ARTICLE IN PRESS

Mechanical Systems and Signal Processing 23 (2009) 1518–1527

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/jnlabr/ymssp

On the study of applying Morlet wavelet to the Hilbert transform for

the envelope detection of bearing vibrations
Yuh-Tay Sheen 
Department of Mechanical Engineering, Southern Taiwan University, 1 Nan-Tai Street, Yung Kang City, Tainan County 71005, Taiwan, ROC

a r t i c l e i n f o abstract

Article history: In this paper, the Morlet wavelet is studied to apply in the envelope analysis for the
Received 16 October 2008 bearing vibration and, in practice, would be easier to apply in the real-time vibration
Received in revised form analyses. The parameter designation of Morlet wavelet is proposed to filter out and
13 January 2009
demodulate one of the resonance modes of a bearing vibration, but the designation of
Accepted 23 January 2009
the filtering passband would not be required. Therefore, the mode vibration and its
Available online 6 February 2009
corresponding envelope could be derived from the real part and the absolute value of
Keywords: the wavelet transform, respectively. In addition, the Morlet wavelet with properly
Envelope detection designating the parameters possesses a very excellent property of fast waveform
Morlet wavelet
convergence and could effectively reduce the computing burden. From theoretical and
Hilbert transform
experimental studies, it is shown that the designation of Morlet wavelet could be
Defect diagnosis
effectively applied in the envelope detection for the vibration signals and could be
useful in the defect diagnosis of bearing vibrations.
& 2009 Elsevier Ltd. All rights reserved.

1. Introduction

In bearing defect diagnosis, the high-frequency resonance technique is always applied to the vibration signal
demodulation [1–4]. In the range of a high-frequency system resonance, this technique takes advantage of the absence of
low-frequency mechanical noise to demodulate a vibration signal and, therefore, provides a low-frequency demodulated
signal with a high signal-to-noise ratio. In practice, the procedure of high-frequency resonance technique could be
implemented in two steps. First, a band-pass filter is applied to the vibration signal around a selected high-frequency band
with the center at a chosen resonance frequency of the mechanical system. Secondly, the band-passed signal is operated
through the Hilbert transform to derive a complex signal and the demodulated signal could be obtained from the absolute
value of the complex signal. The FFT-based Hilbert transform is most commonly used. According to the FFT-based Hilbert
transform, the traditional method for deriving the signal envelope has the advantage of high computing speed. Therefore,
the FFT-based Hilbert transform is adaptable to signal analysis on line. But the FFT-based Hilbert transform could not be
applied in real-time signal processing and the leakage error would exist in the FFT-based Hilbert transform [5].
For solving the above two problems, the wavelet-based enveloping function [5,6] is proposed to implement the high-
frequency resonance technique. In the function, there are three parameters to assign for adjustment and designation of the
filtering passband, which are the low cut-off frequency, the high cut-off frequency and the dilation. Accordingly, it would be
convenient to apply in the high-frequency resonance technique. But there are the passband required to be designated,
which would be a trouble to be determined by the end-user.

 Tel.: +886 6 2533131ext3522; fax: +886 6 2425092.

E-mail address: [email protected]

0888-3270/$ - see front matter & 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2009.01.012
 ARTICLE IN PRESS
Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527 1519

In the paper, another systematic designation of the parameters of Morlet wavelet is proposed for real-time signal
processing. In practice, the Morelet wavelet is not required to designate the filtering passband and would be easier to apply
in the signal analyses of bearing vibrations. The parameter designation for the dilation and the center frequency of Morlet
wavelet is proposed to filter out and demodulate one of the resonance modes of a vibration signal. Thus, the Morelet
wavelet could be an orthogonal filter to perform as a Hilbert transform pair for the envelope detection of the resonance
mode. Therefore, the mode vibration and its corresponding envelope could be derived from the real part and the absolute
value of the wavelet transform, respectively. In addition, the Morlet wavelet possesses a very excellent property of fast
waveform convergence and could effectively reduce the computing burden. From theoretical and experimental studies, it is
shown that the designation of Morlet wavelet could be effectively applied in the envelope detection for the vibration
signals and could be useful in the defect diagnosis of bearing vibrations.

2. Morlet wavelet for analysis of bearing vibration

The vibration of bearing defect is induced by the impulse response of a defect-impact train. Each pulse is modulated by
the system resonances with an exponential decay. Because of the amplitude modulation occurring in the measured
vibration signals, the frequency-translation property would be presented in the vibration spectra. This phenomenon is
because a high-frequency carrier signal of system resonance is varied by a low-frequency modulating signal of the defect
impact. Therefore, its spectrum would be expanded in a frequency band and difficult to find the characteristic frequency of
modulating signal.
The carrier signal is a combination of the resonance frequencies of the bearing or even of the mechanical system, and
thus the vibration signal with amplitude modulation could be represented as [7]
X
n qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
vðtÞ ¼ Ai esi t cosð2p 1  xi f i t þ yi Þ, (1)
i¼1

where Ai, xi, fi, yi and si ¼ 2pxi fi are the amplitude, the damping ratio, the resonance frequency, the phase angle and the
exponential decay frequency of the ith mode, respectively.

2.1. Wavelet transform of vibration signal

It is shown that the wavelet transform by applying the Morlet wavelet for the vibration signal v(t) is a linear
transformation to decompose the vibration signal v(t)AL2(R) into elementary functions [5,8], and could be expressed as
rffiffiffiffi n pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
p X si t jð2p 1x2i f i tþyi 2pasi ða 1x2i f i f c ÞÞ 2p2 ða2 f 2i 2a 1x2i f i f c þf 2c Þþa2 s2i
Wða; tÞ ¼ Ai e e e . (2)
2 i¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
The passband gain and the phase would be slightly distorted by the small value of a2si2 and 2pasi ða 1  xi f i  f c Þ,
2
respectively. In practice, the damping ratio xi would be very small and xi could approach zero. In addition, when
designating a small value of dilation a, the above two terms could be further reduced to improve the problem of distortion.
Thus, the above wavelet transform for the vibration signal v(t) could be simplified as
rffiffiffiffi n pffiffiffiffiffiffiffiffiffi
p X 2pxi f i t jð2p 1x2i f i tþyi Þ 2p2 ðaf i f c Þ2
Wða; tÞ ¼ Ai e e e . (3)
2 i¼1
2 2
According to the Fourier transform of Morlet wavelet, it is found that the term e2p ðaf i f c Þ in Eq. (3) would be proportion
to the passband gain of Morlet wavelet. Thus, the ith mode of v(t) would be possible to filter through Morlet wavelet by
2 2
Eq. (3). When designating the dilation a ¼ fc/fi to be small, the term e2p ðaf i f c Þ in Eq. (3) would be the maximum value and
the ith mode of v(t) could be enhanced. In addition, suppose that the most closely neighboring mode to the ith mode is the
jth mode with the resonance frequency at fj ¼ fi+Df, the increase of |afjfc| would attenuate the passband gain of Morlet
2 2
wavelet for the jth mode. Thus, the term e2p ðaf j f c Þ in Eq. (3) for the jth mode would be decreased and appear to be
negligible. However, it is noted that the passband width of Morlet wavelet should also be wide enough to cover the ith
mode. In another word, |afjfc| should be small enough so that the ith mode could completely filter through Morlet wavelet
by Eq. (3). By satisfying the above conditions, the dilation a and the center frequency fc could be properly designated to
enhance the ith mode and eliminate the other (n1) modes. Thus, the ith mode of v(t) could filter through Morlet wavelet
by Eq. (3).

2.3. Parameter characteristics for envelope detection

On the purpose of applying the wavelet transform in Eq. (3) to filter out the ith resonance mode, the dilation a ¼ fc/fi and
2 2
the term e2p ðaf j f c Þ should be small to enhance the ith mode and eliminate the other (n1) modes, respectively. It is noted
2p2 ðaf j f c Þ2
that e is decrease with the increase of |afjfc| and an increase of |afjfc| would decrease the passband width. In
addition, a decrease of dilation a could increase waveform convergent speed and the passband width but decrease the slope
attenuation. As shown in Fig. 1, the waveform convergent speed for the small-value case of a ¼ 2.5  104 is faster and, on
 ARTICLE IN PRESS
1520 Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527

4000
a=0.00025
a=0.0005
2000

Amplitude
0

-2000

-4000
-1.5 -1 -0.5 0 0.5 1 1.5

4000

2000
Amplitude

-2000

-4000
-1.5 -1 -0.5 0 0.5 1 1.5
Time (ms)
Fig. 1. The waveform of Morlet wavelet with the dilation a ¼ 2.5  104 and 5  104. (a) Real part and (b) imaginary part.

0
a=0.00025
a=0.0005

-10
Gain (dB)

-20

-30

-40
fi -Δ f fi fi+Δf
Frequency (kHz)

Fig. 2. The passband gain of Morlet wavelet with the center frequency at fc ¼ afi.

the other hand, the slope attenuation is decreased as shown in Fig. 2. Thus, increasing the slope attenuation could be a
trade-off for increasing the waveform convergent speed, and vice versa.
Accordingly, let |afjfc| ¼ r be positive and small, the passband gain of the wavelet transform with the center frequency
at fc ¼ afi for the jth mode at its corresponding resonance frequency fj ¼ fi+Df would be 171.45r2 dB. Thus, when the value
of r being 0.483 and 0.592, the passband gains would be 40 dB (or 1%) and 60 dB (or 0.1%), respectively, and would be
small enough to be eliminated the jth mode in practice. In the case of Df being larger than 1 kHz, the Morlet wavelet at the
parameter range of 0.592/DfXaX0.483/Df could be properly designated for filtering out the ith resonance mode of v(t).
Thus, under the conditions of the system vibration without strong closely spaced modes and the range of dilation 0.592/
DfXaX0.483/Df, the wavelet transform at the center frequency fc ¼ afi could filter out the ith mode of v(t) and be
approached by
rffiffiffiffi pffiffiffiffiffiffiffiffiffi
p 1x2i f i tþyi Þ
W i ðtÞ ¼ Ai e2pxi f i t ejð2p . (4)
2
 ARTICLE IN PRESS
Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527 1521

From the above study, it is shown that the ith mode of v(t) could be filtered through Morlet wavelet by Eq. (3). With
properly designating the dilation a and the center frequency fc, the ith mode vibration would be enhanced and the other
(n1) modes could appear to be negligible. In addition, the ith mode vibration of signal v(t) could be derived from the real
part of Wi(t),
rffiffiffiffi
2
vi ðtÞ ¼ realðW i ðtÞÞ (5)
p
and the envelope of signal vi(t) could be obtained from
rffiffiffiffi
2
Ei ðtÞ ¼ jW ðtÞj, (6)
p i
pffiffiffiffiffiffiffiffiffi
where 2=p is the normalized coefficient. It is noted that the quality of above approximations is increased with the
decrease in the coupling to the neighboring modes.

3. Simulation

In this section, the above algorithm implemented in a personal computer is tested on a simulated signal derived from
the impulse response of a pulse train. Each pulse is modulated by four harmonic frequencies with an exponential decay.
Thus, the impulse response could be used to model the modulated signal of a bearing system, and is expressed as
0 0 0 0
xðtÞ ¼ ðA1 es1 t cosð2pf 1 tÞ þ A2 es2 t cosð2pf 2 tÞ þ A3 es3 t cosð2pf 3 tÞ þ A4 es4 t cosð2pf 4 tÞÞ  qðtÞ (7)
with
1
t 0 ¼ modðt; Þ,
fd
where A1 ¼ 0.5, A2 ¼ 1.2, A3 ¼ 1.2, and A4 ¼ 3.0 depict the dimension information of defect and the sensitivity of striking
energy, f1 ¼ 1.8 kHz, f2 ¼ 3.8 kHz, f3 ¼ 5.8 kHz, and f4 ¼ 8.0 kHz are its corresponding resonance frequencies, s1 ¼ 565,
s2 ¼ 1194, s3 ¼ 2042, and s4 ¼ 2608 are its corresponding exponential decay frequencies, q(t) ¼ (1+0.6 sin (2pf0t)) with
f0 ¼ 26.6 Hz describes the transmission path of mode vibration, and fd ¼ 100 Hz is the defect frequency. Fig. 3 shows the
simulation signal in both the time domain and the frequency domain. It is noted that higher the resonance mode would be
wider the passband width. Accordingly, it is shown clear that the third and fourth modes would overlap more serious.
In order to simulate the condition of real data and investigate the effectiveness of the signal processing method shown
in Section 2, the sampling rate for the above simulation signal is 50 kHz and the value of r is 0.483 for deriving the passband
gain 40 dB (or 1%) at the resonance frequency of the most closely neighboring mode. In order to quantify the effect of the
proposed analysis method, an error ratio by measuring the error between the original envelope and the estimated envelope

0.1
Amplitude

0.05

-0.05
0 20 40 60 80 100
Time (ms)
-3
x 10
6
Amplitude

0
0 2000 4000 6000 8000 10000
Frequency (Hz)
Fig. 3. The simulation signal. (a) The time domain and (b) the frequency domain.
 ARTICLE IN PRESS
1522 Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527

is defined as

standard deviation of error

error ratio ¼ . (8)
standard deviation of the original signal

According to the parameter designation of Morlet wavelet proposed in Section 2, the values of dilation a, center
frequency fc and Df for estimating the first mode vibration are 2.42  104, 0.435 and 2000, respectively. It is found that the
dilation a is very small and the waveform of Morlet wavelet could converge within 2 ms. Fig. 4 shows the result of applying
the envelope estimation algorithm. Fig. 4(a) is the estimated signal for the first mode vibration. Fig. 4(b) shows the
comparison between the original envelope and the estimated envelope for the first mode vibration. The error ratio between
the envelope signals is 0.475. It shows that the envelope signal is effectively estimated.
Similarly, the values of dilation a, center frequency fc and Df for estimating the second mode vibration are 2.54  104,
0.966 and 2000, respectively. The error ratio between the original and the estimated envelopes is 0.503. For estimating the
third mode vibration, the values of dilation a, center frequency fc and Df are 2.54  104, 1.449 and 2000, respectively. The
error ratio between the original and the estimated envelopes is 0.577. Finally, the values of dilation a, center frequency fc
and Df for estimating the fourth mode vibration are 2.01 104, 1.630 and 2200, respectively. The error ratio between the
original and the estimated envelopes is 0.976.
From the above study, it shows that the designation of Morlet wavelet could be effectively applied in the vibration
envelope extraction for the simulation signal. However, it is noted that the accuracy of the envelope estimation for the
lower resonance mode is better than that for the higher one. The reason could be that higher the resonance mode would
overlap more serious its neighboring ones. Fig. 5 shows the spectrum of reconstructed signal derived from the four

0.01

0.005
Amplitude

-0.005

-0.01
0 20 40 60 80 100

-3
x 10
10
Original
8 Estimated
Amplitude

0 20 40 60 80 100
Time (ms)
Fig. 4. The estimated signals for the first resonance mode. (a) The first mode vibration and (b) the comparison between the original envelope of the first
mode vibration and the estimated envelope derived from (a).

-3
x 10
6

4
Amplitude

0
0 2000 4000 6000 8000 10000
Frequency (Hz)

Fig. 5. The spectrum of the reconstructed signal.

 ARTICLE IN PRESS
Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527 1523

estimated signals. In comparison Fig. 5 with Fig. 3(b), it is found that the two figures are similar. Thus, the reconstructed
signal could be a good estimation for the simulation signal.

4. Experimental study

In the following, the applications of the proposed method to the vibration signals of tapered roller bearings (SKF type
32208) are studied. The electrical-discharge machining method is applied to produce artificial defect on the surface of
bearing components which are roller, outer race and inner race. The defect sizes are described in Table 1. The vibration
signals are measured on the housing of the test bearing by mounting an accelerometer with the sensitivity 10.41 mV/g. The
sampling rate is 50 kHz and the value of r is 0.483 for deriving the passband gain 40 dB (or 1%) at the resonance frequency
of the most closely neighboring mode. The measured direction is radial to the shaft in the horizontal. The tested bearings
running at 1600 rpm are used to investigate the effectiveness of the proposed diagnosis method for the bearing vibrations.
The characteristic frequencies for roller defect, outer-race defect and inner-race defect are 154.8, 189.8, and 263.6 Hz,
respectively.
Table 2 lists the first four resonance modes for four different bearing types. It is found that the vibration modes for these
bearing vibrations are similar. On the purpose of reducing the distortion caused by the overlap of resonance modes, the
envelope estimation of the second resonance mode would be investigated. The values of dilation a, center frequency fc and
Df for estimating the second mode vibration are listed in Table 3. It is noted that the dilation a is small and, thus, the
waveform of Morlet wavelet could converge within 2 ms. Figs. 6–9 show the investigation result of the second resonance
mode for the roller defect, the outer-race defect, the inner-race defect and normal bearings, respectively. It proves that the
parameter designation of Morlet wavelet could be effectively applied in the vibration envelope extraction for the bearing
vibrations. Fig. 10(a)–(d) shows the envelope spectra for the envelope signals in Figs. 6(b)–9(b), respectively. The
characteristic frequencies for roller defect, outer-race defect and inner-race defect could be found at 154.8 Hz in Fig. 10(a),
189.8 Hz in Fig. 10(b), and 263.6 Hz in Fig. 10(c), respectively. However, there is no characteristic frequency found in
Fig. 10(d). Thus, the designation of Morlet wavelet could be effectively applied in the defect diagnosis of bearing vibrations.
In order to study the effect of the proposed signal processing method at different running speed, the analyses of bearing
vibrations at running speed 800 rpm are also studied. It is found that there are similar resonance frequencies as shown in
Table 2 and, thus, similar parameters as shown in Table 3 for the Morlet wavelet could be designated. Fig. 11(a)–(d) shows

Table 1
Defect sizes of bearing defects.

Defect type Defect size (length  width  depth) (mm3)

Roller defect 16  0.15  0.1

Outer-race defect 14  0.15  0.1
Inner-race defect 18.5  0.15  0.1

Table 2
The resonance frequencies of bearing vibration modes.

Bearing type Mode no.

First mode (Hz) Second mode (Hz) Third mode (Hz) Fourth mode (Hz)

Roller defect 1685 3848 6782 8314

Outer-race defect 1872 3700 6931 7689
Inner-race defect 2317 4029 6679 8078
Normal 2015 4077 6609 7722

Table 3
The parameter designation of Morlet wavelet for the second mode of bearing vibration.

Bearing type Parameter

Dilation, a Center frequency, fc Df

Roller defect 2.234  104 0.8595 2163

Outer-race defect 2.643  104 0.9779 1828
Inner-race defect 2.821  104 1.137 1712
Normal 2.343  104 0.9553 2062
 ARTICLE IN PRESS
1524 Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527

0.01

0.005

Amplitude
0

-0.005

-0.01
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

0.01
Amplitude

0.005

0
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

Fig. 6. The estimated signals for the roller defect bearing. (a) The second mode vibration and (b) the envelope derived from (a).

0.04

0.02
Amplitude

-0.02

-0.04
0 10 20 30 40 50 60 70 80 90 100
Time (ms)
0.04

0.03
Amplitude

0.02

0.01

0
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

Fig. 7. The estimated signals for the outer-race defect bearing. (a) The second mode vibration and (b) the envelope derived from (a).

the envelope spectra for the roller defect, the outer-race defect, the inner-race defect and the normal bearings, respectively.
The characteristic frequencies for roller defect, outer-race defect and inner-race defect could be also found at 77.4 Hz in
Fig. 11(a), 94.9 Hz in Fig. 11(b), and 131.8 Hz in Fig. 11(c), respectively. In addition, there is no characteristic frequency found
in Fig. 11(d). Thus, the designation of Morlet wavelet could be effectively applied in the defect diagnosis of bearing
vibrations even under different running speed.

5. Discussion

The wavelet transform is usually applied to analyze the vibration signal in the time–frequency domain. The derived
figure is called scalogram that describes how energy distribution between the instantaneous time and the instantaneous
 ARTICLE IN PRESS
Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527 1525

0.03
0.02
0.01

Amplitude
0
-0.01
-0.02
-0.03
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

0.03

0.02
Amplitude

0.01

0
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

Fig. 8. The estimated signals for the inner-race defect bearing. (a) The second mode vibration and (b) the envelope derived from (a).

-3
x 10
3

2
Amplitude

-1

-2

-3
0 10 20 30 40 50 60 70 80 90 100
Time (ms)
-3
x 10
3
Amplitude

0
0 10 20 30 40 50 60 70 80 90 100
Time (ms)

Fig. 9. The estimated signals for the normal bearing. (a) The second mode vibration and (b) the envelope derived from (a).

frequency for the vibration signal. According to the pattern of scalogram, the characteristics of analyzed signal could
be classified. Thus, the wavelet transform could be useful in the defect diagnosis of mechanical system. However, the
scalogram would be seriously affected by the parameters of dilation and center frequency. Improper designation of the
parameters could result in the invisibility of characteristic pattern for the analyzed signal.
In practice, how to properly designate the parameters of Morlet wavelet would be the major problem in the application
of signal processing. There are some suggestions for the selection of parameters [8]. Nevertheless, the blind search is
necessary in the selecting process of parameters. In the past, the wavelet-based enveloping function which is the linear
combination of Morlet wavelet is proposed to implement the high-frequency resonance technique and its function
parameters could be properly designated according to the physical sense. In addition, the wavelet-based enveloping
function could be applied in real-time signal processing.
 ARTICLE IN PRESS
1526 Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527

-3
x 10
5 0.9

4
0.6

Amplitude

Amplitude
3

2
0.3
1

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Frequency (Hz) Frequency (Hz)
-5
x 10
0.09 5

4
0.06
Amplitude

Amplitude
3

2
0.03
1

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

Fig. 10. The envelope spectra at running speed 1600 rpm for (a) roller defect bearing, (b) outer-race defect bearing, (c) inner-race defect bearing and
(d) normal bearing.

-3
x 10
1.2 0.06

0.9
Amplitude

Amplitude

0.04
0.6
0.02
0.3

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Frequency (Hz) Frequency (Hz)
-6
x 10
0.015 8

6
Amplitude

Amplitude

0.01
4
0.005
2

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

Fig. 11. The envelope spectra at running speed 800 rpm for (a) roller defect bearing, (b) outer-race defect bearing, (c) inner-race defect bearing and
(d) normal bearing.

In the paper, the designation of the parameters of Morlet wavelet is proposed and easier, in practice, to apply in the
signal analyses of bearing vibrations. The designation of Morlet wavelet possesses the following four advantages. First, the
mode vibration could be direct derived from the real part of the wavelet transform. Secondly, the Morlet wavelet possesses
a very excellent property of fast waveform convergence, which is much better than the wavelet-based enveloping function.
As shown in Fig. 1, the convergence in 2 ms for Morlet wavelet is much better than that in 10 ms for the wavelet-based
enveloping function [6]. Thus, it could effectively reduce the computing burden. Thirdly, the Morelet wavelet could be
 ARTICLE IN PRESS
Y.-T. Sheen / Mechanical Systems and Signal Processing 23 (2009) 1518–1527 1527

applied in the real-time signal processing for the envelope detection of bearing vibrations. Finally, it is not required to
designate the filtering passband. In comparison with the wavelet-based enveloping function, the Morlet wavelet possesses
a robust designation of parameters to filter out one of the resonance modes.

6. Conclusion

In this paper, the Morlet wavelet is applied in the real-time envelope analysis for vibration signals. With properly
designating the function parameters of Morlet wavelet, the wavelet transform could perform as a Hilbert transform to filter
out one of the resonance modes and, thus, the mode vibration and its corresponding envelope could be derived from the
real part and the absolute value of the wavelet transform, respectively. In addition, it is found that the Morlet wavelet
possesses a very excellent property of fast waveform convergence and no filtering passband is designated. From theoretical
and experimental studies, it is shown that the designation of Morlet wavelet could be effectively applied in the envelope
detection and the defect diagnosis for the bearing vibrations.

References

[1] P.D. McFadden, J.D. Smith, Vibration monitoring of rolling element bearings by the high frequency resonance technique – a review, Tribology
International 17 (1984) 1–18.
[2] Y.T. Su, S.J. Lin, On initial fault detection of a tapered rolling bearing: frequency domain analysis, Journal of Sound and Vibration 155 (1992) 75–84.
[3] Y.T. Su, Y.T. Sheen, On the detectability of roller bearing damage by frequency analysis, Proceedings of the Institution of Mechanical Engineers, Part C:
Journal of Mechanical Engineering Science 207 (1993) 23–32.
[4] J. Shiroishi, Y. Li, S. Liang, T. Kurfess, S. Danyluk, Bearing condition diagnostics via vibration and acoustic emission measurement, Mechanical Systems
and Signal Processing 11 (5) (1997) 693–705.
[5] Y.T. Sheen, C.K. Hung, Constructing a wavelet-based envelope function for vibration signal analysis, Mechanical Systems and Signal Processing 18 (1)
(2004) 119–126.
[6] Y.T. Sheen, 3D spectrum analysis for vibration signals by wavelet-based demodulation, Mechanical Systems and Signal Processing 20 (4) (2006)
843–853.
[7] B.F. Yan, A. Miyamoto, E. Bruhwiler, Wavelet transform-based modal parameter identification considering uncertainty, Journal of Sound and Vibration
291 (1–2) (2006) 285–301.
[8] M. Vetterli, J. Kovacevic, Wavelet and Subband Coding, Prentice-Hall PTR, Englewood Cliffs, NJ, 1995.