You are on page 1of 10

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO.

2, APRIL 2005 1273

Modeling and Protection of a Three-Phase Power


Transformer Using Wavelet Packet Transform
S. A. Saleh, Student Member, IEEE, and M. A. Rahman, Fellow, IEEE

Abstract—This paper introduces a novel algorithm for differ- have been different algorithms to carry out the harmonic anal-
ential protection of three-phase power transformers that is based ysis. Among these algorithms, sine cosine correlations, rectan-
on the wavelet packet transform. The wavelet packet transform is gular transform, discrete Fourier transform (DFT), least-square
employed to extract certain features of the differential current to
distinguish between the magnetizing inrush and different internal method, Walsh functions, Haar functions and Kalman filtering
fault currents. The selection of the optimal wavelet analysis that in- technique, etc. are significant [1]–[3]. The main drawbacks of
cludes selecting both the optimal mother wavelet and the optimal the 2nd harmonic restraint approach are: the 2nd harmonic may
number of levels of resolution is carried out using the minimum exist in some internal faults of the transformer windings. In ad-
description length (MDL) data criteria. The proposed algorithm dition, the new low-loss amorphous core materials in modern
is tested off-line using data collected from a prototype laboratory
three-phase power transformer. The test results show reduced com- power transformers is capable of producing magnetizing inrush
putational burden, high speed and high accuracy. currents with low 2nd and 5th harmonic contents [2]–[4]. The
previous work on power transformer protection has included
Index Terms—Daubechies wavelet (db4), differential protection,
discrete Fourier transform (DFT) and harmonic restraint differ- other approaches, among these approaches; transformer induc-
ential protection, internal faults, magnetizing inrush current, min- tance during saturation, artificial neural networks (ANN), flux
imum description length data criteria, power transformer, wavelet and voltage restraints and fuzzy logic [2]–[4]. These approaches
transforms. have high dependence on parameters of the protected trans-
former. In addition, they require complex algorithms to carry
I. INTRODUCTION out the required computations [3].
The frequency analysis can be an effective technique to an-

P OWER transformer protection has been always a chal-


lenging problem for protection engineers. The main
concern in protecting this particular element of power systems
alyze and classify signals with complex characteristics. Such
effective analysis is achieved by employing new and efficient
signal processing tools. The traditional signal processing tools
lies in the accurate and rapid discrimination between mag- used for frequency analysis are based on the conditions of sta-
netizing inrush and different internal faults currents [1]–[3]. tionarity and periodicity. However, disturbances in power sys-
Magnetizing inrush currents sometimes have high magnitudes tems are of a nonperiodic, nonstationary, short duration and
that cause traditional differential relays to initiate trip actions impulse super-imposed natures [5]. Efficient frequency anal-
disconnecting the protected transformer from the system. Such ysis should be able to overcome the limitations of the tradi-
mal-operation of differential relays can affect both the reliability tional signal processing tools. The wavelet analysis is one of
and stability of the whole power system [3]. Many researches the newly applied frequency analysis tools for processing sig-
have been conducted to analyze magnetizing inrush currents in nals with complex characteristics.
three-phase and single phase power transformers. It was found The wavelet analysis and wavelet transforms have emerged
that magnetizing inrush current is dependent on many factors recently as a powerful tool for signal processing in different
including the instant of connecting the supply to the power applications, in particular now for power system applications.
transformer, the residual flux, the type of winding connections, The transient characteristics of wavelets can be employed to
core types, materials and load characteristics [2], [3]. In addition, carry out accurate and effective analysis of signals with complex
magnetizing inrush current contains a high amount of the second frequency-time structure. Moreover, the wavelet analysis acco-
and sometimes the fifth harmonics [1]–[4]. modates a nonuniform bandwidths, such that the bandwidth is
Most of the conventional transformer protection relays em- higher at higher frequencies, making it possible to implement
ploy the harmonic analysis approach to identify the type of the the wavelet analysis through different levels of decimation in
current that flows in the protected transformer. The main idea a filter bank [3]. The applications of wavelet analysis in power
of the harmonic restraint differential relays is to extract the fun- systems include analysis and detection of electromagnetic tran-
damental (1st), the second (2nd) and sometimes the fifth (5th) sients, power quality assessment, data compression, and fault
harmonics and to compare the ratios of the 2nd and 5th har- detection [5]. The wavelet analysis has been recently used in
monics with 1st to a predefined threshold value [2], [3]. There current differential pilot relay, where current diagnosis is based
on comparing the the first level approximation with a predefined
Manuscript received October 16, 2003; revised February 27, 2004. Paper no. threshold value [6]. Another application of wavelet analysis,
TPWRD-00513-2003. which is based on the distribution of energy over different lin-
The authors are with Memorial University of Newfoundland, St. John’s, NL
A1B 3X5, Canada (e-mail: saleh@engr.mun.ca). early divided frequencies was used to identify various types of
Digital Object Identifier 10.1109/TPWRD.2004.834891 currents flowing through a power transformer [8]. The data used
0885-8977$20.00 © 2005 IEEE
1274 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005

TABLE II
TWO-LEVEL MDL INDEX EVALUATIONS FOR UNLOADED MAGNETIZING
INRUSH CURRENT DATA

TABLE III
TWO-LEVEL MDL INDEX EVALUATIONS FOR PRIMARY SIDE SINGLE
PHASE-TO-GROUND FAULT CURRENT DATA

Fig. 1. Experimental setup for differential current data collection.

TABLE I
TWO-LEVEL MDL INDEX EVALUATIONS FOR NORMAL CURRENT DATA

TABLE IV
FOUR-LEVEL MDL INDEX EVALUATIONS USING DAUBECHIES
MOTHER WAVELET (db4)

in simulating these applications is generated from softwares like


EMTP and not collected from real power transformers [7]–[9].
There are different wavelet families, and they are classified
Appendix I. The application of any wavelet transform (WT) on
according to the characteristics of the generated basis functions.
any signal produces details and approximation at each level of
Wavelet families are classified as orthogonal, biorthogonal and
resolution. The absolute value of the second level details of the
nonorthogonal [10]. Daubieches, Coiflet, Symlet and Meyer are
measured differential current samples is compared to zero, and
examples of orthogonal wavelet families, while B-Spline is an
the result of the comparison is used to distinguish between the
example of biorthogonal wavelet families. Morlet, Gaussian and
fault and nonfault currents.
Mexican Hat are examples of the nonorthogonal wavelet fam-
ilies [2], [9]. Appropriate selection of the mother wavelet for
II. MINIMUM DESCRIPTION LENGTH (MDL) DATA CRITERIA
signal representation can maximize the advantages of this tech-
nique. Moreover, the wavelet analysis will be simplified in terms The basic idea of the wavelet analysis is to adapt a wavelet
of the required number of levels of analysis [3], [10]–[12]. One function (a mother wavelet) to the analyzed signal at certain fre-
of the new methods for optimal wavelet analysis selection is the quency at certain time [2], [9]. There are three types of wavelet
minimum description length (MDL) data criteria. This method transform, which are continuous wavelet transform (CWT), dis-
is based on the optimal number of wavelet coefficients to be re- crete wavelet transform (DWT) and wavelet packet transform
tained for the signal reconstruction [7]. (WPT) [10]–[12]. The WPT is briefly described in Appendix I.
This paper proposes a new algorithm based on the wavelet The key for optimal wavelet analysis selection lies in the ap-
packet transform (WPT) to diagnose magnetizing inrush, in- propriate parametrization and the adequate performance mea-
ternal fault and normal (through-fault) currents in three-phase sure as well as the accurate interpretation of the physical phe-
power transformers. At first, a brief introduction of MDL cri- nomenon [7]. A particular setup for the problem of selecting the
teria for selecting the optimal mother wavelet and the optimal mother wavelet can be defined as selecting orthonormal basis
number of levels of resolution have been presented. Then, the functions with compact support, which helps in implementing
data collection procedure and the type of investigated differen- the wavelet transform using filter banks [3], [7]. Such compact-
tial currents are illustrated. Also, the application of the MDL cri- ness provides means of isolation and detection of frequency
teria on the collected data has been applied. Finally, the off-line variations with time regardless of the nature of such variations.
test results for some cases are presented. The same collected These capabilities have proved useful in many signal processing
data are tested off-line using a harmonic restraint approach. The applications [3], [7].
harmonic restraint approach is carried out through extracting the The MDL criteria is one of the new criterion used for op-
1st, 2nd, and the 5th harmonics using discrete Fourier transform timal wavelet analysis selection. The MDL criteria is free from
(DFT) [2]. The wavelet packet transform is briefly presented in any parameter settings or thresholding, which makes it adequate
SALEH AND RAHMAN: MODELING AND PROTECTION OF A THREE-PHASE POWER TRANSFORMER USING WAVELET PACKET TRANSFORM 1275

Fig. 2. WPT first and second level details coefficients of the normal current.

for data compression and denoising applications [5], [10], [11]. between and . It should be noted that the residual en-
The algorithm for this approach is basically picking a function ergy decreases as increases [3], [7]. The number of coefficients
known as index defined for a signal as [7] for which the MDL function reaches its minimum value, is
considered as an optimal one. It should be noted that a wavelet
analysis that is optimal for a certain application, is not neces-
sarily the optimal for another application [10]–[13].
(1)
In this paper, the MDL is applied on the data for differential
where represents the length of the signal current samples, which are collected from a prototype labora-
the number levels of resolution used for decomposing tory three-phase power transformer.
the signal , which is given by [7]
III. DATA COLLECTION
(2)
Many currents are investigated for the purposes of data col-
denotes to the vector of decomposition coefficients of up lection for both selecting the optimal mother wavelet and the
to level that is given by [7] optimal number of levels of resolution. In addition, these col-
lected current data are to be used for testing the performance
(3)
of the proposed algorithm. The following cases are presented in
represents a vector that contains nonzero elements and this paper.
is the hard-thresholding operation that keeps the th largest ele- 1) Fault Currents
ments of in absolute value intact and set all other elements • Secondary Line-to-Ground Fault, Loaded
to zero [10]–[12]. The vectors and have to be normal- • Primary Line-to-Line Fault, Unloaded
ized by so that the magnitude of each coefficient in • Primary Three-phase Fault, Unloaded.
and vectors is strictly less than 1 [7]. The first term of the 2) Normal Current
MDL index represents the penalty function that is increasing 3) Magnetizing Inrush Current, Unloaded.
linearly with the number of the retained wavelet coefficients , In many of the investigated cases, the transformer is not loaded,
while the second term describes the logarithmic energy residual which provides the worst conditions of operation. It is to be
1276 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005

Fig. 3. WPT first and second level details coefficients of the unloaded magnetizing inrush current.

noted that for the case of normal current, a three-phase resistive When the sampled differential currents are downloaded to
load is connected to the transformer that draws 6A. PC, they are converted to data files for further processing using
Mathworks Matlab Wavelet Toolbox.
A. Experimental Setup
The proposed WPT algorithm is to be tested on a three-phase, IV. OPTIMAL WAVELET ANALYSIS SELECTION
5 kVA, 230/550–575–600, 60 Hz, core type -Y laboratory pro- When the samples of the differential current are collected and
totype power transformer. The experimental setup for collecting converted into data form, the MDL criteria can be applied. The
different types of investigated differential currents is shown in application of the MDL criteria is intended to select the optimal
Fig. 1. mother wavelet and the optimal number of levels of resolution.
Three identical current transformers (CTs) are connected in
Y on the primary side, and another three identical CTs are con- A. Selection of the Optimal Mother Wavelet
nected in on the secondary side of the power transformer. The
The MDL index can be calculated after the wavelet coeffi-
differential current is measured at point ‘Q’ using Tektronics
cients are determined. The wavelet coefficients can be deter-
current probe. Magnetizing inrush current may have different
mined using the following relation [7], [11]
waveforms depending on the residual magnetizim and the in-
stant of switching the supply to the transformer. In this work, (4)
different inrush current waveforms are investigated. In all the
investigated cases, the differential current is sampled at 10 kHz where is a square matrix contains the row basis vectors.
and stored using Tektronics 2212 digital storage oscilloscope. These row basis vectors are specified by numbers known as
These waveforms can be downloaded to the computer using wavelet and scaling filter coefficients generated by the mother
Grab 2212 software package through a serial port. Three iden- wavelet [7], [11]. In this work, the square matrix is
tical triac switches and a control circuit to trigger them are used evaluated using the Matlab Wavelet Toolbox for each candidate
to provide a short contact between the supply and the trans- mother wavelet. As mentioned in Section II, there are several
former. The configuration of the switches and the control circuit wavelet families that can be used for power system applications
is shown in Appendix II. [3]. In this paper, the candidate mother wavelets are [10]–[12]:
SALEH AND RAHMAN: MODELING AND PROTECTION OF A THREE-PHASE POWER TRANSFORMER USING WAVELET PACKET TRANSFORM 1277

Fig. 4. WPT first and second level details coefficients of the secondary loaded single line to ground fault current.

• Orthogonal Wavelet Families family has the smallest MDL indices for the three types of cur-
1) Daubechies (db4) rent data. Moreover, the MDL indices for the second level of
2) Coiflet (coif4) resolution also are the smallest. The approximations can contain
3) Symlet (Sym4) most of the information about the signal when moving from a
4) Meyer. lower level of resolution to a higher one. The Daubechies mother
• Biorthogonal Wavelet Families wavelet is selected to be employed in all the wavelet analysis
needed for realizing the proposed wavelet packet-based differ-
1) B-Spline (bior2.2)
ential protection algorithm.
• Nonorthogonal Wavelet Families
1) Morlet B. Selecting the Optimal Number of Levels of Resolution
2) Gaussian (gaus4) The optimal number of levels of resolution indicates the min-
The MDL is applied for each candidate mother wavelet with imum number of changing the scale during the multiresolu-
three types of collected data, which are normal load current, tion analysis (MRA) without violating the synthesis of the de-
loaded secondary single phase-to-ground fault and unloaded composed signal. The number of levels can provide informa-
magnetizing inrush currents. The coefficients are calculated up tion about the signal frequency bands [3], [7], [13]. This condi-
to the second level of resolution for each candidate wavelet. tion for the minimum number of scaling has an effect on the
The results of evaluating the MDL index for the normal current computations needed to decompose the signal, and can pro-
data are provided in Table I. The evaluations of the MDL index vide the ability to realize the MRA. The determination of the
for the unloaded magnetizing inrush current data are provided optimal number of levels of resolution can be done using the
in Table II. The evaluations of the MDL index for the primary MDL criteria, once the mother wavelet is selected. Using the
side single phase-to-ground fault current data are provided in MDL criteria, the optimal level of resolution is reached when
Table III. [3], [7], [11]. Applying this criteria
Tables I–III show the MDL index evaluated for the three types on the same current data using the selected mother wavelet up
of real data representing three different currents. A close eval- to the 4th level of resolution, gives the MDL indices as provided
uation of the MDL indices shows that the Daubechies wavelet in Table IV.
1278 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005

Fig. 5. WPT first and second level details coefficients of the primary unloaded three-phase to ground fault current.

From Table IV, the first level has the highest values due to These figures provide certain features for each current, hence
the jumps of the data at the switching instant. It is clear that making them distinguishable. These features can be thought of
the second level has the lowest MDL index values. It is to be as signatures, which are able to provide accurate and reliable
noted that, at higher levels the values of MDL index tend to de- diagnosis of the differential currents. The desired signatures are
crease. According to Table IV, the second level of resolution is the values and the time locations of the coefficients of the second
selected as the optimal level of resolution for the wavelet packet level details .
transform-based differential relay algorithm. Figs. 2–5 show the
existence of the WPT details coefficients in both the first level V. WAVELET PACKET TRANSFORM ALGORITHM
and the second level and . Also these figures The existence and the time location of the WPT coefficients
show the time location of the WPT coefficients for different can provide the needed signatures of the analyzed current, and
current data using the optimal mother wavelet. These figures
such signatures can be used to identify the current type. As
clearly show the existence of the coefficients in all cases the number of levels of resolution increases, the number of fre-
of internal fault currents data, and their absence in both cases of quency sub-bands increases, offering better, accurate and more
magnetizing inrush and normal current data. Fig. 2 shows the detailed representation of the extracted frequency components.
WPT coefficients , and , and their time location for
The first level of resolution is able to provide means of de-
the normal current. It is to be noted that the bright bands indi- tecting the existence of any disturbance (high frequency com-
cate high values of the evaluated coefficients. ponents superimposed on the signal fundamental frequency).
Fig. 3 shows the WPT coefficients , and , and their Higher levels can provide information about the nature of such
time location for the unloaded magnetizing inrush current. high frequency components according to their frequency sub-
Fig. 4 shows the WPT coefficients , and , and their bands coefficient values and time location.
time location for the secondary loaded single line-to-ground The WPT algorithm for power transformer protection can be
fault current. realized by evaluating the coefficients of the WPT details and
Fig. 5 shows the WPT coefficients , and , and their comparing the their values in the second level highest frequency
time location for the primary unloaded three-phase-to-ground sub-band to zero. The evaluation of the WPT coefficients can be
fault current. done by filtering the signal with filters created by the optimal
SALEH AND RAHMAN: MODELING AND PROTECTION OF A THREE-PHASE POWER TRANSFORMER USING WAVELET PACKET TRANSFORM 1279

Fig. 7. Three-phase currents for the case of normal load current after the
transformer has been energized and both DFT and WPT response signals.

and is the harmonic order. The magnitude


of the th harmonic can be determined as [2]

(7)
The magnitudes are computed at each sampling step when
the vector is updated. If the ratio is higher than
17.7%, the relay will restrain from tripping. If the ratio
is higher than 6.5%, the realy will restrain form tripping.
The value of the ratio may fall below 17.7% under
Fig. 6. Flowchart of the proposed WPT algorithm. over-excitation current [2].
Case 1) Normal Current
mother wavelet. This procedure is realized in the flow chart The differential current for this case is collected when a
shown in Fig. 6. Y-connected balanced resistive load of 100 /phase is con-
nected to the transformer at a line voltage of 230 V, after it
has been energized. Fig. 7 shows the three phase differential
VI. OFFLINE TESTING OF THE WPT ALGORITHM currents Ia, Ib, and Ic and the trip signals representing both
the DFT and the WPT responses. It is to be noted that both
The data collected from the laboratory power transformer can trip signals remain high indicating a nonfault condition.
be used to test the proposed WPT algorithm. MATLAB is used Case 2) Unloaded Magnetizing Inrush Current
for implementing the WPT algorithm using the Daubechies db4 For the case of magnetizing inrush current, the transformer
mother wavelet for both the first and second levels of resolution. is unloaded at a supply line voltage of 230 V. Fig. 8 shows
The harmonic restraint approach for differential protection is the three phase differential currents Ia, Ib, and Ic and the trip
tested with the same collected data for comparison purposes. A signals representing the DFT and the WPT responses. It is
DFT with a window length of 16 samples is used to extract the to be noted that both trip signals remain high indicating a
1st, the 2nd, and the 5th harmonics from the collected samples of nonfault condition.
the differential current [2]. The sine and cosine components of Case 3) Single Phase-to-Ground Fault Current
the Fourier coefficients are computed using the following equa- The single line-to-ground fault (phase C) occurred on the
tions [2]: secondary side with a balanced Y connected load of
/phase connected to the secondary side. It
should be noted that the primary line-to-line voltage is set
(5) to 50 V in all the investigated faults to avoid saturation or
damage in testing equipments. Fig. 9 shows the three phase
differential currents Ia, Ib, and Ic and the trip signals repre-
(6)
senting both the DFT and the WPT responses respectively. It
clearly shows that the WPT trip signal changed its status from
where N is the window length (16 samples), is the vector high to low indicating that an internal fault has been detected
of collected differential current samples, n is the sample number in less than quarter a cycle, while the DFT trip signal changed
1280 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005

Fig. 8. Three-phase currents for the case of unloaded magnetizing inrush Fig. 10. Three-phase currents for the case of primary unloaded phase-to-phase
current and both DFT and WPT response signals. fault current and both DFT and WPT response signals.

Fig. 9. Three-phase currents for the case of secondary loaded single phase to Fig. 11. Three-phase currents for the case of loaded primary three-phase to
ground fault current and both DFT and WPT response signals. ground fault current and the WPT response signal.

its status in more than half a cycle. The trip signals are shown phase differential currents Ia, Ib, and Ic and the trip signals
in Fig. 9(a) and (b). representing both the DFT and the WPT responses. It clearly
Case 4) Line-to-Line Fault Current shows that the WPT trip signal changed its status from high
In the case of line-to-line fault between phases B and C, the to low indicating that an internal fault has been detected in
transformer is not loaded, and the fault takes place on the pri- less than quarter a cycle, while the DFT trip signal changed
mary side of the transformer. Fig. 10 shows the three phase its status in more than half a cycle. The trip signals are shown
differential currents Ia, Ib and Ic and two trip signals repre- in Fig. 11(a) and (b).
senting the DFT and the WPT responses. It clearly shows that These off-line test results show that WPT is an accurate al-
the WPT trip signal changed its status from high to low in- gorithm for building a reliable differential protection scheme
dicating that an internal fault has been detected in less than for three-phase power transformers. In addition, in all cases of
quarter a cycle, while the DFT trip signal changed its status tested internal faults, the fault is identified in less than quarter
in more than three quarters of a cycle. The trip signals are a cycle based on 60 Hz supply. On the other hand, the DFT can
shown in Fig. 10(a) and (b). detect internal faults in more than half a cycle based on 60 Hz
Case 5) Three-Phase to Ground Fault Current system. The successful classification of both the normal and the
The three phase fault has occurred on the primary side of the magnetizing inrush currents as nonfault disturbances shows the
power transformer without a load. Fig. 11 shows the three high reliability of the proposed algorithm.
SALEH AND RAHMAN: MODELING AND PROTECTION OF A THREE-PHASE POWER TRANSFORMER USING WAVELET PACKET TRANSFORM 1281

VII. CONCLUSION
A new method for diagnosing different types of currents for
power transformers protection is developed. The proposed algo-
rithm is a two level of resolution wavelet packet transform. The
algorithm is based on evaluating the WPT coefficients of the
second level highest frequency sub-band . The values of
these coefficients and their time locations represent the needed
signatures to identify the type of the investigated currents. The
selection of the optimal mother wavelet and the optimal number
of levels of resolution is carried out using the MDL data cri-
teria. Several differential currents data are used for testing the
proposed algorithm as well as testing a DFT harmonic restraint
algorithm for comparison purposes. The simulation results show
fast, accurate and reliable capabilities of the WPT algorithm to Fig. 12. WPT Decomposition of the signal f [n].
identify different types of currents flowing in a power trans-
former under various operating conditions. filters, respectively determined by the mother wavelet. It was
found that the Daubechies (db4) is an optimal mother wavelet
APPENDIX I for current diagnosis applications. The coefficients of both
WAVELET PACKET TRANSFORM (WPT) and are given as [3], [7], [11].
The wavelet Packet Transform (WPT) is a generalized ver-
sion of the discrete wavelet transform in a way that each level
of resolution (also known as octave) consists of boxes,
generated by a tree of low pass and high pass filtering opera-
tions. Therefore, the frequency bandwidth of a box decreases (10)
with growing octave number. In other words, with increasing
octave number, the frequency resolution becomes higher while
the time resolution is reduced. Starting with the signal
with length , the first level decomposition will produce
and as any other wavelet transform. The second level de- The coefficients of and are not independent from each
composition will produce four sub-bands due to the decom- other, rather they are related through the following relation [3],
position of both and using the same set of filters [10]–[12].
used in the first level decomposition. These four sub-bands are
(11)
, and [3], [7], [11]. Thus
the ordinary wavelet decomposition is a sub-tree of the wavelet Any filter bank that satisfies the relation in (8) is known as a
packet decomposition tree. The main advantages of the WPT quadrature mirror filter (QMF) bank [5], [7], [10], [12].
over other types of wavelet transforms are accurate and detailed The second level four sub-bands constructed by the WPT are
representations of the decomposed signals. Also, wavelet packet obtained through filtering operations, which can be expressed
basis functions are localized in time offering better signal ap- mathematically as a convolution operation. The first level two
proximation and decomposition. These basis functions are gen- sub-bands can be expressed as [10], [11].
erated from one base function (the mother wavelet or the
scaling function if exists) at scale , oscillation and loca-
tion given as [3], [7]. (12)

(8)
(13)
where is the wavelet transform coefficient matrix. In
wavelet packet analysis, a signal is represented as a sum The first level details represent high frequencies present
of orthogonal wavelet packet basis functions at dif- in the analyzed signal . In general, the details represent
ferent scales , oscillations and locations given as [3], [7]. high frequency components extracted from the analyzed signal
at each level of resolution [7], [13]. The second level four sub-
(9) bands can be expressed mathematically as [7], [13].

The WPT has many applications including pattern recognitions, (14)


image processing, denoising, data compression and others [7],
[10]–[13]. The WPT has a decomposition tree shown in Fig. 12
that employs the DWT as part of its decomposition process [5], (15)
[7]. The vectors and stand for the low pass and high pass
1282 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 2, APRIL 2005

[4] T. S. Sidhu, M. S. Sachdev, H. C. Wood, and M. Nagpal, “Design, im-


plementation, and testing of a micro-processor based high speed relay
for detecting transformer winding faults,” IEEE Trans. Power Del., vol.
7, no. 1, pp. 108–117, Jan. 1992.
[5] S. K. Pandy and L. Satish, “Multiresolution signal decomposition: A
new tool for fault detection in power transformers during impulse tests,”
IEEE Trans. Power Del., vol. 13, no. 4, pp. 1194–1200, Nov. 1998.
[6] C. K. Wang, C. W. Lam, K. C. Lei, and Y. D. Han, “A novel wavelet ap-
proach to current differential pilot protection,” IEEE Trans. Power Del.,
vol. 18, no. 1, pp. 20–25, Jan. 2003.
[7] E. Y. Hamid and Z. I. Kawasaki, “Wavelet-based data compression
for power disturbances using minimum description length data,” IEEE
Trans. Power Del., vol. 17, no. 2, pp. 460–466, Apr. 2002.
[8] M. G. Morante and D. W. Nicoletti, “A wavelet-based differential
transformer protection,” IEEE Trans. Power Del., vol. 14, no. 4, pp.
1351–1358, Nov. 1999.
[9] O. S. Youssef, “A wavelet-based technique for discrimination between
faults and inrush currents in transformers,” IEEE Trans. Power Del., vol.
18, no. 1, pp. 170–176, Jan. 2003.
Fig. 13. Triggering Circuit Connection R1 = 100 k
; C1 = 1 F, R2 = 10 [10] C. K. Chui, Wavelets: A Mathematical Tool for Signal Pro-
k
; C2 = 0:01 F, and T = 1:1R1C1 = 0:1 s. cessing. Philadelphia, PA: SIAM, 1997.
[11] Matlab: Wavelet Tool Box, 1995. Mathworks, Ver. 6.1.
[12] G. Stang and T. Nguyen, Wavelets and Wavelet Filter
Banks. Wellesley, MA: Wellesley-Cambridge, 1996.
[13] Y. Zhuang and J. S. Baras, “Optimal Wavelet Basis Selection for Signal
(16) Representation,” Center for Satellite and Hybrid Communication Net-
works, CSHCN T.R. 94-7, 1994.

(17)
S. A. Saleh (S’03) was born in Kafer Sur, Palestine,
in 1973. He received the B.Sc. degree in electrical en-
It should be noted that the second level details represent gineering from Bir Ziet University, Ramallah, Pales-
tine, in 1996. He received the M.Eng. degree in 2003
the highest frequency at the second level of resolution. and is a graduate student in the Ph.D. degree program
at Memorial University of Newfoundland, St. John’s,
APPENDIX II NL, Canada.
He worked as an Electrical Engineer for two years
CONTROL CIRCUIT and as a Technical College Instructor for three years.
His research interests include wavelets, wavelet
The TRIAC switches used for connecting the supply to the transforms applications, signal processing, power
power transformer are controlled by a triggering circuit con- system protection and control, transient stability, and applications of DSP in
structed mainly from pulse-generator ICs (555 timers), which power systems and power electronics.
is activated by a single push-button switch. The circuit diagram
is shown in Fig. 13. The contact is made for a short time (less
than 5 cycles based on 60 Hz systems) such that the data files M. A. Rahman (S’66–M’68–SM’73–F’88) was
born in Santahar, Bangladesh, in 1941. He received
would have a reasonable size. the B.Sc. degree in electrical engineering from
Bangladesh University of Engineering and Tech-
REFERENCES nology (BUET), Dhaka, the M.A.Sc. degree from
the University of Toronto, Toronto, ON, Canada, and
[1] M. A. Zaman, M. A. Hoque, and M. A. Rahman, “On-line implemen- the Ph.D. degree from Carleton University, Ottawa,
tation of the artificial neural network based protection for power trans- ON, in 1962, 1965, and 1968, respectively.
formers,” in Proc. NECEC, St. John’s, NL, Canada, 1996, pp. 5–11. In 1962, he joined the Department of Electrical En-
[2] M. R. Zaman, “Artificial Neural Network Based Protection of Power gineering, BUET, as a Lecturer. He became an Assis-
Transformer,” Ph.D. dissertation, Memorial Univ. Newfoundland, St. tant Professor in 1969, Associate Professor in 1972,
John’s, NL, Canada, 1996. and Professor in 1975. In 1976, he joined the Memorial University of Newfound-
[3] S. A. Saleh, “A Wavelet Packet Transform-Based Differential Protection land, St. John’s, NL, Canada, where he is a Professor and University Research
of Three-Phase Power Transformers,” Master’s Thesis, Memorial Univ.f Professor. His current research interests are in machines, intelligent controls,
Newfoundland, St. John’s, NF, Canada, 2003. power systems, digital protection, and power electronics.

You might also like