FAULT LOCATION IN POWER TRANSFORMERS

DURING IMPULSE TESTS

Vanaja.R K. Udayakumar, Member, IEEE

Dept. of Electrical Engineering Dept. of High Voltage Engg
Anna University Anna University
INDIA INDIA

Abstract: In this paper we present a new approach to detect, and transferred to an adjacent winding, the tank current or the
investigate the feasibility of locating winding faults with also voltage transferred to a non-tested winding. In its simplest
determining the percentage of short within windings of power form a test arrangement would consist of Marx generator
transformers.There has not been much emphasis on methods of with an arrangement for conducting an impulse test
assessing the extent of failure and in the estimation of the location
of failure during such tests. However, this is a very important chopped on the tail, a voltage divider, a current shunt and
aspect since all faults involving non self-restoring insulation are measuring cables that connect to a recording device as in
destructive in nature and hence have to be necessarily rectified. Fig.1. Traditionally the voltage and supplementary
In case of failure. of a transformer during an impulse test, oscillograms have been recorded on analog impulse
manufacturers have to untank the winding in order to locate the oscilloscopes whose specifications are laid in IEC 790.
fault. Often the destructive 1-minutepower Frequency test is done Here we suggest an important technique for analysis based
in order to visually locate the fault. It would be appropriate to on digital recording.
develop a method for fault location so that results are available
prior to the untanking of the transformer. In principle, the absence of significant difference
This paper discusses methods for fault location based on a between voltage and current transient records constitutes
frequency domain approach. Windings of a transformer behaviour evidence that insulation has withstood the test inspite of
at very high frequency, is essentially that of a capacitive ladder their long usage. The impulse testing is to demarcate the
network. At lower frequency standing wave behaviour is presence of insulation faults and the measurement of
obtained. Based on a choice of filters, the appropriate behaviour parameters.
of the winding during a fault can be studied. In this work, we
develop methods for fault location based on simulations on layer With the advent of digital recorder in the high voltage
windings using a circuit simulation model. The results are environment the requirements of digitizer for impulse
validated experimentally on a specially constructed layer winding. measurement are specified in IEC 1083. A perusal of the
Keywords Transformer, High frequency , Impulse test , Signal standard show that a recording device should have the
Processing ,Transfer function following features for transformer testing.

Vertical resolution 10 bits

I. INTRODUCTION
Sampling rate 60 Mslsec
Accuracy in Voltage measurement : +2%
The impulse test on a power transformer is done in order Accuracy in time 4%
to assess its dielectm integrity when exposed to over
voltages, of atmospheric origin. The shape of the voltage, Digital technique using signal processing are being
their magnitudes and the sequence of application that increasingly used to eliminate drawbacks in time domain
constitutes aproof test are covered by standards such as the techniques. The purpose of this paper is to discuss a
IEC-76, 1980. The generally recommended order of method to observe and study faults in layer windings of a
different impulse applications are transformer coil. It would be appropriate, as a first step, to
one reduced full impulse analyze the behaviour of a uniform singlelayer winding
one 100% full impulse when energized by an impulse voltage. This is not severely
one or more reduced chopped impulse(s) restrictive as single layer windings have several properties
two 100% chopped impulses to elucidate many features of the fhquency response of
two 100% full impulses winding [11. Here we aim to propose
0 methods of location of faults during impulse test
Standards for impulse testing of power transformers such to assess the magnitude of failure of winding
as the IEC - 722, 1982 mainly cater to the recognition of 0 to confirm the utility of the methods with experimental
fault during impulse tests. Such fault recognition is also results.
based on observations recorded on analog impulse
oscilloscopes. Digital data acquisition shows considerable Fault refers to the breakdown across a section of the
promise for fault identification and has led to the winding during impulse test, particularly about non
development of the transfer function method for fault restoring insulation. Several investigators have used
analysis. different principles for analyzing fault during impulse tests.
For example, Heller and Veverka [2] state the appearance
In addition to the record of the applied voltage, it is of the neutral current is similar to that observed when a
mandatory to observe another waveform during a test. This
portion of winding is physically shorted by a piece of wire.
transient could be the winding current, the neutral current

0-7803-5935-6/00/$10.00 (C) 2000 IEEE 2199

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.
 ~

Malewski [3] used a 'diac' across the section for elucidating The experimental simulation to verify the above study
certain features of the transfer function mctliod. was doric os is shown in Fig. 1.

As a tint approximation the high voltage transformer

winding is rcprcscnted by a laddcr network will1 serics
inductance and capacitance as well as the parallel
capacitance to ground. The transfer hnction of such a
nctwork shows poles as the reRonant frcqucncies of the
winding model. Breakdown between turns or coils of
winding under test corresponds to short circuit of one of the
local LC nctworlc [I] with shift in resonant pole to another
frequency.

Detection of turn faults has been based on comparison of

the impulse voltage and neutral current oscillogram taken
with applying I.ZS0 ps. lighting impulse or impulse
chopped on the tail at 2 p to 6 p. Minor difference
between the compared oscillogram can be indicative of the
faults as observed in their time domain records or by Fig.1 .Typical lmpulso Tat Circuit
analysing their frequency spectra using the lTT algorithm
to observe resonant pole shifts. Fig. 3 shows the voltage and current records of the
winding when energized by a Marx circuit test setup for
11. SIMULATION OF FAULTS Lighting impulse of 1.2/50p. Fig. 4, shows the normalized
spectra of the voltage and current records in decibels
In this analysis the short circuit fault is simulated as a obtained using FFT. Fig. 5, shows the transfer function of
switch which remains open for time ti aflcr tlic encrgination the winding. Also superimposed in the figure is the
of coil by m m generator and from ti corilinucs to remain transfer function with short crcated betwccn nodes 3.4 ol'
closed until the end of simulation. the winding. It can be clearly seen that there is a shift of
poles under faulted condition. The model test coil for
In actual situation the breakdown will occur after a finite simulation study is as shown in Fig. 5 whose dimension
instant if the coil has originally not been subjected to am as presented in Appendix 1. A particular advantage of
catastrophic dielectric failure. This justifies the choice of simulation study is that it caters to various input waveforms
using ti to accentuate the difference in, oscillogram as as was also analysed on the 10 section Abetti coil pnsented
occurrence of faults. Since the work mainly caters to the in Appendix 1. For example the study was done using
breakdown of non-self restoring insulation it necessitates to 1.260 p Lighting impulse, and also with 2 p and 6p tail
keep a short circuiting switch closed until breakdown chopped signals to verify experimentation.
phenomena. A 10 section equivalent circuit of the coil [4], 10000
[51 was analyzed using PSPICE, . a circuit-analyzing
Program-

The sequence of simulation is to record the voltage and

current oscillogram when applying the lighting impulse of
12/50 p or the tail chopped lighting impulse on the
winding model, Similarly the hult condition is simulated
with switch closed on portion of the winding at different
locations and for different instants of closure. Since during d :'O 15 20 2 5 ' 30
(Time (micro.eec))
35 40
I
45
experimentation the probability of change in applied
voltage can occur due to test setup the recent trend is to Fig.2. Voltage / Current waveform of' AbeIti Coil energkd by a Man
analyse the signals in ftequency domain. Circuit

FFT (i)
The transfer function approach is T.F ...
-20
FFr (VI
m
w
Since the initial simulation is being done with a standard -40
Lightning impulse voltage constant, firther analysis has
been equivalently done with current spect" only. The
spectral changes are compared for faulted and unfaulted
simulations and analyzed. Analysis was done with respect
to fault identification and measurement of parameters of
measured signal. I .

(c) 2000 IEEE

0-7803-5935-6/00/$10.00 2200

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.
 I

W i t h fault
g .0008 The transfer h c t i o n is computed based on a 3 db cut off
U
“4 of 250 kHz as suggested by Dcgeneff [SI who suggested
$ .0006 that lumped parameter model need be valid upto 50 lcHz for
a
ul a lightning wave and 250 Wz for a chopped wave. The
dl
-0004 analysis considered for the Abetti coil LIS energized by a
U4
Marx circuit is presented with computation of
:
‘0,
.0002 corresponding spectra with simulating winding section
fauln. The design of antialiasing filter is L I ~ S ObaNed on B e
I

CI
0 .5 10 15 20 25 30 fact that filler must have a monotonic response [6].A study
f/2(KHz) of filter application identified that with a Butter worth
digital highpass filter of order 22 and Hamming window
Fig.4. Transfer function with and without fault function, high frequency signals could be identified as in
shown in Fig.6. with the proposed algorithm of transfer
L1
function evaluation as shown in Fig.7
iorvfimn
dcvidmI shun!

60 JB a! SOMhz
f l u . -

%
--
60dB at 500 Khz
Fig.5. Model equivalent of Abetti Coil

The experiment was done with signal acquisition using

an I O bit digitizer and PU12 Marx generator for a lightning
impulse source. The fhquency resolution of the Transfer
function is given by f= sampling fkquency / record length.
This was kept as low as possible by choosing the sampling
frequency as 1 MS/sec. and the record length as IO00
points which is as advocated within IEC standard

111. PARAMETER INVOLVED IN FAULT ANAL,YSIS

The various parameters are obtained from the frequency

spectrum and the relation between the faults and the
parameters are discussed. Fault location new line en4
middle winding, ground end are to be located using
parameters as resonant frequencies, gain, area of
bandwidth. At least the first three resonant frequencies
seem vital. For the coil used a span of 0 to 250 M z , has
been used to analyse percentage faults as low as upto 10%.
From Fig. 8 it is been observed that
-30
72KHz,96KHz,160KHz were the first three resonant
fiequecies and the percentage deviation of gain values of
fault and band width of resonant frequencies wen
calculated with reference to signals without fault. An
arbitrary fault was created across the section of winding as
5%, 10% tappings. It was found that if fault occurred in the
top section the percentage shift in gain is very high but
percentage shifi in area is low than if it was in the middle
section .If fault was in the ground section the percentage
shift in area was too low though gain shifi remained high.
Fig.6. Influence of Digital Filler on M u s u d Signal This was v e r i f d using simulation and experimentation as
(a) Faulty current signal spectrum (b) Filtcrd signs1 (a) using digitul is seen in Fig.9.
Butter WO& filter

(C) 2000 IEEE

0-7803-5935-6/00/$10.00 2201

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.
 C,n = Surge capacitance of coil
U = Amplitude of signal

As presented in Pritindra [7] the voltage distribution in

winding is
v ( x ) ... A , 0'1') I A , c' f" (2)
whcre

I 1

where A,, A2 are boundary condition with origin (x=O)

the line end and V(0) = 1 per unit and V(1) = 0 the ground
t,lO us / division end and A, = -exp (-a)/ 2 Sin h a and
Voltage,current
A2 = exp (a)/ 2 Sin h a

-7 and thus V(x) = Sin h [a (1 -x/l)] / (Sin h a) (3)

The oscillations of the voltage measured across the
length of winding was measured and their spectra evaluated
using FFT routine. The transient voltage distribution in the
coil with grounded neutral for a = 4.0 to a = 14 was
analysed. The percentile variation of frequencies under
fault and no fault condition of the winding subject to
impulse signals L.I. 1.2 /Sops and chop L.1. 2 ps or 6 p
were studied and their spectral distribution along the length
of the winding (x/l) is as seen in Fig.10. for the first three
resonant frequencies which shows the location of fault in
winding. The regions of the spectra prior to the first
resonant fiequencies explains the percentage magnitude of
fault obtained with realisation of a low pass filter whose
critical frequency is the first resonant frequency of the coil
as seen in Fig. 12.

60
0 area Vs 0 Winding - ___
50
40

p0
-2 0
10
0'
- :.

0 10 20 30 40 50 60 70
8 Winding

IV. RESULT AND DISCUSSION

In the previous section, the procedure of simulation of \ gain V s 0 w i n d i n g

70
fault studies was presented. The verification was also done
60
on the specially constructed layer winding model presented
in Appendix XI. It diffm from the Abetti coil in the respect 5
4
$0
that it has a secondary winding which could be used for 6 40
additional measurement and also had higher value Cg due a 30
20
to the presence of additional metallic former in the inner
LO
and outer diameter and 5% tapping arrangement over the
length. As was also represented by Heller [2] the
O0 8 ulndlng
M 0
transformer coil subject to lightning impulses response like
a system of capacitance and inductance. Cenedivc prevails in ?ault: A-lO\.B-ZO\,C-25\
tllc beginning of prv~csswith the voltage across h e ---So*
terminals given as
U -
= 2~ [ I e - ' / ~ .nt~ (1) Fig.9.hmeter Variations under fablkd conditions
where Z = Surge impedance of winding

(c) 2000 IEEE

0-7803-5935-6/00/$10.00 2202

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.
 a=4; 2 u s e c c h o p

Ip 0 0.2 0.4 0.6 0.0 1

Percentile l e n g t h I x / l )

a=4; 6usec c h o p
Fig.11. Magnitude irariation of fault (a) Percentagevariation of first three
rcsonant frequencies (b) Percentage variation with respect to magnitude of
short cuircuit faull

V. CONCLUSION

One of the prime motivation for improved methods of

failure identification was the need for eliminating
- 0.2 0.4 0:6 0.8
P e r c e n t i l e length ( x / l )
1 subjectively associated visual comparisions. The new

220
II/ a=4;1.2/50 US L.I.
method proposed in this paper is used to estimate the
location of fault within windings as a shin in resonant
frequency. The principle emphasis of the location of fault
and percentage of fault during impulse testing was done
using fkquency spectrum on the basis of the parameters
descnied The choice of low pass filter prior to fmt
resonant m u e n c y was used to determine the magnitude of
fault. The results presented h m simulation and
experimental study illustrates this as a reliable method of
location of faults in transformer windings.

V. REFERENCES
rp
”[ 0.2 0.4 0.6 0.8 A-
1. R. Malewski, et.aI, “Five years of monitoring the
P e r c e n t i l e l e n g t h (x/l) impulse test of power transformer with digital
recorders and the transfer function method”, CIGRE,
Fiy.10. Percmlile variation of first threc nsonanl Trqxluccics ( f ~fz.
. 0) August 1992, pp 12-20 I .
along length of winding 2. Heller and Veverka. B, Surge phenomena in electrical
machines, London, Lliffe books.
3. R. Malewski and B. Powlin, “Digital monitoring
technique for HV impulse tests”, IEEE Trans. PAS.
vol. 104,no.ll,April 1985,pp 3108-3116.
4. RC. Degeneff, “A general method of dettrmining
resonance in transformer winding”, IEEE Trans on
PAS, vol. 96,110.2, September 1977, pp. 423430.
5. K. C.DcycncT1; ct.el., “A Methcxl Tor reduced order
transformer model for system studies from detailed
lumped parameter models”, IEEE Trans. Power
Delivery, vol. 7,no. 2, 1992, pp. 649-655.
m V. Jayashankar, et.al., “A new paradigm for the
6.
’“I 0 10 20 30 40 50 60 70 80
impulse testing of power transformers”, third
conference and workshop on EHV technology,
Pig.1l.a. N PAU~T Ban&dore, India, 1995.
Resonance frequency : &-First, B-Second 7. Pritindra, Chowdhuri, Electromagnetic Transients in
C-Third Power Systems, Chapter 12, John Wiley and SOIIS,
1996.

(c) 2000 IEEE

0-7803-5935-6/00/$10.00 2203

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.
 APPENDIX I

The lumped parameter model for the 10 section coil used is

C, =8.5 x 10-"f
& =2.1 x 1o"n
s -22.m
M1-2 = 13.537 mH
M 1-3 =6.23 I d
M1-4 -3.379mH
M1-5 -1.987mH
M 1-6 = 1.242mH
M1-7 =0.817mH
M 1-8 t0.560mH
M1-9 10.398mH
M1-10 ~0.292mH
L11 =28.998 mH
c, =3 . 4 ~ 10"*f
- 1 . 6 5 ~105Q

APPENDIX 11

Specifications of the one 1:l layer winding :

Section per coil 20 nos, 1:1 layer winding

Turns per section lo00
Size of conductor 0.914 diameter, 22 SWG
Mean diameter 409.4
Coil diameter (outer) 429.4
Height of the section 9 150
axial compressed
Depth of winding
radial 0.88
Pemitivity of inter
Turn insulation et = 2.5 (Kraft paper)
Thickness of inter
Turn insulations 0.6 x lV3 m
Thickness of spacers
between discs 3 x ioJ m
Permitivity of insulation
between disc (air) Q= 1
All dimension in mm.

(c) 2000 IEEE

0-7803-5935-6/00/$10.00 2204

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on December 28,2022 at 20:18:27 UTC from IEEE Xplore. Restrictions apply.