21, rue dArtois, F-75008 PARIS

D1-105

CIGRE 2006

http: //www.cigre.org

Synchronous 3-Phase Partial Discharge Detection on Rotating Machines

W. Kalkner, A. Obralic
Technical University Berlin
Berlin, Germany

R. Plath
IPH High Voltage Lab
Berlin, Germany

M. Kaufhold, A. Bethge
Siemens, A&D Large Drives
Germany

SUMMARY
Specific level of partial discharges (PD) is normal and an allowed occurrence in stator insulation of
large high voltage rotating machines. However in certain cases PDs represent symptoms and even
cause of defects of insulation or winding structure. In the stator winding these discharges may appear
in the form of slot discharges from loose coils, slot discharges from deteriorating of semiconductive or
grading coating, internal discharges from insulation voids or delamination, and surface discharges due
to moisture or contamination of the winding [1].
PDs occur in the insulation system, where the local electrical strength is exceeded by the local electric
field stress. In general, in high voltage rotating machines the insulation designers allows for that by the
use of PD resistant components in the system. Nevertheless in certain cases PDs may damage the
voltage grading systems of the winding, the inter-strand insulation, or the mainwall insulation or they
can give a hint, that the original design of the winding has changed by ageing or other effects. A
failure of the stator winding insulation is in majority of cases very costly, in term of both the costs to
repair or replace a stator winding and outage time of the machine. Therefore much effort has been
invested over the years in developing techniques to identify the occurrence of PD in the stator winding
and to classify them in the way to distinguish between normal and harmful occurrences [2]. However,
the main difficulties with the interpretation of PD measurement results are coupling with the electrical
interferences that are inevitably present in industrial surrounding where the machines are mostly
deployed, and recognition and separation of commonly occurring several different parallel active PD
sources within the stator winding insulation. These difficulties can be eliminated through the usage of
a new digital multi-terminal PD measurement system which enables synchronous measurement of PD
impulses on all three phases of the machine combined with hereon based new evaluation technique
that calculates the magnitude ratio of every detected PD impulse with its synchronously measured
inductively and capacitively coupled impulses to the other two phase windings. The result is one
evaluation diagram in which the superimposed PD sources as well as outer noise can be clearly
distinguished as a cluster and separated for subsequent individual analysis. In addition efforts can be
attempted to classify these PD sources toward type, intensity, risk grade and location.

KEYWORDS
Rotating Machine Partial Discharge Synchronous Measurement PD Source Separation

martin.kaufhold@siemens.com

INTRODUCTION
The paper describes the technical characteristics of a new synchronous digital PD measurement
system as well as the theory and abilities of a new evaluation technique, which enables the separation
of multiple PD sources from each other and from outer noise. In addition the measurement results and
basic theory of the complex propagation and coupling characteristics of PD pulses in a stator winding
will be described, achieved through test pulse injection in different parts of stator bar sections.
Subsequently the successful application of the synchronous 3-phase PD detection will be demonstrated
on an example of a PD measurement on a service-aged high voltage machine.

SYNCHRONOUS PD MEASUREMENT SYSTEM

The new synchronous PD measurement technique consists of three (or more) synchronized PD sensor
units, one for each phase and optionally one for the neutral [3]. The circuit diagram of one PD sensor
unit is schematically shown in Figure 1a.

Fig.1:

a) Circuit diagram of one PD sensor unit

b) measurement setup

The measurement system employs the high speed fibre optical network technology, which provides
synchronicity between PD sensor units.
To detect PD events, each sensor unit acquires data from its PD input, which is in the majority of cases
a high voltage capacitor connected to the HV phase terminal of the machine (see Figure 1b). The PD
sensor units are connected to the control station PC by means of a high speed fibre optical network.
Each sensor can resolve PD events that are as short as 2 ns, and occur within as little as 200 ns of one
another. Data from every single PD impulse like apparent charge, polarity, time of occurrence and
relation to the phase voltage can be recorded and stored for later off-line processing.
The system also includes a real time oscilloscope and spectrum analyser view for the PD input as well
as internal calibration output which can be used to determine the frequency response, propagation
delay, and attenuation of the system under investigation as well as to calibrate the apparent charge.
Besides, the centre frequency and the bandwidth can be freely chosen in a wide range (centre
frequency range: 0...20 MHz, bandwidth: 100 kHz3 MHz), which allows an optimum adjustment of
the measurement system to the device under test and the choice of a frequency range of relatively low
disturbances.
All this enables synchronous detection of PD and noise signals on all three phases of rotating high
voltage machines, which allows detection of PD signals originating from the phasewinding connected
and those originating from cross couplings to other phasewindings.

EVALUATION TECHNIQUE
PDs are extremely fast current impulses which rise time and width depends on discharge type. On
their propagation path from the origin to the HV terminal and neutral, PD pulses are faced with all
kinds of deterioration, damping, attenuation and reflection. In addition, PDs tend to be masked by PDs
from other sources within the same machine and from a variety of outer disturbances. Finally the
apparent charge, polarity and at last shape of measured PD impulse also strongly depend on
measurement frequency range. The consequence is that the PDs on their transmission path loose the
information about the physics of the discharge type and gain the information about propagation and
coupling characteristics of the device under test. In other words, the determination of the number of
parallel active PD sources and their classification based upon pulse shape or classical Phase Resolved
Partial Discharge diagram (PRPD diagram) is very difficult even for experienced test personal.
On the other hand the PD impulses tend to make capacitive and inductive couplings to the other two
phasewindings. The result are three almost synchronously or within very small time frame (< 1s)
measured impulses (PD triple). The impulse with the highest measured amplitude often represents
most closely the original PD impulse caused by high voltage stress in the winding insulation, while the
other two result from cross-talk between the phases. Because of the complex propagation and coupling
characteristics of the machine windings, each source of PD impulses has a unique PD-tripel magnitude
relation and consequently a characteristic signature. Hence the importance of the multi-channel
synchronous PD measurement technique is evident, because only usage of this technique (instead of
multiplexer switches between the channels) enables detection of almost simultaneously occurring
impulses building up PD triples. In order to visualise the amplitude ratio of PD triple impulses, the so
called 3 Phase Amplitude Relation Diagram (3PARD) has been developed. There are two possible
construction ways of this diagram. First one, so called addition of magnitude relations, has been
described in detail in [4]. The second one, vectorial addition (two dimensions only) of impulse
magnitudes, is shown in Figure 2.
a)

b)

a) inner PDs: L1>L2, L2=L3

Fig. 2: Construction of 3PARD

b) inner PDs: L1>L2>L3

c)

c) outer noise: L1=L2=L3

One single PD triple is represented by one point which position in the diagram depends on the
magnitude ratio of the related impulses. For example, if an impulse recorded in phase L1 is higher
than the impulses recorded in other two phases which are equal (Fig.2-a: L1 > L2, L2 = L3), then the
point would be displayed exactly on the axis L1. Its distance from the origin of the diagram depends
on the amplitude ratio of the highest impulse in relation to other two ones. On the other hand, if the
coupling to the phases L2 and L3 is not symmetric (Fig.2-b: L1 > L2 > L3), then the position of the
point would be shifted toward axis of the second highest impulse. Electrical interferences and outer
noise, which usually have the pulse-like nature of PD but occur with almost same amplitude on all
three phases of the machine, would be displayed at the origin of the 3PARD (Fig.2-c: L1= L2 = L3).
By repetition of this procedure for a great number of recorded PD impulses and due to the
characteristic PD triple magnitude relation, PD sources within the machine as well as outer noise
appears as a clear distinguishable concentration of points (cluster) in a 3PARD. Subsequently each

cluster can be separated using different clustering algorithms and without superimposition effects
transformed to any phase-voltage related or pulse-to-pulse evaluation diagram [5].

PD PULSE PROPAGATION AND COUPLING CHARACTERISTICS

As mentioned above, the correlation between pulses measured at the terminals of the machine and
original discharge pulse caused by electrical stress is strongly influenced by the pulse propagation path
from PD source to measuring location. Because of their distributed nature, complex geometry, wide
range of materials and various insulation interfaces stator windings of rotating machines can be
considered to be complex propagation networks for signals resulting from PD impulses.
Generally the PD signals in a stator winding of high voltage machines contain two frequency
components (Fig. 3) [6,7,8]. First one, slow or galvanic component, travels through entire winding
conductor and can be measured after transient time that depends on the location of the discharge
origin. The pulse becomes wider and its amplitude smaller with growing propagation length. This
component contains mainly low frequency components. Another one, the fast mode, propagates due
to capacitive and inductive coupling and appears at the measuring terminal, irrespective of PD source
location, without time delay. This pulse component contains mainly higher frequency components.

Fig.3: Pulse shape, when electrical length between pulse origin

and measuring terminal is increased

Fig.4: Experimental setup

In order to verify this propagation theory and to determine its effects on 3PARD evaluation technique
a distinctive type of discharge activity has been simulated on a test high voltage machine. This defect
type, an inadequate endwinding spacing or contamination of the endwinding insulation and support
structure, has been simulated by a calibrator test signal injection between adjacent stator bars
belonging to different phases in the overhang section of the winding (Fig. 4). The pulses were injected
between phases L1 and L3. The output signal and its couplings were synchronously measured on all
three phases at the HV terminals of the machine. The study was performed on an eighteen pole starconnected 10 kV motor (rotor removed) at high-voltage laboratory of Technical University Berlin. A
small hole was drilled in the insulation of each turn of the affected phases in the endwinding to
connect the output of pulse generator with the winding conductor. Three measurement frequency
ranges have been used: 150 kHz 450 kHz, 450 kHz 750 kHz and 750 kHz -1,05 MHz. The results
are similar for all combinations of injected phases because of their symmetrical geometry and
configuration. Figure 5-left shows the measured response at all three HV terminals displayed in a chart
as a function of the injection position. Figure 5-right shows hence resulting 3PARD diagrams. Note
that y-axis are normalized, the maximum value is set to one.

a)

b)

c)

Fig.5: Charges at all three terminals when pulses are injected between turns of the phase L1 and L3 at different
pole numbers (left column), resulting 3PARD diagrams (right column)
a) measurement frequency range 150 kHz 450 kHz
b) measurement frequency range 450 kHz 750 kHz
c) measurement frequency range 750 kHz 1,05 MHz

The measured charges at the HV terminal of the machine strongly depend on two factors: place of
injection and measurement frequency range.
The measured charges at the HV terminals of the phases where the test impulse is injected (L1 and L3)
as well as the couplings to the phase L2 decrease for larger electrical distance between injection and
measuring site. The lower the measuring frequency range the deeper into the winding can be seen.
This indicates that high frequency or fast components of the impulse are heavily attenuated when
pulse is injected further away from measuring terminal. In case of low frequency "slow" component,
although attenuated with electrical length traversed through the winding, this attenuation is much less
severe than by the "fast" component. Within the measuring depth the impulses with the highest and
approximately equal amplitudes are always measured in phases where the test impulse is injected in.
This result allows, related to real on-line measurement of this discharge type, a first categorisation of
the PD source.
Figure 5- right shows the 3PARD diagrams of this measurement for the appropriate frequency ranges.
Each point represents a different injection point. In all three cases the majority of the points are located
between the axis L1 and L3 according to the ratio of the recorded impulses and to 3PARD
construction setup described in chapter above. The higher the measurement frequency the less
distinguishable points can be seen. The reason is the equalisation of all three measured impulse

amplitudes in higher frequency range even for short electrical distances of the PD to the terminal.
Hence PD sources located further away from the detection terminal measurement in high frequency
ranges tends to produce clusters at the origin of the 3PARD diagram similar as outer disturbances. For
classification of these clusters and there differentiation from noise, further investigation has to be
made, like backtransformation into the PRPD diagram. For real on-line PD measurements this may not
always be a significant disadvantage since often partial discharges occur in the high voltage part of the
machine close to the measuring HV terminal. On the other hand the signals to couplings magnitude
ratios measured in low frequency range are still present even for distant injection positions which
display more distinguishable points in 3PARD diagram. For investigation of discharges occurring
deeper into the winding or for localization of PD sources using Time Domain Reflectometry (TDR)
methods, the lower frequency range should be preferred.
Similar conclusions have been also achieved by test pulse injection between phase conductor and
stator core both in endwinding and in slot section of the stator [9].
OFF-LINE PD MEASUREMENT AND 3PARD ANALYSIS
Synchronous 3-phase PD measurements have been performed on a variety of high voltage machines
under on-line as well as under off-line conditions. The differentiation potential and ability of the new
synchronous PD measurement method to distinguish background disturbances form internal PDs can
exemplary be presented on following off-line PD measurement.
Test device was a 10 kV, 7500 kW, 4 pole asynchronous machine. Each phasewinding was energized
successively with rated voltage. Measurement frequency range was 1 MHz 3 MHz. The
measurement was performed in a factory where the maintenance of the machine was carried out. At
the same time a variety of other manufacturing activities took place on site with disturbances
producing electrical tools. During measuring duration of the phase L2 (30 sec) in direct vicinity of the
test setup a large drilling machine was switched on and, after seven seconds, switched off again. These
disturbances coupled to the measuring setup and significantly adulterated the results (see Figure 6).
This case of course symbolically represents noise influence in industrial surrounding but nevertheless
without 3PARD analysis this PD measurement would be hard to accomplish under these
circumstances.
L2

L1

L3

Fig.6: All recorded PD pulses related to the voltage of phase L2 (PRPD) with framed disturbances
a)

b)

cluster 1

c)

cluster 2

2
1

Fig.7: a) 3PARD diagram and backtransformation of b) cluster 1 disturbances; c) cluster 2 inner PDs

a)

b)

Fig.8: Trend analysis of

a) cluster 1 - disturbances b) cluster 2 - inner PDs

Recorded data of every single PD impulse has been evaluated with 3PARD technique. Figure 7 a)
shows the 3PARD diagram with two clear distinguishable clusters each representing a different source
of impulses. Recalculation was performed for both clusters to reconstruct the conventional PD pattern
(PRPD) related to the voltage of phase L2. The result is shown in Figure 7 b) and c).
PRPD pattern of the cluster 1 (Fig. 7-b) can easily be recognized in Figure 6 and identified as outside
disturbances which are coupled in all three phases with almost the same magnitude. Hence this cluster
is displayed at the origin of the 3PARD diagram.
Backtransformation of the cluster 2 is shown in Figure 7 c). It also can be recognised in Figure 6. The
PD pattern of this cluster is symmetric: the discharge magnitude and frequency are equal in negative
and in the positive half AC cycle. Impulses occur with relative low magnitude (< 1,5 nC) but with high
repetition rate which all indicate that the PDs are likely to be between mainwall insulation layers
probably due to delamination (internal insulation micro voids) [10]. Figure 8 shows a trend analysis of
the impulses which constitute cluster 1 (a) and cluster 2 (b) over the entire measurement duration. The
beginning and the end of the disturbances impact can easily be determined (between 7th and 14th
second). Internal PDs are active during the whole period of measurement.
Subsequently a calculation of the time difference between each impulse building up particular clusters
and a selectable number of following impulses which are not displayed in the 3PARD diagram can be
performed for statistical determination of the PD source location (Fig. 9). When a PD impulse appears,
starting at the location of the PD source, two waves travel along the conductor. First one goes directly
to the measuring terminal of the machine. The other part of the impulse travels in opposite direction to
the neutral, where it is reflected, and travels back to the measuring terminal end as well. In
consideration of the propagation velocity and the time difference between the original impulse and the
reflected one, the location of the PD source can be determined. As a result, the Figure 10 shows
exemplary the level of the PD impulses of the cluster 2 and following impulses displayed in a chart as
a function of their detection time difference.

(
dis tan ce = (length of conductor )

2...x

t ) (pulse
1

velocity )

Fig.9: Principle of Time Domain Reflectometry (TDR)

Fig.10: Reflectogram of the cluster 2

Considering the length of the phase conductor from HV terminal to the neutral (297 m) and the pulse
propagation velocity in this frequency range , the histogram shows a significant density at 2,89 s.
This corresponds to the distance of this PD source from the measuring terminal of about 73 m. This is
conforming to total copper strand length of one coil of this machine type.
CONCLUSION
Synchronous multi-terminal PD measurement technique in combination with the new 3PARD
evaluation method enables distinction and separation of several parallel active PD sources within
stator winding of rotating machine and external interference, their assignment to the origin
phasewinding and their further individual investigation. Due to selective isolation of conspicuous
clusters, a wide range of opportunities for classification and localization of corresponding PD sources
has been opened.

BIBLIOGRAPHY
[1]

Kaufhold, M., Schafer, K., Bauer, K., Bethge, A., Risse, J.: Interface Phenomena in Stator
Winding insulation- Challenges in Design, Diagnosis, and Service Experience, Electrical
Insulation Magazine IEEE, Mar/Apr 2002, Volume 18, pages 27-36

[2]

Kaufhold, M et al.: Experiences with Partial Discharge Measurements for the Evaluation of
High Voltage Winding Insulation of Large- and medium sized Rotating Machines. INSUCON
1998, Harrogate, 1998

[3]

Obralic, A., Kalkner, W., Bethge, A., Plath, R.: Synchronous 3-Phase Partial Discharge
Detection on Rotating Machines, Peking, ISH August 2005, page 378

[4]

Schaper, S., Kalkner, W., Heinrich, R., Plath, R., Bethge, A.: Synchronous Three Phase Partial
Discharge Detection on Rotating Machines, Delft ISH, 25-29 August 2003, page 343

[5]

Patsch, R., Benzerouk, D., Berton, F.: Time Differences between Consecutive Partial Discharge
Events - what do they tell us? NORD-IS Tampere Finland, 11-13 Juni 2003

[6]

Pemen, A. J. M.: Detection of partial discharges in stator windings of turbine generators. PhD
Thesis, Technische Universiteit Eindhoven, 2000, pages 23-43

[7]

Zhu, H., Kemp, I.J.: Pulse propagation in rotating machines and its relationship to partial
discharge measurement, IEEE International Symposium on Electrical Insulation, Baltimore, MD
USA, June 7-10, 1992

[8]

McNicoll, Y.: The effect of cross coupling phenomena on partial discharges measurements,
Montral ISH 25-29 August 1997

[9]

Obralic, A., Kalkner, W., Kaufhold, M., Plath, R.: Investigation of PD-Pulse propagation and
coupling between three phases of a stator of a rotating machine, Peking, ISH August 2005, page
104

[10]

Stone, G. C., Boulter, E. A., Culbert, I., Dhirani, H.: Electrical Insulation for Rotating
Machines, IEEE Press Series on Power Engineering, December 2003, page 306, 263

OMICRON is an international company serving the electrical power industry with

innovative testing and diagnostic solutions. The application of OMICRON products
allows users to assess the condition of the primary and secondary equipment on
their systems with complete confidence. Services offered in the area of consulting,
commissioning, testing, diagnosis, and training make the product range complete.
Customers in more than 140 countries rely on the companys ability to supply leading
edge technology of excellent quality. Broad application knowledge and extraordinary
customer support provided by offices in North America, Europe, South and East Asia,
Australia, and the Middle East, together with a worldwide network of distributors
and representatives, make the company a market leader in its sector.

Americas
OMICRON electronics Corp. USA
12 Greenway Plaza, Suite 1510
Houston, TX 77046, USA
Phone: +1 713 830-4660
+1 800-OMICRON
Fax:
+1 713 830-4661
info@omicronusa.com

OMICRON

Asia-Pacific
OMICRON electronics Asia Limited
Suite 2006, 20/F, Tower 2
The Gateway, Harbour City
Kowloon, Hong Kong S.A.R.
Phone: +852 3767 5500
Fax:
+852 3767 5400
info@asia.omicron.at

Europe, Middle East, Africa

OMICRON electronics GmbH
Oberes Ried 1
6833 Klaus, Austria
Phone: +43 5523 507-0
Fax:
+43 5523 507-999
info@omicron.at

www.omicron.at www.omicronusa.com