Experimental Study of Partial Discharge Pulses

Propagation in Stator Winding of Hydro Generator

Yuanlin Luo Zhaohui Li Tian Chen

School of Hydropower and State Key Laboratory of Advanced State Key Laboratory of Advanced
Information Engineering Electromagnetic Engineering and Electromagnetic Engineering and
Huazhong University of Science and Technology Technology
Technology Huazhong University of Science and Huazhong University of Science and
Wuhan, China Technology Technology
ylluo_hust@FRP Wuhan, China Wuhan, China
][email protected] [email protected]

Abstract—The propagation of PD pulses through the winding, of researchers in this area, e.g. Wilson et al.[5], Henriksen et
from the discharge site to the sensor location, is influenced by the al[6], Cheng et al.[7] and Pemen et al.[8], have approached this
frequency of signal and the geometry of generator, which is of problem from a variety of experimental and analytical points
critical importance for accurate measurement of the PD signals. of view. While most of the previous researchHV were carried
The previous researches on PD propagation were mainly made out on turbine generator or motor, rather than large hydro
on turbine generator or motor, little work had been done on generator with a relatively more complicated geometry.
hydro generator. To address this concern, an experimental study
of the propagation of PD pulses in the stator winding of a large Fortunately, since the owner decided to uprate one large
hydro generator is presented in this paper. In the test, the hydro generator after more than 30 years of operation, which
insulation outside slot exit of 22 bars was removed to directly would replace the stator core and stator winding, we had the
inject PD pulses to the inner conductor. The results indicated PD opportunity to investigate propagation characteristics of high
signals manifest two modes when propagating in stator winding frequency signals in stator winding through destructive test,
of large hydro generator, and 5MHz is the corner frequency of other than non-intrusive test carried out by previous researchers.
them. The coupling between bars in slot or overhang will cause It is the purpose of the present work to provide insight into the
crosstalk between phases, which obscures PD source location. propagation of PD pulse in stator winding of large hydro
generator.
Keywords—hydro generator; stator winding; partial discharge;
attenuation; crosstalk
II㸬 TEST SET_UP
Iˊ INTRODUCTION
A. Test Generator
As key equipment of electrical power system, the reliable To study the propagation of PD signals in the stator
operation of the generators rely critically on integrity of its winding of the hydro generator, we performed the test on a 34
stator winding insulation. Partial discharge (PD) is known as year-old 13.8kV hydro generator with the specification in
both a main cause and an indicator of stator insulation TABLE I. The stator winding of this generator was a wave-
deficiencies. Therefore, partial discharge (PD) measurement chorded coils, double-layer winding with a slot pitch of 1-8-17,
aiming at monitoring the stator winding insulation condition is i.e. a distributed fractional-pitch winding. The tests were made
of critical importance. Over the past decades, a number of with the terminal disconnected and parallels connected.
techniques have developed for detecting partial discharge by
measuring the discharge pulse flowing at the terminals or
neural point of the generator[1]-[3]. TABLE I. GENERATOR SPECIFICATION

PD is a localized electrical discharge that only partially Parameter Value

bridges the insulation between conductors, which can or cannot Rated Voltage/UN 13.8 kV
occur adjacent to a conductor[4]. It is impossible to measure a Rated Power/PN 125 MW
PD event directly, remote sensors will always record pulses Core bore/DN1 15000 mm
that are influenced by the propagation along some distance Slots No./Z 792
from the PD origin. A PD pulse has a wide frequency spectrum
Bars/Slot 2
from DC up to GHz, which will propagate through the stator
winding via various paths. Thus, it’s essential to understand the Slot length/l1 1590 mm
effect of PD propagation on the measured PD event, for the Parallels/a 3
purpose of accurate PD measurement and location. A number Insulation class B

978-1-5090-6414-4/17/$31.00 ©2017 IEEE 773

B. Signal Detector In the test, we only tested on the 2nd parallel of phase V,
Although a PD pulse has a rise time of only a few ns at its because nine parallels of three phases are similar in geometry
origin and consequently a wideband spectrum from DC up to and construction. We picked one bar out of every eight bars so
GHz[9], practical measurement shows that the energy totally twenty two bars were selected as test bars. The selected
concentrated below 20MHz after propagation from PD site to bars were listed in TABLE II.
sensor location[10]. Therefore, test systems with high lower As mentioned above, a PD pulse has a rise time of only a
cut-off frequency tend to neglect signals from inner portions of few ns, so it was simulated by an electronic discharger. The
the stator winding. simulated PD pulse is sketched in Fig.2. As can readily be
For the tests described below, three coupling capacitors deduced from Fig.2, the pulse has a rise time of 3ns and a
mounted at generator three phase terminals and one connected duration of 500ns.
to neutral point were used, the measurement frequency range of
which are between 50kHz and 20MHz. The outputs of four
capacitors were recorded on a Tektronix TDS2022B, 200MHz
oscilloscope and a 100MHz, NI-USB5133 computer controlled
digitizer through 50Ω coaxial-cable. Fig.1 illustrates the test
set-up, where LA is arc suppression coil, CC represents the
coupling capacitor.

Fig. 2. Simulation PD pulse

By removing insulation of selected test bars of the 2nd

parallel of phase V, the simulated PD pulses were directly fed
in conductors using an electronic discharger. Fig.3 shows the
procedure of pulses injection.

Fig. 1. Test Set-up

Fig. 3. Injection of simulated PD pulses to conductor directly

C. Injection of Simulated PD Pulses
As listed in TABLE I, the test generator has 792 slots with D. Measurement Procedure
double-layer bars per slot, 3 parallels per phase, then can
readily deduce that every parallel has 176 bars labeled. We had The responses at three high voltage terminals and at the
characterized each stator bar with its phase, parallel and neutral point of the generator could be measured with
number. For instance, bar V2-5 is the 5th bar of the 2nd Tektronix TDS2022B or NI-USB5133. It must be emphasized
parallel of phase V from neutral point. that there are only two input channels of them, which is less
than the number of signals, and that, consequently, the four
response signals cannot be measured simultaneously. In an
TABLE II. TEST BARS attempt to address this concern, we made the first channel
detect response signal from coupling capacitor mounted on the
Series Bar No. Slot No. Series Bar No. Slot No.
neutral point all the time, while signals of other three capacitors
1 V2-2 292 12 V2-89 449 mounted on the three phase terminals shared the second
2 V2-9 253 13 V2-98 409 channel. In other words, we firstly measured the response
3 V2-17 317 14 V2-105 466 signal of phase U by the second channel of Tektronix
4 V2-26 277 15 V2-114 458 TDS2022B, then phase V and phase W
5 V2-33 334 16 V2-121 483
6 V2-42 326 17 V2-138 443 IIIˊ RESULTS AND DISCUSSION
7 V2-49 351 18 V2-146 377
A. Winding coverage
8 V2-58 375 19 V2-154 311
Fig.4 shows the responses of coupling capacitors to the PD
9 V2-65 368 20 V2-162 245 pulses injected at three bars under test. The results demonstrate
10 V2-74 424 21 V2-169 236 that the coupling capacitor used here is capable of detecting PD
11 V2-81 385 22 V2-173 268 throughout the entire stator winding.

774
B. Analysis of propagation mode respectively, when a PD pulse was injected to the bar V2-26.
As can be deduced from Fig.5, the responses of capacitors The fast mode contains mainly the higher frequency signal, and
from both terminal and neutral show two modes, i.e. the fast the slow mode contains the lower frequency signal. The fast
mode and the slow mode. Fig.5a and Fig.5b show the response mode arrives at the capacitor ahead of the slow mode, which
at the high voltage terminal and neutral point of phase V means a shorter time delay from discharge site to sensor
location. In addition, the unit of y axes in Fig.5 is mV.

(a) inject to V2-2

(b) inject to V2-121

(c) inject to V2-169

Fig. 4. Response of four coupling capacitors to simulated PD pulses injected to different bars of the 2nd parallel of phase V

300 300
Raw signal

Raw signal

0 0

-300 -300
300 300
Slow Mode

Slow Mode

0 0

-300 -300
300 300
Fast Mode

Fast Mode

0 0

-300 -300
27.36 27.365 27.37 27.36 27.365 27.37
Time(ms) Time(ms)
(a) Terminal response; (b) Neutral response

Fig. 5. Separation of slow mode and fast mode from original signal

775
 In order to obtain the slow mode, we used a minimal order believe this characteristic arises not only because the stator
Chebyshev Type II zero-phase filter with a cut-off frequency of winding acts as a transmission line, but also because capacitive
5MHz. The fast mode is obtained by subtracting slow mode coupling and inductive coupling between bars. On the other
from original signal. As can be seen from Fig.5, the fast mode hand, these coupling will also make it possible to detect signal
arrives the sensor ahead of the slow mode. The characteristic from terminals of other phases rather than the one injected
arises due to the fact that the slow mode traveled through the simulated PD pulses. The response signal from three terminal
stator winding conductors before it arrived at the capacitor, capacitors and the neutral capacitor left this without doubt. An
other than through the shorter routes made up by inductive and example is given in Fig.7. When injecting simulated PD pulse
capacitive coupling between bars, just as fast mode did[7][8]. in different bars along the parallel tested, not only the terminal
of the phase terminal of the parallel, but also that of the other
In addition, it seems that by measuring the delay time two phases detect the response signal.
between arrival of the fast and slow modes at the PD sensor,
the relative distance to the PD site can be determined.
Unfortunately, before this method can be used, the stator
winding needs extensive testing/calibration where pulses are
injected to different locations and the propagation delays are
measured between the fast and slow modes, or on the other
hand, a mathematical model for stator winding is built.

C. Attenuation during propagation

Using an electronic discharger to simulate the PD pulse
may lead to the output of discharger unstable between test bars
and subsequent response signal at capacitors. To address this
problem, besides the electronic discharger, we also used a
signal generator to generate sine signals with a amplitude of
10V from 1MHz to 20MHz. And we picked up five bars as the Fig. 7. Schematic representation of a stator winding.
signal injection point to investigate the signal attenuation when
travelling from discharge site to sensor location. Fig.8 shows the crosstalk from phase V to other two phases
10 when simulated PD pulses were injected successively to the 22
1M 5M selected bars of the 2nd parallel of phase V. The bar number is
8 10M 20M given along x axis, along the other axis the crosstalk ratio.
3
Amplitude 9

6 U to V
W to V
Crosstalk level

4 2

0 1
0 44 88 132 176
Bar number
0
Fig. 6. Signal attenuation with distance and frequency 0 44 88 132 176
Bar number
As can be seen from Fig.6, the signal with a frequency
higher than 5MHz, do not suffer from an severe attenuation Fig. 8. Crosstalk from phase V to the other two phases
when propagating from discharge site to detector location. This
is identical with the phenomenon mentioned above, that the As easily be deduced from Fig.8, the crosstalk depends on
signal with frequency higher than 5MHz is doing a fast mode, the location of injection, or we can say it depends upon the
which means a less attenuation. geometry of stator winding. The crosstalk ratio is defined as
the ratio of peak value of capacitors mounted on three phase
Obviously, the attenuation of PD signal, when propagating terminal regard to that of the capacitor mounted on terminal of
through stator winding, is frequency dependent and generator phase V. For discharges in the first half bars of the parallel, the
dependent, thus it is impossible to perform a calibration of ratio increased when deeper into the winding from neutral
apparent charge for high frequency or very high frequency point. However, when it comes to the middle bar of the parallel,
range PD measurement system. In addition, amplitude based the ratio is smaller than two sides. This may be due to the
alarm systems are liable to trigger false alarms or overlook geometry of the stator winding. For discharges deeper into the
imminent failures during the PD online measurement. winding from the middle bar, the ratio decreased. It must be
noticed that for most bars, the crosstalk ratio is bigger than 1,
D. Crosstalk Between the Phases which means that the largest signal does not show up in the
As mentioned previously, the responses of capacitors phase where the discharge occurs, but in another phase. For a
consist of two modes. Based on studies of other researchers, we better understanding, the PD location through finding the phase

776
with largest discharge magnitude is not precise, which would 4) Apart from the propagation characteristics of PD pulse,
lead to a negative false indication. the test results demonstrate that the coupling capacitor, which
has a measurement frequency range of 50kHz~20MHz, is
IVˊ CONCLUSION capable of detecting PD signals throughout the entire stator
winding of large hydro generator.
Since the owner of a hydroelectric power plant decided to
uprate a hydro generator, we got an opportunity to apply a
destructive test on this generator to investigate the PD signal REFERENCES
propagation in its stator winding. In the test, we injected [1] Johnson J S, Warre M. Detection of Slot Discharges in High-Voltage
Stator Windings During Operation[J]. American Institute of Electrical
simulated PD pulse and sine signal of different frequency to 22 Engineers, Transactions of the, 1951, 70(2): 1998-2000.
bars of the 2nd parallel of phase V, and measured response at
[2] Kurtz M, Lyles J F, Stone G C. Application of partial discharge testing
three phase terminals and neutral point with four coupling to hydro generator maintenance [J]. IEEE Transactions on Power
capacitors. We found that Apparatus and Systems, 1984 (8): 2148-2157.
1) It’s identical with previous investigations, there exist two [3] Yuan Wan, Zhaohui Li, et al. Study on partial discharge online
monitoring system mounted at neutral point for electric generators[J].
different propagation modeV, i.e. fast mode and slow mode, Water Resources and Power, 2009, 27(4): 197-201.
when PD signal propagating in stator winding, but the corner [4] International Electrotechnical Commission. High 㸫 voltage Test
frequency is about 5MHz, rather than 500KHz, as reported by Techniques: Partial Discharge Measurements[M]. International
previous investigation[8], we believe this is due to the different Electrotochnical Commission, 2015.
geometry of generators under test. [5] Wilson A, Jackson R J, Wang N. Discharge detection techniques for
stator windings[C]//IEE Proceedings B-Electric Power Applications.
2) The different propagation mode have different IET, 1985, 132(5): 234-244.
attenuation characteristics, the fast mode has little attenuation [6] Henriksen M, Stone G C, Kurtz M. Propagation of partial discharge and
between different injected location when propagating from noise pulses in turbine generators[J]. IEEE transactions on energy
discharge sites to detector location, while the slow mode has a conversion, 1986 (3): 161-166.
relatively severe attenuation. However, it should be noticed [7] Cheng Yangchun, Li Chengrong, Wang wei. Research on propagation
that the response of fast mode is relative smaller than that of rule of partial discharge pulses inside generator along stator windings[J].
Power System Technology, 2005, 29(8): 65-70.
slow mode, since it poses a heavy attenuation throughout the
[8] Pemen A J M, van der Laan P C T, De Leeuw W. Propagation of partial
entire stator winding. That’s why 50kHz~20MHz was discharge signals in stator windings of turbine generators[J]. Energy
employed for detection frequency range in this test. Conversion, IEEE Transactions on, 2006, 21(1): 155-161.
3) In addition to attenuation, there exist crosstalk between [9] Devins J C. The 1984 jb whitehead memorial lecture the physics of
partial discharges in solid dielectrics[J]. Electrical Insulation, IEEE
phases, which is much more severe than previous reported. We Transactions on, 1984 (5): 475-495.
believe it is result from the more complex geometry of large [10] Fruth B A, Gross D W. Partial discharge signal conditioning techniques
hydro generator than that of motor or turbine generator used by for on-line noise rejection and improvement of calibration [machine
previous investigations. insulation tests][C]//Electrical Insulation, 1996., Conference Record of
the 1996 IEEE International Symposium on. IEEE, 1996, 1: 397-400.

777