th

The 19 International Symposium on High Voltage Engineering, Pilsen, Czech Republic, August, 23 – 28, 2015

ACOUSTIC LOCATION OF TRANSFORMER PD

UNDER NON-IDEAL CONDITIONS
Detlev W. Gross
Power Diagnostix Systems GmbH
52074 Aachen, Vaalser Str. 250, Germany
Email: gross@pdix.com

Abstract: Partial discharge (PD) activity in large power transformers being detected dur-
ing acceptance testing, or with on-site testing triggered by alarming DGA results, must be
located as precise as possible to start corrective actions.

Here, acoustic location of the activity is a valuable tool and sources offering a direct oil
path to the accessible tank wall are usually an easy find at high precision. However, often
the location is deeper buried in the winding and, hence, acoustic transmission is ham-
pered by attenuation, reflection, dispersion, and transmission following oil channels,
spacer gaps, and other paths. Thus, "textbook solutions", such as solving the sphere
functions produce misleading results and understanding the limitations as well as the re-
maining possibilities of acoustic location is essential.

Additionally, adequate handling of acoustic noise sources helps to improve the sensitivity.
Moreover, the use of an electrical trigger signal in combination with advanced trigger logic
helps to improve detection, separation, and location.

1 INTRODUCTION 2 PARTIAL DISCHARGE PATTERN

Attempts to acoustically locate problems in liquid
filled power transformers are as old as such trans-
formers. Severe problems produce sound that
make them already audible from distance.
Figure 1: "Original Defekt-Sucher" (1930, approx.)
Already in the early days of high voltage insulation
systems, partial discharge activity – then called
corona – was understood as a prominent indicator
for deterioration of insulation materials and incipi-
ent failure [3]. In the early 1990s partial discharge
detectors with pulse counting functions offering the
recording of amplitude-phase-height distributions
(-q-n-pattern) became commercially available [4,
5]. Further understanding of the gas discharge
physics controlling the partial discharge process
was gained with help of this technique [6,7]. Espe-
cially distinguishing and visually separating inde-
pendent discharge activities that occur at the same
time in the specimen became possible. Figure 2,
for instance, shows the superimposal of at least six
spherical gas inclusions in polymeric material.
Here, each activity "writes" its own trace depending
on the cavity's size and location.

Already in the 1930s specialized, stethoscope-like

instruments were available to mechanically guide
and amplify sound signals (Figure 1). Already here,
versions with an isolating rod were offered to in-
vestigate items under (moderate) high voltage. The
piezoelectric effect was already found 1880 by
Jacques and Pierre Curie [1]. Later Marie Curie
used it for her research. Around 1916, the principle
caught the interest of researchers for the detection
of submarines. However, its widely industrial appli-
cation started only in the 1950s. Later, especially
semi-conductor based charge amplifiers supported
the application of this technique as well on liquid
filled power transformer [2]. Figure 2: -q-n-pattern of several voids
Thus, with the limitations of the pulse processing
capabilities due to the IEC60270 [8] frequency
range, partial discharge activity of different sources
can be acquired and visually separated. Figure 3
shows an example of such superimposed inde-
pendent activities, which can be clearly separated.
Figure 3: Several independent activities
2.1 TRANSFORMER ACCEPTANCE TESTING
According to current standards, such as IEC60076-
3 [9] or IEEE C57.113 [10], partial discharge activi-
ties are acquired in parallel on up to nine channels
for HV, LV, and if present, tertiary. Hence, the
cross coupling ratio between the different bushing
taps, i.e., the different windings is already known
during calibration of the multi-channel PD detector.
Figure 4 shows a screenshot of such cross-
calibration matrix as taken during calibration.
Figure 4: Calibration matrix, cross coupling
This calibration cross coupling, of course, reflects
the properties of the winding, the differences be-
tween Y and delta, differences between disc and
layer winding and the overall configuration. Thus,
with inception of PD, comparison of the calibration
cross coupling versus the partial discharge ap-
pearance on the different taps is an indication of
the location. Further information is gained, if com-
paring the time domain signals of the different taps.
Here, signals having a cross talk with an identical
polarity indicate mere coupling, whereas signals
with an opposite polarity show that this part of the
winding forms the opposite electrode of the partial
discharge site and, hence, the discharge activity is
between the two electrodes. Figure 5 shows such
example. Here, the upper trace shows the domi-
nant signal on X1 terminal, while the 2nd trace with
the swapped signal belongs to X3 (tertiary below).
Figure 5: Vs. time signal common 1&3, tertiary 1&2
2.2 ON-SITE PARTIAL DISCHARGE TESTING
AND MONITORING
Generally, on-site measurement conditions are not
as nicely controllable as in a lab environment.
Therefore, the frequency band used for partial dis-
charge detection often cannot meet the require-
ments of the IEC60270 [8] and the relevant trans-
former standards [9,10].
3 ACOUSTIC SIGNALS IN POWER
TRANSFORMERS
Within a power transformer acoustic signals of
various sources are present. Besides the often
quite faint signals of partial discharge activity,
partly strong and dominant signals hamper sensi-
tive acquisition.
The Barkhausen noise [11] is one of the most
prominent sources of such acoustic signals being
present in an energized transformer. This noise
originates from discontinuous changes of the orien-
tation of clusters in ferromagnetic materials during
magnetization and demagnetization. As such
"clicks" of the Barkhausen effect in common ferro-
magnetic materials used in transformer design
have a rise time of a couple of µs, the resulting
signals reach up into several tens of kHz in fre-
quency domain [12]. Of course, Barkhausen noise
does not only originate in core material, but in any
ferromagnetic material that is being magnetized.
Here, as the piezoelectric transducers are usually
placed on the tank wall, stray flux in wall material
and wall shunts is of importance. However, in case
of transformer acceptance testing in the lab, this
effect is less significant, as such tests are typically
made at higher power frequencies, and, hence, the
flux and the resulting Barkhausen noise is propor-
tionally lower.
Additionally, there are other mechanical sources,
such as pumps, tap changer, vibration, and ther-
mally induced clicking sounds. Further, atmos-
pheric sources (rain, ice, and wind) hamper field
tests [13].
3.1 Acoustic Emission of PD Sources
The electrical signal of partial discharge under Ni-
trogen atmosphere has a rise time of about 1 ns
and, hence, posing a literally instantaneous excita-
tion for the acoustic emission under transformer oil,
which is found in a frequency range up to about
300kHz. Thus, at its origin the acoustic signal
shows an even amplitude spectrum and a sharp
pressure increase at the wave front. However, this
ideal situation persists only for a clear oil path with
homogenous properties.
3.2 Properties of Transformer Materials
Transformer oil offers a good signal transmission
of the acoustic pressure wave. However, the
transmission speed is temperature dependent. At
20°C the velocity of propagation is 1415 ms-1 [14],
while it is typically 1200 ms-1 at 80°C. Solid materi-
als, such as steel, wood, pressboard, and fiber-
glass do offer besides the compression wave also
the shear wave, whereas the velocity of propaga-
tion differs for both modes. Moreover, some struc-
tured materials, such as pressboard show different
transmission properties depending on the orienta-
tion of the fibers.
Especially, the speed of the compression wave in
steel, as used for transformer tanks is substantially
higher than the acoustic transmission in oil. Here,
the compression wave (longitudinal wave) travels
at a speed of about 5900 ms-1, which is more than
four times faster than the transmission in oil. The
shear wave (transverse wave) in steel propagates
at about 3200 ms-1 [14, 15].
3.3 Triangulation
Triangulation is based on the differences in arrival
time of signals at different receivers or sensors.
Generally, the location is calculated with solving
the set of equations of the sphere functions (1).
(x-x1)2 + (y-y1)2 + (z-z1)2 = (v • t1)2
(x-x2)2 + (y-y2)2 + (z-z2)2 = (v • t2)2 (1)
2 2 2 2
(x-x3) + (y-y3) + (z-z3) = (v • t3)
This basic, "classical" triangulation requires two
central conditions fulfilled for a tank-type specimen:
Firstly, the interior of the tank must be homogene-
ously filled with a liquid, i.e., otherwise empty and,
secondly, the tank wall must be a thin membrane
without any lateral transmission capabilities. In
case, of course, a location is an easy task.
3.4 Non-Ideal Conditions
Sadly, there is no real-life transformer fulfilling the
above-mentioned conditions and, hence, various
side effects of the non-ideal, real-life transformer
must be taken into consideration, when attempting
an acoustic location:
3.4.1 Tank Wall
As the tank wall has a certain thickness, lateral
transmission has to be considered. Thus, for a
given sound source location, there is a corner an-
gle within which direct oil path signal arrives the
earliest. This angle is a simple function of the ve-
locity of the compression waves in oil vs. steel (2).
 = arctan voil (2)
vsteel
Thus, the fastest signal path is partly in steel for
any incoming signal under an angle of more than
14°, given the aforementioned ratio of about 0.25
between velocities in oil vs. steel. The conse-
quence is that two neighboring sensors having
both a viewing angle of more than 14° cannot pro-
duce a valid location result based on the first arri-
val. In theory, the incoming oil-path signal creates
a distinct increase in magnitude over the leading
tank wall signal, if the signal path deviates more
than 14° from the straight view. The two lower
traces of figures 13 show the signal of such case
with the smaller leading steel signal. However, in
real-life situations, the signal paths often are more
complex and such nice signals are not so common.
3.4.2 Wave Guide
If, for instance, the partial discharge source is lo-
cated at the far tip of a stick, the sound travels in-
side the stick approximately 40% faster than in the
surrounding oil. At the stick's near end, this sound
is then radiated and leads to a false location point-
ing to this near tip, but indicating a closer location.
3.4.3 Acoustic Window
Acoustic emission of hidden partial discharge, for
instance on the core surface, behind a stack of
insulation material on top of a LV winding, a buried
glue drop with an air bubble, or an otherwise hid-
den source may take an angled path to the sensor.
For instance, the sound may emit from an oil
channel or between sticks. In case, the triangula-
tion falsely points to the oil channel emitting the
signal, but indicates the length of the overall path.
Figure 6 (next page) shows the situation of such a
case, where it is essential having drawings at hand
and where experience helps. Here, the location of
the window was precisely known and the remain-
ing path had to lead up the core. Thus, even here
the location can be estimated relatively accurate.

1
2
3 tions. Here, the usual situation is that a transformer
Figure 6: Indirect signal path and false location
3.4.4 Reflection
Of course, acoustic waves undergo reflection
within the structure of a transformer and its tank.
However, in principle reflections do arrive after the
first, direct signal and, hence, are not likely to
hamper a correct triangulation. On the other hand,
identification of reflected signal could help to vali-
date the interpretation of acquired signal. Figure 7
shows the signal of partial discharge at the center
conductor of a tube-like high voltage link to a cable
box. Here, it was possible to calculate the few cm
offset from the center, as the reflected signal trav-
els the distance to the far wall twice.
Figure 7: Direct and reflected acoustic signal
4 TRIGGER METHODS
Two main methods exist to trigger acoustic triangu-
lation. Both have their advantages and disadvan-
tages under different applications.
4.1 All Acoustic
In general, a location is possible based on just
three acoustic signals, when solving the equations
of the sphere functions. However, as mentioned
above, various signals of other sources than partial
discharge are present in a transformer especially in
the field. Here, the placement of sensors is a key
point and it must be avoided to combine signals to
fake location that do not belong together.
However, under field conditions it can be advanta-
geous that besides placing sensors, no further in-
stallation or interference is needed on site [13].
4.2 Standard Electrical Trigger
Using electrical trigger for the acoustic triangulation
is the predominant method for test room applica-
failed to complete the acceptance test and, hence,
focus is on finding the root cause of the partial dis-
charge. Therefore, eliminating any acoustic signal
that is not caused by partial discharge is appreci-
ated. Further, using averaging techniques based
on the precise electrical trigger helps to improve
the signal-to-noise ratio and to reduce the impact
of the signals caused by the Barkhausen effect.
Typically, an analog output of a partial discharge
detector is taken as trigger. Preferably, wide band
detectors are used, as their internal signal process-
ing causes only a delay of few microseconds,
which can be neglected against the acoustic travel
time [13].
In a field situation it may be more convenient using
a split-core RF current transformer for instance on
a neutral or ground lead as trigger source. Like-
wise, UHF signals captured with drain valve sen-
sors can act as trigger source to start the acquisi-
tion. Generally, of course, the electrical signal is
analyzed beforehand to make sure that it reflects
the partial discharge signal in question [16].
Figure 8 shows a typical trace of such a measure-
ment with el. trigger and three sensors in a row.
Figure 9: Averaged acoustic signal with el. trigger
Figure 10: Location result based on figure 9
Figure 10 (previous page) shows the correspond-
ing location result for the "flat problem", as the
three sensors are mounted in a horizontal (or verti-
cal) line. The main point of the third sensor is to
validate the results of the two other sensors. The
distance calculation can be set to reflect the travel
path in the tank wall. In case, the circle changes
into a line for an angle above the corner angle.

In an iterative process, the sensors are replaced to

give a clear signal within the corner angle. With the
then found location result for the said flat problem,
the sensors are swapped from horizontal to vertical
(or vice versa) to solve the second flat problem.
Here, the software always tilts the previous result
layer using the newfound location. With a clear oil
path, this leads to a precision of about 2 cm [16].
4.3 -q-based Electrical Trigger
Figure 12: Independent -q-trigger areas
As this function is implemented in the (hardware-)
state machine routine to accumulate the -q-n pat-
tern of the (electrical PD) trigger channel, the over-
all performance of the instrument does not suffer
from the additional task.

However, the "standard" electrical trigger runs into
problems in case of high magnitude noise, high
repetition rate noise signals, and several parallel
activities competing during averaging. Here, a trig-
ger based on a -q-n-pattern greatly improves the
signal acquisition under difficult situations.
This function is now part of a multi-channel parallel
partial discharge detector (Figure 11). The current
version of this instrument offers a built-in 100
Msample DSO for each channel of the up to nine
parallel acquisition channels.
This allows now firstly to get valid averaging results
for multiple sources even with activities having ex-
tremely different repetition rates and with signals
buried in high-level or scattered impulse noise.
To minimize confusion, all traces (acoustic chan-
nels) belonging to one trigger area are displayed in
the same color as the marked trigger area in the -
q-n-pattern. Currently, this function is limited to
three independent trigger areas, which are shown
in red, blue and green. Figure 13 shows an exam-
ple with two sets of separately averaged traces
shown in blue and red, while the upper black trace
indicates the trigger signal and the t0 reference.

Figure 11: 8-ch. parallel PD and RIV acquisition

Besides the standard mere level-based trigger
function, as found usually with oscilloscopes, here,
the circuits that accumulate the -q-n pattern can
provide a smart trigger to have only signals con-
tributing to the averaging process that are likely to
come from the same source.

Practically, the trigger is set using a -q-editor to

mark areas to provide a trigger within a -q-n-
pattern acquired on the unit under test. Figure 12
shows an example of such partial discharge pat-
tern and the areas to provide a trigger with blue
and red marking, while all other pulses are ne-
glected. Technically, the instrument can handle up
to eight independent trigger condition areas. Sub-
sequently, the instrument acquires the traces of the
up to eight acoustic channels into independent av-
eraging "baskets" for each trigger class, which then
rules out remaining "wrong" traces.
Figure 13: Displaying results of two trigger classes
The blue signal corresponds to the blue trigger
area shown in figure 12. Likewise, the red traces
belong to the red trigger area as marked in figure
12. The blue activity is somewhat left of the three
sensors (~58cm), while the stronger red activity is
located at a closer distance (~40cm), but to the
right hand side and slightly beyond the corner an-
gel with respect to the current sensor position.
Of course, displaying the different trigger classes
can be selected or disabled, as simultaneous dis-
play of activities at a comparable distance can pro-
duce quite confusing results. For instance, figure
14 shows the result of the partly hidden blue trace
of figure 13.
Figure 14: Results of trigger class 2 of figure 13
Instead of showing further trigger classes, the in-
strument can also be set to display the triggered
signals of one class and their complement, i.e.,
every signal that surpasses a basic trigger thresh-
old and does not fall into the marked area. This
helps keeping an eye on less dominant activities
that may be in the view of the sensors, but other-
wise will not make it up in the averaging statistics.
Besides acquiring and averaging signals of acous-
tic sources, this function can also be used to ana-
lyze the time domain properties of electrical partial
discharge signal as described above in paragraph
2.1. Here, likewise, the time domain properties can
be assigned to individual activities selected and
marked in the -q-n-pattern. This gives especially
further insight for deeply buried sources, where no
acoustic signal can be detected.
5 CONCLUSION
Generally, there was always a need to locate par-
tial discharge activity in oil-immersed power trans-
formers. Acoustic location based on triangulation
using piezo-electric sensors and electrical trigger
provides good results under conditions offering a
direct oil path.
A novel -q-trigger logic based on -q-n-pattern
acquisition greatly improves detectability under
non-ideal conditions and under the presence of
high-level noise, or with partial discharge activities
having strongly different repetition rates.
Besides supporting acoustic triangulation, this
method also allows analyzing the time-domain
properties of electrical PD signals under the pres-
ence of multiple partial discharge sources and im-
proves separation and location of sources offering
no possibilities of acoustic location.
REFERENCES
[1] Jacques and Pierre Curie "Développement
par compression de l’électricité polaire dans
les cristaux hémièdres à faces inclinées" (De-
velopment, via compression, of electric polari-
zation in hemihedral crystals with inclined
faces), Bulletin de la Société minérologique
de France, vol. 3, 1880, pages 90-93
[2] Howells, E. and Norton, E. T., "Detection of
partial discharges in transformers using
acoustic emission techniques", IEEE Trans.
Power Appar. Syst. 97, 1978, 1538–1549
[3] Peek, F.W., Dielectric Phenomena in High
Voltage Engineering. McGraw-Hill Book Com-
pany, 1915, pp. 216-217.
[4] Fruth, B., Gross, D. "Partial Discharge Signal
Generation Transmission and Acquisition," in
IEE Proc.-Sci. Meas. Technol., Vol. 142, No.
1, January 1995, pp. 22-28.
[5] Fruth, B., Gross, D. "Phase resolving partial
discharge pattern acquisition and spectrum
analysis", Proc. of the ICPDAM, July 1994,
Brisbane NSW, Australia, pp. 578-581.
[6] Niemeyer, L. "A generalized approach to par-
tial discharge modeling", IEEE Trans DEI, Vol.
2, No. 4, August 1995, pp. 519-528.
[7] Heitz, C. "A generalized model for partial dis-
charge processes based on a stochastic proc-
ess approach", J. Phys. D, Sept. 1999, pp.
1012-1023.
[8] IEC60270, High-voltage test techniques - Par-
tial discharge measurements, CEI/IEC
60270:2000.
[9] IEC60076-3 Ed. 3, Power transformers - Part
3: Insulation levels, dielectric tests and exter-
nal clearances in air, CEI/IEC 60076:2013.
[10] IEEE C57.113-2010 "IEEE Recommended
Practice for Partial Discharge Measurement in
Liquid-Filled Power Transformers and Shunt
Reactors".
[11] Barkhausen, H. “Zwei mit Hilfe der neuen Ver-
stärker entdeckte Erscheinungen.” Physische
Zeitschrift, 1919, year 20, no. 17., 401–403
[12] Clatterbuck, D.M., et al. "An extended model
of the Barkhausen effect based on the ABBM
model" J. of Applied Physics 87 (2000), 4771
[13] IEEE C57.127-2007 "IEEE Guide for the De-
tection and Location of Acoustic Emission
from Partial Discharges in Oil-Immersed
Power Transformers and Reactors"
[14] L.E. Lundgaard, “Partial discharge XIII.
Acoustic partial discharge detection -
fundamental considerations”, IEEE El. Insul.
Mag., Vol. 8, Issue 4, July-Aug. '92, pp. 25-31.
[15] L.E. Lundgaard, “Partial discharge XIV.
Acoustic partial discharge detection-practical
application”, IEEE El. Insul. Magazine, Vol. 8,
Issue 5, Sept.-Oct. 1992, pp. 34-43.
[16] Gross, D., Soeller, M. "On-site transformer
partial discharge diagnosis", 2012 IEEE Inter-
national Symposium on Electrical Insulation",
San Juan, Puerto Rico, June 10-13, 2012.