You are on page 1of 6

IMTC2003 -instrumentation and Measurement

Technology Conference
Vail, CO, USA, 20-22 May 2003

' ' The Application of the Wavelet Analysis of Analytic Signals in Mechanical Fault
Diagnosis of Circuit Breakers

Xiaoguang Hu, Jianze Wang, Yanchao Ji


School of Electrical Engineering and Automation
Harbin Institute of Technology, P.O. Box 404, P.C. 150001
Harbin, Heilongjiang Pravince, P.R. China
Phone:086-45 1-6413603, Ernail:huxghit@yahoo.corn.cn

Absfracl-This paper introduces a faull diagnosis technique using carried out by several research groups. And those approaches
mechanical vibrafion signals of circuit breakers. A new
frequency-domain compuct-supporf dyadic wavelel is constructed
to apply vibration signals analysis for fault diagnosis of
from the point of frequency domain. This wavelet has the circuit breakers are different. In [Z], a modified spectral
characteristic of dejinile frequency-division. A melhod of wavelel
analysis method was introduced. Included in [3] was the
transform lo analylical signals has been presenled Using lhe
constructed wavelel, not only the spread of signal in thefrequency application of automatic timing methods. Noninvasive
baud corresponding 10 fhe scale is oblained, bur also the signal diagnostics, involving short-time spectra, short-time energy
envelope of amplitude, phase and instantaneousfrequency aI each
scnle are afructed These characleristic parameters are used fo
and automatic timing techniques had been applied to assess
diagnose faults. The mechanical vibration signals from closing the mechanical condition of circuit breakers and to determine
operalions of circuit breakers are discussed by using this melhod the timing of events occurring during operations [4]. In [ 5 ] , a
The simulating resulls demonstrafe fhat it is valid Io use the
wavelef transform of analylic signal in mechanicfault diagnosis of noninvasive technique for assessing the condition of circuit
circuit breakers. breakers by considering the acoustic signals was described. A
method called dynamic time warping is applied to analyze
1. INTRODUCTION the acoustic signals. Envelope statistics, events timing
extraction via differentiation and chi-square based shape
According to the statistical data of failures of circuit analysis were addressed in [ 6 ] .
breakers, the mechanical failures are the principle cause of Through testing and surveying, it is known that the
failures that occur during operations of circuit breakers [I]. A mechanical vibration signal is composed of a series of
series of mechanical vibration events resulting from the overlapped attenuation sine signals. For a circuit breaker,
movements and impacts of internal components occur when which is operated repetitively at the same condition, the
closing and opening operations of circuit breakers are made. external vibration signals are relatively stationary at definite
Those mechanical vibration events form vibration signals range. Some vibration events occur during one closing and
including valuable information of mechanical states of circuit opening operation of a circuit breaker, and each vibration
breakers. Consequently, recent years have brought an event corresponds with specific amplitude, frequency and
increasing focus on and interest in the techniques for attenuation exponential. The malfunctions of circuit breakers
diagnostic testing and monitoring of circuit breakers by result of excess vibration events inevitably, and which bring
analyzing vibration signals. changes of signal amplitude, frequency and so on. The
The mechanical vibration signals from closing and characteristic parameters including the signal amplitude
opening operations of a circuit breaker are recorded by using envelope, phase and instantaneous frequency are extracted to
accelerometers and a data acquisition system. And then those assess the mechanical condition of circuit breakers.
vibration signals are compared with a reference recorded The technique of extracting signal phase and instantaneous
earlier from the same breaker. Thus, the mechanical failures frequency is Hilbert transform. But for the mechanical
can be detected
. . as changes in the recorded vibration patterns. vibration signal including multiple frequency bands, the
Based .,on this 'concept,
, .
,many
,.. achievements have been signal phase and frequency extracted don't have exact

0-7803-7705-2/03/$17.00 02003 IEEE 1477


. . ,,...
. .
I ....

physical interpretation any longer, and sometime wrong


Y(@=
5 4 - 0 46cos(nm) WE [4,-2]u[2,4] .: (2)
conclusion is obtained. other
Wavelet transform is an effective tool for nonstational
signal processing and has been used in many research fields. The graphs of !&)and Y ( w are
) shown in Fig. I.
Recently, wavelet transform is widely applied in vibration
signal processing and many new wavelets are constructed for
specific signals [8-to].
0.5
In this paper, a new frequency-domain compact-support
dyadic wavelet is constructed from the point of frequency
domain. The constructed wavelet has many characteristics
-0.5
0-10L---.l;;l:.r;-..-----
-5 0 5 1 10 1

such as definite frequency-division, linear phase and easily


a. Time-domain waveform
computation. Moreover, it is very sensitive to the singularity
of the signals. A method of wavelet transform to analytical V ( 0 )
1
signals has been presented. In this method, not only the
0.5
spread of signal in the frequency band corresponding to the \, \ , - w
0
scale is obtained, but also the envelope of amplitude, phase -10 -5 0 5 10
and instantaneous frequency at each scale are extracted. The Frequency-domainwaveform
b.
mechanical vibration signals from closing operations of Rg. 1 Timedomain and frequency-domain wavefoml of the constructed
circuit breakers are discussed by using this method. wavelet

11. THEORY E. Annlylic signal

A. Frequency-domain compacf-support wavelet The aim of introducing analytic signal is to describe the
signal amplitude and phase. And the instantaneous frequency
In practice, wavelet transform always can’t divide the can be obtained through phase.
frequency band definitely, and which makes the analytical ) a random real signal, then the analytic signal
Let ~ ( tbe
effect bad, The main reasons that cause the problem above corresponding to x ( i ) is written g(t) and described by
are overlapping of frequency band, leaking, lacking of dyadic g ( r )= r(r)+ j;(r) (3)
wavelet frequency-division and so on. Where i(t) is the Hilbert transform of x(t), and which is
In order to divide the frequency band definitely, this paper defined by
has constructed a new frequency-domain compact-support
dyadic wavelet from the point of frequency domain using
Hamming window function. From the equation (3), the signal g(t) can also be
The wavelet function is written y(t)and described by described by
I
y(t)= 4 2 . 1 6 sinc(r)cos3r
2R
- g ( O = A ( f ) exp(jW)) (5)
Where A(r) and s(t) are the amplitude and phase of
0.921sinc (I + r)cos3(lr + t ) + (1) analytic signal respectively.
The instantaneous frequency of signal can be derived from
sinc (I - I ) cos3(z - t ) ] } the phase of analytic signal, which is defined by
sint
Where sinc ( t j is defined as sinc ( t ) = - t
V ( I )=
1 de
--
271 dt
The Fourier transform Of is written Y(’) and Namely the instantaneous frequency is defined as the
described by derivative of phase. This is a common definition’ of
instantaneous frequency. Thus it can be-seen that the signal
envelope of amplitude, phase and instantaneous frequency

1478
are obtained from the analytic signal. Then, let U =t , the equation of (7) is changed to

C. Wavelet analysis of analytic signal

For the mechanical vibration signal including multiple


frequency bands, the signal amplitude, phase and frequency
extracted don't have exact physical interpretation any longer,
and sometime wrong conclusion is obtained.
This paper introduces a wavelet analysis method, which
=- I--[
1 - 1
z b-t

= W,(a,b)
1
01
E.x(c)v(yVr
lit

(10)
adds the time-frequency localization performance of wavelet Hence, the equation (7)is by
transform with the analytic form of signal. And this paper
also proves that the wavelet transform result of an analytic Wa(a,b) = W,(a, b ) + j$r(a,b) (11)
signal is an analytic signal. It can be seen that the result of wavelet transform to an
From the equation (3), the wavelet transform of g(0 is analytic signal is also an analytic signal, Thus the
described by characteristics of analytic signal are also viable.
1
W8(a2b)=-
8c g ( f ) v ( q ) d t 111. SIMULATING AND ANALYSIS

In this paper, the brackets of circuit breakers are selected


as vibration measuring points, and the vibration acceleration
signals are recorded by using acceleration sensors and a data
acquisition system. The simulating vibration acceleration
signal resulting from one closing operation of a circuit
breaker of SN10-10 type in normal is shown in Fig. 2. The
In the equation (7), the first term is written W,(o,b), waveform includes noise interference signal.
which is the wavelet transform of X(I) depending on the

parameters oand b . The second term is written @,(a,b),


which is the Hilbert transform of x(r) to the parameter bat
the scale a . It is proved as follows I
-1001 f/s
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Fig. 2 The normal vibration signal resulting from closing operation

Through Hilbert transform, the analytic signal of the above


r vibration signal is obtained. Based upon the equation (6), the
instantaneous frequencies are derived from the analytic signal,
which is shown in Fig. 3 .
(8)
Let f - b = r - u , hence
v (om

Fig. 3 Signal instantaneous frequency


(9)

1479
Fig. 3 shows that the mechanical vibration frequencies of B (t)lrad
this type circuit breaker are concentrated on the interval of [0,

-
5kHl. But more information can’t be extracted from Fig. 3.
tls
Because the vibration signal of a circuit breaker is composed ‘0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
of many overlapping vibration events and includes multiple c. Phase

frequency bands, it is no practical meaning to extract the U (t)iHz


instantaneous frequency of signal directly. In order to get 2000 1
more characteristic information, the definite 1000
tls
frequency-division must have been done. Then the ‘0 0.05 0.1 0.15 0.2 0.25 0.3 0.358
characteristic parameters are extracted from the analytic d. Instantaneous frequency
Fig. 5 Wavelet decomposition of normal signal at the scale of a = 2.’
signal at each scale.

It has proved that the wavelet transform result of an


R(t)
50 analytic signal is still an analytic signal and its performances
0 are also viable. Based upon the conclusion, firstly, the
-500 +L 0.05 0.1 0.15 0.2 0.25 0.3 0.35
t Is analytic forms of signals are acquired through Hilbert
a. Real pan of signal transform. Then, using the constructed wavelet described
41) above, the frequency-division is done. Using this method, the

0
n tlS
wavelet decompositions of the signal shown in Fig. 2 at the
scale n = 2 + and n=2-’ are shown in Fig. 4 and Fig. 5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 respectively. In order to avoid the impact resulting from
b. Envelope amplitude signal discontinuity, the phase limit [-n,lr]is taken out from
6’ (t)/rad

:::y
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
I Is
the phase curves.
The instantaneous frequencies of signal shown in Fig. 4(d)
fluctuate at the point of 1800Hz, and the iristantaneous
frequencies shown in Fig. 5(d) fluctuate at the point of 500Hz.
c. Phase In fact, Fig. 4 and Fig. 5 correspond with the two main
U (t)m energy-concentrated mechanical vibration events
4000 magnificently. During the closing operation of circuit
2000 +- breakers, a vibration event occurs at the start point of the
0 11s control mechanism, and the frequency of which is about
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
d. Instantaneous frequency
500Hz. From the amplitude envelope of signal shown in Fig.
Fig. 4 Wavelet decomposition of normal signal at the scale of n = Z4 5(b), the time point of this vibration event occurring is
located at f = 0.048s clearly. But a vibration event happening
at the point of contactors closing, the frequency of which is
R(t)
50 around 1 80OHz. And the time point is located at f = 0.169s
0 c from Fig. 4(b). It can be seen that this wavelet is sensitive to
-50
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
tlS the singularity pf signal. In this method, the time points of
a. Real p a t of signal vibration events are extracted comparatively exactly from the
4)

0
0
*
0.05 0.1
--
0.15 0.2 0.25 0.3 0.35
tls
amplitude envelopes, and this is a characteristic parameter
used to diagnose faults of circuit breakers.
In order to verify the validity of this method, the common
malfunction of circuit breakers, namely its foundation bolts
b. Envelope amplitude
becoming loose, is selected to analyze. The closing vibration
signal is shown in Fig. 6 when the foundation bolts of one

1480
circuit breaker have been loose, According to the same in the time-domain forms of instantaneous frequencies
method mentioned above, the wavelet decompositions of this comparing with Fig. 4(d) and Fig. 5(d). At the scale a = 2 4 ,
signal are obtained. Fig. 7 shows the malfunction signal two excess frequency pulse peaks occur at 0.074s and 0.22s
forms of instantaneous frequencies, and Fig. 8 shows the respectively. And at the scale Q = 2" , two excess frequency
amplitude envelope forms. pulse peaks occur at 0.042s and 0.182s respectively. From
Fig. 8, we also can find that the envelope amplitudes of
malfunction signal are large than those o f normal signal at the
x(Ws scale 0 = z-' and Q = 2-* respectively. In fact, the signals
acquired from accelerometers also include a list of vibration
responses resulting from the control mechanisms colliding
Ns with standers, which add excess amplitudes to the normal.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Fig. 6 The abnormal vibration signal resulting from closing operation
IV. DIAGNOSIS

U (t)Rlz Comparing the instantaneous frequency forms of normal


4000, signal in Fig. 4(d) and Fig. 5(d) with the forms in Fig. 7, we

0 k
0
d Y & T P . L .iu@*Mrwqh
0.05 0.1 0.15 0.2 0.25 0.3 0.35
fls
can find that the occurrence of mechanical malfunction
causes excess pulse peaks in time-domain instantaneous
frequency forms. According to the time and number of the
a. At the scale of a = 2.'
excess pulse peaks occurring in instantaneous frequency form
U (t)iHz

I: ; L-&J- rv'
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
tis
at a definite scale, the condition of a circuit breaker can be
assessed.
At a definite scale, let the maximal amplitude of normal
b. At the scale of Q = 2.'
signal envelope be a norm, and then let the maximal
Fig. 7 Instantaneous frequency of abnormal signal amplitudes of malfunction signal envelopes divide the norm,
a list of relative values are obtained. The relative values are
shown in Table I. Another common malfunction of circuit

q
A@)

0
-
- :bzL

Ii s
breakers (Stop-position spacers becoming loose) is given in
Table I .

0 005 0 1 015 02 025 0 3 035 Tahk I The Relaunc Value, of tmtlopr h l a \ ~ n r Ampliludcc
l

a At the scale of a = 2+ The Condition of clrcuil breaker = 2.' a=22

Foundation Bolts Being Loose 1.3376 1.4651


Normal 1 .oooo 1.oooo
- *b"S"".l
20 Stop-position Spacers Being Loose 0.7905 0.8186
0 n m f Js
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

b. At the scale of Q = 2-l In practical technical application, the envelope maximal


Fig. 8 Envelope amplitude of signal amplitudes of normal signals resulting from the each closing
operations of a circuit breaker at a definite scale are recorded,
During the operations of circuit breakers, the mechanical then the statistical criterion-mean is calculated and then put
malfunctions often cause excess vibration events, furthermore into the computer which is used as cyclical detection. At each
the waveform of each vibration event is overlapped and it is diagnosis, let the envelope maximal amplitude divides the
not easy to distinguish in the time-domain. But from Fig. 7, normal criterion, and then the relative value is obtained.
we can find that obvious excess frequency pulse peaks occur According to the value, the condition of the circuit breaker is

1481
assessed. In this way, the technique of automatic diagnosis is for power sircuit breakers’, IEEE Troons Power DeIivery,.l992, I , (I),
achieved. pp. 353-360
In the method presented in this paper, the characteristic 141 DEMJANENKO, V., VALTIN, R.A., SOUMEKH, M., NAIDU, H.,

parameters of event-timing, instantaneous frequency and ANTUR, A., ANTUR, D.P., SOOM, A., TANFRI, ,U.K., PARK, S.Y.,
envelope amplitude are all extracted the mechanical vibration and BENENSON, D.M.: ‘A Noninvsive diagnmis instrument for
signal of a circuit breaker. If those parameters are synthesized, pawer circuit brekers’, IEEE Tlonr. Panw Deliverv, 1992, 7, (2). pp.
the accuracy of failure diagnosis can be improved 656-663
[ 5 ] RUNDE, M., AURUD, T.,LUNDGAARD, L E , OTTESEN, G.E., and
V. CONCLUSION FAUGSTAD, K.: ‘Acousyic diagnostic of high voltagi: circuit breakers’,
IEEE T h n r Power Delivery, 1992, 7, (3),pp. 1306-1315
The mechanical vibration signal is overlapped by a series 161 POLYCARPOU, A.A., SODM, A., SWARNAKAR, \I., VALTM, R.A.,
of vibration events, and the mechanical malfunctions are ACHARYA, R.S., DEMIANENKO, V., SOUMEKH. M..BENENSON,
manifested by the occurrence of excess vibration events and D.M., and PORTER. J.W.: ‘Event timing and sliape analysis of
the change of vibration amplitudes. It is not easy to vibration bursts from power circuit breakers’, IEFE Trans Power
distinguish only in the time-domain. Ddivery, 1996, 11, (2), pp. 848-857
The characteristic parameters of signal amplitude, phase [7] RLJNDE, M., OTTESEN, G.E., SKYBERG B., arid OHLEN, M.:
and instantaneous can be extracted from the analytic signal ‘Vibration analysis for diagnostic testing of circuil.breakers’, IEEE
nicely, but it is not viable to analyze a signal including Trnns. Power Delivery, 1996, 11, (4), pp. 1816-1823
multiple frequency bands. In this paper, the technique of 18) IUERAN, R.A., HAMDAN, MN., and SHABANEH, N.H.: ‘Wavelet
wavelet transform to analytic signals is introduced into the and chaos analysis of flow induced vibration of a single cylinder in
analysis of mechanical vibration signals of circuit breakers. A cross-flow’, lnlernaiional Journalof Engimering Science, 1998.36, pp.
novel frequencydomain compact-support wavelet is 843-864
constructed. Using this wavelet to spread the analytic signals [9] YOSHIDA, A., OHUE, Y., and ISHIKAWA, H.: ‘Dia.gnosis of tooth
of mechanical vibration according to the frequency band at surface faillire by wavelet transform of dynamic characteristics’,
each scale, the signal amplitude, phase and instantaneous ~ibologylnlernoliono(,2000, 33, pp. 273-279
frequency of each scale are extracted to assess the condition [lo] LM, 1.: ‘Feature extraction of machine sound using wavelet and its
of circuit breakers. Using this method, the mechanical application in fault diagnosis’, NDT&E Inrernarional, 2001, 34, pp.
vibration signals from closing operation of circuit breakers 25-30
are analyzed. The results proved that this method is viable in
mechanical fault diagnosis of circuit breakers. Certainly, this
technique can also be applied to analyze the relative transient
signals in power system.

REFERENCES

[I] FOTO, K., SAKAKIRARA, T., KAMATA. L., and IKEDA, S.: ‘On-line
monitoring and diagnostics of gas circuit breakers‘, IEEE Trans. Power
Deliwry, 1989, 4, (I),
pp. 375-381
[2] LAI, M.L., PARK, S.Y., LIN, C.C.,NAIDU, H., SOOM, A., REINHORN,
A.M., LEE, YH., SOONF, T.T., DEMJANENKO. V., BENENSON,
D.M., and WRIGHT, S.E.: ‘Mechanical failure detection of circuit ‘i, .
.. . . .:
breakers’: IEEE Tram. Power Delivery, 1988, 3, (4), pp. 1724-1731
[3J HESS, D.P.. PARK. S.Y.,
TANGRI, M.K., VOUGIOUKAS, S.G. SOOM,
A., DEMIANENKO, V., ACHARYA, R.S., BENENSON, D.M., and
WRIGHT, S.E.: ‘Noninvasive condition assessment and event timing

You might also like