Home Search Collections Journals About Contact us My IOPscience

Detection and wavelet analysis of partial discharges using an optical fibre interferometric

sensor for high-power transformers

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2003 J. Opt. A: Pure Appl. Opt. 5 66

(http://iopscience.iop.org/1464-4258/5/1/310)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 141.210.2.78
This content was downloaded on 11/06/2014 at 12:54

Please note that terms and conditions apply.

INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF OPTICS A: PURE AND APPLIED OPTICS
J. Opt. A: Pure Appl. Opt. 5 (2003) 66–72 PII: S1464-4258(03)52218-8

Detection and wavelet analysis of partial

discharges using an optical fibre
interferometric sensor for high-power
transformers
H Lamela-Rivera, C Macià-Sanahuja and J A Garcı́a-Souto
Optoelectronics and Laser Technology Group, Department of Electronics, Universidad Carlos
III de Madrid, Av. Universidad, 30, 28911 Leganés, Madrid, Spain

E-mail: horacio@ing.uc3m.es

Received 9 August 2002, in final form 21 October 2002

Published 13 December 2002
Online at stacks.iop.org/JOptA/5/66
Abstract
The main problem in power transformers is the degradation of the isolation,
and partial discharges are a major contribution to this shortcoming. Partial
discharges inside medium- and high-power transformers generate ultrasonic
pressure waves which are usually detected by using acoustic piezoelectric
sensors placed outside the transformers. In this paper we describe an optical
fibre interferometric sensor we are developing for the detection and location
of partial discharges by measuring inside the power transformers. This
non-intrusive optical fibre sensor allows a high level of sensitivity to be
achieved for the typical small pressure variations created by partial
discharges, specially if compared with the results obtained by acoustic
piezoelectric external sensors. We present interferometrically detected
levels of partial discharge. These results indicate that the developed optical
fibre sensor is able to detect a partial discharge in the range of 103 pC that
produces an optical phase shift of about 2 rad with a resolution of 30 pC in
charge means. Finally, the wavelet analysis is used to identify the acoustic
patterns instead of the more commonly used Fourier analysis. When
transforming the measured signal to the frequency domain, the temporal
information disappears. But the acoustic signals really contain numerous
transitory pieces of information to be considered. By looking at a
Fourier-transformed signal it is impossible to state when a particular event
took place. The wavelet analysis helps to solve this problem, and some
results are presented here to show this.
Keywords: Interferometry, optical fibre sensors, acoustic waves
measurements, partial discharges, wavelet analysis, power transformers

1. Introduction which are a major shortcoming of the isolation. They

produce quick transfers of charge in localized areas and,
The main problem in power equipment is the degradation consequently, create a high-frequency electric distortion that
of isolation. In medium- and high-power transformers, propagates through the electric circuit. Methods that have been
this electric isolation has a limited duration for acceptable investigated so far for measuring partial discharges are [1]:
conditions, and its degradation during the period in service is
due to the accumulation of mechanical, thermal and electric • Electric detection: detecting the electrical impulses
effects. In fact, partial discharges are electric phenomena generated;

1464-4258/03/010066+07$30.00 © 2003 IOP Publishing Ltd Printed in the UK 66

 Detection and wavelet analysis of partial discharges

inside the transformer allows a high level of sensitivity to be

BS Lens
He-Ne Laser
achieved for the typical small changes of pressure created by
D partial discharges [4, 5]. Optical fibre methods involve optical
phase modulation by the pressure, so interferometry is used
and based on optical fibre intrinsic interferometers such as the
M
Michelson and Mach–Zehnder [4]. However, their outputs
can fade, and the temperature drift induces changes in the path
length difference, as well.
Optical fibres
The motivation behind this investigation is the
development of interferometric optical fibre sensors applied
Pass-Through inside transformers [6, 7]. As a result, we have a laboratory
Three phase transformer environment that provides the specific conditions for inter-
Oil filled cube ferometric measurement inside transformers (figure 1). This
Magnetic core
research is focused on the detection of acoustic waves de-
rived from partial discharges, as they are not affected by elec-
tromagnetic interferences (EMI). The objective of our work
is to develop a measurement system for monitoring partial
PD Windings discharges. We show an optical fibre interferometric sen-
Fibre-Optic
PD Probes sor and its results for partial discharge measurement which is
Acoustic clearly characterized without the impact of other disturbances.
waves In this paper, we demonstrate that optical fibre sensor sys-
tems are well adapted for non-intrusive detection of partial
discharges when being placed inside power transformers and
by using wavelet analysis.
Figure 1. Optoelectronic set-up and measurement concept for The paper is organized as follows. Section 2 presents
interferometric detection of partial discharges inside power the characteristics of the interferometric system given by the
transformers. M: mirror, BS: beamsplitter, D: photo-detector, PD: principle of operation, the basis of the improved sensitivity
partial discharge. sensing and small signal interference output characteristics.
Section 3 presents the calibration used as a reference for
• Acoustic detection: with acoustic sensors outside the the output signal. Experimental results are included here.
transformer; Section 4 is devoted to the wavelet analysis of the signals
• Optical detection: emitted light between 300 and 500 nm; in order to detect the acoustic patterns as events instead of
• Chemical transformation: measurements of the com- a Fourier analysis. As a result, the generation and detection
pounds that arise after a discharge. of partial discharge is proved. Finally, our conclusions are
It is well known that partial discharges generate ultrasonic discussed in section 5.
pressure waves, which can be used to detect the intensity
and location of the partial discharges. The frequency band 2. Characteristics of the interferometric system
of the discharges in oil includes a wide range approximately
between 20 and 300 kHz, which are typical values in real The interferometric sensor of partial discharges is based on
transformers [2]. Partial discharges have been detected using optical phase measurements with an optical fibre intrinsic
piezoelectric acoustic sensors for typical frequencies of about transducer. The experimental set-up is presented in figure 2,
150 kHz placed externally in relation to the transformer [3]. where the partial discharge generation and implemented
The authors in [3] also present a scheme for the location optical fibre measurement system are shown in detail. It is
of partial discharges by means of triangulation based on the based on a Mach–Zehnder interferometer with optical fibres
times of propagation measured at different positions of the in the reference and sensing arms. Both arms have the same
sensor on the outside of the transformer. Since the ultrasonic length and are constructed with identical coils. One fibre, used
signals travel from the origin of a discharge to the sensor as the sensing arm, has its sensor coil submerged in transformer
along different paths and at different speeds, this means oil exposed to the perturbations caused by partial discharge
that low precision is achieved. In fact, this is the main acoustic waves. The other fibre, used as the reference arm of
limitation of the external location, because it is too difficult the interferometer, is isolated from the impact of the acoustic
to distinguish the signal coming from the origin and those wave and is used as a reference for the optical path of the light
corresponding to alternative paths of the reflected signals. If that propagates through the modulated fibre.
the sensors are located inside the transformer then this problem The interferometer is illuminated with a coherent light
is overcome (figure 1). source, a He–Ne laser (633 nm). Polarization control is
The use of optical fibre sensors presents many advantages: accomplished to assure that no signal fades due to the
immunity to electromagnetic fields even with the fibre in birefringence of the fibres. A beamsplitter is used at the laser
contact with parts in the high-voltage environment, along with output, as well as fibre coupling optics, sensor and transmission
their most important physical advantages such as being light, elements of single-mode fibre (4 µm diameter) of identical
small, highly flexible, highly sensitive and having a large length in both arms, and a mixer for the recombination of two
bandwidth. The use of interferometric optical fibre sensors beams focused onto the optical detector. Both arms are 12 m

67
H Lamela-Rivera et al

behaviour of the pressure changes. The intensity response is

He-Ne Laser L1
BS1 quasi-linear to the partial discharge waves in the range of less
Reference than π rad in quadrature operation (φi = π/2) without the need
OFcoil BS2
HV Electrode of an additional external modulation. The phase information is
Cellulosic
L2 Sensing arm extracted from the intensity signal with cos−1 processing [9].
Dielectric
Sensing OF coil
Detector One can assume a linear relationship between the
Transformer High amplitude of the acoustic wave and the discharge magnitude
Voltage
Ground oil tank (in picocoulombs) [10]. Considering that the acoustic pressure
Electrodes
at 10 cm from a 10 pC discharge source is 0.6 Pa [11] and wave
amplitudes are attenuated by 1/D, where D is the distance from
Figure 2. Experimental set-up of the optical fibre interferometric the source [12], in our system with the sensor placed at 0.5 cm
detection of partial discharges. Mach–Zehnder interferometer from the source, 10 pC represents 12 Pa at the fibre sensor.
configuration, where BS: beamsplitter, L: coupling and collimating For this approximation we assume that the fibre optic coil
lens, OF: optical fibre. sensor is a compact sensor, and a sensitivity S of approximately
4.9 × 10−12 Pa−1 is obtained for a single-mode silica fibre [13].
In this way, the pressure increment P corresponding to an
optical phase change φ is determined by

φ φλ
P = = (2)
Sφ S2πnL

where P is the pressure induced onto the fibre, φ

is the associated optical phase change, S is the constant
term (sensitivity) that relates optical phase and pressure for
ultrasonic waves as in [13], λ is the wavelength of the light, n
Figure 3. Detail of the fibre sensor coil. It is made by winding 8 m is the mean refractive index of the fibre and L is the length of
of fibre around a former into a coil of 25 mm in diameter. the fibre exposed to the pressure field.
Concerning partial discharge generation, we assume that a
in length. The sensor arm is made by winding 8 m of fibre partial discharge makes a short circuit between two electrodes
around a former into a coil of 25 mm in diameter (figure 3), under voltage excitation, into small gassy cavities inside the
similar to the reference coil. isolation. We have tried to reproduce this behaviour with the
With this sensor very high sensitivity is reached, especially assembly of two electrodes separated by cellulose insulating
if compared with external piezoelectric acoustic systems. material and submerged in oil (see figure 2). Partial discharges
For the measurements of partial discharges at the tests, this are generated using a flyback transformer, excited with a
sensitivity is enough to detect electric charges of 103 pC, which chopper circuit, which reaches a high voltage of up to 3 kV and
is what we are looking for [8]. Moreover, this sensitivity could produces the perforation of electric isolation of the dielectric
be vastly increased by using more length of fibre at the sensor material between the two electrodes. In this way partial
coil and maintaining a compact solution. It is expected to discharges have been generated at certain instants of time set
detect partial discharges of up to 10 pC or smaller. by the chopper excitation circuit [14].
The transduction of the acoustic stimulus on the fibre The magnitude of electrical charge generated using this
sensor arm is made through changes in the optical phase procedure is higher than the typical values expected for partial
which are proportional to the pressure changes suffered by discharges, which are in the range of tens to hundreds of
the fibre. A pin photodiode is focused onto the main disc of picocoulomb. However, our interest is devoted mainly to the
the fringe pattern resulting from combining the light beams acoustic wave pattern and the design of the optical fibre sensor,
of both single-mode fibre outputs. In this case, the stimulus which are both sufficiently represented in the experiments
forces an interference intensity given by despite the higher amplitude of the discharge.
I0
I = (1 + cos(φi + φ)) (1) 3. Experimental results
2
where I is the intensity output at the detector, I0 /2 is the mean Partial discharge generates an acoustic wave which spreads
value of the intensity at the detector, φi is the initial phase when in a radial direction. It causes a displacement of the
changes are produced, and φ is the optical phase change due interference rings. With the purpose of having a reference of
to the acoustic wave generated by the partial discharge. the interferometric output signal, the reference arm has been
The optical phase has been obtained in a small signal range forced with a controlled stimulus. In this case, the uniform
(φ < π rad) which modulates the intensity around the quasi- displacement of the interferometric output is a sinusoid
static initial optical phase. The disturbance of temperature evolution of the optic power received at the photo-detector.
is mainly due to a very low frequency random drift of the The passing of maximum and minimum levels is used as a
optical phase that is related to the optical path’s stability of reference in order to characterize the changes in the optical
the interferometer’s set-up and to the response of the fibres to phase taken place due to partial discharges and in a range of
thermal changes. Its impact is far slower than the dynamic π rad (figure 4(a)).

68
 Detection and wavelet analysis of partial discharges

Calibration
1

0.9
Optical zero
0.8
Voltage(V)

Voltage (V)
0.7 Minimum level Minimum level
0.6

0.5

0.4

0.3
Maximum level Maximum level
0.2
-1 0 1 2 3 4 5 6
Time (s) Time (s×10-3)
(a) (b)

Figure 4. (a) Output voltage from the calibration curve of the optical-fibre interferometric sensor. The maximum and minimum marks are
used as a reference in order to characterize partial discharge. (b) Partial discharge interferometric response signal. The intensity response is
in the quasi-linear region of the interferometric signal (φ < π rad).

The interferometric output signal, as a response to partial sensors, and partial discharges can be detected in base-band
discharge stimulus, is represented in figure 4(b) where we frequencies such as those presented here.
can appreciate a typical interference pattern. As in regular The phase resolution obtained is around 4 × 10−3 fringes
interferometers, the measurements have a high level of as we have 5 Msamples s−1 over 2 ms of time observation of
sensitivity. However, the measurement would suffer sensitivity partial discharge which means about 80 samples over 2 rad of
reduction when the sensor reaches peaks or valleys of the the faster transition (figure 5(a)). This value is limited by the
fringes because at these points the change in optical intensity is minimum optical phase shift detected by the acquisition system
almost zero. However, as mentioned previously, this problem and the sampling rate. It corresponds to about 30 pC of charge
is overcome by operating the interferometer in quadrature resolution. Compared to other experiments [5, 8, 11, 15], this
during the tests (figure 4(b)). A linear range of half a fringe technique shows robustness without complexity.
is obtained with cos−1 choosing the centre of one interference Typical results in the bibliography concern the frequency
fringe as the initial point [9]. The signal consists of a decaying or spectrum of the signals detected and associated with the
oscillation. We have obtained the phase shift knowing that we partial discharge. The standard acoustic methods used to detect
have a small signal response and that the benchmarks marked partial discharge employ piezoelectric external sensors. As a
by previously interferometer calibration are never reached
result, a typical frequency of around 150 kHz is looked for,
(figure 4(b)). We have adapted the output intensity signal in
far away from the electrical noise. Observing the results of
function of the calibration and applied the cos−1 function to
the above-mentioned references, a wide frequency range is
find the optical phase shift [9].
assumed to be quite adequate for the acoustical detection of
The optical phase shift φ obtained in the tests is up to
partial discharges: from 10 kHz [8] to up to 120 kHz [15].
2 rad, which corresponds to an equivalent global displacement
The main limitations behind these experiments is that they
of the fibre of 120 nm (figure 5(a)). Although the transduction
only obtain frequency information from the acoustic signals
of the acoustic wave onto the fibre sensor is realized through
the pressure of stretching it, the results are presented in terms of because of the very low amplitude output, or they only detect
changes in length in order to compare with the interferometric peaks derived from the stimulus [11]. Sensing length ranges
master reference provided by the light wavelength. Transferred from 10 m [5, 11] to 100 m [5]. Moderate distances from
to equation (2) and knowing the length inside the oil (8 m), the stimulus are considered [11], up to 50 cm with acceptable
a pressure measurement of P = 3530 Pa is obtained. A output levels [15].
first approach at electric charge is detected; assuming a linear Our system is capable of detecting clear acoustic signals
relationship with pressure, the results are about 3000 pC, as with a reasonable length of optical fibre. Consequently, we
was expected in the range of 103 pC [8]. are able to conduct a wavelet analysis, after the frequency
The spectrum of the interferometric output signal analysis, for a better interpretation of the results obtained.
(figure 5(b)), which is generated with a FFT algorithm, shows However, the extrinsic Fabry–Perot sensor presented in [15]
that frequency components are around 15–17 kHz. Recall that has a linear operating range determined by the cavity design.
piezoelectric measurements have to discard a low-frequency The partial discharges obtained are about 120 kHz in different
bandwidth (up to 5 kHz) to prevent disturbing effects caused tests at different locations of the sensor from the origin. The
by harmonics of the transformer voltage. In our system we do main limitation compared to our implementation relates to
not have to worry about this because the fibre sensor is immune increasing the sensitivity, as well as to fabrication of the sensor.
to EMI. Consequently, the use of a high-pass filter applied to Our intrinsic optical fibre transducer is more sensitive simply
the acoustic signals is not necessary, unlike for piezoelectric due to the use of a longer fibre, and its compaction.

69
H Lamela-Rivera et al

A.U.
1

0.875

0.750

0.625
Displacement (nm)

0.500

0.375

0.250

0.125

Frequency (Hz × 10 )
4

Time (s ×10-3)
(a) (b)

Figure 5. (a) Displacement measured. The phase information is extracted from the intensity signal with cos−1 processing. Fibre
displacement is proportional to phase shift. Phase shift is up to 2 rad, which corresponds to a global displacement of 120 nm. (b) Spectrum
of fibre displacement. Frequency components are around 15–20 kHz.

velocity signals (without the superposition of several signals

Signal as in [15]).

4. Wavelet analysis of partial discharges

Filters
If signals do not vary too much in the time domain stationary
lowpasss highpass signal Fourier analysis is useful. However, most of the
acoustic signals contain numerous non-stationary or transitory
characteristics: tendencies, abrupt changes, and start and end
points of the signals. These characteristics are often the most
Approximation Detail important part in the signals, and Fourier analysis does not
Signal Signal allow them to be obtained due to its serious disadvantage: when
transforming to the frequency domain, temporal information
Figure 6. Wavelet decomposition. The original signal passes disappears. It is impossible to look at a transformed Fourier
through two complementary filters and emerges as two signals
(approximation and detail signals). This decomposition process signal and state when a particular event took place.
could be iterated generating the decomposition wavelet tree. A wavelet is a finite waveform with null average value.
Fourier analysis consists of the decomposition of the original
Further, [8] shows values of the charge involved in the signal into sinusoid signals of several frequencies. In
partial discharge. It is acoustically detected for about 200 nC, contrast, wavelet analysis decomposes the original signal into
much larger quantities than the measurements obtained with scaled signals of the original wavelet (or mother). It can
our system. The frequency spectrum is about 10 kHz, and be appreciated intuitively that signals with form changes,
the influence of different characteristics of the fibre and optic characteristic of those generated by partial discharges, can be
intensity output has been studied. better analysed with an irregular wavelet signal than with a
It has been demonstrated that when partial discharges are uniform Fourier sinusoid.
accompanied by ‘buzzes’, which can be heard, the amplitude In wavelet analysis we have to work with two signals: the
of the components with frequencies between 15 and 20 kHz approximation and the detail (figure 6). The approximation
in the presence of this discharge noise is higher than in their corresponds to low-frequency components and the detail
absence [16]. corresponds to high-frequency components. The original
The difference between the frequency values, 120 kHz signal passes through two filters and two signals emerge. These
compared to 15–17 kHz, stems from the diffraction of acoustic are the approximation signal (low frequency) and the detail
waves. Diffraction appears when the acoustic wavelength signal (high frequency). This decomposition process could be
becomes comparable to the probe size. The velocity of acoustic iterated generating the decomposition wavelet tree.
waves in transformer oil is about 1400 m s−1 . For a 100 kHz In the paper published by Masanori [17] the Mexican
ultrasonic wave, the acoustic wavelength is therefore 14 mm. hat and Daubechies are proposed as wavelet functions for
As our fibre coil sensor has a 25 mm diameter, greater than the acoustic partial discharge analysis. Of the two, it concludes
acoustic wavelength, a diffraction phenomenon is produced. that the Daubechies wavelet function is more effective than
Even so, we can detect partial discharge by looking at group the Mexican hat wavelet function for the study of partial

70
 Detection and wavelet analysis of partial discharges

A.U
1
discharges in oil. We have obtained a clear signal between 15
0.8
and 17 kHz corresponding to the acoustic waves propagating
0.6 in the oil generated by a partial discharge.
0.4 Wavelet analysis has provided a tool to identify the shape
0.2 of the partial discharges. As a result, the characteristic pattern
0 used to decompose the signal is obtained. Also it gives us
-0.2
temporal information about the signal that Fourier analysis
-0.4
cannot offer. Moreover, wavelet analysis helps to locate partial
-0.6
discharge.
-0.8

-1
Our system gives excellent results compared to other
Time (ms) acoustic sensors and wavelet analysis [17, 18]. The main
reason is the use of optical fibre placed inside the transformer,
Figure 7. Wavelet decomposition signal from displacement which is thus located in a good place to detect partial discharge.
measured. Daubechies 16 wavelet transform has been used at level 5 Also, compared to the conventional acoustic sensor, the optical
of decomposition. Acoustic patterns obtained attenuate as time
passes, which is a typical behaviour of acoustic signals. fibre sensor has the advantages of being non-electrically
conducting and having immunity to EMI and electrical noise.
The quality of the signals obtained is much better than the
discharges. Thus, the Daubechies 16 wavelet transform has
quality observed by external piezoelectric sensors. We thus
been used, at level 5 of decomposition for the wavelet analysis
applied to the signals obtained in our system. The fundamental have proven observations of partial discharge. Finally, we
reason for this choice is based on its good reproduction of the can distinguish a discharge level of 30 pC which is limited by
partial discharge acoustic pattern that we want to detect. 10−3 rad resolution of the interferometric signal.
The results are shown in figure 7. A typical wave Future work will improve the sensor for real application
train shape is observed, as a result of applying the selected and in-field exploitation. In power transformers low
wavelet pattern in the decomposition. This pattern is typical frequencies have to be prevented, such as core noise and
of acoustic signals and has its characteristic frequency at oil circulation, which are avoided by using a soft fibre
15 kHz. The pickup signal after the wavelet transformation coating which causes sensitivity to decrease to under 2 kHz
is also a substantial improvement compared to other wavelet (similar to a high-pass filter). Future work will deal with
characterization systems [18], since an acoustic signal is the implementation of the system in a real medium-power
represented in a good way, when a pattern that represents the transformer within the projects of our medium-power research
Daubechies db 16 transform wavelet is obtained at the output plant (FUTURE Project), in a similar way as was described
of the analysis. Moreover, as we should expect, the pattern in [19].
attenuates as time passes, which is the typical behaviour of
acoustic signals.
These results can be compared to those presented in [18], Acknowledgments
which were obtained with piezoelectric acoustic sensors, rather
than the optical fibre sensor, when looking for the best wavelet This work is linked to the FUTURE project with the Spanish
application. The noise levels with acoustic sensors are higher electric company UNIÓN FENOSA SA. They are gratefully
than those obtained in our experiment developed with the acknowledged.
optical fibre. It becomes difficult to distinguish between
acoustic signal and noise with the external acoustic sensors. References
Optical fibre results are clearer than others because of the
good properties of the fibre which allows a highly significant [1] Kusumoto S, Itoh S, Tsuchiya Y, Mukae H, Matsuda S and
reduction of EMI. The wavelet analysis carried out by the Takahashi K 1980 Diagnostic technique of gas insulated
authors reveals the appearance of some cleaner signals due substation by partial discharge detection IEEE PAS-99 4
to the filtration effect of this technique, but although the 1456–65
[2] Howells E and Norton E T 1978 Detection of partial
appearance of a wavelet pattern is barely appreciated, it is not discharges in transformers using acoustic emission
clearly observed in [18]. They also propose this technique for techniques IEEE Trans. Power Appar. Syst. 97 1538–49
the location of the exact origin of the partial discharge, with [3] Phung B T, James R E, Blackburn T R and Su Q 1991 Partial
different sensors in different places. Besides this, in [18] they discharge ultrasonic wave propagation in steel transformer
do not look for the pattern used in the wavelet transformation tanks ISH7: 7th Int. Symp. on High Voltage Engineering
pp 131–4
and are also limited in noise filtering and identifying the origin [4] Zargari A and Blackburn T R 1996 Modified optical fibre
of the discharge. sensor for PD detection in high-voltage power equipment
IEEE Int. Symp. on Electrical Insulation pp 424–7
5. Conclusions [5] Zhiqiang Z, MacAlpine M and Süleyman M 2000 The
directionality of an optical fiber high-frequency acoustic
In this paper we present the results that have been obtained sensor for partial discharge detection and location J.
with a Mach–Zehnder interferometric system for measuring Lightwave Technol. 18 795–806
partial discharge as a change of the optical phase of the light [6] Lamela H, Garcı́a-Souto J A and Sanz J 2000 Characterization
of temperature and vibrations into medium-power
in the optical fibre sensing arm. The experimental results transformer using an optical-fiber interferometer Proc. SPIE
indicate that this work can be applied to the detection of partial 4073 316–23

71
H Lamela-Rivera et al

[7] Lamela H, Garcı́a-Souto J A and Sanz J 2000 Measurements [14] Lamela H, Macia C, Garcı́a-Souto J A and Santos J I 2001
of mechanical vibrations at magnetic cores of power First partial discharges tests using an interferometric sensor
transformers with fiber-optic interferometric intrinsic sensor for transformer applications Proc. ODIMAP III pp 343–8
IEEE J. Sel. Top. Quantum Electron. 6 788–97 [15] Deng J, Xiao H, Huo W, Luo M, May R, Wang A and Liu Y
[8] Cosgrave J A, Vourdas A, Jones G R, Spencer J W, Murphy 2001 Optical fiber sensor-based detection of partial
M M and Wilson A 1993 Acoustic monitoring of partial discharges in power transformers Opt. Laser Technol. 33
discharges in gas insulated substations using optical sensors 305–11
IEE Proc. A 140 369–74 [16] Marinescua A, Georgescu G and Filiseanu V 1991 Ultrasonic
[9] Lamela H 1989 Sensores ópticos interferométricos y medidas equipment for detecting, measuring and locating of partial
de longitudes de onda Sensores Ópticos (Madrid: Sociedad discharges at power transformers Proc. 17th Int. Symp. on
Española de Óptica) pp 109–33 High Voltage Engineering pp 127–30
[10] Lundgaard L E 1992 Partial discharge—part XIV: acoustic [17] Masanori M, Okano T, Nishimoto S and Arii K 1995 Study on
partial discharge detection—practical application IEEE application of wavelet analysis for degradation diagnosis of
Electr. Insul. Mag. 8 34–43 partial discharge on a void IEEE 5th Int. Conf. on
[11] Zargari A, Phung B T and Blackburn T R 1995 An ultrasonic Conduction and Breakdown in Solid Dielectrics pp 371–5
optical fibre sensor for partial discharge detection 9th Int. [18] Ramirez-Niño J, Rivera-Castañeda S, Garcia-Colon V R and
Symp. on High Voltage Engineering p 5581 Castaño V M 1998 Analysis of partial electrical discharges
[12] Spaeren P H, Skyberg B, Hegerberg R and Lundgaard L E in isolating materials through the wavelet transform
1985 On site particle detection in gis by acoustic Comput. Mater. Sci. 9 379–88
measurements Int. Symp. on GIS pp 331–40 [19] Garcı́a-Souto J A and Lamela-Rivera H 2002 Comparative
[13] Knudsen S and Blotekjaer K 1994 An ultrasonic fiber-optic analysis of optical-fibre interferometric sensors versus
hydrophone incorporating a push–pull transducer in a accelerometers: application to vibrations inside high-power
sagnac interferometer J. Lightwave Technol. 12 1696–700 transformers J. Opt. A: Pure Appl. Opt. 4 S318–26

72