You are on page 1of 9

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO.

1, JANUARY 2004 221

Vibro-Acoustic Techniques to Diagnose


Power Transformers
Cipriano Bartoletti, Maurizio Desiderio, Danilo Di Carlo, Giuseppe Fazio, Francesco Muzi, Member, IEEE,
Giancarlo Sacerdoti, and Fabio Salvatori

Abstract—This paper deals with new procedures based on • effects due to magnetic shunts and related fasteners lo-
vibro-acoustic techniques to diagnose power medium-voltage/low- cated in the stray magnetic flux.
voltage (MV/LV) transformers. The normal operating machines The acoustic emission depends on the machine anomalies and
can be monitored and checked from a remote, dedicated diagnostic
center where information is received directly from the field. The age. For this reason, the acquired vibrations must be identified
diagnostic method adopted and the experimental test results are and related to the state change of the monitored transformer by
reported. Tests have been performed either in the laboratory or means of comparisons either with a similar new transformer or
directly in the field on transformers in normal operating con- the averaged values of a set of transformers of similar type and
ditions. The laboratory tests related the transformer vibrations age.
to internal anomalies. The suggested monitoring and diagnostic
system allows one to usefully schedule preventative maintenance, The change is linked either to the amplitude of one or more
reduce costs, and improve the quality of power distribution. harmonics, or to the generation of new harmonics.
Index Terms—Acoustic recognition techniques, power distribu-
tion systems, power transformer diagnostic. II. PROCESSING AND RECOGNITION METHOD
The proposed recognition acoustic method is based on the
I. INTRODUCTION following four operations: acquisition, segmentation, parame-
terization, and recognition. The signal segmentation consists of

T HE large number of transformers (some thousand)


installed in a complex power distribution system increas-
ingly requires accurate diagnostics, which can be conveniently
dividing the acquired data into substrings identified by means
of indexes defining location and number of the acquisition. The
subsequent parameterization enables one to identify and esti-
performed in a monitoring center where the information on mate the significant parameters.
the actual state of the machines is received, processed, and Signal processing is performed by means of specific soft-
interpreted. The proposed diagnostic method is based on ware implemented in a LabView environment. The implemented
vibro-acoustic procedures derived from speech-recognition STUD program is able to analyze signals that are coming si-
techniques [1], [2], [5], [6]. The acoustic vibrations of a trans- multaneously from different channels. First, preprocessing op-
former can be generated by the following main phenomena [3], erations, such as software filtering, rectification, etc., are per-
[4]: formed. Afterwards, the signal is processed by means of the fol-
• coil vibrations depending on the current amplitude and lowing procedures:
winding clamping compression; • fast Fourier transform (FFT) and discrete Fourier trans-
• core vibration depending on magnetostriction, the form (DFT) [usually, the DFT is preferred since the
Barkhausen effect, and loosening of core clamping; number of samples often does not match the power of two
• partial discharges either in insulating bushings or winding and computation time is not so important];
insulation; • cross correlation;
• movements of micro-metal chips; • maximum and minimum evaluation.
In addition, other operations can be performed on the stored
vectors such as: extraction of defined vector components (e.g.,
Manuscript received December 6, 2002. This work was supported by ACEA the amplitudes of specific harmonics after the DFT), averages,
SpA, the Rome Water and Electric Utility, that also supplied the transformers to
be tested and other important resources during the experimental tests performed.
ratios between two components, etc. The obtained significant
C. Bartoletti and G. Sacerdoti are with the Electrical Engineering Department, data can be stored in text format, spreadsheet, binary (16 or 32
University of Rome “La Sapienza,” Rome 00184, Italy (e-mail: sacer@elet- b) and in real form (single or double precision). On request, the
trica.ing.uniroma1.it). program can automatically process a large number of signals to
F. Salvatori is with ABBOTT-Italia, Campoverde 04010 (Latina), Italy
(e-mail: fabio.salvatori@abbott.com). extract a desired quantity of interest (for instance the amplitude
M. Desiderio is with the ACEA SpA, Rome Water and Electric Utility, Rome of the 100-Hz harmonic of each signal). Other useful utilities of
00154, Italy (e-mail: m.desiderio@aceaspa.it). the program are:
D. Di Carlo and G. Fazio are with the Electronic Engineering Depart-
ment, University of Rome “Tor Vergata,” Rome 00133, Italy (e-mail: • signal decimation (i.e., reduction in the number of samples
fazio@ing.uniroma2.it). to decrease the necessary computer capabilities);
F. Muzi is with the Electrical Engineering Department, University of
L’Aquila, L’Aquila 67100, Italy (e-mail: muzi@ing.univaq.it). • the transformation of signals into “wave” format to make
Digital Object Identifier 10.1109/TPWRD.2003.820177 the signals audible by an operator; if the frequencies
0885-8977/04$20.00 © 2004 IEEE
222 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004

are out of the audible range, the program operates a


frequency shift bringing the signal frequencies within the
20–20 000-Hz audible range.

III. DETECTING SYSTEM

In regards to acoustic vibrations, medium-voltage/low-


voltage (MV/LV) transformers are less well known than large
HV transformers. For this reason, the PC-based measurement
apparatus developed is very versatile and adjustable in the
setting of the signal acquisition features.

A. Sensors
In a transformer, the frequency band of acoustic emissions is
very large; nevertheless, the acquired experience demonstrated Fig. 1. Sensors positioning (white arrows) on the transformer under
that a reduced frequency range is sufficient to perform a good investigation during a laboratory test.
diagnostic. The frequencies of the chosen sensors were in the
50-Hz—100-kHz range. To detect high frequencies, acoustic
C. Trigger
probes (PAC model R6I—frequency band 20–100 kHz) were
used while for low frequency, two kinds of sensors were The acoustic-source identification is simplified by a phase
adopted: the piezoelectric audio pick-up (Shadow SH-2001) comparison between the transformer supply voltage and signals
and the accelerometer (Endevco 751–10). The R6I probe coming from different channels. For this reason, a reference
is based on piezoelectric resonance obtaining a very high trigger circuit is necessary since the acquisition card requires
sensibility, of course to the detriment of the uniformity of the a signal commutation in the 0–5-V range with a slope of at
frequency response. The preamplifier is characterized by a low least 1 V s. In a medium-voltage/low-voltage (MV/LV)
impedance to allow also for long connecting cables. transforming station, the single-phase voltage is acquired in
The frequency response of the vibrating accelerometer En- the LV side by means of a measuring transformer having the
devco 751–10 is within the 1–15 000-Hz range. In this case, the ratio 230/22 V. With reference to laboratory tests, the acquired
output voltage is a linear function of acceleration. The frequency voltage signal depends on the kind of test performed: an open
response corresponds to a low-pass filter. circuit test, load test, or a short-circuit test.
The pick-up SH-2001 is a commercial transducer able to
detect signals up to 15 kHz. Electromagnetic noise coupled IV. LABORATORY AND FIELD EXPERIMENTAL TESTS
with circuitry may cause disturbances since the transducer
output is not preamplified. The transducer is recommended For the typical frequencies of the phenomena under study, the
for permanent installations. With reference to low frequencies sampling frequency of 500 ksamples/s was used in the measure-
5 kHz , the pick-up SH-2001 response is much better than ments. Due to the signal periodicity, the acquisition is limited to
the accelerometer response. time intervals of 100 ms (i.e., ten periods of the 100-Hz funda-
mental frequency).
With the aim to have a classification of the acoustic frequen-
B. Conditioning and Acquisition Card cies, open circuit and short circuit tests were performed in the
The signals coming from R6I and SH-2001 sensors are am- laboratory. In each of these two kinds of test, only one subset of
plified up to 40 dB with the aim to produce a signal compatible the sources is activated. If the transformer supplies a load, the
with the acquisition card . The amplifier is completed by emitted vibrations are a combination of those emitted during
an anti-aliasing low-pass filter. To guarantee the simultaneous the open circuit test and those caused by the load current. The
acquisition of at least three channels with band higher than 100 tests performed in the laboratory involved both new and used
kHz, the PCI-DAS4020/12 acquisition card was adopted. Its transformers. Fig. 1 shows the detecting system used, requiring
main characteristics are the following: three sensors placed on the transformer tank in correspondence
of each leg.
• resolution 12 b;
Transformers tested in the field were usually supplying
• simultaneous sampling on one, two, or four channels;
a small load (less than 25%); in which case, the generated
• maximum sampling frequency 20 MHz;
emissions are similar to those of an unloaded transformer.
• input ranges and ;
• analog and digital trigger;
• adjustable input impedance (50 or 1.5 M ); A. Laboratory Tests
• autocalibrating; Laboratory tests were performed to characterize signals by
• Fifo of 32 kwords; means of time domain recordings and frequency domain rep-
• PCI interface. resentations obtained after filtering and signal processing. All
BARTOLETTI et al.: VIBRO-ACOUSTIC TECHNIQUES TO DIAGNOSE POWER TRANSFORMERS 223

Fig. 2. Open circuit laboratory tests; lf = low frequency, hf = high frequency; (set a, a ; set b, b ; set c, c = new; used; anomalous transformer).
tests were performed using a supply voltage with a frequency In the plots, is a relative quantity that indicates the signal am-
of 50 Hz. plitude level; the measurement scale is conveniently established
Figs. 2 and 3 show a number of typical recorded acoustic sig- by the operator optimizing the detecting system gain (note that
nals respectively concerning open circuit and short circuit tests. the proposed diagnostic method does not require absolute values
224 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004

Fig. 3. Short circuit laboratory tests (a full-load current in both the primary and secondary circuits); lf , hf , (a, a ), (b, b ) (c, c ), as per Fig. 2.

since it operates on signal shape and on comparisons between depend on the current that is responsible for important coil vi-
comparable processed signals). brations; as a matter of fact, electrodynamic forces depend on
1) New Transformers: For low frequencies (up to 15 kHz) the square of the current. During the open circuit test, the main
and short circuit conditions, the signal shape is sinusoidal at a phenomenon responsible of acoustic emissions was the magne-
frequency of 100 Hz [Fig. 3(a)]. The 100-Hz periodicity is jus- tostriction producing core plate vibrations. Also in this case, the
tified because in short circuit conditions, the main phenomena signal periodicity corresponded to 100 Hz but complex interfer-
BARTOLETTI et al.: VIBRO-ACOUSTIC TECHNIQUES TO DIAGNOSE POWER TRANSFORMERS 225

Fig. 4. DFT of signals concerning used transformers (the meaning of E is similar to " in Figs. 2 and 3).

Fig. 5. Anomalous transformers; (a) comparison between the DFT results of the noise, open circuit signal, and short circuit signal; (b) fitting of a recorded
high-frequency signal.

ence phenomena between different phases appeared. Since mag- For high frequencies, an increase in the ratio signal/noise is
netostriction is a nonlinear phenomenon, the signal is much dif- usually observed. This is more evident during the short circuit
ferent from a pure sinusoid. For high frequency (20–100 kHz), test whereas it is only perceptible during the open circuit test,
the signal is only just higher than the noise in both open circuit as shown in Fig. 5(a).
and short circuit tests [Figs. 2(a), (b) and 3(a)]. To perform a correct diagnostic, a very interesting aspect to
2) Used Transformers: For low frequencies, old trans- be considered is the presence of a 100- (200- or 300-) Hz mod-
formers show an increase of the harmonics with frequencies ulation obtained by the high-frequency signal fitting. The signal
multiple of the fundamental. An example of this is shown amplitude envelope is clearly shown in Fig. 5(b).
in Fig. 4(a). This phenomenon, which can be assumed to Since the same characteristic appears also during the open
define machine aging, is more evident when the transformer is circuit test, the incorrectly working transformer can be detected
unloaded, as shown in Fig. 4(b). On the basis of this criterion, also when the machine supplies a small load.
the machine aging level can be evaluated also in the field when
the transformer supplies a small load. B. Field Tests
3) Anomalous Transformers: Anomalous transformers are Despite the different operating conditions, the considerations
machines that have defects such as loose winding, excessive developed for the laboratory tests can be usefully extended to the
core vibrations, partial discharges, etc. At the moment, the de- behavior of transformers that are working directly in the field.
fect kind is only approximately identified. Comparisons between field and laboratory tests demonstrated
226 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004

Fig. 6. Example of THD values evaluated for different machines.

that transformers in the field supplying a small load give results , , , and . The meanings
similar to those of a transformer during laboratory open circuit of the symbols are defined in the following.
tests.
For old transformers, it is rare to obtain signals of a shape VI. LOW-FREQUENCY PARAMETERS
as simple as in a new machine. As a matter of fact, in old trans-
A very significant parameter is the weighted total harmonic
formers, an increase in the amplitude of high frequencies can be
distortion as defined by the following formula:
noted, probably due to the presence of partial discharges and/or
a generalized increase of the transformer vibrations as a con-
sequence of mechanical weakness increasing. In this case, with (1)
the aim to evaluate the transformer aging, it is possible to make
considerations similar to those in laboratory open circuit tests.
Experimental tests demonstrated that different phenomena In this relation , is the harmonic amplitude and
cause different emissions, which, in turn, can be associated to the weights , . The THD values
different defects. Sources associated with one winding only supply useful information on the signal distortion, that, in turn,
are characterized by remarkable modulation at rated frequency, is linked to the transformer’s aging. The three parameters are re-
while sources associated with all three phases have a third spectively named , , and (where lf
harmonic modulation. means low frequency, and the numbers 1, 2, 3 refer to the trans-
former legs). Fig. 6 shows an example of values
evaluated for different machines (A, B, C, E, G, N, S, T, Y, Z
V. CLASSIFICATION OF TRANSFORMER STATES refer to different transformer trademarks). High THD values for
“old” transformers can be observed, as confirmed by the year of
The identification of significant parameters is of paramount production in the axis.
importance to classify the different states of a transformer. The
laboratory tests performed were very useful to define reference A. High-Frequency Parameters
models based on significant parameters to be used when com-
Typical high-frequency acoustic emissions are mainly due to
paring processed signals acquired from a transformer in service.
the Barkhausen effect and partial discharges in insulation. These
The different, preliminarily classified situations were named
signals are usually acquired using sensors for ultrasounds.
Classes. At the moment, three belonging classes are defined
1) Parameter: The sum of the first four harmonics of
(two for normal conditions and one for anomalous conditions):
the spectrum obtained from the envelope of the high-frequency
• CLASS 1—new transformer (normal condition); signal, supplies useful information on anomalies concerning
• CLASS 2—used transformer (normal condition); high-frequency sources.
• CLASS 3—used, potentially defective transformer With reference to the three legs of the transformer, three sums
(anomalous condition). were computed and named , , and (where hf
On the basis of the experience from performed analyses, a stands for high frequency), respectively. The main anomalies,
transformer’s class was defined by four parameters that must which are of two kinds (related to one phase and all three phases,
be evaluated for each leg of the transformer. In this case, respectively), are shown in Fig. 7 where the experimental re-
the total parameters were 12 and they were named , sults from some transformers are reported. Fig. 7 clearly shows a
, , , , , , , higher value in the parameters on anomalous transformers.
BARTOLETTI et al.: VIBRO-ACOUSTIC TECHNIQUES TO DIAGNOSE POWER TRANSFORMERS 227

Fig. 7. Comparison of A parameters concerning anomalous and nonanomalous transformers.

Fig. 8. Data processing and recognition system.

2) Parameter: This parameter concerns to the total leg of the transformer and were named , , and
harmonic distortion computed with reference to the envelope of , respectively.
the high-frequency signals coming from each leg of the trans-
former.
They are named , , , and give VII. REMOTE MONITORING AND AUTOMATIC DIAGNOSTIC
important information on defects linked to high-frequency
sources. Once both the significant parameters and models defining the
3) Parameter: Also this parameter refers to the en- relevant conditions of the transformer state (pattern-samples)
velope of the high-frequency signals. It concerns the ratio be- are defined, any automatic recognition processes require the im-
tween the amplitudes of the 50- and 100-Hz components and plementation of a specific “vocabulary.”
gives additional information on the anomalies caused by high- The whole diagnostic process is shown in Fig. 8 and can be
frequency sources. explained as follows.
As a matter of fact, wherever these anomalies are present, • The vibro-acoustic signals are acquired by the acquisition
this ratio is very high. The same ratios were evaluated for each system (observer).
228 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004

• From the acquired signals, the significant and synthetic pa- ACKNOWLEDGMENT
rameters are subsequently extracted by means of the pro-
The authors are especially grateful to Dr. M. Cortesi, ACEA
cedure named “features extractor.”
SpA director, for his useful advice and constant encouragement.
• By means of a consultation of the predefined vocabulary,
the analysis process establishes the statistical “closeness”
of the case examined to a reference model in the vocabu- REFERENCES
lary and subsequently identifies the belonging class of the [1] F. Keinosuke, Introduction to Statistical Pattern Recognition. New
transformer under checking. York: Academic, 1994.
[2] C. Bartoletti et al., Metodologie Informatiche nei Processi di Misura,
The design of a remote monitoring and diagnostic center for Siderea et al., Eds. Rome, Italy, 1997.
transformers requires the knowledge of the following informa- [3] Y. Champoux, B. Gosselin, and J. Nicolas, “Application of the inten-
tion [7]: sity technique to the characterization of transformer noise,” IEEE Trans.
Power Delivery, vol. 3, pp. 1802–1808, Oct. 1988.
• daily transportation capacity of each connecting channel; [4] J. S. Mitchell, An Introduction to Machinery Analysis and Moni-
• number of installed transformers on the transforming sta- toring. Tulsa, OK: Penn Well Books, 1993.
[5] A. J. Viterbi, Principles of Spread Spectrum Multiple Access Communi-
tion; cations. Reading, MA: Addison-Wesley, 1995.
• number of the MV/LV transforming stations. [6] M. Alder. An Introduction to Pattern Recognition: Statistical, Neural net
and Syntactic Methods of Getting Robots To See and Hear. [Online].
The acquired signals can be previously stored in local de- Available: http//ciips.ee.uwa.edu.au
vices placed inside the transforming station. The signals to be [7] F. Muzi and F. Panone, “Optimal arrangement of spread automated
stored are usually detected by six sensors placed on the trans- centers for electrical distribution systems,” in Proc. IEEE-Int. Conf.
Harmon. Quality Power, Athens, Greece, Oct. 14–16, 1998.
former tank and come from 3–6 channels (three connecting the
transformer legs, and eventually three connecting the HV insu-
lating bushings). Sensor placement requires great attention since
a transformer is a complicated mechanical assembly varying
significantly from manufacturer to manufacturer and from one
design to another. Vibration spectra and especially signal mag- Cipriano Bartoletti was born in Rome, Italy.
nitude depend on sensor location. In 1966, he joined Olivetti SpA (Research & Development Department),
where he designed digital circuitry for desk-computers and the characterization
The recordings can be performed two to three times a day of DTL logic family to be introduced in Olivetti’s computer manufacturing.
when the load current is less than 1/10 (unload condition) or He also worked in designing new instrumentation, sensors, and measurement
greater than 1/2 (load condition) of the rated current. The record- systems. Many publications were produced from this kind of work.
ings concerning 100-kHz frequencies and 100–120 ms of acqui-
sition time require 100 000 samplings with a 10-b dynamic and
delayed transmission. This protocol can be managed by any kind
of data transmission system.
Maurizio Desiderio was born in Rome, Italy. He received the Ph.D. degree in
electrical engineering from the University of Rome “La Sapienza,” Rome, Italy,
VIII. CONCLUSION in 1992.
Currently, he is with ACEA SpA, Rome Water and Electric Utility, Rome,
Italy, where he deals with the maintenance and management of the local distri-
The acoustic sources of a transformer vary, but only a few of bution network. He also worked in the electronics and electromechanical indus-
them are significant enough to characterize the machine defects. tries (Italtel and OTIS companies).
In laboratory tests, some supply configurations activate only a
subset of sources facilitating their identification. Laboratory and
field tests were useful to describe typical vibro-acoustic signals
on new, used, and defective transformers. In this preliminary
experimental study, it was possible to classify transformers into
Danilo Di Carlo was born in Rome, Italy, in 1973. He received the Ph.D. degree
three reference classes: new, used, and anomalous transformers. in electronic engineering from the University of Rome “Tor Vergata,” Rome,
Processing techniques were identified to recognize the trans- Italy, in 2001 with a thesis on vibro-acoustic monitoring of power transformers.
former state and, therefore, isolate the various “anomalous” Currently, he is working on automatic measuring machines in a production
industry. He also has worked in vibro-acoustic monitoring applied to trains and
cases. The process requires a comparison between the esti- railways.
mated significant parameters of the examined transformer and
pattern-samples summarized in a predefined “vocabulary.” The
research study showed the opportunity to perform the diagnosis
in a remote monitoring center where all of the transformers
belonging to a distribution system are contemporarily checked
through daily tests (e.g., twice a day when the transformer is Giuseppe Fazio was born in Naples, Italy, in 1951. He received the Ph.D. degree
in electronic engineering from the University of Rome “Tor Vergata,” Rome,
unloaded and at least at 50% of the rated load). The acquired Italy, in 1979.
and processed information allowed to establish the actual Currently, he is a Professor of electrical measurements at the “Tor Vergata”
state and residual life expectation of each transformer. This University. He was also a Researcher at the Electronic Department of the Uni-
versity of Rome “La Sapienza.” His main research interests include micropro-
information can be usefully applied to make the best decision cessor applications, signal acquisition and processing, remote control systems,
in the management of the transformer population. and acoustic control techniques. He has also a number of patents in these fields.
BARTOLETTI et al.: VIBRO-ACOUSTIC TECHNIQUES TO DIAGNOSE POWER TRANSFORMERS 229

Francesco Muzi (M’90) was born in L’Aquila, Italy. He received the Ph.D. de- Fabio Salvatori was born in Rome, Italy. He received the Ph.D. degree in elec-
gree in electrical engineering (hons.) from the University of L’Aquila, L’Aquila, trical engineering from the University of Rome “La Sapienza,” Rome, Italy, in
Italy, in 1981. 2001.
Currently, he is an Associate Professor of electrical distribution systems in Currently, he is with ABBOTT, Campoverde, Italy, where he works on the
the Electrical Engineering Department at the University of L’Aquila. In 1984, maintenance and management of the gas turbine generating plant and the fac-
he was appointed Researcher in power systems in the same university. His main tory’s electrical system. He was with the Department of Electrical Engineering
research interests include power systems analysis, power systems reliability, of Rome “La Sapienza” and with the Department of Electronic Engineering of
electromagnetic analysis using the finite element method, and power quality Rome “Tor Vergata” University in the field of acoustic control techniques.
in distribution systems.

Giancarlo Sacerdoti was born in Bologna, Italy. He received the Ph.D. degree
in engineering from the University of Bologna, Italy, in 1952.
Currently, he is a Full Professor of electrical measurements at “La Sapienza”
University, Rome, Italy. He contributed to the development of Nb3Al supercon-
ductor cables from 1970 to 1975, electric energy meters based on oversampling
technique as sigma-delta modulator in 1966, electric-charged particle lenses for
particle beam transport channel from 1960 to 1970, and techniques to charac-
terize nonhomogeneous dielectrics for MV-cable diagnostics in 1999.

You might also like