10
8 (a)

c(1)
6 0
4
2
Input

0 −10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−2
−4
−6
−8 5
−10

c(2)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0
−3
Time, t(s) ×10
−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

5
5000
(c)
Frequency (Hz)

c(3)
4000 0
3000
−5
2000 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1000
0 5
0 1 2 3 4 5
−3
Time, t(s) ×10
c(4)
0

−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
5
(d)
4 1
Amplitude

3
Residue

0
2
1 −1
0 −2
0 1 2 3 4 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−3 −3
Time, t(s) ×10 Time, t(s) ×10

FIGURE 10.2 (a) Dynamic response of a beam; (b) IMFs of response signal after EMD; (c)
frequency; (d) energy spectra of first IMF component.

TABLE 10.2
Comparison of Results Based on HHT Method and Wavelet Analysis
Wavelet Analysis
HHT Analysis Scale 12 Scale 12.2
Estimated Estimated Estimated
Damage Error Damage Error Damage *Error
End Condition Position (%) Position (%) Position (%)

Fixed ended 437.5 –2.78 492.0 9.33 450 0.00


Simply Supported 471.7 4.83 493.0 9.56 457 2.00
Cantilever 403.8 –10.26 542.0 20.00 457 2.00

Note: Actual damage position is 450 mm from left of the beam.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 219 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 219

D C
600

500
Sensor

400 Actuator

50 mm
300
75 mm

200 Line crack

y
100

0 X
A 0 100 200 300 400 500 600 B

FIGURE 10.3 Schematic view of PZT positions on plate.

example, Table 3 of Quek et al. [7]). This is due to the shortcomings of WT, which
produces false harmonics depending on the scale adopted, although WT also has
the ability to achieve high resolution in both frequency and time domain. The
advantage of HHT over WT in this case is that the HHT procedure is more direct
and can be performed to obtain the desired accuracy for the results without prior
knowledge of the excitation frequency (which in this case of impact load is broad-
band). It should be noted that there are other applications where CWT may be more
appropriate.

10.3.2 DETECTION OF EDGES IN ALUMINUM PLATE


In this example, we adopted the time-of-flight analysis of acoustic Lamb wave
propagation in a 600 mm × 600 mm aluminum plate to detect the plate’s edges, In
the experiment, a through crack, 1 mm wide and 30 mm long, inclined at 30º to one
side of the plate (side AB in Figure 10.3), was cut with a milling machine. The
location of the center of the crack is (270, 290) mm. Piezoelectric sensors and
actuators (PZTs) were used to excite and receive direct and reflected waves at a
specific frequency. A Lamb wave with a narrow-band (600 kHz) actuation pulse
(Figure 10.4) was activated based on the following equation:

( )
X (t ) = cos 2πω t e − a (t −t0 )
2
(10.1)

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 220 Sunday, May 8, 2005 2:25 PM

220 The Hilbert-Huang Transform in Engineering

20

15

10
Amplitude (V)
5

−5

−10

−15

−20
1.5 2 2.5 3 3.5 4
Time (s) ×10−5

FIGURE 10.4 Actuation pulse using Equation 10.1 at frequency 600 kHz.

where ω is the actuation frequency (Hz), and a and t0 are constants. This was done
with a function generator (Yokogawa FG300 15 MHz Synthesized FG) to feed a
PZT actuator (8.0 mm × 8.0 mm × 0.5 mm). The responses were captured by a PZT
sensor placed at a distance away and channeled to an oscilloscope (Yokogawa DL716
16-Channel) for monitoring and analysis. Figure 10.3 shows the positions of the
actuator and sensor. By estimating the flight time due to the propagation of either
a S0 or an A0 Lamb wave, the distance traveled by the wave from the actuator to
the sensor can be computed as

li = ∆ ti cg, (10.2)

where ti is the time of flight referenced from the actuation time, and cg is the group
velocity of either the S0 or A0 mode, determined either theoretically or experimen-
tally.
The response signal captured by the PZT sensor (Figure 10.5a) was processed
by both HHT and Gabor WT. The resulting energy spectra are given in Figure 10.5d
and Figure 10.6. In the HHT analysis, the peaks in the energy–time spectrum for
the IMF component containing the highest energy (in this case the first IMF) give
the wave arrival times of interest. Based on these energy peaks, the distances traveled
by the wave from the actuator to the sensor can be computed; these are summarized
in Table 10.3. It can also be observed from the frequency spectrum (see Figure
10.5b) that the frequency corresponding to the energy peaks is very close to the
actuation frequency, which indicates the accurate representation of localization
events by HHT. On the other hand, Gabor WT analysis on the same signal response
was performed based on the prior knowledge that the actuation frequency is 600
kHz, which corresponds to a scale of 20.2. Similar energy peaks were obtained,
which accounted for the wave pulses on the different paths received by the sensor.
The results obtained by WT analysis are also given in Table 10.3. These indicate

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 221 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 221

1 (b)
0.8 (a) 1

c(1)
0.6 0
0.4 −1
0 1 2
Amplitude

0.2 0.05

c(2)
0 0
−0.2 −0.05
−0.4 0 1 2
0.02
−0.6

c(3)
0
−0.8 −0.02
−1 0 1 2
0 1 2 0.02

c(4)
−4 0
Time, t(s) ×10
−0.02
0 1 2
0.01

c(5)
0
5 −0.01
×10 0 1 2
15 0.01

c(6)
(c) 0
Frequency, (Hz)

10 −0.01
0 ×10−3 1 2
5

c(7)
5 0
−5
0 0 1 2
0.01

c(8)
−5 0
0 1 2 −0.01
−4 0 1 2
Time, t(s) ×10 0.2
c(9)

0
−0.2
0 1 2
0.02
c(10)

1 A0/S0 0
(d) −0.02
0.8 (incident) 0 1 2
Amplitude

A0 0.1
c(11)

0.6 0
(crack) A0
0.4 (boundary, CD) −0.1
0 1 2
0.2 0.1
Res

0
0 −0.1
0 1 2 0 1 2
−4 −4
Time, t(s) ×10 Time, t(s) ×10

FIGURE 10.5 (a) Response signal captured by sensor on aluminum plate; (b) IMFs of
response signal after EMD; (c) frequency; (d) energy spectra of first IMF component.

1.2E–03
A0/S0
(incident)
1.0E–03

8.0E–04
Amplitude

A0
6.0E–04 (crack)
A0
4.0E–04 (boundary, CD)

2.0E–04

0.0E+00
0.0E+00 5.0E–05 1.0E–04 1.5E–04 2.0E–04
Time (s)

FIGURE 10.6 Gabor wavelet analysis of response signal given in Figure 10.5a.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 222 Sunday, May 8, 2005 2:25 PM

222 The Hilbert-Huang Transform in Engineering

TABLE 10.3
Results of Wave Propagation in Aluminum Plate
Distance Traveled Distance Traveled
Theoretical Distance Traveled by Reflected Wave by Reflected Wave
Velocity by Incident Wave from Line Crack from Boundary
Signal (m/s) (mm)a (mm)b (CD; mm)c
Analysis Error Error Error
Technique A0 A0 (%) A0 (%) A0 (%)

HHT 3134 52.9 5.80 196.1 –1.95 425.0 –1.62


WT 52.7 5.40 196.4 –1.80 424.0 –1.85
a Actual distance traveled by incident wave is 50 mm.
b Actual distance traveled by crack-reflected wave is 200 mm.
c Actual distance traveled by boundary (CD)-reflected wave is 432 mm.

that HHT and WT give similar results in terms of precision and accuracy. HHT does
not require prior knowledge of the activation frequency.

10.3.3 DETERMINATION OF MODAL FREQUENCIES OF ALUMINUM BEAM


In this part of the study, we consider an aluminum beam under three support
conditions, shown in Figure 10.7, with material properties similar to the plate in
Section 10.3.2 (see Table 10.1). The theoretical modal frequencies of the beam were
first obtained (see Table 10.4) by using the following equation [16]:

(β nl )2 EI
fn = , n = 1, 2, 3, ... (10.3)
2πl 2 ρA

where l, E, I, ρ, and A are the length, Young’s modulus, moment of inertia, density,
and cross-sectional area of the beam, respectively; n corresponds to the vibration
modes; and βn is solved from the following equations, respectively, for cantilever,
simply supported, and fixed-fixed conditions.

cos β n l cosh β n l = −1 (10.4a)

sinβ n l = 0 (10.4b)

cos β n l cosh β n l = 1 (10.4c)

For the experiment, the free vibration responses of the beam under the three different
support conditions (Figure 10.7) were monitored and collected by an oscilloscope
(Yokogawa DL716 16-Channel; see Figure 10.8). FFT was first performed directly on

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 223 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 223

Aluminum
Piezoelectric beam 6 mm
x Sensor

(a)
32 mm
y

920 mm
0 mm 150 mm
Aluminum
Piezoelectric beam 6 mm
x Sensor

(b) 32 mm
y

800 mm
0 mm 150 mm
Aluminum
Piezoelectric beam 6 mm
x Sensor

(c)
32 mm
y

800 mm
0 mm 150 mm

FIGURE 10.7 Schematic setup of aluminum beam under different support conditions: (a)
cantilever; (b) simply supported; (c) fixed-fixed for modal frequency analysis.

the response signal. For example, Figure 10.9a and b shows the linear and logarithmic
plots of the Fourier spectrum, respectively, for the cantilevered beam. In the linear
plot of the Fourier spectrum (Figure 10.9a), the first two modal frequencies of the
beam can be easily identified (5.99 Hz and 34.93 Hz respectively). However, the
third mode may be easily overlooked due to the low energy of the displayed peak.
On the other hand, the logarithmic scale plot of the spectrum (Figure 10.9b) exhibits
the first three modal frequencies clearly. This is also true for the fixed-fixed support
condition (see Figure 10.10). However, for the simply supported condition, it is not
easy to distinguish the modes from the logarithmic scale plot of the Fourier spectrum.
It can be observed in Figure 10.11b that besides the three fundamental modal
frequencies, there are also a number of energy peaks at other frequency ranges
having amplitudes of order close to the modal frequencies. Hence for beams with
closely spaced frequencies, the Fourier spectrum may not be appropriate.
EMD was performed on the responses given in Figure 10.8. Consider the results
for the cantilevered beam: three decomposed IMF components and one residual
component are obtained, as depicted in Figure 10.12a. The individual IMFs are
subjected to FT, and the resulting spectra are shown in Figure 10.12b through d.
The residual component is not subjected to FT as it represents the mean or residual
trend of the captured response signal. Figure 10.12b through d clearly identifies the

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 224 Sunday, May 8, 2005 2:25 PM
224
© 2005 by Taylor & Francis Group, LLC

TABLE 10.4
Analysis of Modal Frequencies of Aluminum Beam under Different Support Conditions Obtained
by FFT, EMD Followed by FT, HHT, and WT
FFT EMD + FT HHT WT
Theoretical Freq Error Freq Error Freq Error Freq Error
Support Conditions Mode Frequency (Hz) (Hz) (%) (Hz) (%) (Hz) (%) (Hz) (%)

The Hilbert-Huang Transform in Engineering


Cantilever 1st 5.86 5.99 2.20 5.76 -1.72 5.77 1.52 5.64 3.73
2nd 36.73 34.93 –4.91 35.23 –4.09 34.63 5.72 35.86 2.39
3rd 102.86 96.80 –5.89 98.00 –4.73 96.62 6.06 101.56 1.26
Simply supported 1st 22.77 22.89 5.17 22.95 5.44 22.80 4.75 23.20 6.59
2nd 87.03 81.80 –6.01 82.85 –4.81 77.88 –10.52 81.25 –6.65
3rd 195.84 165.80 –15.34 167.50 –14.47 159.18 –18.72 162.50 –17.02
Fixed-fixed 1st 49.33 48.00 –2.69 48.00 –2.69 46.96 4.80 49.85 1.06
2nd 135.97 138.00 1.49 138.00 1.49 135.34 0.47 135.42 –0.41
3rd 266.56 268.00 0.54 — — 263.19 1.26 270.83 1.60
4th 440.64 446.00 1.22 460.00 4.39 — — 451.39 2.44
DK342X_book.fm copy Page 225 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 225

25
(a)
20
15
10
Amplitude

5
0
−5
−10
−15
−20
−25
0 0.2 0.4 0.6 0.8 1
Time, t(s)

15
(b)

10

5
Amplitude

−5

−10

−15
0 0.2 0.4 0.6 0.8 1
Time, t(s)

25
(c)
20

15

10
Amplitude

−5

−10

−15

−20
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

FIGURE 10.8 Free vibration response of aluminum under (a) cantilever; (b) simply sup-
ported; (c) fixed-fixed conditions.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 226 Sunday, May 8, 2005 2:25 PM

226 The Hilbert-Huang Transform in Engineering

×104
2.5
1st Mode (a)
= 5.99 Hz
2

1.5
Amplitude

0.5 2nd Mode


rd
= 34.93 Hz 3 Mode
= 96.8 Hz
0
0 50 100 150 200 250
Frequency (Hz)

105 1st Mode


= 5.99 Hz (b)
4
10 2nd Mode
= 34.93 Hz rd
103 3 Mode
= 96.8 Hz
Amplitude

102

101

100

−1
10
0 50 100 150 200 250
Frequency (Hz)

FIGURE 10.9 FFT analysis of response of cantilever beam given in Figure 10.8a, plotted on
(a) linear scale and (b) log-scale.

first three modal frequencies of the beam (5.76 Hz, 35.23 Hz, and 96.8 Hz respec-
tively). Even for the third mode, the appearance of the model frequencies is very
distinctive because the EMD process has effectively decomposed the response signal
into its respective components, which are narrow band. Hence, the energy peak of
the third mode is not obscured by the first and second modes, which contain relatively
higher energies. Similar analyses are done for the same beam under the two other
support conditions, and the results (see Figure 10.13 and Figure 10.14) obtained are
comparable to those obtained by the conventional direct FT. However, for the case
of the fixed-fixed condition, the third modal frequency could not be identified by
the FT method (Figure 10.14). Instead, the first and second modal frequencies are

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 227 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 227

1200 st
1 Mode (a)
= 22.89 Hz
1000

800
Amplitude

600

400

200 nd
rd
3 Mode
2 Mode
= 165.8 Hz
= 81.8 Hz
0
0 50 100 150 200 250
Frequency (Hz)

st
1 Mode
rd
3 Mode (b)
2 nd
10 = 22.89 Hz 2 Mode = 165.8 Hz
= 81.8 Hz

0
10
Amplitude

−2
10

−4
10

0 50 100 150 200 250


Frequency (Hz)

FIGURE 10.10 FFT analysis of response of fixed-fixed beam given in Figure 10.8c, plotted
on (a) linear scale and (b) log-scale.

each contained in two IMF components (fourth and fifth IMFs and second and third
IMFs respectively; see Figure 10.14). On closer observation, the third and fourth
modal frequencies can be spotted in the Fourier spectrum of the second IMF com-
ponent, but they are obscured by the stronger presence of the second modal fre-
quency. This setback of the proposed method can be explained by the fact that IMFs
may contain mix modes. This may occur when the frequency of interest in an IMF
is only contained along a particular segment of the signal and not throughout the
entire time domain. This is the so-called intermittency problem discussed by Huang
et al [17].
The original HHT was also performed on the decomposed IMF components of
the response signals for the beam under the different support conditions. However,
as mentioned earlier, the frequency of interest in an IMF may be contained only

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 228 Sunday, May 8, 2005 2:25 PM

228 The Hilbert-Huang Transform in Engineering

1400 nd
2 Mode (a)
1200
st
1 Mode = 138 Hz
= 48 Hz
1000
Amplitude

800

600

400

200 3rd Mode


= 268 Hz
0
0 100 200 300 400 500
Frequency (Hz)

4
10 st nd
1 Mode 2 Mode (b)
3 = 48 Hz = 138 Hz
10
rd
3 Mode
2
10 = 268 Hz
th
4 Mode
Amplitude

1
10 = 446 Hz

0
10

−1
10

−2
10

10−3
0 100 200 300 400 500
Frequency (Hz)

FIGURE 10.11 FFT analysis of response of simply supported beam given in Figure 10.8b,
plotted on (a) linear scale and (b) log-scale.

along a particular segment of the signal; hence, a weighted average using the
amplitude is performed on segments of the frequency spectrum from HT for each
IMF to obtain the modal frequencies of the beam. To illustrate, spectra for the beam
under the three support conditions will be presented in detail individually.
First, consider the HT of the IMFs for the cantilever beam (Figure 10.15). From
the Hilbert spectra of the three IMF components, it can be observed that the EMD
has effectively decomposed the original response signal into its different compo-
nents, which are well contained within a narrow band, as the fluctuation of the
frequency along the time axis is rather small. For this case, weighted averaging was
performed over the entire time domain for each of the IMF components. The modal
frequencies obtained are 101.56 Hz, 35.86 Hz, and 5.64 Hz from the first, second,
and third IMFs, respectively, which corresponds well with the theoretical values.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 229 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 229

c(1)
0

−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
c(2)

−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (a)
20
c(3)

−20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
Residue

5
0
−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time, t(s)

180 400
rd (b) nd (c)
160 3 Mode 2 Mode
350
= 96.8 Hz = 35.23 Hz
140 300
120
Amplitude

Amplitude

250
100
200
80
150
60
40 100

20 50
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Frequency (Hz) Frequency (Hz)

4
×10
3.5
st (d)
1 Mode
3 = 5.76 Hz
2.5
Amplitude

1.5

0.5

0
0 50 100 150 200 250
Frequency (Hz)

FIGURE 10.12 (a) IMF components of response of cantilever beam given in Figure 10.8a;
FT analysis of (b) first; (c) second; (d) third IMF component.

Next, for the simply supported beam, it can be observed in the HT spectra shown
in Figure 10.16 that the frequency is stable only over a certain time range for each
IMF. For example, for the first IMF, the frequency has a small fluctuation between

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 230 Sunday, May 8, 2005 2:25 PM

230 The Hilbert-Huang Transform in Engineering

c(1)
0

−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
5
c(2)

−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (a)
10
c(3)

−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
Residue

0
−1
−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time, t(s)

45 4.5
rd (b) nd (c)
40 3 Mode 4 2 Mode
= 167.5 Hz = 82.85 Hz
35 3.5
30 3
Amplitude

Amplitude

25 2.5
20 2
15 1.5
10 1
5 0.5
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Frequency (Hz) Frequency (Hz)

900
rd (d)
800 1 Mode
= 22.95 Hz
700
600
Amplitude

500
400
300
200
100
0
0 50 100 150 200 250
Frequency (Hz)

FIGURE 10.13 (a) IMF components of response of simply supported beam given in Figure
10.8b; FT analysis of (b) first; (c) second; (d) third IMF component.

0.1 sec and 0.3 sec; hence, a weighted average on the frequency is done over this
time range. The frequency obtained is 159.18 Hz, which is closer to the third
vibration mode of the beam. Similar analysis is done for the second IMF (between

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 231 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 231

10

c(1) 0
−10
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
5
c(2)

0
−5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
10
c(3)

0
−10
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 (a)
5
c(4)

0
−5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
2
c(5)

0
−2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0.5
Residue

−0.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time, t(s)

0.7 90
(b) nd (c)
80 2 Mode
0.6 = 138 Hz
70
0.5
60
Amplitude

Amplitude

rd
0.4 50 3 Mode
(268 Hz)
0.3 40
30 th
0.2 4 Mode
20 (460 Hz)
0.1 10
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Frequency (Hz) Frequency (Hz)

600 350
nd (d) st (e)
2 Mode 1 Mode
500 = 138 Hz 300 = 48 Hz
250
400
Amplitude
Amplitude

200
300
150
200
100
100 50

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Frequency (Hz) Frequency (Hz)

FIGURE 10.14 (a) IMF components of response of fixed-fixed beam given in Figure 10.8c;
and FT analysis of (b) first; (c) second; (d) third; (e) fourth IMF; (f) fifth IMF component.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 232 Sunday, May 8, 2005 2:25 PM

232 The Hilbert-Huang Transform in Engineering

45 st
1 Mode (f )
40 = 48 Hz
35

30
Amplitude

25

20

15

10

0
0 100 200 300 400 500 600
Frequency (Hz)

FIGURE 10.14 (continued)

0.12 sec and 0.28 sec), and the result (77.88 Hz) gives the second vibration mode.
For the third IMF, a rather stable fluctuation of the frequency is observed over the
entire time domain; hence, a weighted average is taken over the entire data range. This
obviously gave the result corresponding to the first fundamental frequency (22.80 Hz).
The Hilbert spectra for the IMF components of the fixed-fixed condition are
shown in Figure 10.17. Considering the first IMF, a large fluctuation of the frequency
over the entire time domain is observed. This can be explained by the fact that the
filtered component effectively contains the noise in the original response signal;
hence, meaningful modal parameters of the beam can be extracted from it. The
second IMF reveals a stable frequency region with high amplitude over the time
domain between 0.05 sec and 0.12 sec. Taking the weighted average over this time
range gives the second modal frequency (135.34 Hz). For the third IMF, two stable
frequency fluctuation regions with high amplitude can be identified, namely, from
0 to 0.16 sec and from 0.16 to 0.46 sec. By performing a weighted average on the
two regions, two fundamental frequencies are obtained (138 Hz and 263.19 Hz). A
similar stable region is observed for the fourth IMF, from 0.04 to 0.15 sec; this
corresponds to an averaged frequency of 46.96 Hz, which gives the first modal
frequency of the beam. In the fifth IMF, the two stable regions can be pinpointed,
from 0 to 0.15 sec and from 0.24 to 0.48 sec. These two regions give a weighted
average frequency of 42 Hz and 43.3 Hz respectively.
Last, wavelet analysis using the Morlet function was also performed on the three
signals of the beam under the three support conditions. WT was first done on the
original response signals for the beam over a large scale range. By observing the
dominant energy contents in this overall spectrum, different scales were zoomed in
to obtain the dominant frequency contents. For example, WT was first performed
for the vibration response of the cantilevered beam given in Figure 10.8a over a
wide scale range of 1 to 256. This spectrum plot (see Figure 10.18a) shows three
regions with a consistent existence of a corresponding frequency throughout the

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 233 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 233

500 2.0
450
400
350
Frequency, (Hz)

300
250 1.0 (a)
200
150
100
50
0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

100 8.0
90
80
70
Frequency, (Hz)

60
50 4.0 (b)
40
30
20
10
0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

60 20

50
Frequency, (Hz)

40

30 10 (c)

20

10

0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

FIGURE 10.15 HT spectrum for (a) first, (b) second, and (c) third IMF component for
response of cantilever beam as given in Figure 10.12a.

entire time domain, namely, in scales 105 to 183, 14 to 27, and 4 to 12. By focusing
toward these regions, a more refined dominant frequency can be deduced, namely
at scales of 144, 22.66, and 8, which corresponds to frequencies of 5.64 Hz, 35.86

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 234 Sunday, May 8, 2005 2:25 PM

234 The Hilbert-Huang Transform in Engineering

500 4.0
450
400
350
Frequency, (Hz)

300
250 2.0 (a)
200
150
100
50
0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

300 3.0

250
Frequency, (Hz)

200

150 1.5 (b)

100

50

0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

350 8
300

250
Frequency, (Hz)

200
4 (c)
150

100

50

0 0
0 0.2 0.4 0.6 0.8 1
Time, t(s)

FIGURE 10.16 HT spectrum for (a) first, (b) second, and (c) third IMF component for
response of simply supported beam as given in Figure 10.13a.

Hz, and 101.56 Hz respectively. These results give the three fundamental frequencies
of the beam. The wavelet analyses for the other two support conditions are presented
in Figure 10.19 and Figure 10.20.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 235 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 235

5000 10
4500
4000
3500
Frequency, (Hz)

3000
2500 5 (a)
2000
1500
1000
500
0 0
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

2500 5

2000
Frequency, (Hz)

1500
2.5 (b)
1000

500

0 0
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

2000 10
1800
1600
1400
Frequency, (Hz)

1200
1000 5 (c)
800
600
400
200
0 0
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

FIGURE 10.17 HT spectrum for (a) first, (b) second, (c) third, (d) fourth, and (e) fifth IMF
component for response of fixed-fixed beam as given in Figure 10.14a.

The results obtained by the different signal analysis methods are compared in
Table 10.4. It can be observed that all the methods produce results of approximately
the same order of error. The methods are equally feasible in determining the modal

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 236 Sunday, May 8, 2005 2:25 PM

236 The Hilbert-Huang Transform in Engineering

800 6
700
600
Frequency, (Hz)

500
400 3 (d)

300
200
100
0 0
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

300 2

250
Frequency, (Hz)

200

150 1 (e)

100

50

0 0
0 0.1 0.2 0.3 0.4 0.5
Time, t(s)

FIGURE 10.17 (continued)

frequencies of the beam. However, these methods do contain imperfections. For


example, the direct FT method may not be very effective for cases where the modal
frequencies have energy of same order as that of other frequency ranges. Similarly,
for the modified FT with EMD process, the identification of certain modal frequen-
cies may not be possible when the IMF components contain mixed modes. For HHT,
the definition of the stable frequency region over the time domain, although subjec-
tive, is least problematic. For WT, identification of the precise scale from a range
of scales for the dominant frequency can be ambiguous.

10.4 CONCLUDING REMARKS


A comparison of three signal-processing techniques is presented. In the first study of
flexural wave propagation in an aluminum beam, HHT is shown to be a more direct
method compared to WT when no knowledge of the actuation frequency is available.
In the second experiment, involving acoustic Lamb wave propagation in an aluminum
plate, HHT and WT give similar results for the analysis of the propagating Lamb waves
actuated and received by the PZTs. Good WT results hinge on the fact that the actuation

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 237 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 237

248
235
222
209
196
183
170
157
144
Scale

131 (a)
118
105
92
79
66
53
40
27
14
1
Scale of colors from MIN to MAX
Time
Scale

144

Scale of colors from MIN to MAX


(b)
Coefficients Line - Ca, b for scale a = 144 (frequency = 5.642)
300
200
Amplitude

100
0
−100
−200
−300
100 200 300 400 500 600 700 800 900 1000
Time

FIGURE 10.18 Morlet WT analysis for response of cantilever beam given in Figure 10.8a
at scales (a) 1–256, (b) 144, (c) 22.66, and (d) 8 with sampling rate 1000 samples/sec.

frequency is narrowband and known a priori. For the last case, the combination of the
EMD process from HHT with FT on the decomposed IMF components has proven to
be capable of revealing modal frequencies of the aluminum beam having relatively
lower energy contents. These modal frequencies may be overlooked by the traditional
direct FFT, unless a logarithmic plot is done for the energy spectrum. However, the
problem of mixed modes within individual IMFs needs to be addressed. The original

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 238 Sunday, May 8, 2005 2:25 PM

238 The Hilbert-Huang Transform in Engineering

Scale

22.66

Scale of colors from MIN to MAX


(c)
Coefficients Line - Ca, b for scale a = 22.66 (frequency = 35.856)
50
Amplitude

−50
100 200 300 400 500 600 700 800 900 1000
Time
Scale

Scale of colors from MIN to MAX


(d)
Coefficients Line - Ca, b for scale a = 8 (frequency = 101.563)
15
10
Amplitude

5
0
−5
−10
100 200 300 400 500 600 700 800 900 1000
Time

FIGURE 10.18 (continued)

HHT with knowledge of the problem does provide good results when it is used with
an averaging concept. These examples show that HHT is a suitable tool for processing
nonstationary wave propagation signals encountered in damage detection studies. It
provides a good representation of localized events in both the frequency and energy of
any transient signal collected. Indeed, it is a simple signal-processing tool to use and
provides reasonably good results.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 239 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 239

248
235
222
209
196
183
170
157
144
Scale

131
118
105 (a)
92
79
66
53
40
27
14
1
Scale of colors from MIN to MAX
Time
Scale

35

Scale of colors from MIN to MAX


(b)
Coefficients Line - Ca, b for scale a = 35 (frequency = 23.214)
40

20
Amplitude

−20

−40
100 200 300 400 500 600 700 800 900 1000
Time

FIGURE 10.19 Morlet WT analysis for response of simply supported beam given in Figure
10.8b at scales (a) 1–256, (b) 33.5, (c) 10, and (d) 5 with sampling rate 1000 samples/sec.

ACKNOWLEDGMENTS
The authors wish to thank the National University of Singapore for providing the
support to perform this study.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 240 Sunday, May 8, 2005 2:25 PM

240 The Hilbert-Huang Transform in Engineering

Scale

10

Scale of colors from MIN to MAX


(c)
Coefficients Line - Ca, b for scale a = 10 (frequency = 81.250)
4
3
2
Amplitude

1
0
−1
−2
−3
−4
100 200 300 400 500 600 700 800 900 1000
Time
Scale

Scale of colors from MIN to MAX


(d)
Coefficients Line - Ca, b for scale a = 5 (frequency = 162.500)
8
6
4
Amplitude

2
0
−2
−4
−6
−8
100 200 300 400 500 600 700 800 900 1000
Time

FIGURE 10.19 (continued)

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 241 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 241

1958
1855
1752
1649
1546
1443
1340
1237
1134
Scale

1031 (a)
928
825
722
619
516
413
310
207
104
1
Scale of colors from MIN to MAX
Time
Scale

163

Scale of colors from MIN to MAX

(b)
Coefficients Line - Ca, b for scale a = 163 (frequency = 49.847)
80
60
40
Amplitude

20
0
−20
−40
−60
−80
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time

FIGURE 10.20 Morlet WT analysis for response of fixed-fixed beam given in Figure 10.8c
at scales (a) 1–2042, (b) 163, (c) 60, (d) 28, and (e) 18 with sampling rate at 10000 samples/sec.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 242 Sunday, May 8, 2005 2:25 PM

242 The Hilbert-Huang Transform in Engineering

Scale

60

Scale of colors from MIN to MAX


(c)
Coefficients Line - Ca, b for scale a = 60 (frequency = 135.417)
80
60
40
Amplitude

20
0
−20
−40
−60
−80
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time
Scale

30

Scale of colors from MIN to MAX


(d)
Coefficients Line - Ca, b for scale a = 30 (frequency = 270.833)
15
10
Amplitude

5
0
−5
−10
−15
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time

FIGURE 10.20 (continued)

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 243 Sunday, May 8, 2005 2:25 PM

Comparison of Hilbert-Huang, Wavelet, and Fourier Transforms 243

Scale

163

Scale of colors from MIN to MAX


(e)
Coefficients Line - Ca, b for scale a = 163 (frequency = 49.847)
80
60
40
Amplitude

20
0
−20
−40
−60
−80
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time

FIGURE 10.20 (continued)

REFERENCES
1. Crema L B, Peroni I and Castellani A (1985). Modal Tests on Composite Material
Structures Application in Damage Detection. Proc. Int. Modal Anal. Conf. Exhibit
Vol. 2, pp. 708–713.
2. Cook D A and Berthelot Y H (2001). Detection of Small Surface-Breaking Fatigue
Cracks in Steel Using Scattering of Rayleigh Waves. NDT & E Int. 34:483–492.
3. Hurlebaus S, Niethammer, Jacobs L J and Valle C (2001). Automated Methodology
to Locate Notches with Lamb Waves. Acous. Soc. Am. ARLO 2(4):97–102.
4. Staszewski W J (1998) Structural and Mechanical Damage Detection Using Wavelets.
Shock Vibr. Digest 30(6):457–472.
5. Kishimoto K, Inoue H, Hamada M and Shibuya T (1995). Time Frequency Analysis
of Dispersive Waves by Means of Wavelet Transform. ASME J. Appl. Mech. 62:
841–846.
6. Lee Y Y and Liew K M (2001). Detection of Damage Locations in a Beam Using
the Wavelet Analysis. Int. J. Struct. Stab. Dyn. 1(3):455–465.
7. Quek S T, Wang Q, Zhang L and Ong K H (2001). Practical Issues in Detection of
Damage in Beams Using Wavelets. Smart Mater. Struct. 10(5):1009–1017.
8. Huang H, Norden E, Zheng S, Long S R, Wu M C, Shih H H, Zheng Q, Yen N C,
Tung C C and Liu H H (1998). The Empirical Mode Decomposition and Hilbert
Spectrum for NonLinear and Nonstationary Time Series Analysis. Proc. R. Soc. Lond.
A 454:903–995.
9. Huang H, Norden E, Zheng S and Long S R (1999). A New View of Nonlinear Water
Waves: The Hilbert Spectrum. Annu. Rev. Fluid Mech. 31:417–457.

© 2005 by Taylor & Francis Group, LLC


DK342X_book.fm copy Page 244 Sunday, May 8, 2005 2:25 PM

244 The Hilbert-Huang Transform in Engineering

10. Huang H, Norden E, Chern C C, Huang K, Salvino L W, Long S R and Fan K L


(2001). A New Spectral Representation of Earthquake Data: Hilbert Spectral Analysis
of Station TCU129, Chi-Chi, Taiwan, 21 September 1999. Bull. Seism. Soc. Am.
91(5):1310–1338.
11. Zhang R R, King R, Olson L and Xu Y (2001). A HHT View of Structural Damage
from Vibration Recordings Proc. ICOSSAR ’01 (San Diego) – CD.
12. Shim S H, Jang S A, Lee J J and Yun C B (2002). Damage Detection Method for
Bridge Structures Using Hilbert-Huang Transform Technique. 2nd Int. Conf. Struct.
Stab. Dynam., Dec. 16–18, 2002, Singapore.
13. Quek S T, Tua P S and Wang Q (2003). Detecting Anomaly in Beams and Plate
Based on Hilbert-Huang Transform of Real Signals. Smart Mater. Struct. 12(3):
447–460.
14. Tua P S, Jin J, Quek S T and Wang Q (2002). Analysis of Lamb Modes Dominance
in Plates via Huang Hilbert Transform for Health Monitoring. Proc. 15 KKCNN Symp.
Civil Eng., Ed. S T Quek and D W S Ho, Dec. 19–20, 2002, Singapore.
15. Hahn S L (1996). Hilbert Transform in Signal Processing. London: Artech House
Boston.
16. Graff K F (1975). Wave Motion in Elastic Solids. Oxford: Clarendon Press.
17. Huang H, Norden E, Zheng S and Long S R (1999). A New View of Nonlinear Water
Waves: The Hilbert spectrum. Annu. Rev. Fluid Mech. 31:417–457.

© 2005 by Taylor & Francis Group, LLC

Potrebbero piacerti anche