Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Optics & Laser Technology 33 (2001) 305–311

www.elsevier.com/locate/optlastec

Optical ber sensor-based detection of partial discharges


in power transformers
Jiangdong Denga; ∗ , Hai Xiaob , Wei Huoa , Ming Luob , Russ Maya ,
Anbo Wanga , Yilu Liua
a Photonics Laboratory, Bradley Department of Electrical and Computer Engineering, Virginia Polytechnic Institute and State University,
Blacksburg, VA 24061-0111, USA
b Lucent Technologies, Optical Ampli*er Development Group, 9999 Hamilton Blvd, Breinigsville, PA 18031-9359, USA

Received 15 April 2000; received in revised form 25 November 2000; accepted 23 December 2000

Abstract

In this paper, a ber optic acoustic sensor system is designed and tested for on-line detection of the partial discharges inside high voltage
power transformers. The ber optic sensor uses a silica diaphragm and a single mode optical ber encapsulated in a silica glass tube to form
an extrinsic Fabry–Perot interferometer. Test results indicate that the developed ber optic sensors are capable of detecting the acoustic
signals propagating inside the transformer oil with high resolution and high frequency.  c 2001 Elsevier Science Ltd. All rights reserved.

Keywords: Optical ber sensors; Acousto-optics sensors; Power transformers

1. Introduction
laboratory conditions, but it is diCcult to be applied in the
eld on in-service transformers because of the high envi-
Power transformers are the most critical and costly com-
ronmental noise level and lack of accurate calibrations. The
ponent in power transmission systems. Catastrophic failures
current chemical approach detects PDs in transformers by
of power transformers can occur without warning, result-
taking gas or oil samples from the transformer. More recent
ing in serious oil spills, res, extensive damage to adjacent
research includes the development of gas sensors and apply-
equipment, and major disruption of service. The cost of fail-
ing them in on-line gas monitors [3–5]. Problems associated
ures can easily drive the total cost of a single transformer
with chemical methods are the fact that there can be a long
failure into the tens of millions of dollars [1]. Dielectric
time delay between the initiation of a PD source and the
breakdown is a major cause of power transformer failures,
evolution of enough gas to be detectable. For the electrical
and very often, partial discharges (PDs) are involved in those
and chemical methods, a further limitation of them is that it
dielectric breakdowns by degrading the insulation property
is generally not possible to allocate the exact location of a
of the transformer oil [2]. Hence, it is important that the par-
detected PD source. Generally speaking, a PD results in a lo-
tial discharge activity being studied and monitored to detect
calized, nearly instantaneous release of energy. It produces
incipient insulation problems, to prevent catastrophic fail-
ultrasonic waves propagating through the insulate medium.
ures, and to prevent extensive costs.
By placing a suitable sensor, the acoustic wave can be
In general, three approaches can be used to detect partial
detected to generate information relevant to the PDs
discharges inside the power transformer including electri-
[6 –8]. One obvious advantage of the acoustic methods is
cal, chemical and acoustic methods. The electrical method
that the site of a PD can be located by studying the phase
can provide accurate recordings of partial discharges under
delay or the amplitude attenuation of the acoustic waves.
∗ Corresponding author. Tel.: +1-540-231-4355; fax: +1-540-231- Furthermore, acoustic methods have the potential advan-
4363. tage of better noise immunity for on-line PD detection
E-mail address: jdeng@vt.edu (J. Deng). applications.

0030-3992/01/$ - see front matter  c 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 3 0 - 3 9 9 2 ( 0 1 ) 0 0 0 2 2 - 6
306 J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311

Acoustic PD detection can be realized by mounting piezo-


electric acoustic sensors externally on the walls of the power
transformer, and very often a suitable couplant is used to
ensure good transmission of the acoustic waves. The exter-
nally mounted piezoelectric acoustic sensor method oJers
the advantage of easy installation and replacement. How-
ever, the piezoelectric sensor suJers from corruption of the
signal from environmental noises such as electro-magnetic
interference (EMI). Another problem associated with the ex-
ternally mounted piezoelectric sensor is that the multi-path
of the acoustic wave transmission makes it diCcult to locate
the exact site of the partial discharges [9,10]. It is thus desir-
able to have such sensors that can be reliably operated inside
a transformer, even deep within the transformer windings,
to pick up clean PD-induced acoustic signals. These sensors Fig. 1. Illustration of the principle of ber optic acoustic sensor.
need to be chemically inert, electrically non-conducting, op-
erating passively, and small in size.
Optical ber-based sensors have been shown to be at- are environmentally induced in the ber connecting, the op-
tractive to measure a wide range of physical and chemical tical source, the sensor head and the detectors, are common
parameters because of a number of inherent advantages, mode and therefore do not aJect the signal phases.
including small size, light weight, high sensitivity, high In this paper, we demonstrate a ber optic EFPI prototype
frequency response, and immunity to electromagnetic inter- sensor system for inside detection of partial discharges in
ference. These advantages make optical ber sensors an ex- power transformers.
cellent candidate for transformer PD detection. Fiber optical
acoustic sensors have shown useful in many applications,
such as under water hydrophones [11–14], material prop- 2. Principle of operation
erty analysis, and civil structure non-destructive diagnosis
[15,16], vehicle detection and traCc monitoring [17]. The The basic principle of the ber optic acoustic sensor is
earlier methods of acoustic signal detection were based on illustrated in Fig. 1. The system consists of a sensor probe,
ber optic intrinsic interferometers like all- ber Michelson a laser diode, an optoelectronic signal-processing unit,
interferometer and Mach–Zehnder interferometers. These and a single mode ber linking the sensor probe and the
intrinsic ber sensors usually use single mode ber and co- signal-processing unit. The light from a distributed feedback
herent sources, such as lasers. The light from the source is (DFB) laser diode is launched into a 2×2 ber coupler and
split into two bers with equal intensity by a 3 dB ber cou- propagates along the ber to the sensor head (to reduce the
pler. One ber, referred to sensing arm, is exposed to the optical feedback eJect to DFB laser, an anti-reMection con-
acoustic signal while the other, referred to reference arm, nector is used after DFB laser). As shown in the enlarged
is shielded from the impact of the acoustic wave. The light view of the sensor head, the lead in=out ber and the thin
beams propagating in the two arms are recombined either silica glass diaphragm are bonded together in a cylindrical
by their reMections (Michelson) or transmissions (Mach– silica glass tube to form a Fabry–Perot sensing element.
Zehnder) to generate interference signals which are modu- The incident light is rst partially reMected (∼ 4%) at the
lated by the acoustic waves. The intrinsic ber interferomet- endface of the lead in=out ber due to the Fressnel reMec-
ric sensors have shown very high sensitivities when a long tion at the glass–air interface. The remainder of the light
ber in the sensing is used. However, they suJer from the propagates across the air gap to the inner surface of the
fringe fading problems resulting from the random polariza- diaphragm. Because the inner surface of the diaphragm is
tion rotation. They are also unstable because of the source coated with gold which reMects nearly all the incident light
wavelength and the temperature-induced changes in the path (∼ 96%), the ber sensor is thus optically self-contained in
length diJerence. any environments. This means that the optical signal is only
More recently, ber optic extrinsic Fabry–Perot interfer- a function of the length of the sealed cavity, and is immune
ometric (EFPI) sensors are under development for the ap- to the diaphragm outer surface contamination resulting from
plications of acoustic signal detection [18 –21]. Fiber EFPI the contact with transformer oil. The two reMections travel
sensors are implemented on a very small sensing element back along the same lead-in ber through the same ber
known as Fabry–Perot cavity which is formed by two reMect- coupler to the photodetection end. The interference of these
ing surfaces. Compared to Michelson and Mach–Zehnder two reMections produces sinusoidal intensity variations,
ber sensors, EFPI sensors are compact in size and there- referred to as interference fringes, as the air gap is continu-
fore realize virtually single point measurement. More impor- ously changed. One period of fringe variation corresponds
tantly, those polarization rotation and phase changes, which to an air gap change of one-half of the optical wavelength,
J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311 307

3. Sensor frequency response and sensitivity design

As shown in the enlarged view of Fig. 1, the sensor head


is fabricated by bonding the ber, ferrule, silica tubing and
silica diaphragm together to form a sealed ber optic extrin-
sic Fabry–Perot interferometer. The diaphragm vibrates at
the presence of an acoustic wave which imposes a dynamic
pressure on the diaphragm. It is very important to design
the sensor head to ensure high enough frequency response
and sensitivity to achieve optimum detection of partial
discharges.
In the partial discharge detection, the sensitive frequency
of the sensor system is suggested to be in the range of
20 –200 kHz [6 –10]. This is basically due to the fact that
Fig. 2. Illustration of a linear operating range of the sensor response curve.
magnetostriction-introduced noise of transformers usually
exists in the low frequency range below 60 kHz, and the
acoustic emission frequency of PD is mostly around 150 kHz
which in our case is about 0:66 m. As indicated in the en- [8]. According to Ref. [22], the frequency response fn of
larged view of the sensor head, the diaphragm is tilted at an the sensor in terms of dynamic pressure load is given by the
angle with respect to the lead-in ber endface so that only following equation:
about 4% of the second reMection is captured by the ber 
 Dg
to ensure a maximum visibility of the interference fringes. fn = × ; (1)
2R2 hw
Fig. 2 illustrates the output interference fringes typically
obtained from the ber optic EFPI sensor. In principle, con- where  is a constant related to the vibrating modes of the
tinuous tracking of phase changes in the interference fringes diaphragm, and takes a value of 10.21 for the fundamental
can yield information about air gap changes in the sensor mode (n = 0); R is the eJective radius de ned by the inner
element. The acoustic signal generated by partial discharges diameter of the silica sensor-housing glass tubing; h is the
causes the deformation of the diaphragm and modulates the thickness of the diaphragm; g is the gravitational constant;
sealed air gap length. The sensor therefore gives outputs w is the speci c weight of the diaphragm material; and D is
that correspond to the acoustic signals. Like regular interfer- the Mexural rigidity of the diaphragm de ned by
ometers, the measurement will have ultra-high sensitivity. Eh3
D= ; (2)
However, the measurement would suJer from the disadvan- 12(1 − 2 )
tages of sensitivity reduction and fringe direction ambigu- where  is the Poisson’s ratio ( = 0:17 for silica glass
ity when the sensor reaches peaks or valleys of the fringes. material at 25◦C); E is the Young’s modulus of the silica
Sensitivity is reduced at the peak or valley of a fringe since glass material (E =73:73 GPa or 7:49×109 kg=m2 at 25◦C):
at that point the change in optical intensity is nearly zero For a fused silica glass diaphragm at 25◦ C; w is 2:202 ×
for a small change in the air gap. Fringe direction ambigu- 103 kg=m3 . The frequency response of the sensor can thus be
ity refers to the diCculty in determining whether the air gap calculated by combining Eqs. (1) and (2) into the following:
is increasing or decreasing by detecting the optical inten-
h
sity. If a measurement starts with an air gap corresponding fn = 2:749 × 109 2 (Hz); (3)
to the peak of a fringe, the optical intensity will decrease, R
regardless of whether the gap increases or decreases. One where h and R are in microns. As indicated by Eq. (3), the
of the key successes of our sensor is that a controlled bond- frequency response is proportional to the thickness of the
ing technique is used to fabricate the sensor so that the sen- diaphragm and inversely proportional to the square of
sor operates only over the linear range of a half fringe. As the eJective diaphragm radius. In practice, the diaphragm
shown in Fig. 2, the initial operating point is chosen to be fundamental frequency is usually designed larger than
the central point of the interference fringe, and the thickness 200 kHz to ensure the sensor have relative uniform response
of the diaphragm is designed in such a way that the imposed for the PD acoustic wave.
acoustic signal only deforms the thin diaphragm within the The diaphragm will be deMected when there is a diJeren-
linear part of the sensor response curve. On the other hand, tial pressure p between the inside and outside sealed cavity.
since the maximum amplitude of acoustic wave depend on The out-of-plane deMection of the diaphragm y is a function
the PD source and the damaged oil properties, the sensitiv- of the pressure diJerence at any radium position r. The ratio
ity of the sensor is selected carefully for matching diJerent between the deMection and the pressure diJerence is de ned
amplitude of the PD acoustic wave. Therefore the sensitiv- as the sensitivity of the sensor, which can be expressed as
ity reduction and fringe direction ambiguity problems can y 3(1 − 2 )
= = × (R2 − r 2 )2 : (4)
be completely avoided. p 16Eh3
308 J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311

Fig. 3. Predicted sensor sensitivity (m=Psi) versus diaphragm thickness Fig. 4. Image of the ber optic acoustic sensor head.
at R = 0:5 mm.

In our sensor con guration, the ber is positioned to the cen- tubing is chosen to be 1 mm, which limits the eJective
tral part of the diaphragm so that only the center deMection diaphragm radius to be 0:5 mm. To satisfy the frequency
y0 is of interest, which is given by response requirement, we choose the diaphragm thickness
to be 20 m, which provides a upper frequency response of
y0 R4 220 kHz, therefore, the sensor provides a response of about
= = 1:71 × 10−8 3 ; (5)
p h 0:134 m=psi.
where y0 is in microns, and p is in the pounds per square Several sensor heads were made by bonding a single mode
inch (psi), so  is in m=psi. Fig. 3 shows a typical sensor ber, a silica glass ferrule, a silica glass tubing and a thin
sensitivity curve at R = 0:5 mm. silica diaphragm together as shown in to Fig. 1. The ferrule
Based on these equations, the sensitivity and the fre- used has an inner diameter of 127 m and outer diameter
quency response of the sensor can be designed to t diJer- of 995 m; the tubing used has an inner diameter of 1 mm
ent application requirements either by choosing diaphragm and outer diameter of 6 mm. The sensor head is nished by
materials with diJerent  and E or by changing the ge- polishing the diaphragm to the thickness of 20 m, as shown
ometric parameters of the sensor head with desired eJec- in Fig. 4. In the sensor fabrication, the initial air gap between
tive diaphragm size R and thickness h. In general, a di- the ber and the inner surface of the silica diaphragm was
aphragm with a larger radius and a smaller thickness will adjusted to obtain the highest interference fringe visibility.
render more sensitive detection of the acoustic signals. How- The initial operating point was also adjusted to the central
ever, as indicated above, the operating range of the EFPI point of a fringe for linear operation and the highest detection
sensor head needs to be limited within the linear range, sensitivity.
which is a fraction of an interference fringe to avoid the To validate the feasibility of using the designed ber optic
sensitivity reduction and fringe direction ambiguity prob- sensors for the detection of partial discharges inside power
lems. This imposes a limitation on the thickness of the silica transformers, a eld test was performed at J.W. Harley, Inc.,
diaphragm. in Twinsburg, Ohio, where a specially designed testing fa-
cility (Fig. 5) is available to study the partial discharge
phenomena power transformers. The testing facility is basi-
4. Experiments and results cally a simulated partial discharge environment inside high
voltage power transformers, where a controlled needle-plate
A prototype ber optic sensor system was built and tested partial discharge generator (PDS) is immersed in the trans-
to demonstrate the feasibility of on-line detection of partial former oil.
discharges in power transformers. The prototype system uses The ber optic acoustic sensor (OFS) was immersed in the
9
a DFB laser at 1315 nm pigtailed with standard 125 single transformer oil with the diaphragm towards the needle-plate
mode ber as the source. A high speed InGaAs detector partial discharge generator, as shown in Fig. 5. The sensor
is used to detect the optical signal from the sensor, and probe was connected to the signal demodulation unit through
its output is fed to a low noise high gain trans-impedance a single mode ber cable and an FC connector. After the
ampli er. The frequency response of the electronic circuit photo-detection and the signal processing, the output data
is limited within a range from 30 to 300 kHz, de ned by an were digitized and recorded through a digital oscilloscope.
additional band-pass lter. Fig. 6 shows the typical acoustic signal output from the ber
The analysis described above provides a clear guideline optic sensor at the events of partial discharges.
to the design of the ber optic acoustic sensor for partial A comparison test was performed between the Physical
discharge detection. The inner diameter of the silica glass acoustic sensor (PZT shown in Fig. 5) and the ber sensor,
J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311 309

Fig. 5. Partial discharge test setup at J.W. Harley Inc. at Twinsburg, Ohio, U.S.A. OFS: Optical ber sensor, PZT: physical acoustic sensor, PDS: partial
discharges sources, HVT: high voltage transformer.

Fig. 6. Typical PD acoustic signals from ber optic sensor.

arranged side by side. Their typical output signals upon par-


tial discharges are given in Fig. 7, where, (a) is the output
from the Physical acoustic sensor and (b) is the output from
the ber sensor. The two sensors’ outputs resemble each
other in details. Both the outputs indicate that the acoustic
signal generated by a partial discharge comprise of sinu-
soidal pulses with a gradually decreasing amplitude. Fig. 7
also shows that the two acoustic signal pulses detected by
the two sensors have the same time period of about 8 s.
This means that both sensors detected the partial discharge
acoustic signal at a frequency of about 120 kHz, which is
the typical frequency of partial discharge acoustic waves, as
indicated by the literature [6 –10].
When an acoustic wave generated by the partial discharge Fig. 7. Typical partial discharge acoustic signals detected by the ber
travels in the insulation medium of a transformer, its am- optic sensor and the physical acoustic sensor. (a) Physical acoustic sensor
plitude will attenuate over distance. Therefore, one of the output, and (b) ber optic sensor output. Both sensors are located at a
possible methods to locate the site of the partial discharge distance of 10 cm away from the partial discharge source.
310 J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311

facilities. The reported research work is funded by the Elec-


trical Power Research Institute (EPRI) and National Science
Funding (NSF).

References

[1] An international survey on failures in large power transformers in


service. Final Report of Working Group 05 of Study Committee
12 Transformers, ELECTRA, No.88, May 1983.
[2] Kogan VI, Fleeman JA, Provanzana JH, Shih CH. Failure analysis
of EHV transformers. IEEE Transactions on Power Delivery
1988;3(2):672–83.
[3] Inoue Y, Suganuma K, Kamba M, Kikkawa M. Development of
oil-dissolved hydrogen gas detector for diagnosis of transformers.
IEEE Transactions on Power Delivery 1990;5(1):226–32.
[4] Pledger WA, Pyke SC. Gas monitor update: review of progress
in the development of a microelectronic in-situ transformer
fault gas analyzer. Third EPRI Substation Equipment Diagnostics
Fig. 8. Amplitude dependence on the distance between sensor location Conference, New Orleans, Louisiana, 1–3 November 1994.
and the PD source. [5] Abbott JW, Chu D, Diamond AE, ElBadaly HA, Slemon CS
Development of an automated transformer oil monitor (ATOM).
Third EPRI Substation Equipment Diagnostics Conference, New
Orleans, Louisiana, 1–3 November 1994.
source is to monitor the amplitude attenuation of the re-
[6] Lundgaard LE. Partial discharge—part XIII: acoustic partial
ceived acoustic signals. We measured the amplitudes of discharge detection—fundamental considerations. IEEE Electrical
the acoustic signals detected by the ber sensor at diJer- Insulation Magazine 1992;8(4):25–31.
ent distances away from the needle-plate partial discharge [7] Lundgaard LE. Partial discharge—part XIV: acoustic partial
generator. The results are shown in Fig. 8, which indi- discharge detection—practical application. IEEE Electrical
cate that the amplitude of the acoustic signal has a strong Insulation Magazine. 1992;8(5):34.
[8] Kawada H, Honda M, Inoue T, Amemiya T. Partial
dependence on the location distance between the sensor discharge automatic monitor for oil- lled power transformer.
and the PD source. The largest distance tested was 55 cm IEEE Transactions on Power Apparatus and Systems
which is limited by the size of the oil tank of the testing 1984;PAS-103(3):422–28.
facility. [9] Howells E, Norton ET. Location of partial discharge sites in on-line
transformers. IEEE Transactions on Power Apparatus and Systems
1981;PAS-100(1):57.
[10] Eleftherion PM. Partial discharge XXI: Acoustic emission-based PD
5. Conclusions source location in transformers. IEEE Electrical Insulation Magazine
1995;11(6):22.
[11] Dakin J, Culshaw B. Optical ber sensors: principles and
In conclusion, this paper has presented the develo- components. Norwood (MA): Artech House, Inc., 1988.
pment and testing of ber optic acoustic sensor for inside- [12] Bucaro JA, Dardy HD, Carome E. Fiber optic hydrophone. J Acoust
transformer partial discharge detection. The test results Soc Am 1977;62:1302.
clearly demonstrated the feasibility of the developed ber [13] Dandridge A. Acoustic Sensor Development at NRL. Acoustic
Society of America Annual Meeting, Miami, Florida, November
sensor for the detection of partial discharges inside electri-
1987.
cal power transformers. Compared with the conventional [14] Dandridge A. Kersey AD. Overview of Mach–Zehnder sensor
acoustic sensors, the ber optic sensor has the advantages technology and applications. SPIE Proceedings 985, Fiber optic and
of non-electrically conducting, high frequency response laser sensor VI, Boston, September 1988.
(220 kHz), immunity to the electro-magnetic interference [15] Gunther MF, Wang A, Fogg BR, Murphy KA, Claus RO. Fiber
(EMI), chemical inertness, and small size, and the capabil- optic impact detection and location system embedded in a composite
material. Proceedings of the SPIE OE=FIBERS’92, Fiber Optic
ity of multiplexing more than one sensor in a single ber, Smart Structures and Skins V, Boston, vol. 1798, September 1992.
enabling on-line monitor of many power transformers at p. 262–9.
low cost. [16] Greene JA, Tran TA, Bhatia V, Gunther MF, Wang A, Wang KA,
Claus RO. Optical ber sensing technique for impact detection and
location in composites and metal specimens. J Smart Mater Struct
1995;4:93–9.
Acknowledgements [17] Furstenau N, Schmidt M, Horack H, Goetze W, Schmidt W.
Acoustic sensor system for airport ground traCc monitoring. IEE
Proceedings of the Optoelectronics 1997;144(3):134.
The authors would like to thank John W. Harley at J.W. [18] Claus RO, Gunther MF, Wang A, Murphy KA. Extrinsic Fabry–
Harley Inc, Twinsburg, OH, for the useful discussions and Perot sensor for strain and crack opening displacement measurement

kindly oJering the access to the partial discharge studying from −200 to 900 C. J Smart Mater Struct 1992;1(3):30.
J. Deng et al. / Optics & Laser Technology 33 (2001) 305–311 311

[19] Wang A, Ridraraju S, Murphy KA, Claus RO. Detection of [21] Wang J, Zhao W, Xiao H, Wang A. Self-calibrated
acoustic emission location using optical ber sensors. SPIE, Smart interferometric=intensity-based optical ber sensors. SPIE
Materials and Structures 1994: Smart Sensing, Processing, and Conference on Sensors and Controls for Advanced Manufacturing,
Instrumentation, February 1994. Pittsburgh, PA, 14 –15 October 1997.
[20] Wang A, Miller MS, Plante AJ, Gunther MF, Murphy KA, [22] Mario Di Giovanni, Flat and corrugated diaphragm design
Claus RO. Split-spectrum intensity-based optical ber sensors for handbook. New York: Marcel Dekker, 1982.
measurement of microdisplacement, strain and pressure. Appl Opt
1996;35(15):2595–601.

You might also like