SCIENCE CHINA

Technological Sciences
• RESEARCH PAPER • January 2013 Vol.56 No.1: 262–270
doi: 10.1007/s11431-012-5040-1

Study of partial discharge characteristics at repetitive square

voltages based on UHF method
WANG Peng, WU GuangNing*, GAO Bo, LUO Yang & CAO KaiJiang
School of Electrical Engineering, Southwest Jiaotong University, Chengdu 610031, China

Received June 26, 2012; accepted August 27, 2012; published online October 10, 2012

Due to overvoltage produced by inverter output, inverter-fed motor insulation systems often experience fast electrical aging
process, especially when partial discharge (PD) is incepted. Before putting into use, the PD detection should be performed on
inverter-fed motors at repetitive square voltages to avoid the PD caused insulation deterioration when the motors are collected
to inverters having specific characteristics. However, unlike PD tests at AC/DC voltages proposed in IEC 60270, the PD detec-
tion at repetitive square voltages is much more complex because of serious interference generated by impulse generator. To
solve the problem, ultra-high frequency (UHF) method seems recommendable for its preferable signal-to-noise ratio (SNR).
The chief aim of this study is to investigate PD pulse and statistical characteristics of turn-to-turn insulation for inverter-fed
traction motors. A square-shaped Archimedes antenna, specially designed for the PD detection at repetitive square voltages of
fast rise times, was used to perform PD tests on turn-to-turn insulation models. Time and frequency analysis results indicate
that energy component of generator disturbance and PD pulses are mainly distributed in the 0–0.5 GHz and 0.6–1.5 GHz range,
respectively. Based on the results, suitable filter was designed for power disturbance suppression. Additionally, resorting to the
sensor unit (i.e. antenna and filter) and the PD test system, the PD statistical features at square voltages of different frequencies
were obtained. Experimental results show that higher frequency will give rise, statistically, to PD of lower magnitudes distrib-
uting at smaller phases. A reasonable interpretation of this phenomenon was presented. Lastly, according to the PD statistical
features, some suggestions for the PD detection system design, generator parameter optimization and the PD pulse extracting were
given. The results of this work would be beneficial to the increase of the sensitivity when performing the PD detection on insula-
tion systems for inverter-fed motors at repetitive square voltages and thus, improving the reliability of inverter-fed motors.

inverter-fed motors, insulation system, partial discharge, UHF, square voltage, antenna, filter

Citation: Wang P, Wu G N, Gao B, et al. Study of partial discharge characteristics at repetitive square voltages based on UHF method. Sci China Tech Sci,
2013, 56: 262270, doi: 10.1007/s11431-012-5040-1

1 Introduction partial discharge (PD) in stator insulation systems [1–3],

especially in the positions shown in Figure 1. For low-
voltage (<700 V) inverter-fed motors, insulation degrada-
Inverter-fed motors, fed by pulse width modulation (PWM)
tion rate could be very high when PD is incepted [4–7], be-
inverters, are used in more and more industry applications,
cause the insulation is generally realized using purely or-
because of many outstanding advantages, such as energy
ganic materials (e.g. polyimide-imide). During the applica-
efficiency and good performance of adjustment. Overvolt-
tion of inverter-fed motors, many unexpected insulation
age caused by impedance mismatch (between supply cable
failures have been reported in recent years [4, 8–10].
and motor) and uneven potential distribution could induce
To solve the above problems, the International Electro-
technical Committee (IEC) proposed measurements that
*Corresponding author (email: [email protected])

© Science China Press and Springer-Verlag Berlin Heidelberg 2012 tech.scichina.com www.springerlink.com
 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1 263

This study aims to investigate PD and disturbance pulse

characteristics when performing the PD detection on turn-
to-turn insulation (for high voltage form wound motors) at
repetitive square voltages with fast rise times. Using Ar-
chimedes antenna and PD detection system specially de-
signed, the PD measurements were carried out on turn-to-
turn insulation used for high-speed railway traction motors.
Time and frequency features of PD and disturbance pulses
were analyzed respectively. Based on the frequency features,
digital filters were designed and applied to suppress gener-
ator disturbance. Statistical results of the PD tests at square
voltages of different frequencies show that higher frequency
give rise to the PD pulses of smaller magnitudes and ap-
pearing phases. Furthermore, based on the PD physical
Figure 1 Typical PD positions inner stator slot of inter-fed motor. mechanism, the explanation for this phenomenon was given.
Lastly, according to the PD statistical features provided in
this study, some suggestions were proposed for the PD de-
should be carried out on inverter-fed motor stators, in order
tection systems design, optimum generator parameters se-
to verify that the Partial Discharge Inception Voltage (PDIV)
lection and the PD pulse extracting when the PD detection
and Repetitive Partial Discharge Inception Voltage (RPDIV)
was performed on inverter-fed motors at repetitive square
are high enough when the motors are connected to inverters
voltages.
with specified characteristics [11, 12]. According to the
related standards of IEC, when performing the PD tests on
inverter-fed motors, motor stator insulation systems should 2 Experimental setup and sample preparation
be stressed by impulsive voltages to reproduce surge volt-
ages, similar to the voltage applied to motor stator insula-
2.1 PD detection system
tion.
However, compared with the PD tests under AC and DC The PD detection system used in this study is sketched in
voltage conditions, PD detection at repetitive square volt- Figure 2. R1 and R2 denote the voltage dividers (ratio 1000:
ages becomes much more complex because of serious dis- 1, 50 MHz bandwidth), whose output was connected to one
turbance induced by impulse generators [13, 14]. PD is of- channel of a high speed digital oscilloscope (Tektronix
ten submerged in the strong disturbance, indicating that DPO7254C, maximum sample rate 10 Gs s1, 2.5 GHz
special measures should be taken to extract the PD signal bandwidth). Cx is turn-to-turn insulation sample, made out
from power noise. of electromagnetic wires insulated by corona-resistance
In order to detect PD with high signal-to-noise ratio films manufactured by DuPont company. An Archimedes
(SNR), i.e. suppressing the disturbance from generators, antenna (see Figure 3(a)), whose output is connected to an-
some improved methods were proposed, including improv- other channel of the oscilloscope, was used to receive the
ed high-frequency current transformer [15], ultra-high fre- PD signals. The gain profile of the antenna is reported in
quency (UHF) antenna [13], digital signal processing tech- Figure 3(b) to increase SNR of the PD detection system at
nologies [14, 16] and optical test method [8, 17]. Among impulsive voltages [13], the response gain of the antenna is
these methods, UHF method is a feasible choice because of higher than 3.5 dB in the high frequency range (>1.5 GHz).
its high capability of de-noising disturbance [7, 13, 18]. The length, width and height of the Archimedes are 6, 6
To design high performance antennas used in the UHF and 4.5 cm, respectively. The antenna was placed at a dis-
tests, the time and frequency features of PD and generator tance of 15 cm from and directed at the sample.
disturbance should be known. However, the PD frequency The data recorded by the oscilloscope was transferred to
features are dependent on many factors such as sample personal computer (PC) cycle by cycle utilizing a General
types [19], voltage parameters [19, 20] and aging process Purpose Interface Bus (GPIB) card and IEEE 488.2 com-
[21]. As a result, before performing the PD tests on speci- munication protocol.
fied models, its frequency features should be investigated to
choose appropriate antenna and increase SNR as high as
2.2 Generator waveform
possible. However, at impulse voltages of fast rise times,
although some PD detections on twisted-pairs/crossed-pairs The repetitive square voltage generator is designed using 8
samples for low voltage random wound motor insulation are series Insulated Gate Bipolar Transistors (IGBT). It can
reported [15], less work has addressed the PD characteris- produce bipolar impulsive voltages from 500–7000 V
tics of high voltage form wound insulation models. (peak-to-peak). The minimum rise time of the waveform is
264 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1

Figure 2 Sketch of PD detection system.

Figure 4 Waveform output by the generator (peak-to-peak voltage 2000
V, frequency 100 Hz, duty cycle 50%). (a) Square waveform output by the
generator; (b) rise time of the waveform.

3 Time and frequency features of PD and dis-

turbance pulses

3.1 At low voltage

Firstly, the PDIV tests were performed on the insulation

models. Experimental results show that the PDIV value of
this type of samples is around 2.05 kV. In order to avoid PD
and disturbance overlapping with each other at high voltage
[15], a voltage, 2.2 kV, not much higher than the PDIV
value, was stressed on the sample. The frequency of the
square voltage was 50 Hz. The antenna output signal was
reported in Figure 5. It can be seen that disturbance, which
was at the rise time of impulsive voltage, had a value
around 300 mV and, PD was incepted at around 40 μs after
the rise time, with a value around 200 mV.
The time waveform and frequency spectra of the PD
Figure 3 Archimedes antenna and its gain profile. (a) Archimedes an-
tenna; (b) antenna gain profile. pulse and disturbance are reported in Figures 6(a) and 6(b),

100 ns (maximum slew rate, 70 kV μs1), duty cycle 50%

and frequency 2–20 kHz. The output waveform of the gen-
erator is reported in Figure 4.
Due to impedance mismatch between the cable and sam-
ple, overvoltage occurs at the rise and fall times of the
square voltage waveform (see Figure 4(a)). The overvoltage
ratio k can be evaluated by
Vov
k  100%,
Vf

where Vov and Vf are overvoltage value at rise/fall time and

voltage magnitude at flat part. In this study, k≈15%. Figure 5 PD and disturbance pulses (voltage 2.2 kV, frequency 50 Hz).
 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1 265

Figure 6 Time and frequency analysis of PD and disturbance.

respectively. Clearly, both the PD pulse and disturbance

have much energy component in the low frequency range
(around 100 MHz). It can be known, in Figure 6(b), that the
disturbance energy is mostly distributed in 0–300 MHz
range. While PD has considerable energy component in 300
MHz to 1 GHz range as shown in Figure 6(a).

3.2 At high voltage

When voltage was 3.2 kV (frequency 50 Hz), the signal

received by the antenna was reported in Figure 7. It can be
seen that here the PD pulse overlaps with disturbance. It is
hard to distinguish PD from disturbance. In order to obtain
time and frequency features of the overlapped signal, short-
time Fourier transformer (STFT) calculation was used,
Figure 8 STFT results of the overlapped signal in Figure 7.
whose results were reported in Figure 8. We can see that the
energy of both PD and disturbance is distributed in the
range of 0–0.4 GHz, while only PD has considerable energy 3.3 Digital filter design
component when frequency is higher than 0.4 GHz (around).
The results are in agreement with the results shown in Fig- According to the above time and frequency analysis results,
ure 6, which means that the PD and disturbance frequency digital high-pass filter can be designed to suppress the dis-
features do not change significantly with square voltage turbance from generator. A sixty-orders Butterworth digital
increasing. filter with 350 MHz pass band and stop band attenuation of
80 dB was designed. The PD signal (reported in Figure 7)
after passing from the filter is shown in Figure 9. Clearly,
after suppressing disturbance in the range of 0–350 MHz,
the output signal energy is mostly distributed in 450–1.2
GHz.

4 PD features at different frequencies

Before performing the PD detection on inverter-fed motors,

impulse voltage frequency is one of the most important pa-
rameters and should be considered. However, the IEC
standards only suggest the range of the frequency. For ob-
jective evaluation of insulation status, suitable frequency
Figure 7 PD and disturbance pulses (voltage 3.2 kV, frequency 50 Hz). should be selected.
266 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1

ures 10(b)–10(d)).

4.2 PD delay time and magnitude

In order to get statistical features of the PD delay time and

magnitude at different frequencies the box charts of delay
time and magnitude were reported in Figure 11. In Figure
11(a), it can be known that, indeed, the PD delay time de-
creases significantly with the increasing of frequency.
While in Figure 11(b), the average magnitude of 2 Hz is
little smaller than 200 Hz. When the PD patterns were
reported in Figures 10(a) and 10(b), it can be known that
this phenomenon was caused by the influence of oscillating
voltage, indicating that the firing voltage, i.e. the instanta-
neous voltage magnitude when PD occurs, could affect the
Figure 9 PD time pulse and spectrum after applying high-pass filter. PD magnitude significantly. For giving further support to
this influence, Figure 12 shows all the PD peak vs. firing
voltage recorded in Figure 10. It is clear that, statistically,
In order to study the PD statistical characteristics at
the PD magnitude increases with the increasing of firing
square voltages of different frequencies. The PD measure-
voltage.
ments were carried on at repetitive square voltages of 4
frequencies, i.e. 2, 20, 200, 2000 Hz. During each test, the
PD signals output from the high pass filter was saved into 5 Discussions
PC resorting to the GPIB interface card and virtual instru-
ment technology. To get statistical features of the PD pulses,
5.1 Interpretation of PD statistical features
at least 500 cycles’ data was saved into database in each
test. For the PD present inner insulation bulk or along insulation
surface, two conditions must be satisfied: (1) The electric
4.1 PD patterns field in the gas should be greater than PD inception electric
field (Einc); (2) initial electrons should present for an ava-
The phase resolved PD (PRPD) patterns at four voltage lanche generating one PD [22].
frequencies are reported in Figure 10. It can be seen that PD Assume that PD occurs two times in one same cycle of
delay time (the time between PD appearing position and 0 square voltage, as shown in Figure 13. Just after the first PD,
voltage) and magnitude decrease dramatically with fre- the sketch of electric field and charge distribution is shown
quency increasing. At low frequency (Figure 10(a)), most in Figure 14(a), at this time,
PD occurs at the flat part of the waveform, while almost all
PD is distributed at the rise time at higher frequencies (Fig- Ei  Eres  E0  Eq , (1)

Figure 10 PD Patterns at different frequencies. (a) Frequency 2 Hz; (b) frequency 20 Hz; (c) frequency 200 Hz; (d) frequency 2000 Hz.
 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1 267

decay caused by, for example, neutralization by drifting

along the cavity walls and diffusion/hopping into the die-
lectric bulk [24], makes that, when the second PD appears,
the electric field Eq will be
tdecay

Eq  Eq e 
, (2)

where  is relaxation time, a constant evaluating the charge

relaxation process, tdecay is the time after the first PD. If
PD presents at the rise time and fall time of square voltage
(as shown in Figure 13) respectively, tdecay can be expressed
by,
1
tdecay  , (3)
2f

where f is the frequency of repetitive voltage.

When the second PD occurs, the electric field inside the
gas is shown in Figure 14(b). Because of the voltage polari-
ty inversion, here, the electric field Ei' becomes,
Ei  E0  Eq . (4)

Using eqs.(2) and (3), eq. (4) can be written as,

1

Ei  E0  Eq e 2 f
. (5)

From eqs. (3) and (5), it can be concluded that, higher fre-
quency will make the decay time shorter and, consequently,
Figure 11 PD delay time and magnitude box charts (voltage 3.5 kV). (a)
PD delay time; (b) PD magnitude.
the decaying of surface charge produced by the first PD will
become smaller. When the second PD appears, it just needs
a relatively small background electric field and, the total
electric field inside the gas would reach the PD inception
electric filed.
Moreover, when considering the initial electrons needed
by PD, the electrons emission probability P can be estimat-
ed by
P  1  e   t , (6)

where  and t are electron hazard rate and evaluated time,

respectively. After the first PD, initial electron emission is

Figure 12 PD firing voltage vs. PD magnitude.

where Ei, Eres, E0 and Eq are total electric field, residual elec-
tric field after PD, background electric field in absence of
discharge and surface charge induced electric field, respec-
tively. Because of the surface charge produced by the first
PD, the total electric field in the gas drops to Eres, which is
very small but not zero [23]. Thereafter, the surface charge Figure 13 PD occurs in a same square voltage cycle.
268 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1

can be expressed by the following equation [23]:

   a  
 q   0 2 1   r  K    1  ( Ei  Eres ), (8)
   b  

where q is the PD charges; K(a/b) the size parameter of the

PD position; r and 0 permittivity in vacuum and relative
permittivity respectively. Ei is the external electrical field
when PD occurs and Eres is the residual electrical field after
one PD. Clearly, PD magnitude largely depends on the in-
stantaneous electrical field Ei, determined by firing voltage,
when PD occurs. Therefore, the different magnitudes
caused by frequencies can be explained.

5.2 Suggestions to PD detection system design

According to the related standards, the PD detection should

be performed on insulation systems of inverter-fed motors,
to make sure that PDIV and RPDIV are high enough when
the motor is connected to a inverter. PDIV is the minimum
voltage value of the first PD at repetitive square voltage,
and RPDIV the minimum voltage value of the repetitive
PD.
To obtain objective PDIV and RPDIV values, the sensi-
tivity of the PD detection system should be high enough, in
other words, for the PD detection at repetitive square volt-
ages with fast rise times, the power disturbance should be
Figure 14 Electric field and charge distribution during PD in one same suppressed to a very low value, in order to make sure that
voltage cycle.
PD can be extracted from power noise signals. When re-
sorting to the UHF test method, this can be achieved by:
mainly produced by Schottky-Richardson emission, which (1) Optimizing the sensor and filter design. UHF sensor
can be estimated by [22] should have better PD pulse coupling performance accord-
ing to the frequency spectrum of PD and disturbance, i.e.
E
  qe
4  0
the antenna should receive the energy from PD pulses as
dn(t ) 
much as possible, other than the noise from the generator.
  n(t ) 0 e KT
, (7)
dt Additionally, the high-pass filter must be suitable for in-
creasing the SNR.
where n(t) and E are the space charge and electrical field at
(2) Selecting the most favorable square voltage generator,
the PD position;  is the height of the emission barrier (in
in order to excite PD at low square voltage and make PD
eV); ν0 the base phonon escape frequency; qe the electron
pulses detectable.
charge; K, T Boltzmann constant and temperature respec-
For (1), according to the frequency features of PD and
tively. At higher frequency, n(t) has larger value because of
generator disturbance reported in Section 3, the antenna
smaller decaying, therefore, according to eqs. (6) and (7),
should have higher response profile in the range of 0.5–1.5
there will be larger initial emission probability and thus,
GHz, where the energy of power disturbance is very small
easier PD exciting.
while parts of PD pulse energy are distributed. For de-
From the above discussion, it can be singled out that, at
noising the residual disturbance, the pass-band of the high-
higher frequency, there will be higher electric field induced
pass filter should be around 0.4–0.5 GHz.
by surface charge and higher initial electrons emission
As reported in Figure 10, higher frequency will give rise
probability for exciting succeeds PD. These two factors will
to PD pulses of smaller magnitudes and appearance phases.
cause the firing voltage to decrease, as shown in Figure 10.
Table 1 lists the minimum PD magnitude at 4 frequencies. It
In Figure 12, the PD magnitude increases with the in-
can be seen that, at frequency 2000 Hz, the sensitivity of the
creasing of firing voltage. It is well known that the PD
PD test system should be higher than 0.1350 mV in order to
magnitude is strongly correlated to the electric field in the
detect the smallest PD pulses.
PD position at the moment the initiatory electron appears
Moreover, at higher frequency, the PD pulses are easier
[25, 26]. The apparent discharge during one PD in a cavity
to overlap with power disturbance, as shown in Figure 15.
 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1 269

Table 1 The minimum PD magnitude at different frequencies According to the above results, when designing the PD
No. Frequency Minimum PD magnitude detection systems used at square voltages and selecting the
1 2 0.3038 frequency of the square voltage generator for the PD detec-
2 20 0.2288
tion on inter-fed motor insulation systems, in order to get
3 200 0.1800
higher SNR, the following suggestions should be followed.
4 2000 0.1350
(1) Antenna should have high gain in the range of
0.5–1.5 GHz to increase SNR, i.e. avoiding the strong en-
ergy components from generator interference.
(2) Relatively low frequency square voltage (<2000 Hz)
is recommended for exciting the detectable PD pulses with
higher phases and larger magnitudes.

This work was supported by the National Natural Science Foundation of

China (Grant Nos. 51177136, 50377035) and Zhuzhou Electric Motor
Company of South China Locomotive ＆ Rolling Stock Corporation Lim-
ited.

1 Erik P. Transient effects in application of PWM inverters to induction

motors. IEEE T Ind Appl, 1992,28(5): 1095–1101
2 Stone G C, Heeswijk R G, Bartnikas R. Investigation of the effect of
Figure 15 PD pulse overlapping with disturbance (frequency 2000 Hz, repetitive voltage surges on epoxy insulation. IEEE T Energy Conver,
generator peak-to-peak voltage 3.0 kV). 1992, 7(4): 754–760
3 Wheeler J C G. Effects of converter pulses on the electrical insulation
in low and medium voltage motors. IEEE Electr Insul M, 2005, 21(2):
In this case, if the sensitivity of the PD detection system is 22–29
not high enough, PD would be missed, consequently, the 4 Yin W J. Failure mechanism of winding insulations in inverter-fed
higher PDIV or RPDIV values would be obtained. There- motors. IEEE Electr Insul M, 1997, 13(6): 18–23
fore, for (2), the frequency of the square voltage generator 5 Zhou K, Wu G N, Guo X X, et al. Partial discharge activity of
should be not very high (e.g. less than 2000 Hz), in order to enameled wires under different parameters of square wave voltage.
IEE J T Fund Mater, 2008, 129(12): 922–930
make PD appear after a short time (e.g. 0.5 s) from the 6 Hayakawa N, Morikawa M, Okubo H. Partial discharge inception and
disturbance, with relatively large magnitude. In this way, propagation characteristics of magnet wire for inverter-fed motor
PD is easier to be distinguished and extracted. under surge voltage application. IEEE T Dielect El In, 2007, 14(1):
39–45
7 Tozzi M, Cavallini A, Montanari G C. Monitoring off-line and
6 Conclusions on-line PD under impulsive voltage on induction motors—Part 1:
Standard procedure. IEEE Electr Insul M, 2010, 26(4): 16–26
8 Hayakawa N, Shimizu F, Okubo H. Estimation of partial discharge
In this study, an Archimedes antenna was specially de- inception voltage of magnet wires under inverter surge voltage by
signed for the PD detections at repetitive square voltages volume-time theory. IEEE T Dielect El In, 2012, 19(2): 550–557
with fast rise times, resorting to which the time and fre- 9 Zhou L R, Wu G N, Gao B, et al. Study on charge transport
mechanism and space charge characteristics of polyimide films. IEEE
quency features of PD and generator interference were in-
T Dielect El In, 2009, 16(4): 1143–1149
vestigated. The PD tests results, carried out on turn-to-turn 10 Kaufhold M, Borner G, Eberhardt M, et al. Failure mechanism of the
insulation models, indicate that when the rise time of the interturn insulation of low voltage electric machines fed by pulse-
square voltage is around 100–200 ns, the energy of the controlled inverters. IEEE Electr Insul M, 1996, 12(5): 9–16
power disturbance and PD pulses is distributed in the range 11 IEC 60034-18-41: Rotating electrical machines—Part 18-41:
of 0–0.4 GHz and 0–1.5 GHz, respectively. Based on the Qualification and type tests for Type I electrical insulation systems
used in rotating electrical machines fed from voltage converters, 2006
obtained frequency domain features, the pass band of the
12 IEC TS 60034-18-42: Rotating electrical machines—Part 18-42:
high-pass filter should be around 0.4–0.5 GHz. Qualification and acceptance tests for partial discharge resistant
From the experimental results on turn-to-turn insulation electrical insulation systems (Type II) used in rotating electrical
models of inter-fed motors under square voltages of differ- machines fed from voltage converters, 2008
ent frequencies, it can be concluded that, higher frequency 13 Fabiani D, Cavallini A, Montanari G C. A UHF technique for
advanced PD measurements on inverter-fed motors. IEEE T Power
(e.g. 2000 Hz) can give rise to the PD pulses with lower
Electr, 2008, 23(5): 2546–2556
magnitudes close to generator disturbance (i.e. low phases). 14 Lindell E, Bengtsson T, Blennow J, et al. Evaluation of a PD
These different statistical features can be explained by that, measuring system for repetitive steep voltage waveforms. IEEE T
at higher frequency, shorter decay times of the PD produced Dielect El In, 2011, 18(1): 246–255
surface charge can cause larger electric field and higher 15 Guastavino F, Diadano A, Torello E. Measuring partial discharges
initial electron emission probability. under pulsed voltage conditions. IEEE T Dielect El In, 2008, 15(6):
270 Wang P, et al. Sci China Tech Sci January (2013) Vol.56 No.1

1640–1648 21 Nyamupangedengu C. Time-varying partial discharge spectral

16 Lindell E, Bengtsson T, Blennow J, et al. Influence of rise time on characteristics in solid polymer insulation. IET Sci Meas Technol,
partial discharge extinction voltage at semi-square voltage waveforms. 2012, 6(2): 85–95
IEEE T Dielect El In, 2010, 17(1): 141–148 22 Niemeyer L. A generalized approach to partial discharge modeling.
17 Kimura K, Ushirone S, Koyanagi T, et al. PDIV characteristics of IEEE T Dielect El In, 1995, 2(4): 510–528
twisted-pair of magnet wires with repetitive impulse voltage. IEEE T 23 Gutfleisch F, Niemeyer L. Measurement and simulation of PD in
Dielect El In, 2007, 14(3): 744–750 epoxy voids. IEEE T Dielect El In, 1995, 2(5): 729–743
18 Florkowska B, Roehrich J, Zydroń P, et al. Measurement and analysis 24 Montanari G C. Bringing an insulation to failure : the role of space
of surface partial discharges at semi-square voltage waveforms. IEEE charge. IEEE T Dielect El In, 2011, 18(2): 339–363
T Dielect El In, 2011, 18(4): 990–96 25 Illias H A, Chen G, Lewin P L. Partial discharge behavior within a
19 Nyamupangedengu C, Jandrell I R. Partial discharge spectral spherical cavity in a solid dielectric material as a function of
response to variations in the supply voltage frequency. IEEE T frequency and amplitude of the applied voltage. IEEE T Dielect El In,
Dielect El In, 2012, 19(2): 521–32 2011, 18(2): 433–443
20 Hammarström T, Bengtsson T, Blennow J, et al. Evidence for 26 Forssén C, Edin H. Partial discharges in a cavity at variable applied
changing PD properties at short voltage rise times. IEEE T Dielect El frequency part 2 measurements and modeling. IEEE T Dielect El In,
In, 2011, 18(5): 1686–1692 2008, 15(6): 1610–1616