Sunteți pe pagina 1din 17

Hindawi Publishing Corporation

EURASIP Journal on Advances in Signal Processing


Volume 2007, Article ID 73629, 14 pages
doi:10.1155/2007/73629

Research Article
Rolling Element Bearing Fault Diagnosis Using
Laplace-Wavelet Envelope Power Spectrum

Khalid F. Al-Raheem,1 Asok Roy,2 K. P. Ramachandran,1 D. K. Harrison,2 and Steven Grainger2


1 Department of Mechanical and Industrial Engineering, Caledonian College of Engineering, P.O. Box 2322, CPO Seeb, PC 111, Oman
2 School of Engineering Science and Design, Glasgow Caledonian University, Glasgow G40BA, UK

Received 1 July 2006; Revised 19 December 2006; Accepted 1 April 2007

Recommended by Alex Kot

The bearing characteristic frequencies (BCF) contain very little energy, and are usually overwhelmed by noise and higher levels of
macro-structural vibrations. They are difficult to find in their frequency spectra when using the common technique of fast fourier
transforms (FFT). Therefore, Envelope Detection (ED) has always been used with FFT to identify faults occurring at the BCF.
However, the computation of the ED is suffering to strictly define the resonance frequency band. In this paper, an alternative ap-
proach based on the Laplace-wavelet enveloped power spectrum is proposed. The Laplace-Wavelet shape parameters are optimized
based on Kurtosis maximization criteria. The results for simulated as well as real bearing vibration signal show the effectiveness of
the proposed method to extract the bearing fault characteristic frequencies from the resonant frequency band.

Copyright © 2007 Khalid F. Al-Raheem et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

1. INTRODUCTION The rolling elements experience some slippage as the


rolling elements enter and leave the bearing load zone. As a
The predictive maintenance philosophy of using vibration consequence, the occurrence of the impacts never reproduce
information to lower operating costs and increase machinery exactly at the same position from one cycle to another, more-
availability is gaining acceptance throughout industry. Since over, when the position of the defect is moving with respect
most of the machinery in a predictive maintenance program to the load distribution of the bearing, the series of impulses
contains rolling element bearings, it is imperative to establish is modulated in amplitude. However, the periodicity and the
a suitable condition monitoring procedure to prevent mal- amplitude of the impulses experience a certain degree of ran-
function and breakage during operation. domness [1–4]. In such case, the signal is not strictly peri-
The hertzian contact stresses between the rolling ele- odic, but can be considered as cyclo-stationary (periodically
ments and the races are one of the basic mechanisms that time-varying statistics), then the cyclic second-order statis-
initiate a localized defect. When a rolling element strikes a tics (such as cyclic autocorrelation and cyclic spectral den-
localized defect, an impulse occurs which excites the reso- sity) are suited to demodulate the signal and extract the fault
nance of the structure. Therefore, the vibration signature of feature [5–7]. All these make the bearing defects very diffi-
the damaged bearing consists of exponentially decaying sinu- cult to detect by conventional FFT-spectrum analysis which
soid having the structure resonance frequency. The duration assumes that the analyzed signal to be strictly periodic. A
of the impulse is extremely short compared with the interval method of conditioning the signal before the spectrum es-
between impulses, and so its energy is distributed at a very timation takes places is necessary.
low level over a wide range of frequency and hence, can be To overcome the modulation problem, several signal
easily masked by noise and low-frequency effects. The peri- envelope demodulation techniques have been introduced.
odicity and amplitude of the impulses are governed by the In high-frequency resonance technique (HFRT), an en-
bearing operating speed, location of the defect, geometry of velope detector demodulates the passband filtered signal
the bearing, and the type of the bearing load. The theoret- and the frequency spectrum is determined by FFT tech-
ical estimations of these frequencies are denoted as bearing nique [8]. Another well-established method is based on the
characteristics frequencies (BCF); see the appendix. Hilbert transform [9, 10]. The inconvenience of the envelope
2 EURASIP Journal on Advances in Signal Processing

demodulation techniques is that the most suitable passband The response of the bearing structure as an under-
must be identified before the demodulation takes place. damped second-order mass-spring-damper system to a sin-
The wavelet transform provides powerful multiresolu- gle impulse force is given by
tion analysis in both time and frequency domain and thereby
becomes a favored tool to extract the transitory features √  
of nonstationary vibration signals produced by the faulty S(t) = Ce−(ξ/ 1−ξ 2 )ωd t
sin ωd t , (1)
bearing [11–16]. The wavelet analysis results in a series of
wavelet coefficients, which indicate how close the signal is where ζ is the damping ratio and ωd is the damped natural
to the particular wavelet. In order to extract the fault fea- frequency of the bearing structure.
ture of signals more effectively, an appropriate wavelet base As the shaft rotates, this process occurs periodically every
function should be selected. Morlet wavelet is mostly ap- time a defect hits another part of the bearing and its rate of
plied to extract the rolling element bearing fault feature be- occurrence is equal to one of the BCF. In reality, there is a
cause of the large similarity with the impulse generated by slight random fluctuation in the spacing between impulses
the faulty bearing [17–20]. The impulse response wavelet is because the load angle on each rolling element changes as the
constructed and applied to extract the feature of fault vibra- rolling element passes through the load zone. Furthermore,
tion signal in [21]. A number of wavelet-based functions are the amplitude of the impulse response will be modulated as
proposed for mechanical fault detection with high sensitiv- a result of the passage of the fault through the load zone,
ity in [22], and the differences between single and double-
sided Morlet wavelets are presented. An adaptive wavelet fil-   
ter based on single-sided Morlet wavelet is introduced in x(t) = Ai S t − Ti + n(t), (2)
[23]. i

The Laplace wavelet is a complex, single-sided damped


exponential formulated as an impulse response of a single where S(t − Ti ) is the waveform generated by the ith im-
mode system to be similar to data feature commonly encoun- pact at the time Ti , and Ti = iT + τi , where T is the aver-
tered in health monitoring tasks. It is applied to the vibration age time between two impacts, and τi describe the random
analysis of an actual aircraft for aerodynamic and structural slips of the rolling elements. Ai is the time varying amplitude-
testing [24], and to diagnose the wear fault of the intake valve demodulation, and n(t) is an additive background noise
of an internal combustion engine [25]. which takes into account the effects of the other vibrations
In this paper, an alternative approach for detecting local- in the bearing structure.
ized faults in the outer and inner races of a rolling element Figures 1 and 2 show the impulses and the acceleration
bearing using the envelope power spectrum of the Laplace signals (d2 x(t)/dt 2 ) generated by the model in (2) with ran-
wavelet is investigated. The wavelet shape parameters are op- dom slip (τ) of 10 percent of the period T and signal to noise
timized by maximizing the kurtosis of the wavelet coefficients ratio of 0.6 dB for outer-race and inner-race bearing faults,
to ensure a large similarity between the wavelet function and respectively.
the generated fault impulse.
This paper is organized as follows. In the next section, 3. ENVELOPED WAVELET POWER SPECTRUM
the vibration model for a rolling bearing with outer- and
inner-race faults is introduced. Then in Section 3, the pro- The waveform x(t) in (2) can be viewed as a carrier signal at a
cedures of the proposed approach are set up. In Section 4, resonant frequency of the bearing housing (high frequency)
the implementation of the proposed approach for detection modulated by the decaying envelope. The frequency of inter-
of localized ball bearing defects for both simulated and actual est in the detection of bearing defects is the modulating fre-
bearing vibration signals is presented. Conclusions are finally quency (low frequency). The goal of the enveloping approach
given in Section 5. is to replace the oscillation caused by each impact with a sin-
gle pulse over the entire period of the impact.
The Laplace wavelet is a complex, analytical, and single-
2. VIBRATION MODEL FOR ROLLING ELEMENT sided damped exponential, and it is given by
BEARING LOCALIZED DEFECTS
⎧ √
⎨Ae−(β/ 1−β2 )ωc t e− jωc t , t ≥ 0,
Every time the rolling element strikes a defect in the raceway Ψ(t) = ⎩ (3)
0, t < 0,
or every time a defect in the rolling element hits the raceway,
a force impulse of short duration is produced which in turn
excites the natural frequencies of the bearing parts and hous- where β is the damping factor that controls the decay rate
ing structure. The structure resonance in the system acts as of the exponential envelope in the time domain and hence
an amplifier of low-energy impacts. Therefore, the overall vi- regulates the resolution of the wavelet, and it simultaneously
bration signal measured on the bearing shows a pattern con- corresponds to the frequency bandwidth of the wavelet in the
sisting of a succession of oscillating bursts dominated by the frequency domain. The frequency ωc determines the number
major resonance frequency of the structure. of significant oscillations of the wavelet in the time domain
Khalid F. Al-Raheem et al. 3

6 5
4
4
3
2
Acceleration (m.s−2 )

Acceleration (m.s−2 )
2
1

0 0
−1
−2
−2

−3
−4
−4

−6 −5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time (s) Time (s)
(a) (b)

Figure 1: The simulated impulses (a) and the vibration signal (b) for a rolling bearing with outer-race fault.

5 8
4 6
3
4
Acceleration (m.s−2 )

Acceleration (m.s−2 )

2
1 2

0 0
−1 −2
−2
−4
−3

−4 −6

−5 −8
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time (s) Time (s)
(a) (b)

Figure 2: The simulated impulses (a) and the vibration signal (b) for a rolling bearing with inner-race fault.

and corresponds to the wavelet centre frequency in frequency value and the intensity distribution for a white noise signal,
domain, and A is an arbitrary scaling factor. Figure 3 shows pure impulsive signal, and impulsive signal mixed with noise.
the Laplace wavelet, its real part, imaginary part, and its spec- The objective of the Laplace wavelet shape optimization
trum. process is to find out the wavelet shape parameters (β and
It is possible to find optimal values of β and ωc for a ωc ) which maximize the kurtosis of the wavelet transform
given vibration signal by adjusting the time-frequency reso- output;
lution of the Laplace wavelet to the decay rate and frequency  
Optimal β, ωc
of impulses to be extracted. Kurtosis is an indicator that re-  N 
4
flects the “peakiness” of a signal, which is a property of the n=1 WT x(t), ψβ,ωc (t) (4)
impulses and also it measures the divergence from a funda- = max ·
N 2  2 .
n=1 WT x(t), ψβ,ωc (t)
mental Gaussian distribution. A high kurtosis value indicates
high-impulsive content of the signal with more sharpness in The wavelet transform (WT) of a finite energy signal x(t),
the signal intensity distribution. Figure 4 shows the kurtosis with the mother wavelet ψ(t), is the inner product of x(t)
4 EURASIP Journal on Advances in Signal Processing

1
0.8
0.6
0.4
0.2

Amplitude
6
0
4 1
Time

−0.2
2
0 0.5 −0.4
1
−0.6

rt
0
pa
0.5 Re
al −0.8
Im 0
a g in −0.5 −1
ar y −0.5
par −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
t
−1 −1 Time (s)
(a) (b)

×10−4
1 3.5

0.8
3
0.6
0.4 2.5
Power spectrum

0.2
Amplitude

2
0
1.5
−0.2

−0.4 1
−0.6
0.5
−0.8

−1 0
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 0 10 20 30 40 50 60 70 80 90 100
Time (s) Frequency (Hz)
(c) (d)

Figure 3: (a) The Laplace wavelet, (b) the real part, (c) the imaginary part, and (d) its spectrum.

∗ √
with a scaled and conjugate wavelet ψa,b , since the analytical 1/ a is used to ensure energy preservation. The time-varying
and complex wavelet is employed to calculate the wavelet function A(t) is the instantaneous envelope of the resulting
transform. The result of the WT is also an analytical signal, wavelet transform (EWT) which extracts the slow time vari-
ation of the signal, and is given by

   1
WT x(t), a, b = x(t), ψa,b (t) = √ x(t)Ψ∗a,b (t)dt
a A(t) = EWT(a, b)


(5)  (6)
= Re WT(a, b) + j Im WT(a, b)
2
2
= Re WT(a, b) + Im WT(a, b) .
= A(t)eiθ(t) ,
For each wavelet, the inner product results in a series of coef-
where ψa,b is a family of daughter wavelets, defined by the di- ficients which indicate how close the signal is to that partic-
lation parameter a and the translation parameter b, the factor ular wavelet.
Khalid F. Al-Raheem et al. 5

150 350

100 300

250
50
200
0
150
−50
100

−100 50

−150 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 −150 −100 −50 0 50 100 150
(a)

4 500
3 450

2 400
350
1
300
0
250
−1
200
−2
150
−3 100
−4 50
−5 0
0 200 400 600 800 1000 1200 1400 1600 1800 −5 −4 −3 −2 −1 0 1 2 3 4 5
(b)

5 1400
4 1200
3
1000
2
1 800

0 600
−1
400
−2
200
−3
−4 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 −4 −3 −2 −1 0 1 2 3 4 5
(c)

Figure 4: (a) The noise signal (kurtosis = 3.0843), (b) the overall vibration signal (kurtosis = 7.7644), and (c) outer-race fault impulses
(kurtosis = 8.5312) with the corresponding intensity distribution curve.

To extract the frequency content of the enveloped corre- The total energy of the signal x(t),
lation coefficients, the scale power spectrum (WPS) (energy  
 
per unit scale) is given by TWPS = x(t)2 dt = 1 WPS(a, ω)da. (8)

∞
  4. IMPLEMENTATION OF WPS FOR BEARING
WPS(a, ω) = SEWT(a, ω)2 dω, (7)
−∞ FAULT DETECTION

To demonstrate the performance of the proposed approach,


where SEWT (a, ω) is the Fourier transform of EWT(a, b). this section presents several application examples for the
6 EURASIP Journal on Advances in Signal Processing

detection of localized bearing defects. In all the examples, the


Laplace wavelet is used as a WT base-function. The wavelet ×104
X: 0.8
Y : 18
parameters (damping factor and centre frequency) are opti- 10 Z: 8.211e + 004
mized based on maximizing the kurtosis value for the wavelet
8
coefficients; see Figure 5.

Kurtosis
6

4.1. Application to the simulation signals 4

2
For a rolling element bearing with pitch diameter of
51.16 mm, ball diameter of 11.9 mm, with 8 rolling elements 0
20
and 0◦ contact angle, the calculated BCF for an outer-race Cen 15 2
fault is 107.36 Hz, and for an inner-race fault is 162.18 Hz tre 1.5
freq 10 1
uen
with shaft speed of 1797 rev/min. Figures 1 and 2 show the cy, 5 0.5 ctor, β
ω ing fa
simulated time domain fault impulses and the overall vi- c (H 0 0 Damp
z)
bration signal for the bearing with outer-race and inner-
race faults, respectively, based on the model described in (a)
Section 2.
To evaluate the performance of the proposed method,
a scale-wavelet power spectrum comparison for the Laplace
wavelet and the widely used Morlet wavelet was carried out;
×107 X: 0.6
see Figure 6. It can be found that the amplitude of the power 2 Y : 12
Z: 1.956e + 007
spectrum increases further for the faulty bearing than the
normal one, and the power spectrum is concentrated in the 1.5
scale interval of [15–20] for Laplace wavelet compared with
Kurtosis

speared power spectrum in wide range scales for Morlet 1


wavelet. That shows the increased effectiveness of the Laplace
wavelet over the Morlet wavelet for bearing fault impulses ex- 0.5
traction.
0
The FFT spectrum, the envelope spectrum using Hilbert 20
transform, and the Laplace-wavelet transform envelope spec- Cen 15 2
tre 1.5
trum for the simulated outer- and inner-race fault vibration freq 10 1
uen
signals are shown in Figure 7. Figure 7 shows that the BCFs cy, 5 0.5 ctor, β
ω ing fa
are unspecified in the FFT spectrum and unclearly defined in c (H 0 0 Damp
z)
the envelope power spectrum but it is clearly identified in the
(b)
Laplace-wavelet power spectrum for both outer- and inner-
race faults. The TWPS effectively extracts the BCF, 105.5 Hz
for outer-race fault and 164.1 Hz for inner-race fault and its
harmonics, with side bands at rotational speed for inner-race
fault as a result of amplitude modulation and it is very close
×104
to the calculated BCF. 5 X: 0.9
To evaluate the robustness of the proposed technique to Y : 17
extract the BCF for different signal to noise ratio (SNR), and 4 Z: 4.817e + 004

randomness in the impulses period (τ) as a result of slip vari-


Kurtosis

3
ation, Figure 8 shows the TWPS for outer-race fault simu-
lated signals for different values of SNR, and τ as a percentage 2
of the pulses period (T). 1

0
20
4.2. Application to experimental data
Cen 15 2
tre 1.5
freq 10 1
A B&K 752A12 piezoelectric accelerometer was used to col- uen 5 or, β
cy,
ω
0.5
in g fact
lect the vibration signals for an outer-race defective, deep c (H 0 0 Damp
z)
groove, ball bearing (with same simulated specifications) at
different shaft rotational speeds. The vibration signals were (c)
transferred to the PC through a B&K controller module type
7536 at a sampling rate of 12.8 KHz. Based on the bearing Figure 5: The optimal values for Laplace wavelet parameters based
parameters, the calculated outer-race fault characteristic fre- on maximum kurtosis for (a) simulated outer-race fault, (b) the
quency is 0.05115x rpm; see Figure 9. measured outer-race fault, (c) the CWRU vibration data.
Khalid F. Al-Raheem et al. 7

×106 ×104
9 18

8 16

7 14

6 12

Power spectrum
Power spectrum

5 10

4 8

3 6

2 4

1 2

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Wavelet scale, a Frequency (Hz)

(a) New bearing

×106 ×106
6 4

3.5
5
3
Envelope spectrum

Envelope spectrum

4
2.5

3 2

1.5
2
1
1
0.5

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Wavelet scale, a Wavelet scale, a
(b) Outer-race defective bearing

Figure 6: The wavelet-level power spectrum using (left column) Morlet wavelet, (right column) Laplace wavelet for new and outer-race
defective bearing.

With application of the TWPS, the power spectrum peak shaft speed (Hz) for inner race. The time course of the vi-
values at the position of the outer-race characteristic fre- bration signals for normal bearing and bearings with outer
quency and its harmonics are easily defined; see Figure 9. It is and inner race faults at shaft rotational speed 1797 rpm with
shown that TWPS is sensitive to the variation of the BCF as a its corresponding TWPS are shown in Figures 10–12, re-
result of variation in the shaft rotational speeds; see Table 1. spectively. The calculated BCF are 107.36 Hz for outer-race
fault and 162.185 Hz for inner-race fault. The TWPS for
4.3. Application TWPS to vibration data the vibration data show spectrum peak values at 106.9 Hz
for outer- race fault and its harmonics (Figure 11), and
Using the data given by the CWRU bearing centre website 161.1 Hz for inner-race fault with its harmonics and side-
[26], for rolling bearings seeded with faults using electro- bands at shaft speed (30 Hz) as a result of amplitude mod-
discharge machining (EDM). The calculated defect frequen- ulation (Figure 12), which are very close to the calculated
cies are 3.5848x shaft speed (Hz) for outer race and 5.4152x BCF.
8 EURASIP Journal on Advances in Signal Processing

×10−3 ×10−3
25 30
X: 1594 X: 1623
Y : 0.02441 Y : 0.02859
25
20
Power spectrum

Power spectrum
20
15
15
10
10

5
5

0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Frequency (Hz) Frequency (Hz)

(a) FFT spectrum

×10−3 ×10−3
18 50
16 45

14 40
35
Power spectrum

Power spectrum

12
30
10
25 X: 193.4
8 Y : 0.0202
20 X: 164.1
6 Y : 0.01482
15
4 10
2 5
0 0
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)
(b) ED spectrum

×10−3 ×10−3
8 X: 105.5 X: 29.3
Y : 0.007132 Y : 0.007967
7 8
X: 316.4
X: 210.9 Y : 0.006115 7 X: 164.1
6 Y : 0.005609
Y : 0.005984
Power spectrum

Power spectrum

6 X: 322.3
5 Y : 0.006526
5
4 X: 486.3
4 Y : 0.003657
3
3
X: 515.6
2 2 Y : 0.002931

1 1
0 0
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)
(c) Laplace wavelet spectrum

Figure 7: The simulated vibration signal power spectrum, the envelope power spectrum, and the Laplace-wavelet transform power spec-
trum, respectively, for rolling bearing with (left column) outer-race fault and (right column) inner-race fault.
Khalid F. Al-Raheem et al. 9

×105 ×105
2.5 4

X: 107.4 3.5 X: 107.4


2 Y : 2.278e + 005 Y : 3.596e + 005
X: 214.8 3
Y : 1.685e + 005
Power spectrum

Power spectrum
1.5 2.5
X: 322.3 X: 214.8
Y : 1.201e + 005 Y : 1.985e + 005
2
X: 322.3
1 X: 429.7
1.5 Y : 1.873e + 005
Y : 7.834e + 004

1 X: 429.7
0.5 Y : 1.12e + 005
0.5

0 0
0 100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)
(a) SNR = 3.165 dB (b) τ = 1%

×105 ×105
3 4.5
4 X: 107.4
2.5 X: 107.4 Y : 4.189e + 005
Y : 2.605e + 005 3.5
Power spectrum

Power spectrum

2 3

X: 214.8 2.5
1.5 Y : 1.339e + 005
2 X: 214.8
Y : 1.602e + 005
1 X: 322.3 1.5
Y : 1.18e + 005 X: 322.3
1 Y : 8.311e + 004
0.5
0.5
0 0
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)
(c) SNR = 0.6488 dB (d) τ = 5%

×105 ×105
12 4

3.5
10 X: 107.4
X: 107.4
Y : 2.974e + 005
Y : 1.02e + 006 3
Power spectrum
Power spectrum

8
2.5
X: 322.3
6 Y : 5.165e + 005
2
X: 214.8
Y : 3.655e + 005 1.5
4
X: 214.8
1 Y : 7.543e + 004
2
0.5

0 0
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)
(e) SNR = 0.384 dB (f) τ = 10%

Figure 8: The TWPS for bearing with outer-race fault for different (left column) SNR and (right column) slip variation (τ).
10 EURASIP Journal on Advances in Signal Processing

8 0.1
0.09
6 X: 46.88
0.08 Y : 0.07639
4 X: 93.75
Acceleration (m.s−2 )

0.07 Y : 0.07927

Power spectrum
2 0.06 X: 140.6
0.05 Y : 0.06141
0
0.04
−2 X: 187.5
Y : 0.04188
0.03
−4
0.02
−6 0.01
−8 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(a) 984 rpm

40 18

30 16 X: 99.61
X: 304.7
Y : 16.72
Y : 13.54
14 X: 404.3
20
Acceleration (m.s−2 )

Y : 12.1
Power spectrum

12 X: 205.1
10
Y : 12.3
10
0
8 X: 503.9
−10 Y : 5.844
6
−20 4
−30 2
−40 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(b) 1389 rpm

100 2000
80 1800 X: 175.8
Y : 1871
60 1600
Acceleration (m.s−2 )

40 1400
Power spectrum

X: 345.7
20 1200 Y : 1087

0 1000
X: 521.5
−20 800 Y : 712.7

−40 600 X: 697.3


Y : 429.1
−60 400
−80 200
−100 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(c) 3531 rpm

Figure 9: The measured vibration signals for rolling bearing with outer-race fault at different shaft rotational speed (a) 984 rpm, (b)
1389 rpm, and (c) 3531 rpm.
Khalid F. Al-Raheem et al. 11

×10−7
0.5 5

0.4 4.5 X: 29.3


Y : 4.629e − 007
0.3 4

0.2 3.5
Acceleration (m.s−2 )

Power spectrum
0.1 3

0 2.5

−0.1 2

−0.2 1.5

−0.3 1

−0.4 0.5

−0.5 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(a) (b)

Figure 10: The vibration signal (a) and the corresponding TWPS (b) for new rolling bearing (CWRU data).

×10−3
4 4

3 3.5
X: 106.9
Y : 0.00357
2 3 X: 215.3
Y : 0.002646
Acceleration (m.s−2 )

Power spectrum

1 2.5

0 2

−1 1.5

−2 1 X: 322.3
Y : 0.0005621
−3 0.5

−4 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(a) (b)

Figure 11: The vibration signal (a) and the corresponding TWPS (b) for rolling bearing with outer-race fault (CWRU data).

5. CONCLUSIONS Table 1
Shaft speed (rpm) Calculated BCF (Hz) TWPS peak(Hz)
984 50.331 50
A new approach based on the Laplace-wavelet enveloped
power spectrum (TWPS) is proposed. The wavelet shape pa- 1389 71.047 68.750
rameters (damping factor and the centre frequency) are op- 3541 181.17 187.5
timized by maximizing the kurtosis value for the vibration
signal wavelet transform coefficients. The application of the
proposed technique for both the simulated and real bear- the extraction of the BCF and its harmonics, for outer- and
ing vibration signals has shown the effectiveness of TWPS in inner-race defective bearings from noisy vibration signals.
12 EURASIP Journal on Advances in Signal Processing

×10−4
2 1.5

1.5 X: 161.1
Y : 0.0001185

1
Acceleration (m.s−2 )

Power spectrum
0.5

−0.5
0.5 X: 323.7
Y : 3.647e − 005
−1
X: 131.8 X: 484.9
Y : 1.569e − 005 Y : 1.555e − 005
−1.5

−2 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 100 200 300 400 500 600 700 800 900 1000
Time (s) Frequency (Hz)
(a) (b)

Figure 12: The vibration signal (a) and the corresponding TWPS (b) for rolling bearing with inner-race fault (CWRU data).

APPENDIX where z is the number of rollers, d is the diameter of the


rollers, D is the pitch diameter, α is the contact angle, and
FAULT RELATED BEARING CHARACTERISTIC f is the rotating speed of shaft.
FREQUENCIES (BCF)
REFERENCES
Each bearing element has its own characteristic frequency of
defect. Those frequencies can be calculated from the kine- [1] Z. Kiral and H. Karagülle, “Simulation and analysis of vibra-
matics relation, that is, the geometry of the bearing and its tion signals generated by rolling element bearing with defects,”
rotating speed. For a bearing with a stationary outer race, the Tribology International, vol. 36, no. 9, pp. 667–678, 2003.
above defect characteristic frequencies can be obtained [14] [2] N. Tandon and A. Choudhury, “An analytical model for the
as follows. prediction of the vibration response of rolling element bear-
ings due to a localized defect,” Journal of Sound and Vibration,
(i) Characteristic frequency of the outer race vol. 205, no. 3, pp. 275–292, 1997.
[3] J. Antoni and R. B. Randall, “Differential diagnosis of gear
  and bearing faults,” ASME Journal of Vibration and Acoustics,
d
fo (in.Hz) = 0.5z f 1 − cos α . (A.1) vol. 124, no. 2, pp. 165–171, 2002.
D
[4] P. D. McFadden and J. D. Smith, “Model for the vibration pro-
duced by a single point defect in a rolling element bearing,”
(ii) Characteristic frequency of the inner race Journal of Sound and Vibration, vol. 96, no. 1, pp. 69–82, 1984.
[5] I. Antoniadis and G. Glossiotis, “Cyclostationary analysis of
  rolling-element bearing vibration signals,” Journal of Sound
d and Vibration, vol. 248, no. 5, pp. 829–845, 2001.
fi (in.Hz) = 0.5z f 1 + cos α . (A.2)
D [6] L. Li and L. Qu, “Cyclic statistics in rolling bearing diagnosis,”
Journal of Sound and Vibration, vol. 267, no. 2, pp. 253–265,
(iii) Characteristic frequency of the rollers 2003.
[7] R. B. Randall, J. Antoni, and S. Chobsaard, “The relationship
between spectral correlation and envelope analysis in the di-
  2
D d agnostics of bearing faults and other cyclostationary machine
fr (in.Hz) = f 1− cos α . (A.3) signals,” Mechanical Systems and Signal Processing, vol. 15,
d D no. 5, pp. 945–962, 2001.
[8] P. D. McFadden and J. D. Smith, “Vibration monitoring of
(iv) Characteristic frequency of the cage rolling element bearings by the high-frequency resonance
technique—a review,” Tribology International, vol. 17, no. 1,
pp. 3–10, 1984.
 
f d [9] R. B. Randall, J. Antoni, and S. Chobsaard, “A compari-
fC = 1 − cos(α) , (A.4) son of cyclostationary and envelope analysis in the diagnosis
2 D
Khalid F. Al-Raheem et al. 13

of rolling element bearing,” in Proceedings of IEEE Interna- sis using Laplace wavelet,” Key Engineering Materials, vol. 293-
tional Conference on Acoustics, Speech, and Signal Processing 294, pp. 183–190, 2005.
(ICASSP ’00), vol. 6, pp. 3882–3885, Istanbul, Turkey, June [26] CWRU, Bearing Data Center, seeded fault test data. http://
2000. www.eecs.case.edu/.
[10] D. Ho and R. B. Randall, “Optimization of bearing diagnostic
techniques using simulated and actual bearing fault signals,”
Mechanical Systems and Signal Processing, vol. 14, no. 5, pp. Khalid F. Al-Raheem received his B.S.
763–788, 2000. (1987), and M.S. (1990) degree in pro-
duction and metallurgical engineering from
[11] H. Qiu, J. Lee, J. Lin, and G. Yu, “Wavelet filter-based weak
University of Technology, Iraq. He is cur-
signature detection method and its application on rolling el-
rently a Lecturer at Mechanical and In-
ement bearing prognostics,” Journal of Sound and Vibration,
dustrial Engineering Department, Caledo-
vol. 289, no. 4-5, pp. 1066–1090, 2006.
nian College of Engineering (CCE), Oman.
[12] D. F. Shi, W. J. Wang, and L. S. Qu, “Defect detection for bear- Prior to joining CCE he was a Lecturer at
ings using envelope spectra of wavelet transform,” ASME Jour- Mechanical Engineering Department, UOT,
nal of Vibration and Acoustics, vol. 126, no. 4, pp. 567–573, Iraq (1991–2001). His research and teach-
2004. ing interests are in machinery condition monitoring and control
[13] C. J. Li and J. Ma, “Wavelet decomposition of vibrations for system engineering. He is currently Ph.D. research student at Glas-
detection of bearing-localized defects,” NDT & E Interna- gow Caledonian University.
tional, vol. 30, no. 3, pp. 143–149, 1997.
[14] R. Rubini and U. Meneghetti, “Application of the envelope Asok Roy is Lecturer at the Glasgow Cale-
and wavelet transform analyses for the diagnosis of incipient donian University. He holds an honors de-
faults in ball bearings,” Mechanical Systems and Signal Process- gree in mechanical engineering, M.S. degree
ing, vol. 15, no. 2, pp. 287–302, 2001. in IT and completed his Ph.D. in Computer
Vision Systems from Glasgow Caledonian
[15] C. Junsheng, Y. Dejie, and Y. Yu, “Time-energy density analysis
University. He has years of industrial expe-
based on wavelet transform,” NDT & E International, vol. 38,
rience in process industry and his current
no. 7, pp. 569–572, 2005.
research interests are in mechanical condi-
[16] W.-X. Yang and X.-M. Ren, “Detecting impulses in mechan- tion monitoring, computer vision systems,
ical signals by wavelets,” EURASIP Journal on Applied Signal and AI applications.
Processing, vol. 2004, no. 8, pp. 1156–1162, 2004.
[17] J. Vass and C. Cristalli, “Optimization of Morlet wavelet for K. P. Ramachandran has the B.E., M.Tech,
mechanical fault diagnosis,” in Proceedings of the 12th Inter- and Ph.D. degrees. He is currently working
national Congress on Sound and Vibration (ICSV ’05), vol. 1, as an Associate Dean (Academic), Caledo-
Lisbon, Portugal, July 2005. nian College of Engineering, Muscat, Sul-
[18] J. Lin and L. Qu, “Feature extraction based on Morlet wavelet tanate of Oman. He has been working more
and its application for mechanical fault diagnosis,” Journal of than 23 years in engineering institutions
Sound and Vibration, vol. 234, no. 1, pp. 135–148, 2000. and as a consultant to many industries. He
joined Caledonian College of Engineering
[19] H. Qiu, J. Lee, J. Lin, and G. Yu, “Robust performance degra-
(a university college) in 1997 where he has
dation assessment methods for enhanced rolling element bear-
had variety of roles. He has obtained pro-
ing prognostics,” Advanced Engineering Informatics, vol. 17,
fessional qualification in mechanical engineering and specialized in
no. 3-4, pp. 127–140, 2003.
the area of maintenance engineering and management. He has re-
[20] N. G. Nikolaou and I. A. Antoniadis, “Demodulation of vibra- search interest in the vibration instrumentation and measurement,
tion signals generated by defects in rolling element bearings analysis and control; condition monitoring of rotating machines.
using complex shifted Morlet wavelets,” Mechanical Systems He has many publications in his credit and has been conferred
and Signal Processing, vol. 16, no. 4, pp. 677–694, 2002. Sir C. V. Raman award for the best technical paper published in
[21] C. Junsheng, Y. Dejie, and Y. Yu, “Application of an impulse re- the Journal of Vibration and Acoustics (1997). He is a Member of
sponse wavelet to fault diagnosis of rolling bearings,” Mechan- ASME, Life Member in the Indian Society of Technical Education,
ical Systems and Signal Processing, vol. 21, no. 2, pp. 920–929, Member of Institution of Plant Engineers India.
2007.
[22] W. J. Wang, “Wavelets for detecting mechanical faults with D. K. Harrison received the B.S. degree with
high sensitivity,” Mechanical Systems and Signal Processing, honours, M.S., Ph.D., CEng FIEE, FIMechE,
vol. 15, no. 4, pp. 685–696, 2001. FIES, CITP, and MBCS degrees. He is cur-
rently the Acting Dean of the School of En-
[23] J. Lin and M. J. Zuo, “Gearbox fault diagnosis using adap-
gineering, Science and Design within Glas-
tive wavelet filter,” Mechanical Systems and Signal Processing,
gow Caledonian University. He has spent
vol. 17, no. 6, pp. 1259–1269, 2003.
most of his working career in the manufac-
[24] L. C. Freudinger, R. Lind, and M. J. Brenner, “Correlation fil- turing industry. A graduate of UMIST, he
tering of modal dynamics using the Laplace wavelet,” in Pro- has also worked within the Department of
ceedings of the 16th International Modal Analysis Conference Mechanical Engineering at UMIST, where
(IMAC ’98), vol. 2, pp. 868–877, Santa Barbara, Calif, USA, he was a Consultant to many blue chip manufacturing companies.
February 1998. He joined Glasgow Caledonian University in 1994 where he has
[25] Z. Yanyang, C. Xuefeng, H. Zhengjia, and C. Peng, “Vibration had a variety of roles. He is currently involved with a number of
based modal parameters identification and wear fault diagno- research projects, principally with industrial organisations, all of
14 EURASIP Journal on Advances in Signal Processing

which have the common theme of design and manufacturing op-


timization. This activity has led to him having successfully super-
vised 23 Ph.D. students through to graduation. He is the joint au-
thor of the widely used textbook “Systems for Planning and Control
in Manufacturing.” He has also edited several other books and con-
ference proceedings and has published his work widely.

Steven Grainger is a Lecturer in Signal Pro-


cessing at Glasgow Caledonian University.
He holds undergraduate degrees in elec-
tronic engineering and computing and re-
ceived his Ph.D. degree from Glasgow Cale-
donian on the application of DSP for the
optimization of motor drives. His teach-
ing and research interests lie in the areas
of DSP, motor drives, condition monitoring
and digital watermarking and he has been a
Visiting Lecturer at Wismar University, Germany; Tomsk Polytech-
nic University, Siberia, and the University of Adelaide, Australia.
EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING

Special Issue on
Emerging Machine Learning Techniques in Signal
Processing

Call for Papers


In the era of knowledge-based society and machine automa- • Adaptive algorithms for machine learning
tion there is a strong interest in machine learning techniques • Applications to speech, audio, video, image, commu-
in a wide range of applications. The attention paid to ma- nications, and so forth
chine learning methods within the DSP community is not • New data representations and learning (independent
new. Speech recognition is an example of an area where DSP component analysis, feature selection, feature learn-
and machine learning “mix” together to develop efficient and ing, kernel PCA, etc.)
robust speech recognizers. Channel equalization is another
area at the intersection of machine learning and DSP tech- Authors should follow the EURASIP Journal on Advances
niques. After all, deciding upon the transmitted information in Signal Processing manuscript format described at the jour-
symbol is nothing but a class assignment task. In cognitive nal site http://www.hindawi.com/journals/asp/. Prospective
radio, DSP techniques and machine learning methods can authors should submit an electronic copy of their complete
work together for developing algorithms for the efficient uti- manuscript through the journal Manuscript Tracking Sys-
lization of the radio spectrum. Image/video/music recogni- tem at http://mts.hindawi.com/ according to the following
tion and retrieval are some more typical examples where SP timetable:
and machine learning tie together. Another problem at the
heart of the DSP community interests is the regression task Manuscript Due October 1, 2007
that can be cast as a machine learning problem.
Over the past years a number of new powerful machine First Round of Reviews January 1, 2008
learning techniques have been developed, which are suitable Publication Date April 1, 2008
for nonlinear processing and for the general case of non-
Gaussian data, and also for nondifferentiable cost functions
or cost functions referring to robust statistics. Adaptive ver-
sions of some of these powerful techniques have only recently Guest Editors:
started being studied. This is an area in which the DSP com-
munity has a lot to say and contribute. Theodoros Evgeniou, INSEAD, Boulevard de constance,
The focus of this special issue is twofold: (a) to consider 77300 Fontainebleau, France;
novel theoretical results in machine learning methods and theodoros.evgeniou@insead.edu
algorithms in the light of typical DSP applications and (b) to Aníbal R. Figueiras-Vidal, Universidad Carlos III de
report novel results obtained by the application of ML tech- Madrid (UC3M), Avenida Universidad 30, 28911 Leganés,
niques in some typical DSP tasks. Madrid, Spain; arfv@tsc.uc3m.es
The special issue is intended to cover topics such as:
Sergios Theodoridis, Department of Informatics and
• Kernel methods (support Vector and Gaussian pro- Telecommunications, National and Kapodistrian University
cess machines and their modifications and exten- of Athens, Panepistimiopolis, Ilissia, 15784 Athens, Greece;
sions) stheodor@di.uoa.gr
• Bayesian networks
• Ensembles: committees, mixtures, boosting, and so
forth

Hindawi Publishing Corporation


http://www.hindawi.com
EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING

Special Issue on
3DTV: Capture, Transmission and Display of 3D Video

Call for Papers


Capturing three-dimensional visual information of a real-life • 3D transmission
scene and creating an exact (except the scale) optical du- ◦ Systems, architecture, and transmission aspects
plicate of it at a remote site instantaneously, or at a later of 3D
time, are ultimate goals in visual communications. All core ◦ 3D streaming
and peripheral components related to this goal are collec- ◦ Error-related issues and handling of 3D video
tively referred to as “three-dimensional television (3DTV).” ◦ Hologram compression
Main functional components of 3DTV are “capture and rep- ◦ Multiview video coding
resentation of 3D scene information,” “complete definition ◦ 3D mesh compression
of digital 3DTV signal,” “storage and transmission of this sig- ◦ Multiple-description coding for 3D
nal,” and finally the “display of the reproduced 3D scene.” ◦ Signal processing for diffraction and holo-
For a successful consumer accepted operation of 3DTV, all graphic 3DTV
these functional components must be carefully designed in • 3D visualization
an integrated fashion by considering the harmonious inter- ◦ 3D mesh representation
action among them. This kind of large-scale integration nat- ◦ Texture and point representation
urally involves a large group of researchers with diverse back- ◦ Object-based representation and segmentation
grounds, and therefore has a highly multidisciplinary na- ◦ Volume representation
ture. ◦ 3D motion animation
◦ Dense stereo and 3D reconstruction
The objective of the proposed special issue is to present, ◦ Stereoscopic display techniques
in a well-coordinated fashion, the works and efforts of re- ◦ Holographic display technology
searchers with rather diverse experience and activity in dis- ◦ Reduced parallax systems and integral imaging
tinct, yet related and complementary areas for achieving full- ◦ Underlying optics and VLSI technology
scale three-dimensional television. The latest research con- ◦ Projection and display technology for 3D videos
tributors in the field and authors of selected papers of the ◦ Human factors
3DTV CON 2007 (www.3dtv-con.org) will be invited to this • 3D applications
special issue. ◦ 3D imaging in virtual heritage and virtual ar-
Papers on the following and related list of topics are so- chaeology
licited: ◦ 3D teleimmersion and remote collaboration
◦ Augmented reality and virtual environments
• 3D Capture and processing ◦ 3D television, cinema, games, and entertain-
ment
◦ 3D time-varying scene capture technology
◦ Medical and biomedical applications
◦ Multicamera recording
◦ 3D content-based retrieval and recognition
◦ 3D photography algorithms
◦ 3D watermarking
◦ Synchronization and calibration of camera
arrays
◦ 3D view registration Authors should follow the EURASIP Journal on Ad-
◦ Multiview geometry and calibration vances in Signal Processing manuscript format described
◦ Holographic camera techniques at the journal site http://www.hindawi.com/journals/asp/.
◦ 3D motion analysis and tracking Prospective authors should submit an electronic copy of
◦ Surface modeling for 3D scenes their complete manuscript through the EURASIP Journal
◦ Multiview image and 3D data processing on Advances in Signal Processing Manuscript Tracking Sys-
tem at http://mts.hindawi.com/, according to the following
timetable:

Manuscript Due October 1, 2007


First Round of Reviews January 1, 2008
Publication Date April 1, 2008

Guest Editors:
Levent Onural, Department of Electrical and Electron-
ics Engineering, Bilkent University, 06800 Bilkent, Ankara,
Turkey; onural@ee.bilkent.edu.tr
Aljoscha Smolic, Fraunhofer-Institut für Nachrichtentech-
nik Heinrich-Hertz-Institut, 10587 Berlin, Germany;
smolic@hhi.de
A. Enis Cetin, Department of Electrical and Electronics En-
gineering, Bilkent University, 06800 Bilkent, Ankara, Turkey;
cetin@ee.bilkent.edu.tr
John Watson, Department of Engineering, University of
Aberdeen, Scotland AB24 3UE, UK; j.watson@abdn.ac.uk
Georgios A. Triantafyllidis, Informatics and Telematics In-
stitute, Center for Research and Technology-Hellas/CERTH,
57001 Thermi-Thessaloniki, Greece; gatrian@iti.gr
Thomas Sikora, Institut für Telekommunikationssysteme,
Technische Universität Berlin, 10587 Berlin, Germany;
sikora@nue.tu-berlin.de

Hindawi Publishing Corporation


http://www.hindawi.com

S-ar putea să vă placă și