21, rue d’Artois, F-75008 PARIS A2 - 205 CIGRE 2018

http : //www.cigre.org

Localization of Partial Discharge Sources in Transformers by Analysis of

Signals in Time- and Frequency-Domain

J. FUHR *, TH. ASCHWANDEN

AF Engineers + Consultants (AFEC) GmbH
Switzerland

SUMMARY

Partial discharge (PD) measurement is a non-destructive method to detect local defects (PD sources) in
insulation systems of high voltage apparatus. In complex insulation structures, such as those in large
power transformers, PD sources can be located deep inside the active part (core and windings), in
particular inside the winding blocks or in the barrier insulation system between windings. For this type
of PD sources, the value of the apparent charge measured at the terminals of the transformer, does not
sufficiently reflect the real risk of such internal defects. Moreover, the calibration procedure usually
performed at transformer bushings will not deliver meaningful results regarding the discharge intensity
of such PD defects due to serious integration errors in the PD measuring system.
PD sources located inside the transformer active part typically produce oscillating PD signals at the
measuring terminals (bushing tap). Under such conditions, the waveform of the original PD impulse is
attenuated and shaped on its way from the PD source to the measuring system. In general, oscillating
PD signals detected at the bushings of a transformer represent the responses of the active part of the
transformer to the excitation of a PD impulse generated at a specific (unknown) position inside the
transformer. Therefore, PD signals contain information about both, the physical process at the origin
of the PD source, and the attenuation and reflection phenomena along the signal propagation path.
Based on this background, the evaluation of PD signals in the time- and frequency-domain offers a
considerable potential to localize internal PD sources in large power transformers.
A three-step procedure for PD localization in power transformers based on the analysis of PD signals
is proposed in this contribution. A multi-terminal measuring set-up is used to detect internal PD
activity simultaneously at all bushings of a power transformer. PD signal waveforms arriving at
different terminals may differ significantly due to their specific signal transfer paths and due to
different signal coupling modes between windings or between terminals, respectively. The advantages
and limits of PD signal analysis in the time- and frequency-domain and their application to PD source
localization in power transformers are discussed using practical examples.

KEYWORDS

Power Transformer - High Voltage Insulation System - Partial Discharge - PD - PD Measurement –

Time Domain Analysis - Frequency Domain Analysis - PD Source - PD Localization.

[email protected]
1. INTRODUCTION

Power transformers are indispensable elements in all AC-systems used for power generation,
transmission and distribution. The construction of a specific transformer is mainly determined by its
function, e.g. generator step-up transformer, transformer for transmission or distribution, and by the
specified voltage levels of the windings. A common requirement for all transformer designs is the
ability of the electrical insulation system to withstand the permanent and temporary operating stresses
without damage during the expected lifetime of 40 to 50 years. In the focus of this contribution are
power transformers filled with mineral oil. However, most issues of PD localisation apply also to dry
type transformers.
During service, the insulation system of a power transformer ages continuously, primarily due to the
combination of different stresses, such as dielectric stress (lightning and switching over-voltages),
mechanical stress (caused by inrush currents and short circuits), thermal stress (due to overload and
hot-spots) and chemical decomposition of the insulating materials (formation of gas, depolymerisation
etc.). These ageing processes may generate regions with low dielectric strength in the high voltage
insulation system. At an early stage of ageing such week regions are the origin of PD activity, which
may finally lead to an insulation damage or electrical breakdown. According to the latest CIGRE
reliability survey, more than 40% of the unplanned transformers outages are caused by insulation
damages [1].
The main sources of partial discharge activity are (1) weak regions or spots with reduced dielectric
strength in aged insulation systems, or (2) local defects with excessive field stress in new insulation
systems, caused by inadequate design or by poor manufacturing quality. It is well established, that
permanent PD activity deteriorates the insulation materials, i.e. the insulating oil and the cellulose in
oil-filled transformers. This may finally end in a complete electrical breakdown of the deteriorated
insulation system. The time needed to develop a complete electrical breakdown depends mainly on the
location of the PD source and on the safety margin in the dielectric design of a specific insulation
system. When PD is detected in a transformer, neither experts nor sophisticated monitoring systems
can predict the time to breakdown. Instead, practical experience shows that in most cases complete
electrical breakdown in oil/cellulose insulation systems caused by dangerous PD sources develops
slowly. Therefore, sensitive PD detection, reliable identification and localisation of PD sources at an
early stage of a transformer’s life can avoid serious damage of the insulation system and prevent an
unplanned (forced) outage [2].
PD measurements in new or repaired transformers are carried out during the factory acceptance tests
(FAT) or also in site acceptance tests (SAT). With the development of both, advanced measuring
techniques such as phase resolving partial discharge analysers (PRPDA) for recording and evaluation
of so called PD pattern [3], and mobile HV-test sets based on modern power electronics [4],[5]
sensitive PD tests are possible on site and can be applied successfully even for large transformer units.
In fact, PD monitoring systems are currently installed for the surveillance and for the condition
assessment of strategically important transformers to avoid unplanned outages [5].
Requirements for the PD measuring system, the measuring and calibration procedures, along with the
recommended acceptance criteria for PD tests are described in International Standards of IEC [6] and
IEEE [7]. In these Standards, the value of the “apparent” charge in pico-Coulomb (pC) detected at the
measuring terminals is used to quantify the PD intensity. Admissable values of PD intensity in [6] and
[7], i.e. the acceptance criteria for PD tests are based on both, experience collected during FAT from
transformer manufacturers, and on explicitly identified PD defects of damaged or failed units.
However, for complex and large insulation systems, there is no direct relationship between the real
charge released at the PD source and the apparent charge measured at the transformer terminals. In
addition, the PD intensity calibration as defined in [6] and [7] is valid only for PD sources, which are
located close to the bushing used for calibration. A PD calibration carried out according to these
Standards can have an error of more than 50% when the PD source is located far away from the
calibrating bushing. Hence, for PD sources hidden in the transformer active part, the value of the
apparent charge in pC measured at the terminals does not sufficiently reflect the real risk of these
defects. Consequently, it is mandatory for new or repaired transformers to identify and localize any
PD activity with an inception below or at nominal voltage. The strict requirement, that a new

2
transformer must be PD free at nominal voltage during FAT or SAT assures that continuous PD
activity will be absent under service conditions.
Despite of widespread experience with PD detection methods and with PD tests on power
transformers, there is still a lack of reliable procedures to localize an internal PD source in large and
complex insulation structures. First concepts and ideas to localise PD sources in high voltage
components using time resolved PD signals were proposed already in 1962 [8]. Recent advances in the
localisation of PD in (homogeneous) transformer windings using high frequency winding models and
PD impulse measurements at both terminals of a specific winding were reviewed in [9]. The concept
and the application of multi-terminal PD detection and localisation techniques in the time- and
frequency-domain are described in [2] and [10].
Alternative to the direct method of measuring electrical PD current impulses at the transformer
terminals (at both terminals of a specific winding or at all terminals, simultaneously), indirect
diagnostic methods are used in practise, namely the acoustic method and the UHF method. The
physical principle behind the acoustic method are ultrasonic pressure waves, which are generated by
PD. These pressure waves can be detected by piezo-electric sensors placed at the tank wall of a
transformer [2]. On the other hand, the UHF method is based on the detection of electromagnetic
waves emitted by the PD source in the ultra-high frequency (UHF) range [14].
All PD detection techniques have their inherent limitations, it is therefore unlikely that a single
diagnostic method is able to provide a reliable assessment of a PD problem in a complex insulation
system. In such cases, a combination of appropriate detection methods is the most promising
approach. Typical applications and inherent limits for different PD detection and localisation methods
are summarized in Table 1.

Method Focus of Application Limits

PD source inside or outside the active
Electrical method: part Bushings (terminals) without tap-off may limit
detection of PD
PD source inside main insulation and detection sensitivity and location accuracy
impulses at all
between windings (HV/LV) (PD signal distortion)
terminals [2]
Application off-line
Electrical method: Applicable to single windings only, no
detection of PD PD source in (homogeneous)
coupling between windings considered
impulses at both windings
terminals of a Low PD detection sensitivity for defects far
Application off-line
specific winding [9] away from terminals
PD inside winding blocks cannot be detected
Acoustic method: PD source outside active part, in Poor localisation due to multi-path
detection of particular in oil, on leads or on the propagation and due to material dependent
ultrasonic pressure outside of the insulation system wave propagation velocities
waves at the tank
wall [2] Application off-line and on-line Presence of solid barriers inside the
transformer (core, supporting structure)
UHV method: Poor localisation due limited number and
detection of fixed positions of antennas (ports in tank wall)
PD source inside or outside the active
electromagnetic part Poor localisation due to multi-path
waves using UHF propagation and reflections at the tank wall
antennas installed in Application off-line and on-line
Presence of metallic structures inside the
the tank wall [14] transformer (shielding, reflection, attenuation)

Table 1: Tyical application and inherent limits of methods for PD detection and localisation.
The present contribution describes the practical experience with a three-step procedure to localize PD
sources using electrical PD signals, which are simultaneously measured at all bushings of a
transformer. The proposed PD localisation procedure is mainly based on information about PD
impulse propagation and PD signal coupling between transformer windings. This information can be
drawn from consistent sets of PD signals in the time- and frequency-domain.

3
2. TIME RESOLVED PD SIGNALS

The electrical breakdown in a weak region (a PD source) at an unknown location in the insulation
system of the transformer generates current impulses of very short duration in the Nanosecond (ns) or
Microsecond (μs) range. The shape of these PD current impulses is strongly dependent on the
insulating medium (gas, solid, liquid) and on the local dielectric stress at the PD source. Examples of
such current impulses measured at the origin of typical PD defects are shown in Figure 1.

Figure 1: PD current impulses measured at the source location [2]

a) Partial discharge in a void in solid insulation
b) Partial discharge in transformer oil

The waveform of the original PD signal is attenuated and shaped on its way from the PD source to the
measuring terminals (bushings). Therefore, in the classical PD test circuit, which consists of the
coupling capacitance Ck, (for transformers: Ck capacitance of bushings) connected in parallel to the
HV- and LV- windings and to the neutral terminal, only attenuated and deformed PD current impulses
i(t) can be measured. These current impulses at the measuring terminals are the responses of the active
part of the transformer to the excitation by the PD source at an unknown position.
At each bushing where PD activity is detected, the shape of the measured PD current impulse contains
information about the type of the PD source and the PD signal propagation path from the PD source
(unknown location) to a specific bushing. In consequence, for any PD source at any position in the
insulation system, a specific signal transfer function exists for each transformer terminal.
Figure 2 illustrates the diversity of possible PD signal paths and coupling modes within a transformer
with three windings. The equivalent circuit with lumped elements, as shown in Figure 2, is widely
used to study the electrical stress in a transformer subjected to lightning impulse voltages. However,
such lumped parameter models do not satisfactorily describe the PD signal transfer along the windings
and coupling between them, because their useful frequency range is limited to about 500 kHz.
Due to the very short rise times of typical PD current impulses at their origin (see Figure 1), high
frequency transformer models with a frequency range up to several MHz are necessary to describe the
PD propagation and coupling in winding structures of power transformers. The application of high
frequency winding models, such as distributed parameter models, capacitive ladder network, multi-
conductor transmission line models and hybrid models has been discussed in [9].
Due to the unequal PD signal transfer functions and due to the various coupling modes between the
windings or terminals (bushings), the PD signal waveforms recorded at several bushings differ
significantly. Consequently, the analysis of the time resolved shape of PD signals recorded
simultaneously at the terminals, together with the knowledge of the transformer design, offers an
interesting opportunity to perform a coarse localisation of PD sources in the insulation system of a
transformer.

4
 i(t) i(t)

C L

RW

i(t)
a) b)

Figure 2: Winding system and equivalent circuit of a power transformer [12]

a) Winding system: LV: low voltage winding, HV: high voltage winding, RW:
regulating winding
b) Equivalent circuit: NT: neutral terminal, C: capacitance, L: inductance, DO:
digital oscilloscope, SPEC: Spectrum analyser, i(t) : PD current impulse

The time-resolved signal of a PD arriving at a terminal of a power transformer contains three main
signal components, which are superimposed in the measured PD signal (Figure 3a):
 Capacitive component due to direct coupling of a PD source located very close to a specific
terminal (fast response of the capacitive network, Figure 3b)
 Traveling wave component caused by signal propagation along conductors (galvanic path),
typically seen when PD sources are located in windings directly connected to the specific
terminal (Figure 3c)
 Oscillating component (band pass filter response), typically observed when PD sources are
located far away from a specific terminal (Figure 3d).

5
Figure 3: Time-resolved PD signal and its components [11]
a) signal measured at the bushing
b) capacitive component
c) traveling wave component
d) oscillating component

These PD signal components are strongly influenced by the design of the transformer (size and type of
windings etc.), by the PD defect location with respect to the bushing where the PD current impulse
was recorded, and finally by the type of the PD source (electrical breakdown in gas or in liquid). In
most cases, all three components are reflected in PD current impulses recorded at the bushings of the
transformer. Despite the huge variety in the shape of measured PD currents, one can clearly
distinguish between two PD signals, which are delivered at different distances between the PD source
and the bushings (see Figure 4).
PD sites, which are close to a specific bushing, generate PD signals with a strong capacitive
component (Figure 3b). Measured signals from PD sources, which are located far away from a
specific bushing, are characterized by a strong oscillating component (Figure 3d). This oscillating
signal shape represents the step response of the winding system (band pass filter), which is excited by
the sharp PD impulse at its origin.
In a set of simultaneously recorded PD signals, there is at least one terminal where the PD signal
contains a predominant capacitive component. In this case, the search for the PD source location starts
from this bushing. The situation, where only oscillating wave shapes (band pass filter type, see Figure
3d) are measured at the terminals, indicates that the PD source is located inside the winding block, e.g.
in the winding structure or in the insulation system between the windings. To verify such PD defects,
it is necessary to disassemble the active part (core and windings) of the transformer.

6
Figure 4: Typical time resolved PD signals recorded at the bushings of a transformer [2]

3. ADVANCED SYSTEM FOR ANALYSIS OF PD SIGNALS IN TIME- AND

FREQUENCY-DOMAIN

Generally, PD signals can be recorded either in the time-domain (amplitude and time) using a digital
oscilloscope, or in the frequency-domain (magnitude and frequency) using a spectrum analyser or using
Fast Fourier Transformation (FFT) of the measured time-domain signal (see Figure 5).

A f

A … amplitude
f … frequency
t … time

t
A A

t f
time domin frequency domain

Figure 5: Signal presentation in time- and frequency-domain [2]

In linear systems, time- and frequency-domain are linked via Fourier transformation. For a given non-
periodic PD current impulse i(t), the complex frequency spectrum F(jω) is obtained from the Fourier
integral. According to this relationship, registered signals in the time-domain can be transferred into the
frequency-domain and vice versa.

7
The correct analysis of PD current impulses in the time- and frequency-domain requires wideband-
coupling devices directly connected to the terminals of the power transformer. High frequency current
transformers (HF-CT) with a frequency range between 0.1 to 30 MHz were successfully used for this
purpose [4]. It must be mentioned here, that a conventional PD measuring impedance Zm is not suitable
for time resolved measurements because of its narrow band pass filter characteristics, which is required by
IEC Standard 60270 for the “quasi integration” of PD-current impulses.
To provide multi-terminal detection capability, which is important for PD analysis in the complex
insulation system of transformers, HF-CTs are directly connected to the capacitive tap of the bushings
(Figure 6), or to external coupling capacitors, when bushing taps are not available.

a) connection of HF-CT to HV bushing tap b) connection to the external coupling

capacitor at LV bushings

Figure 6: Connection of high frequency current transformer (HF-CT) to HV- and LV-bushings [13]

(2)

(1)

(3)

Figure 7: Advanced PD system [4]

Advanced PD detection systems [4],[13] are suitable for recording PD pattern (statistical analysis of the
amplitudes of apparent charge) and for the acquisition of PD current impulses in the time- and frequency-
domain (Figure 7).

8
The main components of an advanced PD system are (see Figure 7):
(1) Spectrum analyser with a typical frequency span between 10 kHz and 20 MHz. The spectrum
analyser is used for the analysis of PD signals in the frequency-domain (full span mode) or for the
quasi-integration of PD signals to evaluate the amplitude of the apparent charge using the variable
band-pass filter (zero span mode). The band pass filter of the spectrum analyser can be used as the
front-end of a PRPDA-system to record the PD-pattern. In this case, the variable band pass filter
(zero span mode) of the spectrum analyser offers high flexibility for efficient noise suppression
[4].
(2) A multi-channel digital oscilloscope is employed for the analysis of PD signals in the time-
domain. The oscilloscope must have at least two channels with a bandwidth larger than 100 MHz.
(3) The PD analyser (PRPDA-system) is a computer-controlled impulse acquisition and digital signal
processing system, which performs a statistical analysis of recorded PD data (PD pattern). In a
typical measuring set-up for power transformers, multi-channel PD analysers are used to record
and process the signals from all terminals simultaneously.
In many cases, a digital oscilloscope with an integrated FFT function is sufficient to investigate PD
signals in the frequency-domain. However, the use of a spectrum analyser has the advantage of its
extremely large dynamic range. This feature is important when PD signals with a large range of
amplitudes must be analysed.

4. THREE-STEP PD LOCALISATION PROCEDURE

The localisation of PD activity detected in power transformers is based on the analysis of both, the
frequency spectra and the time-domain signals generated by a PD source in the insulation system. A set of
real PD data in the time- and frequency-domain is compared with the “characteristic time-domain signals”
and the “characteristic frequency spectra”, obtained in the procedure described below. Besides the
analysis of PD signals, the understanding of the specific transformer design is essential to correctly locate
and assess a PD source.
Step 1: A set of “real PD data” is established in the time- and frequency-domain by recording PD signals
simultaneously at terminals where PD activity (recorded PD pattern) was detected during
previous tests. A “PD signal cross-coupling matrix“ is derived from the real PD data set.
Step 2: A “characteristic data set” is built for the transformer under test: a calibrating signal is injected
(usually 1000 pC) at the bushing, where either the highest amplitude of apparent charge was
detected during previous PD tests, or where the shape of the real PD signal in the time- domain
indicates that this bushing is closest to the internal PD source.
The responses of the active part to the calibrating signal injected at the specific terminal are
recorded in the time- and frequency-domain at all other terminals. This result represents the
“characteristic PD signal cross-coupling matrix” for an assumed PD source, which would be
located close to the bushing, where the calibrating signal was injected.
When two or more PD sources are present, this characterization procedure should be repeated for all
terminals where significant PD signal amplitudes were detected during the previous PD tests.
A comparison of the “real PD data set” with the “characteristic data set” produces first hints of the
possible location of the PD source. By comparing with the design of the transformer, in particular
the electrical field at possible locations of the PD source, several potential locations of the PD
source can be excluded. In some cases, simultaneous detection of acoustic signals may help to
localize the supposed PD source.
Step 3: A steep current impulse (PD calibrator) is injected directly in the active part of the transformer
(which does not contain oil) at the place where the PD defect is assumed to be. This simulation
delivers a representative response of the transformer windings and insulation system to the
excitation by the fast calibrating impulse, i.e. by the artificial PD source at the supposed location.

9
 If the position of the PD source in the active part corresponds to the injection point of the artificial
PD impulse (calibrator), the waveforms of both, the PD signal recorded at the specific bushing
during the PD tests, and of the signal measured at the same bushing in the simulation procedure,
are very similar. However, it must be considered in the comparison of the time-resolved signals in
a transformer without oil, the measured signals oscillate at higher frequencies due to the lower
dielectric permittivity (air instead of oil).

5. PRACTICAL EXAMPLES

To demonstrate the ability of the localisation procedure described above, selected PD problems in power
transformers and the corresponding activities for PD localisation are discussed in this section. The detailed
analysis of the results in each case will also comment the limitations of the proposed localisation method.

5.1 PD Source on the Press Plate close to the HV Bushing

Transformer: 250 MVA, 220/150/24 kV

During the site acceptance test (SAT), a PD activity of 1500 pC was detected at the HV bushing of the
phase U1 already at 50% of the nominal voltage. The PD pattern, as shown in the left column of Table 2
are typical for floating conducting particles.

PD pattern at HV bushing U1 Location of the PD source

Table 2: PD pattern and location of the PD source

The comparison of real PD signals in the time-domain, recorded at bushings U1 (HV bushing) and U2
(LV bushing), with the calibrating signal (1000 pC) injected at the bushing U1 and its coupling to the
bushing U2 are shown in the Table 3.
Time resolved PD signals were recorded with a time base of 50 ns/div. Due to the very fast rise-time of
the real PD signal (left column in Table 3), compared with the much slower rise-time of the calibrating
signal, and due to the difference in the shape of both signals, it is evident, that the PD source must be very
close to the measuring tap of the HV bushing. To confirm this, a calibrating signal was directly injected at
the measuring impedance, which was still connected to the measuring-tap of bushing U1. Both signals
have the same shape (see Table 4), which confirms that the PD source is very close to terminal U1.

10
 Real PD data set in the time-domain Characteristic data set in the time-domain
1000 pC injection at U1

Table 3: Comparison of PD signals in the time-domain

Real PD data set in the time-domain Characteristic data set in the time-domain
1000 pC injection at the measuring imepdance
connected to bushing tap U1

Table 4: Simulation of the real PD signal

The comparison of real PD signals in the frequency-domain, recorded at bushings U1 (HV bushing) and
U2 (LV bushing), with the frequency spectra of the calibrating signal (1000 pC) injected at the bushing
U1 and its coupling to the bushing U2 and V2 is shown in the Table 5.

Real PD data set in the frequency-domain Characteristic data set in the frequency-
domain, 1000 pC injection at U1

Table 5: Comparison of PD signals in the frequency domain

11
 Unlike the signals in the time-domain, the frequency spectra of the real PD signals and the characteristic
data set in the frequency-domain are consistent. This can be explained with the sweep method used for
recording of the frequency spectra with a duration the sweep of 20 s. Due to the inherent variation in the
PD signals, in particular the variation in the rise-time, the frequency spectrum of real PD signals gives a
better agreement with the characteristic data set, than the comparison of discrete signals in the time-
domain. The fairly even frequency spectra in Table 5 reaching 10 MHz further confirm that the PD source
is close to the measuring tap of the bushing U1.
Based on these results, the first troubleshooting decision was to change bushing U1. This was much easier
and cheaper than lowering the oil level and opening the transformer for the inspection of the active part.
However, changing bushing U1 did not reveal the expected result.
In the next step, the level of the oil was lowered to perform a visual inspection through the manhole.
Conducting particles laying on the press plate were discovered, very close to the exit of the HV bushing
U1 (see right column in Table 2). After removing these particles, refilling the oil and final reassembly, the
transformer passed the second PD test without problems.
Conclusions drawn from this example are:
 The long cables necessary to connect the calibrator with the bushing head, lead to slower rise time
and to much more pronounced oscillations of the injected impulse as compared with the real PD
signal from the PD source located very close to the measuring tap.
 It is difficult to distinguish between the simulated PD source at the measuring tap of the bushing
and real PD source very close to the HV exit.
 The application of FFT signal processing routines to single PD signals clarified the observed
discrepancy between the consistent frequency spectra (close agreement) and the differences in the
time-resolved signals.

5.2 PD Source between the LV Leads

Transformer: 75 MVA, 132/50 kV

After one month of operation, the Buchholz relay tripped several times. Analysis of dissolved gases in
oil (DGA) indicated intense internal PD activity. During the on-site high voltage test, a PD pattern
with very high amplitudes was detected well below the rated voltage at two terminals 2U and 2V (see
Table 6). The PD pattern of terminal 2U exhibits two phase-shifted but similar PD signal clusters,
representing PD activities in two phases. One PD source was close to terminal 2U and a second PD
activity was coupled from phase 2V (phase-shift 120°). Nearly identical PD pattern were recorded at
terminal 2V (not shown here). Consequently, a second PD source was identified close to the bushing
2V. The recorded PD-pattern is typical for local humidity in an oil-cellulose insulation system.
The comparison of real PD signals in the time-domain, recorded at bushings 2U and 2V (LV
bushings), with the calibrating signal (1000 pC) injected at bushing 2U, and its coupling to the other
bushings, is shown in the Table 7. The real PD signal at terminal 2U is comparable to the injected
calibrating signal at bushing 2U. The corresponding signals at the bushing 2V were similar. This
means, that two nearly identical PD sources are located close to the bushings 2U and 2V.
The opposite polarity of the simultaneously recorded PD signals at terminals 2U and 2V is a strong
indication, that the PD source is located between both terminals.
The comparison of real PD signals in the frequency-domain, recorded at bushings 2U and 2V, with the
frequency spectrum of the calibrating signal (1000 pC) injected at the bushing 2U is shown in the
Table 8. Frequency spectra of the real PD signals recorded at terminals 2U and 2V show close
agreement with the characteristic frequency spectrum recorded at bushing 2U. This confirms the
postulation that both PD sources must be close to terminals 2U and 2V.

12
 PD pattern at LV bushing 2U Location of the PD source

PD-source

Table 6: PD pattern and location of the PD source

Real PD data set in the time-domain Characteristic data set in the time-domain
1000 pC injected at 2U

2U

Table 7: Comparison of signals in the time-domain

Real PD data set in frequency domain Characteristic data set in frequency domain,
1000 pC injected at 2U

Table 8: Comparison of signals in the frequency domain

The very similar shape of the time signals and frequency spectra recorded at terminals 2U and 2V
support the explanation that the two different PD signals run through a similar propagation path inside
the transformer. Obviously, the two PD sources are located at the same place, close to terminals 2U
and 2V. Additionally, acoustic PD signals detected at the location of the 50 kV leads further approved
the position of both PD sources (see right column in Table 6).
Finally, the recorded PD activity was attributed to high local moisture in both, the thick paper
insulation of the 50 kV leads, and in the laminated wood used for clamping the leads (see Table 6).
After a drying process on site with hot oil circulation and low frequency heating, the transformer
passed the PD test. This confirmed that the localisation and the repair of the defect were successful.

13
Conclusions drawn from this example are:
 When superimposed PD signals due to multiple PD sources exist in the insulation system, it is
not possible to perform reliable localisation based only on the analysis of PD signals in the
time-domain. Variation in the PD inception voltage and different amplitude levels of PD
signals are a clear indication for multiple PD sources.
 From the close agreement in the frequency spectra recorded at terminals 2U and 2V can be
concluded that the PD sources must be located very close to their bushings. Opposite polarity
of simultaneously recorded PD signals of two bushings indicates, that the PD source must be
located between these terminals.

5.3 PD Source in the Main Insulation

Transformer: 987 MVA, 345/24.8 kV

During the factory acceptance test (FAT), PD activities with similar PD pattern were recorded at all
three HV terminals. The typical PD pattern of this case is shown in Table 9. The pattern are
characterised by two superimposed PD activities: (1) voids in the insulation system, (2) conducting
particle lying on the surface of the insulation material.
PD pattern at HV bushing H3 Location of the PD source

(1)

(2)

Table 9: PD pattern and location of the PD source

The comparison of the real PD signals in the time-domain, recorded at terminals H3 (HV bushing) and
H0 (bushing of neutral terminal), with the calibrating signal (1000 pC) injected at bushing H3 and its
coupling to the neutral terminal H0 is shown in Table 10. There is no similarity between the PD
signals measured at terminals H3 and H0 and the corresponding characteristic data set (injection of the
calibrating signal at terminal H3). The much slower rise-time (approx. four times slower) of the real
PD signals is an indication that the PD source is located at some distance from bushing H3. Similar
signals were recorded at the other HV terminals H1 and H2. The opposite polarity between PD signals
recorded at H3 and H0 gives evidence that the PD source must be located between terminals H3 and
H0.
The comparison of the real PD signals H3 and H0 in the frequency-domain with the frequency
spectrum of the calibrating signal (1000 pC) injected at the bushing H3, is shown in Table 11. (Please
note the different frequency spans of the recorded spectra). The agreement between the two frequency
spectra is rather poor. Hence, the location of the PD source cannot be close to terminal H3, as already
stated in the analysis of the time-domain signals.
14
 Real PD data set in the time-domain Characteristic data set in the time-domain,
1000 pC injected at H3

H3

H0

Table 10: Comparison of signals in the time-domain

Real PD data set in the frequency-domain Characteristic data set in the frequency-
domain, 1000 pC injected at H3

5 MHz

Table 11: Comparison of signals in the frequency-domain

The application of the acoustic PD detection method did not produce any results. This was an
important clue that the PD source must be located inside the active part of the transformer.
Due to the fact, that this transformer had already been opened twice for visual inspection without
solving the PD problem, it was decided to apply Step 3 of the proposed localisation procedure, i.e. to
simulate the PD source in the active part of the transformer without oil.
A steep impulse signal (PD calibrator, 10 nC) was directly injected at different locations in the active
part. The best alignment between the real PD signal and the simulated PD signal was reached when the
artificial PD was injected in the main insulation between the windings. It was not possible to compare
the simulated signals unambiguously with the real PD signals, as the amplitude ratio and the rise-time
were different.
The comparison of real PD signals in the time- and frequency-domain with the responses recorded at
terminals H3 and H0 to the injected impulse signal at the suspected position of the PD source in the
active part is shown in Tables 12 and 13.
After disassembling the active part, a visual inspection finally revealed carbonised tracks in the barrier
system of the main insulation in all three phases (see left column in Table 9). The origin of this PD
activity in the main insulation was a wrong type of glue, which was used for the fixing the distance
ledges in the main insulation system during the manufacturing process.

15
 Real PD data set in the time-domain Characteristic data set in the time-domain
10 nC injected on the active part without oil

Table 12: Comparison of signals in the time-domain

Real PD data set in the frequency-domain Characteristic data set in the frequency-
domain, 10 nC injected on the active part
without oil

Table 13: Comparison of signals in the frequency domain

Conclusions drawn from this example are:

 A discrepancy between the real PD signals and the characteristic data in time- and frequency-
domain is a strong indication that the PD source is located inside the active part of the
transformer. The absence of acoustic PD signals supports this statement.
 The simulation of the PD source by the injection of an impulse signal at suspected PD sites in
the active part (Step 3 of the localisation procedure) led to the conclusion that the PD source
must be located in the main insulation between HV- and LV windings.
 The options to simulate a PD source inside the active part of a transformer are rather limited,
as direct access to the interior of the winding blocks is generally not possible. Nevertheless,
with the systematic application of Step 3 of the proposed procedure, the search of a PD source
will be focussed (several assumed positions of PD sources can be excluded) to a specific
substructure of the active part (e.g. windings or insulation systems).
 The applied localisation procedure was successful to find the PD source (carbonized tracks) in
the barrier system of the main insulation between HV- and LV-windings.

16
6. CONCLUSIONS

PD activity detected in a power transformer at or below nominal voltage may be harmful for the
insulation system. Regardless which value of PD intensity (apparent charge) is measured in a
particular transformer, PD defects should generally be identified and localised to assess their severity
including the risk of damage or of a forced outage.
Besides the widely used acoustic method for localisation of PD sources in oil-filled transformers and
novel detection and localisation techniques in the ultra-high frequency (UHF) range, the evaluation of
electrical PD signals in the time- and frequency-domain offers a considerable potential to localise all
categories of PD defects in the complex insulation system of power transformers.
PD signals measured at the terminals of a transformer contain information about both, the type of a PD
activity (discharge in oil or gas), and the propagation path from the unknown location of a PD source
to a specific terminal including associated phenomena, such as coupling modes between windings,
signal attenuation and reflection. The present contribution demonstrates that the localisation of a PD
source is possible when the detected PD signals are systematically analysed in the time- and frequency
domain. Advanced PD measuring systems are available today, to detect PD signals at several terminals
simultaneously (multi-terminal detection) with high sensitivity and appropriate bandwidth, and to
analyse them in the time- and frequency-domain.
The proposed PD localisation procedure compares a consistent set of time-resolved PD signals and
frequency spectra, recorded at the transformer terminals, with the responses of the transformer’s active
part (windings and core) to the excitation by an artificial PD current impulse (calibrator), injected at
the external terminals (bushings), or at the assumed PD site in the active part (oil removed from the
transformer). Although the options to simulate an artificial PD source inside a winding block are rather
limited (no direct access to main insulation system), the proposed localisation procedure successfully
revealed a PD-defect, which was located in the barrier insulation system of the main insulation.
Even when the proposed procedure for PD localisation delivers promising results in many cases, the
efficiency of PD localisation could be substantially improved with a simulation model for PD signal
propagation inside a transformer, including all relevant coupling modes between the windings. It is
mandatory to use a high frequency transformer model for such simulations, built from the design data
of the transformer unit to be investigated. The development of high frequency models suitable for PD
localisation in power transformers is an attractive topic for future work, which could be of interest for
both, the universities and the manufactures.

BIBLIOGRAPHY

[1] CIGRE Technical Brochure 642, Transformer Reliability Survey, WG A2.37, 2015.
[2] CIGRE Technical Brochure 676, Partial Discharges in Transformers, WG D1.29, 2017.
[3] B. Fruth, J. Fuhr, Partial Discharge Pattern Recognition - a Tool for Diagnosis and Monitoring
of Aging, CIGRE Main Session 1990, Paris, Paper 15/33-12.
[4] R. Bräunlich, M. Hässig, J. Fuhr, Th. Aschwanden, Assessment of Insulation Condition of Large
Power Transformers by On-site Electrical Diagnostic Methods, Conference Record of 2000
IEEE International Symposium on Electrical Insulation, IEEE Publication 00CH37075, pp. 368-
372.
[5] D. Gross, M. Soeller, On-Site Transformer Partial Discharge Diagnosis, Conference Record of
2012 IEEE International Symposium on Electrical Insulation, San Juan, June 10-13, 2012.
[6] IEC Standard 60076-3, Power Transformers – Part 3: Insulation levels, dielectric tests and
external clearances, Annex A, Application Guide for Partial Discharge Measurements during
AC Voltage Withstand Test on Transformers according to 12.2, 12.3 and 12.4, International
Electrotechnical Commission, Geneva, 2014.
[7] IEEE/ANSI Standard C57.12.90, Test Code for Liquid-Immersed Distribution and Regulating
Transformers, clause 10.8 and 10.9, Induced Voltage Tests, Institute of Electrical and
Electronics Engineers, Inc., New York, 1999.

17
[8] V. Kreisinger, A Method of Locating the Starting Point of Corona in Transformer on test with
Induced Voltage, Acta Technik, CZA, Vol. 7, pp. 296-305, 1962.
[9] B.T. Phung, Detection and Location of Partial Discharges in Transformers Based on High
Frequency Winding Responses, Chapter 14 in Electromagnetic Transients in Transformers and
Rotating Machine Windings, Information Science Reference (IGI Global), Su, C.Q. (Editor),
Hershey, PA, USA, 2013, pp. 521-539.
[10] J. Fuhr, Th. Aschwanden, Identification and Localization of PD-Sources in Power-Transformers
and Power-Generators, IEEE Transactions on Dielectrics and Electrical Insulation, Vol. 24,
No.1, 2017, pp. 17-30.
[11] J. Fuhr, M. Hässig, P. Boss, D. Tschudi, R.A. King, Detection and Localisation of Internal
Defects in the Insulation of Power Transformers, IEEE Transaction on Electrical Insulation,
Vol.28, No. 6, 1993, pp 1057-1067.
[12] A. Carlson, J. Fuhr, G. Schemel, F. Wegscheider, Testing of Power Transformer and Reactors,
ABB Handbook, Second Edition 2010,
[13] Th. Heizmann, R. Bräunlich, Th. Aschwanden, J. Fuhr, M. Hässig, P. Müller, The Swiss
experiences of On-site High Voltage Tests and Diagnostic assessment on Large Power
Transformers, CIGRE Main Session 2010, Paris, Paper A2-211.
[14] M.D. Judd, Partial Discharge Detection and Location in Transfomers Using UHF Techniques,
Chapter 13 in Electromagnetic Transients in Transformers and Rotating Machine Windings,
Information Science Reference (IGI Global), Su, C.Q. (Editor), Hershey, PA, USA, 2013, pp.
457-520.

18