You are on page 1of 5

2009 IEEE Electrical Insulation Conference, Montreal, QC, Canada, 31 May - 3 June 2009

Partial Discharge Location in


High Voltage Transformers
R. Mohamed and P. L. Lewin
The Tony Davies High Voltage Laboratory
Electronics and Computer Science
University of Southampton, SO17 1BJ
United Kingdom

Abstract- Partial discharge (PD) location is a common problem using a digital filtering technique [2]. However there may be
within Power Transformers diagnostics and is important for some dispute with respect to the choice of model technique
assessing the severity of any detected fault. The technique used, as it is not suitable for certain types of signal that have
presented in this paper is suitable for any type of discharge signal steep rise times, possibly 10/20 ns wavefronts and 20/30 ns
through determination of the predominantly capacitive operating
width. It is to be established whether such an approach is
region in the frequency domain of the transformer.
Theoretically, any components signal in this frequency region suitable for various types of transformer which may have
will retain their waveshape and phase without significant different properties and parameters [2-3].
attenuation. To obtain the predominantly capacitive region of a
transformer winding, the relationship between the bushing Considerable experimental work has been undertaken by the
terminal and neutral terminal has to be determined. In the authors using an interleaved transformer winding model [2-5].
predominantly capacitive region the two terminals will have a Analysis has been performed using impulse-like signals
linear relationship in dB. To obtain the model of the transformer having a range of rise times and pulse widths. Since for this
winding under test, it is injected at one end with a low voltage experiment the only available terminals for measurement are
calibrated impulse to determine its initial parameters. Based on
at the bushing and neutral lines (grounded or ungrounded), the
the measurement at the bushing and neutral terminals, the linear
relationship of the two terminals is found by a fixed distribution linear relationship of the two terminals has to be established
equation derived from a lumped circuit parameter model. For before attempting to locate a discharge source. This is
verification, experiments within the laboratory have used various achieved by injecting an impulse signal at the bushing
artificial PD signals injected at different points along a model terminal and detecting the response at the neutral point. This
winding to represent different PD sources. By considering the experiment is repeated by injecting at the neutral with
relative magnitude of a discharge signal at predominantly measurement at the bushing terminal. The time-based
capacitive frequencies of the winding it is possible to evaluate the response is measured and transformed into the frequency
position of the discharge source. domain. The capacitive operating frequency region is
extracted based on identifying ranges of constant magnitude
I. INTRODUCTION and phase. This is verified through the use of network
The identification of PD in high voltage transformers is an analyser to obtain a frequency response of the whole winding.
important tool for transformer diagnostics. The estimation of
PD and its location is essential in predicting the health of II. THE TRANSFORMER EXPERIMENTAL MODEL
transformers and assists in selection of suitable maintenance
or replacement strategies. However, the techniques of A. The Lumped Circuit Parameter Model
estimating PD location can result in error because of the Consider the lumped circuit parameter model as shown in
complexity of the transformer and the routes by which signals Figure 1. The transformer model is a disc-type interleaved
propagate from the source to the measurement point. winding with 7 discs. The end connections of the winding are
connected to end plates which are noted as EP in Figure 1. At
The modelling propagation of high frequency signals within a high frequency, the lumped circuit model becomes a series
transformer winding may demonstrate attenuation or be loss connection of capacitances K/dx and inductances Ldx arranged
free depending on the model adopted [1]. The capacitive in parallel and series respectively. Losses are included in the
ladder network of a transformer winding which can be model (winding resistance and core losses) by the addition of
determined by the initial distribution equation has a property rdx and Rdx. Cdx and Gdx are the series capacitance and shunt
of constant output to input gain. This can be easily conductance that represent the damping behaviour from
determined in the frequency domain. winding to the ground. The relationship between each
winding is modelled by its mutual inductance denoted as a
Previous studies have shown that the predominantly capacitive matrix of inductance [L].
region in the frequency domain of a transformer frequency
response is approximately constant in magnitude and phase. The fundamental relationships of the parameters at high
This can be realized by extracting the frequency of interest frequency can be represented by the partial differential
equation [1,6] shown in Equation 1.

978-1-4244-3917-1/09/$25.00 ©2009 IEEE 200


Bushing tap point is also modelled using an equivalent
capacitor CB, and the Neutral line is always referred as a
ground connection. Therefore, a linear relationship can be
found between the two terminals [2], as follows:

VL KC cosh( αx) + C N sinh( αx)


= (5)
VN KC cosh[α (l - x)] + C L sinh[α (l - x)]

IL KC C N cosh( αx) + sinh( αx)


= (6)
Fig. 1. Lumped circuit parameter model at high frequency. IN KC C L cosh[α (l - x)] + sinh[α (l - x)]

∂ 5e r ∂ 4e ∂ 4e where IL and VL are current and voltages at the bushing, IN and


rK 4
+ ( 1+ ) 4 − LK 2 2
∂x ∂t R ∂x ∂x ∂t VN are current and voltages at the neutral, CN and CL are
capacitance of the neutral and bushing respectively. Thus in
L ∂ 3e ∂ 2e ∂e (1) the case of ungrounded transformers, Ap will be large or if the
− (rC + ) 2 + LC 2 + LG = 0
R ∂x ∂t ∂t ∂t transformer is grounded AP will be very small.

By solving equation (1) and also considering the half range III. DETERMINATION OF A WINDING
sine series the complete solution for a grounded transformer TRANSFER FUNCTION
becomes:
A. The Frequency Response
Measurement of the frequency response of the interleaved
cosh( βx) − γ st πx disc windings was carried out using a network analyzer,
e= E
cosh( βl )
+E ∑A e s cosωs tcos( 2s − 1 )
2l
(2)
Agilent 4395A (Figure 2). With reference to Figure 2, ZP is
infinite for an ungrounded transformer winding and zero for
and for an ungrounded transformer: the grounded case. During the test, the measurement
frequency range was set from 100Hz to 500MHz, with a
sample bandwidth of 30Hz. It is necessary to measure the
sinh( αx) − γst sππ
e= E
sinh(αl )
+E ∑A e s cos(ω s t )sin
l
(3) frequency response at the highest frequency possible, in which
it is implicitly represents the system in order to accurately
simulate an impulse response. Therefore, due to the limitation
Where, As is an amplification factor, γs the damping factor, ωs of the network analyser, the data was taken many times and
the harmonic frequency, s the harmonic order, x the distance resampled to further refine the complete frequency response
from the neutral line, l the total length of the transformer which contains 1400 sample points.
winding, and β and α are fixed distribution factors for
grounded and ungrounded connections respectively. E is the B. The Impulse Response
input voltage, e a voltage level at distance x, and t the time The impulse response can give direct clues about multipath
vector. The equation contains two functions, the fixed propagation, of a high frequency signal or echoes and
distribution equation and the damped harmonic equation. reverberations within a transformer. From a theoretical point
Overall the solutions portray the high frequency oscillation at of view an impulse response experiment can also be used to
different points along the winding, where at any instant the determine the frequency response and transfer function of a
signal transient oscillation can be determined from the system.
harmonic damping equation.

Equations 2 and 3 have been simplified from Equation (1). By


considering the fixed distribution equation only, the full fixed
distribution equation can be written as follows:
AP cosh( αx) + sinh( αx) (4)
e= E
AP cosh( αl ) + sinh( αl) ZP

where AP is the corresponding equivalent impedance


coefficient at the end windings. However in the case of real
transformers, the only available terminals for measurement are Fig. 2. Measurement Setup for Interleaved Windings
at the Bushing tap point and Neutral lines, for which the

201
Fig. 3. Ideal Impulse Response and Black Box Model

Consider an impulse function δ(t), injected at terminal 1 on


one side of a transformer. By taking the fourier transform, it
is equivalent to a constant magnitude response in the
frequency domain for which it will excite all the system
frequencies. Figure 3, shows an illustration of an impulse
injected into a transformer model which also can be Fig. 4. Linear Relationship of Terminal Response from Frequency Response
considered as a 'Black Box' model and the related response. Measurements, averaged over 300kHz - 1MHz
The response represents the frequency response of the
transformer. Therefore the impulse response and measured
frequency response can be compared any difference being due
to the inability to experimentally generate a perfect dirac
impulse wave shape.

IV. TERMINAL LINEAR RELATIONSHIP

A. Frequency Response - Terminal Relationship


The frequency response measurement of the experimental
model (Figure 2), consists of two sets of measurements. The
first set is eight measurements from terminal 1 to 8, and the
second set measurements from terminal 8 to 1. From
theoretical point of view, if the system is linear, and the discs
are identical, comparison of the measurement data should
indicate the same magnitude and phase for sections of the
winding irregardless of direction of measurement.
Fig. 5. Fourier Transform of Impulse Response with 0.5µs pulsewidth and
The linear relationship of the bushing terminal to neutral 50ns transition
terminal (from equation 5) can be extracted by considering a
range of frequency of interest, and taking the mean of it. Figure 5 indicates that the critical frequency of this winding is
Figure 4 shows the linear combination of terminal response about 1MHz at which point the magnitude of the gain, |H(ω)|
over frequency range from 300kHz to 1MHz. The obtained (with impulse injected at bushing), increase by 15dB. At the
results clearly indicate that for the model winding used in the same frequency, the impulse response transfer function for
experiment that there is a very linear response in terms of gain impulse injected at Neutral line has a reduction of
as a function of distance along the winding. approximately -10dB.

B. Impulse Response - Terminal Relationship Between 300kHz to 1MHz, both derived frequency response
Similarly, the terminal response linear relationship can be have almost the same absolute constant value of gain |H(ω)|
found from impulse response in the frequency domain. To do equivalent to 0.8 and a constant phase angle. This shows that
this experimentally, a low voltage high transition pulse is the impulse signal has almost the same propagation path in
injected at the bushing terminal and measurement taken at the both directions from bushing to neutral terminals and from
neutral point, the same impulse signal is then injected at the neutral to bushing terminals. Therefore the capacitive
neutral terminal and the response is measured at the bushing distribution coefficient α can be determined from equation 4,
terminal. Using the FFT, the time based measurements can be for ungrounded winding as follows:
converted into the frequency domain. Figure 5 shows the ⎧ 1 ⎫
obtained frequency responses derived from the impulse α = cosh −1 ⎨ ⎬ = 0.68 (7)
response measurements. ⎩ (ω ) | ⎭
| H

202
For ZP = 0Ω, simulated PD pulses were injected with very low
voltage with different variation of pulsewidth and transition.
From Figure 8, the transformer was fully grounded, therefore
the factor AP is approximately equivalent to zero which leads
(a)
to a very low value of relative gain (-28dB to -6dB) as can be
seen from figure 6. Hence the evaluation of voltage
measurements may lead to error because the voltage values are
relatively small. Current measurements were used as an
alternative for the grounded winding measurement, which
(b)
have larger values and are more representative of discharge
magnitude.
(c)
The ground connection was now connected via a 50Ω
resistance, and current measurements taken for different types
of PD signals. There is a change in the capacitive straight line
distribution as shown in Figure 6(b), except that at bushing
terminal the gain is the same as it is still connected via the tap
point to ground. The 50Ω resistor in this experiment provides
Fig. 6. Linear Terminal Relationship with different ZP. The Bushing
a potential offset to the overall gain increasing towards the
capacitance, CB, used is 1000pF. (a) ZP ≅ ∞, (b)ZP ≅ 0Ω, (c)ZP ≅ 50Ω.
neutral terminal, which significantly changes the terminal
(Bushing to Neutral) response relationship as also shown in
From Figure 4 and Figure 5, the capacitive network of the
Figure 9.
tranformer winding can be extracted over a frequency range
from 300kHz to 1MHz. Meanwhile Figure 6 shows the
Finally, the winding was tested with the same range of
theoretical straight line bushing to neutral linear relationship,
simulated PDs for ungrounded connection. Both ends are
for different possible values of ZP (0, 50Ω, ∞), calculated
open circuited, hence the only suitable measurement is of
using the value of α obtained in Equation 7. In the grounded voltages. Obviously, for an ungrounded winding the voltage
case (ZP = 0), the range of attenuation is largest over the measurement is easier, and the ratio of linear relationship still
winding length, inferring that this condition is preferable for holds within the range of 300kHz to 1MHz, as illustrated in
PD location experiments. Figure 10.

It also can be seen that, the linear relationship closely matched


V. RESULTS AND DISCUSSION the theoretical line with the PD measurement (injected at
A. Partial Discharge Location. different terminals) indicating that there is a direct path of
To visualize the PD location process, several experiments signal propagation in this frequency region. Since the ratio of
were carried out to inject an artificial pulse that closely its VL/VN is linear, the signal travels to the line end terminals
matches PD pulses at various points along the experimental experiencing a constant gain.
winding. Figure 7 shows the measurement arrangement for
this experiment.

Pulses were injected using a pulse generator and measured CB


using a digital storage oscilloscope, with sampling interval of CT
5 × 10−7 to 5 × 10−9 seconds and 1000 sample points per
Pulse Generator
measurement. Data were both recorded at Terminal 1 (T1 or
bushing) and Terminal 8 (T8 or neutral). Two Radio
Frequency Current Transformers (CTs) were used in the T1
experiment for current measurement.

Several steep pulses with different transitions and pulsewidths


were used in the experiment to mimic PD signals with initial T8
positive peaks. Figure 8 to Figure 10 show the graphical CT
results of partial discharge location for different line end
terminations of ZP. In all cases the gain represents the ratio of
the average magnitude of the measured current pulses over the
frequency range 300kHz – 1MHz. Fig. 7. Measurement Setup for partial discharge location using Pulse
Generator,

203
VI. CONCLUSION

There are several conclusions that can be made from this


study:
1. There exists a capacitive operating region under over
frequency ranges for a transformer winding. In
which in this experiment the gain is almost constant,
and the phase is constant for ωL ≤ ωC ≤. ωU. Where
ωL is the lower frequency limit, ωU is the upper limit
and ωC is the capacitive operating frequency.
2. The PD signal propagation behaves the same under
PulseWidth=0.5µ Transition=0.5µ
different winding connections, i.e, when the winding
PulseWidth=0.5µ Transition=50n
PulseWidth=5µ Transition=5µ
is ungrounded, grounded or grounded via a 50Ω
Theoretical Line resistor.
3. The level of gain changes with choice of termination
impedance, which consequently changes the
windings capacitive distribution level.
Fig. 8. PD Location from Current Measurement for ZP ≅ 0Ω

ACKNOWLEDGMENT

The first author wishes to express his gratitude to Universiti


PulseWidth=0.5µ Transition=0.5µ Kebangsaan Malaysia for giving him the opportunity to pursue
PulseWidth=0.5µ Transition=50n his Phd study and for financial support.
PulseWidth=5µ Transition=5µ
Theoretical Line
REFERENCES
[1] L.V. Bewley, Travelling Waves on Transmission Systems, John Wiley &
Sons, Inc. New York, Chapman & Hall Ltd. London 1951, 2nd Ed.
[2] R. E. James, B. T. Phung and Q. Su, “Application of Digital Filtering
Techniques to the Determination of Partial Discharge Location in
Transformers”, IEEE Transaction on Electrical Insulation, Vol. 24, No.
4, August 1989, pp. 657-668.
[3] J. Fuhr, M. Haessig, P. Boss, D. Tschudi, and R. A. King, “Detection
and location of internal defects in the insulation of power transformers”,
IEEE Transaction on Electrical Insulation, Vol. 28, Issue. 6, Dec 1993,
pp. 1057-1067
[4] J. Fuhr, “PD-Source Localization in the Insulating System of Large
Power Transformers”, IEEE International Symposium on Electrical
Fig. 9. PD Location from Current Measurement for ZP ≅ 50Ω Insulation 2006, 11-14 June 2006, pp. 424-427
[5] J. Fuhr, “Procedure for identification and localization of dangerous PD
sources in power transformers”, IEEE Transaction on Dielectrics and
Electrical Insulation, Vol. 12, Issue. 5, Oct 2005, pp. 1005-1014
[6] A. Akbari, P. Werle, H. Borsi, and E. Gockenbach, “Transfer function-
based partial discharge localization in power transformers: a feasibility
study” IEEE Electrical Insulation Magazine, Vol. 18, Issue. 5, Sept-Oct
2002, pp. 22-32.
[7] G. B. Gharehpetian, H. Mohseni and K. Moller, “Hybrid Modelling of
Inhomogeneous Transformer Windings for Very Fast Transient
Overvoltage Studies”, IEEE Trans. Power Delivery, Vol. 13, No. 1,
January 1998, pp. 157-163.
[8] Z. D. Wang, S. N. Hettiwatte, and P. A, Crossley, “A measurements-
based discharge location algorithm for plain disc winding power
transformers”, IEEE Trans.On Dielectrics and Electrical Insulation,
Vol. 12, Issue. 3, June 2005, pp. 416-422.
PulseWidth=0.5µ Transition=0.5µ [9] S. N. Hettiwatte, Z. D. Wang, and P. A. Crossley, “Investigation of
PulseWidth=0.5µ Transition=50n propagation of partial discharges in power transformers and techniques
PulseWidth=5µ Transition=5µ
for locating the discharge”, IEE Proceedings Science, Measurement and
Technology, Vol. 152, Issue. 1, Jan 2005, pp. 25-30
Theoretical Line
[10] A. Morched, L. Marti and J. Ottenvangers, “A High Frequency
Transformer Models for the EMTP”, IEEE Trans. Power Delivery, Vol.
8, No. 3, July 1993, pp. 1615-1626.
[11] S. M. H. Hosseini, M. Vakilian, and G. B. Gharehpetian, “Comparison
Fig. 10. PD Location from Voltage Measurement for ZP ≅ ∞
of Transformer Detailed Models for Fast and Very Fast Transient
Studies” IEEE Trans. Power Delivery, Vol. 23, No. 2, April 2008, pp.
733-741.

204

You might also like