FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
As amended by the Patents (Amendment) Act, 2005
AND
The Patents Rules, 2003
As amended by the Patents (Amendment) Rules, 2006
COMPLETE SPECIFICATION (See section 10 and rule 13)
TITLE OF THE INVENTION
A system for locating Partial Discharge in windings.
APPLICANT
Crompton Greaves Limited, CG House, Dr Annie Besant Road, Worli, Mumbai 400 030, Maharashtra, India, an Indian Company
INVENTOR(S)
Jeyabalan Velandy of Crompton Greaves Limited, High Voltage Product Technology (HVPT) Centre, Global R&D Centre, Kanjur (E), Mumbai - 400 042, Maharashtra, India; an Indian National.
PREAMBLE TO THE DESCRIPTION:
The following specification particularly describes the nature of this invention and the manner in which it is to be performed:

FIELD OF THE INVENTION:
This invention relates to the field of electrical and electronics engineering.
Particularly, this invention relates to the field of high voltage engineering.
Particularly, this invention relates to partial discharges in transformer windings.
Specifically, this invention relates a system using Matched Adaptive filter approach for Partial discharge location in Transformer Windings.
More specifically, this invention relates to a system for locating Partial Discharge in windings.
BACKGROUND OF THE INVENTION:
The efficacy of a power system depends mainly on the reliability of the power equipment connected to the system. The power transformer is one of the most essential components of power generation and transmission system.
A Transformer is an electrical device that transfers energy from one circuit to another by magnetic coupling without any moving parts. It works on the Principle of Faradays Law of Electromagnetic Induction together with other laws of electricity. When a magnetic flux linked with an electric circuit varies, an electro motive force (voltage) is generated in the electric circuit, proportional to the rate of variation of flux with time.
The basic functional components of the transformer are a core made of magnetic material, a primary winding, and a secondary winding. The primary winding is connected to a source of an alternating current and the secondary winding is connected to the output. As the alternating current flows through the primary winding, the change in current in the primary windings creates a time-varying magnetic flux in the core, which induces a voltage in the secondary windings.
There is insulation medium between the windings and the windings and core of the transformer which forms the core coil assembly of the transformer. This insulation may be gas insulation, the gas being sulfur hexafluoride gas (SF6).

Any failure in the power transformer will have a crucial impact on the system operation. The detailed survey of transformer failure reports show that more than 50 percent of the failures in power transformers are due to insulation failures in transformer windings.
The winding insulation plays a vital role in the design of transformer. The main function of winding insulation has to withstand the electric stress without any major damage during the expected life time.
For a successful operation, the transformer winding insulation has to withstand the electric stress during operation. Most of the insulation failures are due to the gradual deterioration of insulation, which are mainly due to partial discharge (PD). PD may occur anywhere within the transformer winding, where the insulation becomes weak. When the transformer insulations are highly stressed, the voids present within the solid or liquid or gaseous insulation generates PD. PD can be prevented through careful design and proper selection of insulation material. Once PD is initiated. PD causes progressive deterioration of insulating materials ultimately, leading to electrical breakdown. PDs are very short pulses of durations from ns to 5 (is. Identification and locations of PD sources is an important diagnostic tool for reliable operation of power transformers.
For the location of PD, a number of electrical methods and non electrical methods have been used in the literature for identification and location of PD in transformer windings. Generally, the electrical methods are more sensitive and preferable for online PD detection and Electrical methods are standardized by International Electrotechnical Commission (IEC) and Institute of Electrical and Electronics Engineers (IEEE).
Amongst the various electrical methods that have been developed for PD detection, transfer function method have gained some acceptance. The fundamental issues in transfer function method have always been frequency resolution and Signal-to-noise ratio (SNR) at higher frequencies. It would be useful to develop methods that can work under circumstances of improving the SNR too. Hence, in communication theory, the correlation method is known to be the optimum method for detecting a pulse of known shape contaminated by additive noise with known spectral density of the signal. In other words, the correlation method is optimal for

detection of PD signal in the presence of noise provided that only the reference responses are known.
OBJECTS OF THE INVENTION:
An object of the invention is to provide a system for location of partial discharge in transformer windings during impulse test and power frequency test (high voltage test).
Another object of the invention is to provide an efficient and accurate system for location of partial discharge in transformer windings.
Yet another object of the invention is to provide a system and method for location of partial discharge in transformer windings, such that there is improved frequency resolution in determination techniques.
Still another object of the invention is to provide a system for location of partial discharge in transformer windings, such that there is improved Signal-to-noise ration in determination techniques, at higher frequencies.
SUMMARY OF THE INVENTION:
According to this invention, there is provided a system for locating partial discharge in windings, said system comprises:
a. reference database means adapted to store reference winding response signals for partial
discharge pulse durations of time durations selected from a range of pre-defined time
durations, said signals being correlative to various pre-defined sections of windings;
b. means adapted to provide winding response signals in correlation to said various pre
defined sections of windings in order to obtain signals correlative to said pre-defined
sections of windings;
c. matched adaptive filter adapted to filter said obtained winding response signals;
d. time domain analysis means adapted to obtain time domain winding response signals
from said filtered winding response signals;
e. frequency domain analysis means adapted to obtain frequency domain winding response
signals from said filtered winding response signals;

f. time based normalization means adapted to normalise said time domain winding response
signals;
g. frequency based normalization means adapted to normalise said frequency domain
winding response signals;
h. time based correlation means adapted to correlate said normalised time domain winding
response signals with said reference winding response signals for obtaining a numerical
value; and
i. frequency based correlation means adapted to correlate said normalised frequency
domain winding response signals with said reference winding response signals for
obtaining a numerical value.
Typically, said system comprises a first selection means adapted to select a value from said correlated time based values in accordance with pre-defined parameters of selection in order to obtain partial discharge location correlative to said selected value.
Typically, said system comprises a second selection means adapted to select a value from said correlated frequency based values in accordance with pre-defined parameters of selection in order to obtain partial discharge location correlative to said selected value.
Typically, said pre-defined time duration is in the region of 0.1 us to 5us.
Typically, said matched adaptive filter is a band pass filter.
Typically, said matched adaptive filter is a Butterworth band pass filter.
Typically, said reference database means comprises means to consider winding responses corresponding to pulse duration tref=T us is as reference response signals.
Typically, said system comprises mapping means adapted to map winding response signals to said pre-defined sections of windings.

Typically, said -matched adaptive filter is governed by cut off frequencies of said band pass filter being adaptive based on said winding responses and covering the entire frequency spectrum.
Typically, said time based normalization means comprises means to compute normalized correlation or time cross - correlation coefficient (pix) between peak value of said response
windings, said coefficient being defined as
Typically, said system comprises a first selection means adapted to select a value from said correlated time based values in accordance with pre-defined parameters of selection in order to obtain partial discharge location correlative to said selected value, said location selection being selected by said time based normalization means comprising means to compute normalized correlation or time cross - correlation coefficient (pix) between peak value of said response
windings, said coefficient being defined as such that the location
criterion is the section 'i' that yields maximum pix.
Typically, said frequency based normalization means comprises means to compute normalized correlation or frequency cross - correlation coefficient (pix) between peak value of said response
windings, said coefficient being defined as
Typically, said system comprises a first selection means adapted to select a value from said correlated time based values in accordance with pre-defined parameters of selection in order to obtain partial discharge location correlative to said selected value, said location selection being selected by said time based normalization means comprising means to compute normalized

correlation or time cross - correlation coefficient (pix) between peak value of said response
windings, said coefficient being defined as such that the location
criterion is the section 'i' that yields maximum pix.
Typically, said system comprises standard deviation computation means adapted compute standard deviation values as a measure of variability or diversity between responses.
Typically, said system comprises standard deviation computation means adapted compute standard deviation values as a measure of variability or diversity between responses, said standard deviation value between peak value of said response being defined as
such that the location criterion is the section 'i' that yields minimum Six.
Typically, wherein said system comprises absolute sum of log error computation means adapted compute absolute sum of log error values between peak value of the response Si (f) and rx'(f) by
such that the location criterion is the section 'i' that yields minimum ASLEix
Typically, wherein said system comprises maximum absolute deviation means adapted compute maximum absolute deviation values between peak value of the response Sj(f) and rx(f)
by such that the location criterion is the section T that
yields minimum DAB Six

Typically, said system comprises means to select minimum value of SDOMix, ASLEix and DABSix n order to obtain corresponding partial discharge location,
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
The invention will now be described in relation to the accompanying drawings, in which:
Figure 1 illustrates injection of pulses and winding responses;
Figure 2 illustrates Impulse responses of the layer winding for tpulse=l us wherein Figure 2a illustrates Time domain responses and Figure 2b illustrates Frequency domain responses;
Figure 3 illustrates Impulse responses of the continuous disc winding for tpuls=l us wherein Figure 3a illustrates Time domain responses and Figure 3b illustrates Frequency domain responses;
Figure 4 illustrates Impulse responses of the interleaved winding for tpuls=l JJ.S wherein Figure 4a illustrates Time domain responses and Figure 4b illustrates Frequency domain responses;
Figure 5 illustrates Enlarged view of interleaved winding impulse response for tpuls=l us wherein Figure 5a illustrates Time domain responses and Figure 5b illustrates Frequency domain responses;
Figure 6 illustrates Scatter diagram of interleaved winding responses for PD of tpuls=lus and
ttest=0.5us;
Figure 7 illustrates Circuit diagram of an impulse voltage generator and experimental setup for live discharge;
Figure 8a Responses of the interleaved winding in the presence of discharge (with air medium between the electrodes); and
Figure 8b Enlarged view of the oscillation due to PD (with air medium between the electrodes).

DETAILED DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
According to this invention, there is provided a system for locating partial discharge in windings.
In this system and method of this invention, correlation method (matched filter) in normalized principle is used for PD location and it is improved by using Butterworth band pass filter to remove unwanted signal from winding responses.
In accordance with an embodiment of this invention, in order to study the responses of transformer windings due to PD pulse, experimental studies were performed by injecting a voltage source of PD pulse with different pulse durations across each section of the windings. The energy distributions of the measured winding responses were calculated to decide on the time period to be considered for analysis. From the measured winding responses (in both time and frequency domain) due to injection of a 1 us pulse across the windings are considered as a reference responses of the winding for PD location. If the PD has occurred anywhere within the winding, matching coefficient has to observe the test PD winding responses due to PD pulse and make a 'best estimate / best matches' with reference impulse PD responses of the winding. The best matches will give the originating source of PD. The matched adaptive filter approach is applied for all types of transformer windings, demonstrated by measurements and validated with live discharges.
According to an exemplary embodiment, the commonly used typical transformer windings namely layer, continuous disc and interleaved winding (typically, 22kV transformer windings) were considered to study the responses of the transformer winding due to PD. All the windings were provided with tappings at every 10% of the winding. The layer winding had 500 turns and tapping were taken across every 50 turns (10% of the winding). The continuous disc and interleaved winding (with the order of interleaving as two) have 40 discs with 12 turns per disc and it was designed in such a way to have the same R, L and Cg by having equal number of turns, conductor dimension and height of the windings. Tapping were brought out at every 4 discs (10% of the winding). Earthed galvanised iron shields were placed inside the windings to represent the core.

A Voltage pulse Vpulse of duration tpuls (0.1 µs-5 jis) was injected across each section 'i' (where i =1 to Number of sections, N) and the corresponding impulse PD winding responses were measured across current measuring resistor (as shown in Figure 1 of the accompanying drawings). Reference numeral FG refers to Function Generator. Reference numeral DSO refers to digital storage oscilloscope. The measurement was done using a 10 bit Digital storage oscilloscope at the sampling rate of 5 Ms/sec. The measured winding responses were digitized into 10000 discrete samples. The measured winding responses were plotted using MATLAB functions (M-file) utilizing the discrete sample data obtained from experimental study. Figures 2, 3 and 4, of the accompanying drawings, show the impulse responses (in time domain and frequency domain) of the layer, continuous disc and interleaved windings for tpuls=l |is across the sections respectively. The similar analyses were carried out for injection of different tpuls (0.1 u,s - 5 us) across the windings.
The PD pulse suffers distortion and attenuation as it travelled from location of PD (Vpulse injected section) to the measuring terminals. Each winding had its own resonance frequencies based on the physical dimensions of the winding. For the same physical dimension of layer, continuous disc and interleaved winding, the resonance frequencies were different due to the difference in winding capacitance (series capacitance). The transient responses of the windings were mainly dependent on the resonance frequencies and it was clearly reflected in the responses of the layer, continuous disc and interleaved winding. From the Figures, the winding responses due to PD across each section were different from each other and it can be observed from the responses in both time and frequency domain. Figures 5a and 5b, of the accompanying drawings, show the enlarged view of the interleaved winding response for tpuls-l us. It is observed from the Figures that, there was a definite difference between the winding responses in magnitude, phase shift and resonant frequencies for different sections of the windings. The measured winding responses were then fed into a computer where MATLAB software routines were used to calculate the energy of the winding responses. Table 1 shows the percentage energy distribution of the transformer windings for injection of PD of tpuls=l u,s.
Table 1, below, shows Energy distribution of the transformer windings:

Transformer windings % Energy distribution of the winding response for PD of
tpulse=1 µs

99% 100%
Layer winding 1.4 µ.s 2 us
Continuous disc winding 1.6 us 2 us
Interleaved disc winding 2.3 us 3 us
Table 1
It was observed from the table, 99% of the energy was within 1.4 us for the case of layer winding and remaining 1% was at 1.4 us to 2 us. For the case of PD pulse 0.1 us, 99% of the energy was within 1.38 us and remaining 1% was from 1.38 us to 2 µs. Hence, the winding responses were considered up to maximum energy region (99%) to improve the efficacy of the matched adaptive filter approach.
In accordance with an embodiment of this invention, there is provided a reference database means adapted to store reference winding response signals for partial discharge pulse durations of time durations selected from a range of pre-defined time durations, said signals being correlative to various pre-defined sections of windings.
In general, a set of reference winding responses (signals) are necessary to locate the PD by using matched adaptive filter approach. Generally, one can consider any Vpulse of duration tpulse for reference signals. A reference voltage Vpule=Vrcf of pulse duration tpulse=tref (1 us) is applied across each section T (i =1 to 8) and the corresponding PD winding responses Sj(t) where (i =1 to 8) are considered as a reference winding responses.
In practice, the location of PD occurrence and shape of PD pulse is random in nature. Hence, a test PD signal Vtest of pulse durations ttest(ttest need not be same as tref) occur across any section 'x' (PD location) and the test PD impulse winding response be rx(t). In the system and method of

this invention, ttest is considered in the range of 0.1 us - 5 us. In practice, if a real PD response (test PD response) of the winding rx(t) is measured at the current measuring resistor across High Frequency Current transformer (HFCT), by an appropriate method one has to identify / estimate the section 'x' using s,(t) and rx(t). The system and method of this invention can establish a direct mapping between the response Sj(t) and the location of PD.
In accordance with another embodiment of this invention, there is provided a means adapted to provide winding response signals in correlation to said various pre-defined sections of windings in order to obtain signals correlative to said pre-defined sections of windings.
Initially, 'm' discrete sample data Sj[m] of Sj(t) are plotted against corresponding 'nr discrete sample data rx[m] of rx(t) to study the different correlation function on scatter diagram. Figure 6 shows the scatter diagram between si[m] and rx[m] of interleaved winding response for trcf=lu.s and ttest=0.5us. When the response si[m] is plotted against rx[m], the scatter diagram follows a complete correlation (linear) compared to remaining signal (S2[m], ... ,S8[m]) arid it is inferred that the PD location is in section 1 (as shown in Figure 6 of the accompanying drawings). The similar analysis is performed for all sections of interleaved winding and analyses are extended to continuous disc and layer windings. In practice, as the scatter diagrams do not provide conclusive evidence for PD location graphically, an attempt has been made to find the degree of similarity in numerically for PD location.
In accordance with yet another embodiment of this invention, there is provided a time domain analysis means adapted to obtain time domain winding response signals from said filtered winding response signals.
In accordance with yet another embodiment of this invention, there is provided a time based normalization means adapted to normalise said time domain winding response signals.
In accordance with yet another embodiment of this invention, there is provided a time based correlation means adapted to correlate said normalised time domain winding response signals with said reference winding response signals for obtaining a numerical value.

For estimating the location of PD numerically, the time domain correlation method is proposed earlier using a lumped layer winding. Initially, time domain correlation method is used for different PD pulse durations (ttest)- The time domain correlation method places restrictions for PD identification with different pulse durations of test PD winding response. For signal analysis point of view, if the sequence of test PD signal with difference form is presented, then the detection of PD signal matched with a priori reference signal is difficult.
In the system and method of this invention, principle of normalization is used in correlation method for PD location. In general, normalization principle of correlation finds the degree of linearity between rx(t) and S[(t).
It is sometimes convenient to study the winding response in a normalized manner to compare voltages at different ttest and voltages. As ttcst can have any duration from 0.1 us to 5 us with the different magnitude of Vtcst, normalization is required for correlation. The normalized correlation or more precisely time cross - correlation (reference cross correlation) coefficient (piX) between peak value of the response Sj(t) and rx(t) can be defined as

The location criterion is the section T that yields maximum piX.
The value ofpiX varies from -1 to 1. The highest value of piX indicates the stronger correlation (i.e., closer to 1). The system and method, of this invention, is tried for ttest and ttest==ttest with ttest=0.1 |JS -5 us. For example, Table 2 shows the maximum value of piX(bold with under lined data) when ttest =1 us, ttest=0.1 us, 0.5 us, 1 \is and 5 us. The winding section pair (i=x) corresponding to the maximum value is an estimate of PD location.
Table 2, below, illustrates Normalized time domain correlation outputs (pjX) of interleaved winding

Sections vref =1V, tref=l us
ttest- 1 2 3 4 5 6 7 8
5|IS 0.665 0.981 0.819 0.777 0.748 0.790 0.885 0.783

6 0.514 0.802 0.925 0.924 0.952 0.990 0.929 0.617

8 0.739 0.613 0.415 0.402 0.418 0.516 0.714 0.988
luS 2 0.657 1.000 0.871 0.832 0.800 0.817 0.868 0.707

6 0.541 0.817 0.941 0.948 0.950 1.000 0.925 0.608

8 0.771 0.707 0.522 0.505 0.518 0.608 0.785 1.000
0.5us 2 0.591 0.974 0.881 0.845 0.807 0.810 0.824 0.591

6 0.475 0.769 0.941 0.957 0.962 0.975 0.843 0.472

8 0.746 0.753 0.564 0.544 0.556 0.647 0.822 0.987
0.1 us 2 0.407 0.831 0.850 0.826 0.771 0.718 0.623 0.249

6 0.297 0.548 0.786 0.821 0.830 0.728 0.468 0.164

8 0.618 0.934 0.821 0.788 0.786 0.828 0.881 .7610
Table 2
From the Table 2, it can be seen that the system and method of this invention locates the PD for the duration of 0.5 us-5 us and it is found that the method fails to locate the PD for the case of below 0.5 us. Hence, to improve the method normalized correlation in frequency domain is attempted.
In this case, a Fast Fourier Transform (FFT) of the time domain response at the measuring terminal indicates the frequency content of the PD pulse and its location. Hence, the frequency domain winding response Sj(f) are used. The frequency responses of the winding will always contain noise floor at high frequencies. The often encountered difficulty in PD location method is to discriminate PD signals from various noises. The sophisticated signal filtering method is required to extract the correct frequency components alone before applying method. Hence, Butterworth band pass filter is used. Because, the frequency characteristics of most PD detector

systems as seen by a PD pulse can be considered as band pass in nature. Thus, a band pass filter is considered. For signal analysis point of view, the advantage of Butterworth filter has a monotonically changing magnitude with frequency in the pass band and / or the stop band compared to other filters. The Butterworth filter has a slower rolls-off around the cutoff frequency and it will eliminate the ripple contents.
The cut off frequencies of the band pass filter is adaptive based on the winding responses and cover the entire frequency spectrum. After observing the winding response s;(f) and rx(f), the frequency band in which maximum signal strength is identified. Frequencies outside this band are filtered and the signal shape within this pass band is considered for further analysis. In this case, the cut off'frequency of the filter is 0.1 xlO4 kHz (lower frequency) - 2xl04 kHz (upper frequency). Hence, here after the filtered responses of the windings are represented as Sj(f) & TX (f) and it is used for normalized correlation method.
The peak value of the winding responses Si(f) and rx(f) are normalized such that the maximum value of each response occurs at OdB. Thus, the correlation coefficient (pix) between peak value of the response Sj(f) and rx(f) can be calculated as

The location criterion is the section T that yields maximum pjX.
In accordance with this invention, the normalized frequency domain correlation with butterworth band pass filter locates the PD for ttest below 0.5 us and is shown in Table 3. The maximum value of pix (bold with under lined data) gives the PD location. Tables 4 and 5, below, show the output of continuous disc and layer windings, respectively. The method locates the PD source for ttest other than ttestfor all the sections.
Table 3, below, illustrates Normalized Frequency domain correlation outputs (pix) of interleaved winding

Sections
Vref=lV,tref=l US
ttest- 1 2 3 4 5 6 7 8
0.1 µs 2 0.816 0.986 0.951 0.957 0.922 0.910 0.923 0.877

6 0.709 0.907 0.966 0.977 0.951 0.992 0.966 0.848

8 0.867 0.974 0.931 0.918 0.911 0.922 0.951 0.987
Table 3
Table 4, below, illustrates Normalized Frequency domain correlation outputs (pjX) of continuous disc winding

Sections vref =1V, tref=l us
ttest" 1 2 3 4 5 6 7 8
5 µs 1 0.999 0.968 0.962 0.964 0.956 0.967 0.968 0.936

3 0.963 0.964 0.999 0.964 0.972 0.952 0.920 0.964

8 0.924 0.946 0.955 0.941 0.941 0.954 0.938 0.998
1 µS 1 1.000 0.965 0.958 0.961 0.954 0.964 0.965 0.928

3 0.958 0.943 1.000 0.964 0.912 0.964 0.931 0.964

8 0.928 0.955 0.964 0.950 0.954 0.964 0.948 LOOO
0.1µs 1 0.994 0.954 0.956 0.921 0.954 0.963 0.952 0.917

5 0.952 0.950 0.994 0.962 0.979 0.960 0.936 0.948

8 0.931 0.964 0.952 0.961 0.965 0.973 0.957 0.992
Table 4
Table 5, below, illustrates Normalized Frequency domain correlation outputs (pix) of layer winding

Sections Vref=lV,tref=l US
ttest= \i x\ 1 2 3 4 5 6 7 8
1 0.998 0.883 0.922 0.957 0.916 0.963 0.896 0.947
5|is 5 0.921 0.961 0.930 0.937 0.995 0.958 0.959 0.963

7 0.917 0.921 0.959 0.960 0.956 0.916 0.993 0.942

1 (J.S
1 1.000 0.886 0.930 0.956 0.921 0.956 0.899 0.942

5 0.921 0.963 0.928 0.953 1.000 0.952 0.963 0.966

7 0.899 0.981 0.961 0.961 0.963 0.920 1.000 0.946
O.lfxs 1 0.987 0.918 0.956 0.957 0.920 0.953 0.942 0.952

5 0.887 0.967 0.927 0.967 0.990 0.937 0.963 0.958

7 0.882 0.950 0.955 0.926 0.918 0.885 0.985 0.915
Table 5 In accordance with another embodiment of this invention, there is provided means for obtaining co-efficient of determination.
The correlation coefficient (pix) represents the linear relationship between two signals. If the correlation coefficient is squared, then the resulting value (p^ , the coefficient of determination) will represent the proportion of common variation in the two signals (i.e., the "strength" or "magnitude" of the relationship). In order to evaluate the correlation between signals, it is important to find "magnitude" or "strength" as well as the significance of the correlation. For example, if two signals have an p;x value of 0.998, the coefficient of determination is 0.996. Hence, coefficient of determination states that only 99.6% of the change in Si(f) can be explained by a change in rx(f). The larger the correlation coefficient, the larger the coefficient of determination and the more influence changes in the Si(f) have on the rx(f). Table 6 shows the interleaved winding output of coefficient of determination.
Table 6, below, illustrates Coefficient of determination outputs (pjX2) of interleaved windings

Sections ■ - \ vref=rv,w=iµs '
ttesl= 1 2 3 4 5 6 7 8
5µs 1 99.60 77.96 85.0 91.58 83.9 92.73 80.28 89.68

5 84.82 92.35 86.49 87.79 99.0 91.77 91.96 92.73

7 84.08 84.82 91.96 92.16 91.39 83.9 98,6 88.73
0.1 µs 1 97.47 84.27 91.39 91.58 84.64 90.82 88.73 90.63

5 78.67 93.50 85.93 93.5 98.01 87.79 92.73 91.77

7 77.79 90.25 91.20 85.74 84.27 78.32 97.02 83.72
Table 6

In accordance with another embodiment of this invention, there is provided a standard deviation computation means adapted compute standard deviation values.
Standard deviation is a used measure of variability or diversity between the responses. It shows how much variation exists from the average value. A low value of standard deviation indicates that the discrete sample region of the winding responses tend to be very close to the mean, whereas high standard deviation indicates that the discrete sample region are spread out over a large range of values. Thus, the standard deviation (Six) between peak value of the response Sj(f) and rx(f) can be calculated as

The location criterion is the section 'i' that yields minimum S-tx.
The standard deviation approach locates the PD pulse from the durations of 0.1µ.s-5µs for all sections. For example, Table 7 shows the interleaved winding output of standard deviations for the duration of 0.5µ.s. The minimum value of Six (bold with under lined data) gives the PD location within the windings.
Table 7, below, illustrates Standard deviations outputs (SDjx) of interleaved windings

Sections Vref=lV, tref=l µS
ttest V 1 2 3 4 5. ;6 7 8.
0.5us 1 39.4 79.6 78. 1 80.2 94.3 88.2 91.9 119.5

5 106.2 65.2 58.9 56.9 39.5 55.3 65.1 114.0

7 95.0 59.4 51.0 52.1 46.4 50.4 41.6 97.8
Table 7
In accordance with another embodiment of this invention, there are provided computation means for Standard deviation of mean (SDOM) and absolute sum of log error (ASLE) approach.

The accuracy of the discrete sample regions between the responses of the windings can be further examined by looking at the standard deviation of the means (SDOM), absolute sum of log error (ASLE) and maximum absolute deviation (DABS)
Thus, the absolute sum of log error (ASLEjx) and DABSjx between peak value of the response Sj (f) and rx (f) can be calculated as
The location criterion is the section T that yields minimum ASLEix and DABSiX.
After calculating the mean of each set of discrete data, the deviation between the means of the data can be calculated by taking the standard deviation of all the mean values. The standard deviation of mean and absolute sum of log error approach locates the PD pulse from the durations of 0.1µs-5vµs for all sections. For example, Tables 8 and Table 9 show the interleaved winding output of standard deviations of mean and absolute sum of log error for the duration of 0.5(is respectively. The minimum value of SDOMiX, ASLEix and DABSjx (bold with under lined data) gives the PD location.
Table 8, below, illustrates Standard deviations of means outputs (SDOMjx) of interleaved windings

Table 9, below, illustrates, Absolute sum of log error outputs (ASLEii:) of interleaved windings

Sections Vref=lV, tref=l µS
ttest" 12 3. 4 5 6 7 8
0.5us 1 4.837 7.050 7.185 6.943 8.097 7.803 7.891 9.092

5 8.393 6.692 6.529 6.189 4.749 5.798 6.369 8.690

7 7.504 6.136 5.470 5.482 5.449 5.181 4.994 7.587
Table 9
Table 10, below, illustrates Standard deviations outputs (DABSjX) of interleaved windings

Sections Vref=lV,trerl "s
ttest~ 1 2 3 4 5 6 7 8
1 2.99 6.28 6.09 6.27 7.59 7.26 7.27 9.09
0.5us 5 8.42 5.38 4.86 4.74 3JZ 4.48 5.03 8.82
7 7.30 4.77 3.97 4.11 3.67 3.86 2.35 7.26
Table 10
To validate the method, the real experimental winding responses due to PD injection by live discharges are considered for the PD location. The live discharges are generated across the different sections of the winding by applying standard lightning impulse voltage (LI) across the entire winding and the corresponding winding responses are measured. Figure 7 shows the test setup to create the live discharge during impulse voltage excitation. To create a live discharge across a section 'x', the plane-needle electrode configuration is used, which has a distinct advantage of visual indication of discharge between the electrodes. The plane electrode is designed with a smooth surface and curved edges to avoid corona losses. The type of discharge across the section depends on the dielectric medium between the electrodes, applied voltage, temperature, inter electrode gap, etc.

Figure 7 illustrates Circuit diagram of an impulse voltage generator and experimental setup for live discharge.
T refers to Transformer, 230V/I00kV.
D1, D2 refers to Diodes,l40kV,20mA.
Cs refers to Charging Capacitor, 2500pF,
S refers to Sphere Gap.
CDI, CD2 refers to Capacitive divider, 1200pF.
12 refers to impulse voltage generation setup.
14 refers to live discharge test setup.
A 120kV, 250kJ impulse voltage generator (MWB-make) is used for generating the standard lightning impulse voltage. The real impulse PD responses are measured for the following cases:
- Medium between the electrodes: air, press board
- Location of discharges: across all sections of the windings
- Windings: 22kV transformer windings
The impulse voltage is increased till a discharge across the electrodes occur, thus generating PD across a section of the winding. The occurrence of PD is observed as oscillations in the winding response. Different types of discharges with air and press board between the electrodes are generated. Figures 8a and 8b show the responses of the interleaved winding due to the presence of discharge (with air medium between the electrodes) across the sections x=l,4 and the enlarged view of the oscillation due to PD respectively. The matched adaptive filter approach locates the PD pulse for all sections- For example, Table 10 shows the outputs of the interleaved winding for injection of live discharge (air medium between the electrodes) across the section 1 and 4.
Figure 8a Responses of the interleaved winding in the presence of discharge (with air medium between the electrodes).
Figure 8b Enlarged view of the oscillation due to PD (with air medium between the electrodes).

Table 10, below, illustrates Normalized Frequency domain correlation outputs (p;x) of interleaved winding for live discharge.

Sections vref=iv,trert =µs
ttest= s 12 3 4 5 6 7 8
Live PD 1 0.971 0.934 0.902 0.937 0.925 0.942 0.925 0.912

4 0.911 0.942 0.916 0.985 0.902 0.903 0.943 0.936
Table 10
In the previous studies, the principle of correlation (in frequency domain) with Gram Schmidit (G-S) orthogonalization had been formulated to detect the PD location for the duration of 0.5 µ,s to 5µ.s. Hence, to improve the efficacy of the method Butterworth band filter is adopted in the winding responses. The location of PD has been done successfully for the durations of 0.1 us to 5\is.
The normalization principle considerably simplifies practical implementation of the procedure for PD location and investigates free-form signals as opposed to orthogonalization method, which cannot use to signal with flat regions.
The major focus of the system and method of this invention is in identifying the location of PD for different PD pulse durations during impulse test within the transformer windings using matched adaptive filter approach. The probabilities of error between reference responses and PD response is based signal energy. The shape of the PD signal and its duration will not effect the PD location.
The location of PD using matched adaptive filter approach is considerably difficult problem as without priori known information exists regarding the reference winding response. Hence, one can use any one of the following general case for reference responses.

• The reference responses are estimated theoretically using an equivalent winding circuit model by injecting reference / known pulse across the different tapping of the transformer winding.
• The references responses can measure from any identical windings (unfault of other phase winding) by injecting reference pulse across the different tapping of the winding.
To analyse the winding behaviour for PD with different durations, commonly used transformer windings are used. The winding responses are measured for the PD pulse durations of 0.1 µs to 5µs. From the winding responses in time and frequency domain, one can observe the definite difference in the winding responses for PD at different sections. The ' winding responses corresponding to pulse duration tref=l µs is considered as reference responses for matched adaptive filter approach.
Already proposed time domain correlation method locates the PD for ttest = ttest. As the PD pulses are of different pulse durations in practice, correlation with normalized signals in time domain is attempted and it is found that the method locates the PD for the duration of 0.5 µ.s - 5 µ.s. In order to improve the method the analysis are carried out in the frequency domain with ButterwortK band pass filter and locates the PD pulse durations of below 0.5 µs. The efficacy of the method for different discharge pulse durations is validated by using live discharges. The proposed invention is validated on all types of transformer windings and the method locates the PD successfully for all the sections (tappings) with difference PD pulse durations.
The technical advancement of this invention lies in provisioning the system for not only detection partial discharge but also locating it's position in transformer windings using a matched adaptive filter and correlation techniques which use normalized frequency domain outputs of sensed signals along with normalized time domain outputs of sensed signals. Pulses used in this system for sensing are of durations from 0.1 us to 5 us. Also, the current system and method is validated on all types of transformer windings and the method locates the partial discharge successfully for all the sections (tappings) with different partial discharge pulse durations.

While this detailed description has disclosed certain specific embodiments of the present invention for illustrative purposes, various modifications will be apparent to those skilled in the art which do not constitute departures from the spirit and scope of the invention as defined in the following claims, and it is to be distinctly understood that the foregoing descriptive matter is to be interpreted merely as illustrative of the invention and not as a limitation.

We claim,
1. A system for locating partial discharge in windings, said system comprising:
a. reference database means adapted to store reference winding response signals for
partial discharge pulse durations of time durations selected from a range of pre
defined time durations, said signals being correlative to various pre-defined sections
of windings;
b. means adapted to provide winding response signals in correlation to said various pre
defined sections of windings in order to obtain signals correlative to said pre-defined
sections of windings;
c. matched adaptive filter adapted to filter said obtained winding response signals;
d. time domain analysis means adapted to obtain time domain winding response signals
from said filtered winding response signals;
e. frequency domain analysis means adapted to obtain frequency domain winding
response signals from said filtered winding response signals;
f. time based normalization means adapted to normalise said time domain winding
response signals;
g. frequency based normalization means adapted to normalise said frequency domain
winding response signals;
h. time based correlation means adapted to correlate said normalised time domain
winding response signals with said reference winding response signals for obtaining a
numerical value; and i. frequency based correlation means adapted to correlate said normalised frequency
domain winding response signals with said reference winding response signals for
obtaining a numerical value.
2. The system as claimed in claim 1, wherein said system comprising a first selection means adapted to select a value from said correlated time based values in accordance with predefined parameters of selection in order to obtain partial discharge location correlative to said selected value.
3. The system as claimed in claim 1, wherein said system comprising a second selection means adapted to select a value from said correlated frequency based values in

accordance with pre-defined parameters of selection in order to obtain partial discharge location correlative to said selected value.
4. The system as claimed in claim 1, wherein said pre-defined time duration is in the region of 0.1 µs to 5 µs.
5. The system as claimed in claim 1, wherein said matched adaptive filter is a band pass filter.
6. The system as claimed in claim I, wherein said matched adaptive filter is a Butterworth band pass filter.
7. The system as claimed in claim 1, wherein said reference database means comprising means to consider winding responses corresponding to pulse duration tref=l us is as reference response signals.
8. The system as claimed in claim 1, wherein said system comprising mapping means adapted to map winding response signals to said pre-defined sections of windings.
9. The system as claimed in claim 1, wherein said -matched adaptive filter being governed by cut off frequencies of said band pass filter being adaptive based on said winding responses and covering the entire frequency spectrum.
10. The system as claimed in claim 1, wherein said time based normalization means comprising means to compute normalized correlation or time cross - correlation coefficient (pix) between peak value of said response windings, said coefficient being
defined as
11. The system as claimed in claim 1, wherein said system comprising a first selection means
adapted to select a value from said correlated time based values in accordance with pre
defined parameters of selection in order to obtain partial discharge location correlative to

said selected value, said location selection being selected by said time based normalization means comprising means to compute normalized correlation or time cross -correlation coefficient (pix) between peak value of said response windings, said
coefficient being defined as such that the location criterion is
the section 'i' that yields maximum piX.
12. The system as claimed in claim 1, wherein said frequency based normalization means
comprising means to compute normalized correlation or frequency cross - correlation
coefficient (piX) between peak value of said response windings, said coefficient being
defined as
13. The system as claimed in claim 1, wherein said system comprising a first selection means
adapted to select a value from said correlated time based values in accordance with pre
defined parameters of selection in order to obtain partial discharge location correlative to
said selected value, said location selection being selected by said time based
normalization means comprising means to compute normalized correlation or time cross -
correlation coefficient (piX) between peak value of said response windings, said
coefficient being defined as such that the location criterion is
the section 'i' that yields maximum piX.
14. The system as claimed in claim 1, wherein said system comprising standard deviation computation means adapted compute standard deviation values as a measure of variability or diversity between responses.
15. The system as claimed in claim 1, wherein said system comprising standard deviation

computation means adapted compute standard deviation values as a measure of variability or diversity between responses, said standard deviation value between peak
value of said response being defined as such that the location
criterion is the section T that yields minimum Six.
16. The system as claimed in claim 1, wherein said system comprising absolute sum of log
error computation means adapted compute absolute sum of log erronvalues between peak
value of the response s\ (f) and rx (f) by
such that the location criterion is the section 'i' that yields minimum ASLEiX
17. The system as claimed in claim 1, wherein said system comprising maximum absolute
deviation means adapted compute maximum absolute deviation values between peak
value of the response Sj (f) and rx (f) by such that the
location criterion is the section T that yields minimum DABSjX
18. The system as claimed in all preceding claims wherein, said system comprising means to
select minimum value of SDOMiX, ASLEiX and DABSiX n order to obtain corresponding
partial discharge location.