Partial Discharge Detection In Medium
Voltage Stators Using an Antenna
Akshay Bhure Elias G. Strangas
bhureaks@egr.msu.edu strangas@egr.msu.edu
Department of Electrical and Computer Engineering
Michigan State University
East Lansing, MI, USA
Abstract—Partial Discharges, if initiated give rise to
progressive deterioration of insulation material and lowers its
life expectancy. If allowed to persist further, PD can even lead to
the electric breakdown of insulation. However, in spite of its
damaging effects on machine insulation, its application during
the production phase of machines, to avoid their permanent
failure has gained importance in this research field.
This paper discusses a Partial Discharge detection technique
developed for medium voltage stator insulation utilizing an
antenna to sense the discharge currents and record PD events
produced inside a stator insulation. To reduce the effect of noise
on detections process, some grounding techniques incorporated
and later Undecimated Discrete Wavelet Transform was used to
extract the discharge events from antenna output. To extract the
best out of the technique, outputs were analyzed on a repetitive
basis to make a confident decision on the presence of PD.
This technique of detecting PD shows high sensitivity, as the
detections are independent of the impedance of test setup used
for experiments.
I. INTRODUCTION
A partial discharge (PD) occurs within the insulation
material when a high voltage stresses it, initiating ionization
within imperfections/voids of the insulation, due to the lower
dielectric strength of the void compared to the insulation
material. When the voltage across the insulation approaches
its peak value, these discharges bridge the gap between
imperfection and the ground partially and attenuate quickly
due to its high frequency [1].
When PD continues to cause ionization within the
imperfection/void, it causes local heating and carbonization.
If the same insulation is continued to be used for any
application, the progressive carbonization/ tracking will
ultimately lead to complete breakdown of insulation when an
arc bridges the gap between conductor and ground completely
[1].
With the rapidly increasing use of PWM inverter-fed
machines, detecting PD events at an early stage has become
very important for industries in order to avoid downtime
caused due to insulation failure [3-5]. Hence, PD detection is
seen as a preventive measure to check the health of insulation
before installation and/or as part of prediction of the
remaining life of insulation after a certain period of use [2].
 ‹ ,(((
John Agapiou Ronald M. Lesperance
john.agapiou@gm.com ronald.m.lesperance@gm.com
Research & Development
General Motors Technical Center
Warren, MI, USA
The PD signals have a high-frequency bandwidth,
ranging from few to hundreds of MHz. In the presence of
high-frequency noise, it becomes difficult to obtain a clean
PD detection. High-frequency noise, if present, also makes
it challenging to define thresholds for the PD levels hence
making PD detection scheme ineffective to detect smaller
PD events [6].
Some of the presently used PD detection techniques use
the integral of PD currents, in pico-Coulombs (pC) to
measure the PD signal magnitude by using an analog
oscilloscope [6], by calculating the charge transferred [9] or
by implementing a factory measurement system using a
shielded room [10]. However, through recent advancements
in PD testing measures along with better digital signal
processing technologies, PD was measured in millivolts
(mV) which avoided calibration problems experienced
while measuring in pC [11].
Many detection schemes have been proposed, such as
conventional detection of PD using a coupling capacitor [6-
8], detection based on the broadband impedance of the motor
[7], hi-pass filtering [8] etc. The physical attributes of
discharge such as electromagnetic radiation using RF antenna
[14 and 16-19], acoustic signals, visible and UV light
emissions have been used before to sense discharges within
the test insulation in many PD sensing techniques [6, 12 and
15]. Ultra wide band (UWB) detection, using wavelet
transform has also become popular in recent times to extract
PD from IGBT driven motors [15].
Figure 1: Impedance diagram of PD test setups which use voltage
measurements techniques

 It is not always possible to rely entirely on results from a to any particular frequency band. The Baker instrument
PD detection method based on voltage measurement at the model DT05-3 was used as impulse generator, producing
terminals of a machine to make a confident decision on the high voltage impulse with a very short rise time of 400 ns.
presence of PD. As depicted in fig. 1, measuring the terminal The output of the antenna along with the applied impulse
voltage during discharges and extracting information about voltage was analyzed on a high-resolution oscilloscope.
PD level using filter [16] is highly dependent on the internal
impedances of the high voltage source (impulse generator),
the connections, and the distributed inductances and
capacitances of motor under test as seen from point of
measurement [7].
On the other hand, PD measured by an antenna is unaffected
by these impedances, as it is not the terminal voltage that is
measured but the electromagnetic radiation from the discharge
currents taking place inside the insulation. In order to enhance
the reliability in the decision-making process for PD, in this
paper a new PD detection technique was developed, which
enables no contact detection of PD events by an antenna
Figure 2: Laboratory setup for PD detection using filter at stator ends
through grounding arrangements devised for the test setup.
These grounding arrangements ensure minimum system and
background noise during the test, enabling the antenna to
measure PD by sensing the electromagnetic radiation caused
by discharge currents without any interference.
PD detection techniques already published [17-19] have
provided ample evidence to support that antenna can be used
efficiently to recognize PD events by extracting the high-
frequency detections using wavelet transform. Some of these
used artificial defect models as laboratory setups by carrying
out either the fractional dimensions with energy parameters
[18 and 20], or feature extraction of PD (applicable to GIS and
transformer insulation) through kernel principal component
analysis (KPCA) [17], or by deploying 4 different types of
antenna to sense PD events. Figure 3: Grounded steel plate setup for reliable PD detection using a filter
and an antenna simultaneously.
The motive of this work was to develop an accurate and
repeatable standard technique to test actual medium voltage
To minimize the system noise and interference, a
stators for any presence of PD when tested by applying high
common grounding steel plate was placed as shown in fig.
voltage impulses. This method is to be used as standard PD
3. The plate served as a common grounding node for each
test to be conducted on stator alone, before assembling them
equipment used for testing.
with rotors units. Therefore, the developed method aims to
facilitate PD testing of motors at the end of the production
line and does not intend to serve as a predictive maintenance
tool. It was made sure that the developed method provides
high confidence to declare if a stator under test is affected by
the presence of PD or healthy to use for a normal operation.

II. LABORATORY SETUP

During the developing stages of the proposed detection

technique for medium voltage stator insulation, an antenna
was positioned at the axis of the stator to sense high-frequency
partial discharges occurring within the stator insulation as
shown in Fig. 2. As the procedure is to be developed for the
production line of motors and not for prognosis purposes, the
tests on stator were performed in a controlled environment
with minimum influence of surrounding objects.
The stators to be tested were rated at a low voltage of 300 Figure 4: Grounded steel plate setup for partial discharge detection using
V. To capture every detail of signal provided by the antenna, an antenna.
it was decided to examine the raw signal first, by not tuning it


 The connecting leads were kept short in length and were process as applied to raw antenna outputs confirms the
laid parallel to common grounding plate to minimize the effect presence of PD for the applied impulse voltage and presents
of interference due to inductive effect (Fig. 4). a reliable way to sense PD activity inside the stator under
test.
To collect the data from a large number of tests and analyze
them quickly and effectively, a data acquisition system was
set up using NI PXI 8119 controller to acquire desired signals
namely applied impulse voltage (L-G), ground current, raw
filter output and raw antenna output.

III. SIGNAL CONDITIONING IN LABVIEW

A. Extraction of Partial Discharge signal

A high voltage impulse, as shown in Fig. 5C, was applied

to the stator. The antenna senses the discharges occurring
around the time of the peak of impulse waveform by capturing
the electromagnetic radiation due to discharge current (Fig.
5B). The output signals from impulse generator and antenna
were digitized (0.5 G-samples/s) and then analyzed together
using LabVIEW, as shown in Fig. 5.

Figure 6: Raw PD response on antenna (A) and extracted PD signal from

antenna (B)

B. Second Level Denoising process

While testing stators for PD, on some occasions, the

antenna output even after UDWT processing, had some
spurious random peaks, similar to one shown in Fig. 7. As
these exceeded the threshold set for PD detection, and in
some cases resulted in false positives, it became necessary
to eliminate them. Alternative signal processing methods
were considered, but the simplest was based on the duration
of the pulse On observing that this noise lasted for at most
80 ns, which was much shorter than the actual duration of
peaks for PD, such cases were separated and rejected, based
on the duration of the peaks in the antenna output.

Figure 5: PD response from antenna (A) for applied impulse of 2.25KV (B)
from test setup shown in figure 2 and 3

As seen in Fig. 5, the peak of the applied HV impulse (plot

B) is around the 350th sample, whereas the antenna raw output
(plot A) appears too noisy, making it difficult to spot the PD
event, which roughly starts at that time. Therefore, the raw
signals from antenna were de-noised using Undecimated
Discrete Wavelet Transform (UDWT) to extract cleaner PD
signals, as shown in Fig. 6.
By applying multilevel UDWT (Daubechies 4) to the raw
Figure 7: Noisy peak in UDWT treated antenna output
antenna output (Fig. 6A), the antenna signal was decomposed
into 6 levels. The selected level 4 was adequate to extract the
IV. EXPERIMENTAL PROCEDURE
PD component from the raw antenna signal (Fig. 6B).
As seen from above, the raw antenna output (Fig. 6A) was
As the PD analysis was to be carried out for a number of
successfully denoised after UDWT operation (Fig. 6B) with
times on different kinds of stators, it was not feasible to
clear PD detection at the 350th sample. Hence, the UDWT


carry out the analysis manually looking at the waveforms after
each test. In order to automate the PD detection process, an
algorithm was programmed in LabVIEW (fig. 8), which can
process a large amount of test data and perform PD detection
analysis easily and in lesser time.
The stator under was tested for PD by applying a train of
rising impulses starting from few hundred volts up to 3 to 3.5
KV. After sampling all the outputs, the de-noised waveforms
of each, filter and antenna were compared with fixed
thresholds, which provided information about the presence of
PD.
Stepwise increment of Impulse voltage
across one phase of stator and ground
Data acquisition of
three signals
at 0.5 G.sa/sec
Ground
Impulse Current
Voltage Raw Antenna output
De-noising raw antenna
by DWT (Daubechies 4)
Threshold operation on denoised antenna output
Noise separation by
checking peak duration
Partial Discharge?
Yes/No
Figure 8: Flow Chart showing Signal Reconditioning process in LabVIEW.
Figure 9: Threshold reference levels for antenna output
To determine the threshold reference level for the filtered
outputs of the antenna, the maximum peak-to-peak value of
noise was calculated from the data from five different stators
below PD inception voltages for the antenna. After
considering a safety margin above the noise levels, the
threshold reference level for the antenna was set to 100 mV
(fig. 9).
As can be seen from fig. 9, the signal on antenna
represents a partial discharge signal with minimum
influence of noise, hence the developed algorithm in
LabVIEW was used in a NI PXI controller to analyze PD
events and make comparisons.
V. EXPERIMENTAL RESULTS
The algorithm developed in LabVIEW (Fig. 8) was
applied to the test data from 13 different stators tested within
a period of 8 months with variations of the phase under test,
axial antenna position and the inner diameter of the stator.
The stators were selected from trial production line with
internally and externally introduced defects to simulate PD
at lower voltages.
The PD detections out of the antenna were much cleaner
with the grounded steel plate in place as seen from fig. 11.
The level of noise in antenna output for test setup with no
steel plate was found to be higher than noise levels with
grounded steel plate. Hence, the modification in the test
setup resulted in cleaner PD signals as they were detected
by the antenna with minimum interference.
Figure 10: PD detection comparison for antenna with and without grounded
steel plate for stator A
An important factor considered for PD analysis is PD
inception voltage (PDIV), which in this case is solely
determined by the antenna output. The lowest value of
impulse voltage, which initiated a detection, was recorded.
Fig. 11 shows the results after UDWT processing for 20
successive tests trials conducted on one stator, plotted in
terms of PD inception voltages in order to examine the
consistency of the technique.

 As seen in Fig. 11, after PD detection using the UDWT
output, most of the trials show the inception voltage to be
around 2250 V. The irregularities similar to those observed in
the case of 4th and 14th trials were due to the presence of
spurious noise in antenna as explained in Fig. 7. Hence,
UDWT processing for extracting PD was not accurate alone
to provide high consistency for recording the PD inception
voltages.
Figure 11: PD Inception voltages before and after eliminating noise by
checking peak durations for stator A (20 trials)
In order to improve the PDIV decision made by the antenna,
peak duration algorithm was applied after UDWT processing
of antenna output, such that after threshold processing PD
signals extracted by the antenna, the duration of the peak was
checked.
Figure 12: PD Inception voltages for stator B (20 trials) after applying peak
duration criteria.
As seen from Fig. 11, the cases of noise, which lowered the
PDIV for trials 4 and 14 before applying peak duration
criterion (fig. 11), were eliminated. This made the PD
detection scheme more consistent. It can be seen that the
PDIV for this stator A is around the mean of 2200 V. Similar
trend is followed for similar comparisons for another stator B
in Fig. 12.
VI. CONCLUSIONS
The PD detections made using an antenna served as
a reliable detection technique for medium voltage stators.
The inception voltages for PD reported by this technique
show considerable accuracy in detection after denoising was
performed using both UDWT and pulse duration check on
antenna outputs. The paper represents a unique way to use
antenna for PD detection by placing it along with the
grounded metal plate with shorter connecting leads, which
helped to eliminate background noise.
Results can be further improved if the thresholds are
optimized for antenna output, depending on the type of
stator to be tested. The PD detection technique presented in
this paper provides higher reliability with respect to the PD
detection techniques based on voltage measurement, as it is
unaffected by the system impedances as seen from point of
measurement. Moreover, unlike artificial test setups used in
the past for PD detection, a practical PD detection technique
is presented based on performing tests on actual stators to
estimate their health accurately, with higher consistency and
confidence.
REFERENCES
[1] G. Paoletti, A. Golubev, “Partial discharge theory and applications to
electrical systems,” Conference Record of 1999 Annual Pulp and Paper
Industry Technical Conference (Cat. No.99CH36338), 1999, pp. 124-138.
[2] G. C. Stone, E. A. Boulter, I. Culbert, and H. Dhirani, “Electrical
insulation for rotating machines - design, evaluation, aging, testing, and
repair”. IEEE Press, John Wiley and Sons, Volume: 20, Issue: 3, 2004,
pp. 65 – 65.
[3] F. Guastavino, E. Torello, M. Di Lorenzo del Casale, L. Egiziano:
“Voltage distortion effects on insulation systems behavior in ASD
motors”, 2003 Annual Report Conference on Electrical Insulation and
Dielectric Phenomena, 2003, pp. 608-611.
[4] H. Okubo, et al., “Technical development on partial discharge
measurement and electrical insulation techniques for low voltage motors
driven by voltage inverters” IEEE Transactions on Dielectrics and
Electrical Insulation, Vol. 14, No. 6, 2007, pp.1516-1530.
[5] G. Stone, S. Campbell, and S. Tetreault, “Inverter-fed drives: which
motor stators are at risk?” IEEE Industry Applications Magazine, vol. 6,
no. 5, Sep/Oct 2000, pp. 17-22.
[6] G. C. Stone, “Partial discharge diagnostics and electrical equipment
insulation condition assessment,” IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 12, no. 5, Oct. 2005, pp. 891-904.
[7] Prabhakar Neti and Stefan Grubic, “Online broadband insulation
spectroscopy of induction machines using signal injection,” IEEE
Transactions on Dielectrics and Electrical Insulation, 2014 IEEE Energy
Conversion Congress and Exposition (ECCE), Pittsburgh, PA, 2014, pp.
630-637.
[8] Jinkyu Yang, Tae-june Kang, Byunghwan Kim, and Sang Bin Lee,
“Experimental evaluation of using the surge pd test as a predictive
maintenance tool for monitoring turn insulation quality in random wound
ac motor stator windings,” IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 19, no. 1, February 2012, pp. 53-60.

[9] R. Bartnikas, “Partial discharges. their mechanism, detection, and
measurement”, IEEE Transactions on Dielectrics and Electrical Insulation,
vol. 9, no. 5, Oct 2002, pp. 763-808.

[10] Yong-Joo Kim, Dong-Sik Kang, Don-Ha Hwang, Ioan Haiduc,

“Factory measurement system for sub pico coulomb partial discharge
signal,” Electrical Insulation Conference and Electrical Manufacturing and
Coil Winding Technology Conference (Cat. No.03CH37480), 2003, pp.
347-351.

[11] G.C. Stone, V. Warren, “Advancements in interpreting partial

discharge test results to assess stator winding condition,” IEEE-IAS/PCA
2004 Cement Industry Technical Conference (IEEE Cat. No04CH37518),
2004, pp. 61-65.

[12] IEEE guide to the measurement of partial discharges in rotating

machinery, IEEE Std 1434-2000, Aug. 15 2000, pp.1-64.

[13] G. C. Stone and V. Warren, "Advancements in interpreting partial

discharge test results to assess stator winding condition," IEEE-IAS/PCA
2004 Cement Industry Technical Conference (IEEE Cat. No04CH37518),
2004, pp. 61-65.

[14] B. Sarkar, D. K. Mishra, C. Koley and N. K. Roy, "Microstrip patch

antenna based UHF sensor for detection of Partial Discharge in High
Voltage electrical equipments," 2014 Annual IEEE India Conference
(INDICON), Pune, 2014, pp. 1-6.

[15] L. Angrisani, M. Di Lorenzo del Casale, M. D'Arco and C. Petrarca,

"Extraction of PD data from UWB measurements on motor stator bars fed
by an IGBT inverter," Annual Report Conference on Electrical Insulation
and Dielectric Phenomena, 2002, pp. 844-847.

[16] Duan Dapeng; Teng Letian; Qian Yong; Wang Jiang; Huang Chengjun;
Zeng Yi; Jiang Xiuchen, “Experimental study and feature extraction on
UHF partial discharge detection for typical model in air,” 2008 International
Conference on Condition Monitoring and Diagnosis, Beijing, 2008, pp.
1040-1043.

[17] Jian Li; Tianyan Jiang; Robert F. Harrison; Stanislaw Grzybowski,

“Recognition of ultra high frequency partial discharge signals using multi-
scale features,” IEEE Transactions on Dielectrics and Electrical Insulation,
Volume: 19, Issue: 4, August 2012, pp.1412 - 1420

[18] G. Robles, M. Sánchez-Fernández, R. Albarracín Sánchez, M. V.

Rojas-Moreno, E. Rajo-Iglesias and J. M. Martínez-Tarifa, "Antenna
Parametrization for the Detection of Partial Discharges," in IEEE
Transactions on Instrumentation and Measurement, vol. 62, no. 5, pp. 932-
941, May 2013.

[19] T. Jiang, J. Li, M. Chen and S. Grzybowski, "Recognition on ultra-

high-frequency signals of partial discharge by support vector machine,"
2010 International Conference on High Voltage Engineering and
Application, New Orleans, LA, 2010, pp. 634-637.