Get documents about "DWT-BASED DETECTION OF HIGH IMPEDANCE FAULT DUE TO LEA
Document Sample
CIRED 19th International Conference on Electricity Distribution Vienna, 21-24 May 2007
Paper 0174
DWT-BASED DETECTION OF HIGH IMPEDANCE FAULT DUE TO LEANING TREES IN
COMPENSATED MV NETWORKS
Nagy I. ELKALASHY Matti LEHTONEN
Helsinki University of Technology– Finland Helsinki University of technology- Finland
nagy.elkalashy@tkk.fi matti.lehtonen@tkk.fi
Hatem A. DARWISH Abdel-Maksoud I. TAALAB Mohamed A. IZZULARAB
Minoufiya University – Egypt Minoufiya University– Egypt Minoufiya University- Egypt
h_a_darwish@yahoo.com taalab3@yahoo.com mizzularab@yahoo.com
ABSTRACT PROPOSED TECHNIQUE PRINCIPLES
Features of faults due to leaning trees are extracted using The proposed technique mainly depends on DWT and
discrete wavelet transform (DWT) and an absolute sum of wireless sensor concept. As shown in Figure 1, phase
the detail coefficient d3 over a period of power frequency voltages and branch phase currents are measured at each
cycle is used as a detector. DWT is processed on the measuring node. The corresponding residual current and
residual voltage at different measuring nodes allocated in a voltage are computed and they are extracted using DWT.
wide area of the network and such correlation of DWT The absolute sum of the residual voltage detail d3
performance at different nodes can be carried out using coefficient over a power cycle is computed for the fault
distributed wireless sensors. Therefore, the fault detection detection purpose. A timer is used for determining the fault
is confirmed by numerous detectors. Other fault features period and it can be implemented using a samples counter.
that can enhance the detection security are that the initial In order to track the fault, the detail d3 of the residual
transients are frequently repeated and therefore localized voltage and current at each measuring node is multiplied to
with each current zero crossing. The fault detection compute the residual power of the frequency range 12.5-
selectivity is carried out considering the multiplications of 6.25 kHz where the sampling frequency is 100 kHz. Using
DWT detail coefficients of the residual current and voltage the sum over a period of two power cycles, the power
at each measuring nodes. A sum over two cycles is then direction in the form of its polarity is utilized for
computed to estimate the direction of the transient power determining which branch leads to the fault point. The fault
and therefore to discriminate between the healthy and tracking process is considered since the fault features
faulty sections. Test cases prove with evidence the efficacy appeared on the residual voltage details. The protection
of proposed technique. technique behaviour over a wide area of the network is
collected using wireless sensors.
INTRODUCTION The wireless sensor concept is a modern insight used for
various tasks with the objective of saving time and expense.
The electrical fault caused by a leaning tree is considered a
Wireless sensors are distributed throughout the electrical
high impedance arcing fault due to the high resistance of
tree (several hundred ohms) and associated arcs [1]. network. The electrical quantities are then regularity
Detection if high impedance faults are still major challenges transmitted from the different measuring nodes and
for protection engineers [2]. investigated for several purposes such as load monitoring,
There are several earthing concepts such as solidly, fault detection and location. The availability of sensing
compensated and unearthed networks. The compensated devices, embedded processors, communication kits and
networks are increasingly applied in Nordic Countries. Due power equipment enables the design of wireless sensors as
to small earth fault currents in compensated networks, the depicted in the four major blocks in Figure 2 [6]. This paper
transients’ phenomena are considered for detection such will not explore for these issues in more depth. The point is
faults [3]. The best signal processing techniques for
extracting these features is the wavelet transform. that the wireless concept can be considered to gather data
In [4-5], the fault due to leaning trees have been detected from different measuring nodes in the network.
using DWT. It was found that associated arc reignitions
with this fault type contributed to repeated initial transients SIMULATED SYSTEM
in the network. In this paper, the study of this fault detection
is extended to be discussed when it occurred in The system model can be divided into two main parts: the
compensated networks. The fault due to a leaning tree MV network model and the representation of the high
occurring in 20 kV network (80% under compensation) is impedance arcing fault. Figure 3 illustrates the single line
simulated by ATP/EMTP and the arc model is implemented diagram of an unearthed 20 kV, 5 feeders distribution
using the universal arc representation. The system model is network simulated using ATP/EMTP, in which the
processed using ATPDraw. processing is created by ATPDraw [7]. The feeder lines are
CIRED2007 Session 3 Paper No 0174 Page 1 / 4
CIRED 19th International Conference on Electricity Distribution Vienna, 21-24 May 2007
Paper 0174
represented using the frequency dependent JMarti model Power Supply Communication Processing Unit Sensing
consistent with the feeder configuration given in Appendix. Radio,
ADC
The neutral of the main transformer is earthed through a coil
DC-DC
Battery
Laser or MCU Sensors
Infrared
to achieve earth fault compensation degree of -31%.
Memory
The faults due to leaning trees are modeled using two series
parts: a dynamic arc model and a high resistance. For the Figure 2 Architecture of the sensor node system [6].
considered case study, the resistance is equal to 140 kΩ and A Feeder 5 Sum mation length
the arc is modeled depending upon thermal equilibrium that Feeder 4 equal to 198 km
Feeder 3
is adapted as following [1]: Feeder 2
C Load J Load
dg 1 66/20kV, ∆/Υ
C ircuit 10 km 5 km
= (G − g ) (1) Breaker
B
D I
dt τ 5 km 5 km 7 km
Load
G = i Varc (2) W ireless 7 km
Fault point 4 km
sensors H Load
τ = Ae Bg (3)
E 5 km
K
where g is the time-varying arc conductance, G is the 5 km
Load
stationary arc conductance, |i| is the absolute value of the arc F Load
current, Varc is a constant arc voltage parameter, τ is the arc Figure 3 Simulated system (5 feeders).
time constant and A and B are constants. In [1], the 0.16
parameters Varc, A and B were found to be 2520V, 5.6E-7 0.12
and 395917, respectively. Considering the conductance at Current (A) 0.08
each zero crossing, the dielectric is represented by a 0.04
variable resistance until the instant of reignition. It is 0.00
-0.04
represented using a ramp function of 0.5 M /ms for a -0.08
period of 1 ms after the zero-crossing and then 4 M /ms -0.12
until the reignition instant. -0.16
0.00 0.03 0.06 0.09 0.12 [s] 0.15
The universal arc representation is used for implementing 0.150
Time (s)
the arcing equations (1), (2) and (3) [8]. Where the fault 0.125
ir(AB)
ir(BE)
current is transposed into the TACS field using type 91 0.100
Current (A)
sensors. Therefore, the arc model is solved in the TACS 0.075
ir(EF)
exploiting integrator device type 58. In the next step, the 0.050
ir of healthy sections
0.025
computed arc resistance is sent back into the network using
0.000
TACS controlled resistance type 91 and so on. Control -0.025
signals are generated to distinguish between arcing and -0.050
[ms]
60.5 61.5 62.5 63.5 64.5 65.5 66.5 67.5
dielectric periods and therefore to fulfill the reignition Time (ms)
a- Enlarged view of residual current waveforms (ir).
instant after each zero-crossing. 900
Start 600
1
Voltage (V)
300
Currents (ia, ib, ic)
Currents and voltages at 0
Voltages (ua, ub, uc) the measuring node
-300
(ua, ub, uc) (ia, ib, ic) -600
-900
Σ 0.00 0.03 0.06
Time (s)
0.09 0.12 [s] 0.15
-615
Residual voltage ur ir Residual currents
-625
DWT
Features Extraction
Voltage (V)
-635 ur(B)
Detail d3 of d3Ur d3Ir Detail d3 of
residual voltage residual currents -645
ur(A)
Absolute sum over d3UrΧ d3Ir -655
one power cycle Transient power over ur(E)
a frequency band of -665
Detector Sd3 detail d3 ur(D)
Sum over two
power cycles -675
62.0 62.4 62.8 63.2 63.6 [ms] 64.0
Pd3 Time (ms)
No If
1 Sd3 > 0 b- Enlarged view of residual voltage waveforms (ur).
Check +ve
1
Pd3 Polarity Figure 4 The residual waveforms when fault occurred in section EF.
Yes
1 Timer 1
-ve The aforementioned MV network and the fault modeling are
Fault Detection and fault The Fault is The Fault is not combined in a single arrangement, as shown in the
tracking consideration behind the node behind the node
ATPDraw circuit illustrated in the Appendix. When phase-a
Figure 1 The proposed detection technique.
to ground fault occurred in section EF, corresponding
CIRED2007 Session 3 Paper No 0174 Page 2 / 4
CIRED 19th International Conference on Electricity Distribution Vienna, 21-24 May 2007
Paper 0174
d3
residual voltage and current waveforms are shown in Figure 2
Ur(A)
4. The fault instant is at 40 ms. The initial transients due to 0
-2
arc reignitions are obvious in the residual waveforms and it 0 0.02 0.04 0.06 0.08
d3
0.1 0.12 0.14 0.16
Ur(B)
is required to extract them using a suitable signal processing 5
0
technique such as DWT. -5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
d3Ur(D)
1
DWT-BASED FAULT DETECTION 0
-1
Wavelets are families of functions generated from a single 0 0.02 0.04 0.06 0.08
d3
0.1 0.12 0.14 0.16
Ur(E)
function, called the mother wavelet, by means of scaling and 10
0
translating. The scaling operation is used to dilate and -10
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
compress the mother wavelet to obtain the respective high Time(s)
and low frequency information of the function to be Figure 5 Details d3 of the residual voltages at nodes A, B, D and E.
100
analyzed. Then the translation is used to obtain the time 90
S
d3(A)
Sd3(B)
information. In this way a family of scaled and translated 80 S
d3(D)
wavelets is created and it serves as the base for representing 70 Sd3(E)
the function to be analyzed. The DWT is in the form: 60
m
k − nbo ao
1 50
m ∑
DWTψ f (m, k ) = x(n)ψ ( m
) (4) 40
ao n ao 30
20
where ψ(.) is the mother wavelet that is discretely dilated
10
and translated by aom and nboaom, respectively. ao and bo are 0
fixed values with ao>1 and bo>0. m and n are integers. In the 0 0.02 0.04 0.06 0.08
Time(s)
0.1 0.12 0.14 0.16
case of the dyadic transform, which can be viewed as a Figure 6 The detector Sd3 of the voltage details at nodes A, B, D and E.
special kind of DWT spectral analyzer, ao=2 and bo=1. 0.8
d3Ur(A)
Several wavelet families were tested to extract the fault 0.6
d3Ir(Healthy Feeders)×100
features using the Wavelet toolbox incorporated into the 0.4
MATLAB program [9]. Daubechies wavelet 14 (db14) is 0.2
found appropriate for localizing this fault. The Details d3 0
including the frequency band 12.5-6.25 kHz of the residual -0.2
voltages are investigated at different measuring nodes as
-0.4
shown in Figure 5. It is obvious that the initial transients due d3Ir(Faulty Feeder) ×100
0.0625 0.063 0.0635 0.064 0.0645
to arc reignitions are frequently localized. The absolute sum Time(s)
value of the voltage detail d3 over a period of the power Figure 7 Enlarged view of residual details at node A.
frequency is computed in a discrete form at each measuring k
node, as in [10]:
k
Pd 3 (k ) = ∑ d3
n = k − 2 N +1
Ur (n) × d 3 Ir (n) (6)
Sd 3 (k ) = ∑ d3 Ur ( n) (5) where Pd3(k) is used for the discrimination and its polarity is
n = k − N +1 used to track the fault point. The discriminator performance
where Sd3(k) means the detector in the discrete samples. n is P at different measuring nodes is shown in Figure 8. Its
used for carrying out a sliding window covering 20 ms and polarity is positive for healthy feeders and negative for
N is a number of window samples. Sd3 performance is shown faulty feeder as shown in Figure 8-a, Figure 8-b points out
in Figure 6. The detectors are high not only at the starting the fault track is in section BE and Figure 8-c illustrates that
instant of the fault occurrence but also during the fault the fault is in section EF. So, the fault route is estimated.
period, which improves the protection security.
To estimate the faulty section, Figure 7 can help for CONCLUSION
illustrating the proposed technique, which is an enlarged
DWT-based detection of high impedance arcing fault due to
view of the details d3 of the residual voltage and currents at
leaning trees has been investigated in compensated MV
node A. It is recognizable that the details d3 of the voltage
network. The fault detector was carried out using the
and currents of the healthy feeders are in-phase. However,
absolute sum over power cycle for the residual voltage
the detail of the faulty feeder residual current is out of
detail d3 coefficients. The periodicity of the arc reignitions
phase. This shifting can be supervised by multiplying the
gives a specific performance for the DWT with this fault
details of the residual voltage (d3Ur) and current (d3Ir). It
type and the results ensure the fault detection. The fault
can be considered to be the harmonic-band power over the
tracking has been estimated using the polarities of the power
frequency range 12.5-6.25 kHz. Then its polarity is
computed by multiplying the detail d3 coefficients of the
estimated using summation over a period of two power
residual voltage and current.
frequency cycles. This power is computed as:
CIRED2007 Session 3 Paper No 0174 Page 3 / 4
CIRED 19th International Conference on Electricity Distribution Vienna, 21-24 May 2007
Paper 0174
0.01
Pd3 of other healthy Feeders [5] N. Elkalashy, M. Lehtonen, H. Darwish, A. Taalab and
0.005
M. Izzularab, 2007, “DWT-Based Detection and
0
Transient Power Direction-Based Location of High
-0.005
Impedance Faults Due to Leaning Trees in Unearthed
-0.01
MV Networks” Submitted to International Conference
-0.015
on Power Systems Transients, IPST.
-0.02
Pd3 of faulty Feeder
[6] M. Vieira, C. Coelho, D. da Silva, J. da Mata, 2003,
-0.025
“Survey on Wireless Sensor Network Devices”
-0.03
0 0.02 0.04 0.06 0.08
Time(s)
0.1 0.12 0.14 0.16 Emerging Technologies and Factory Automation,
a- The discriminator Pd3 at node A to determine the faulty feeder. ETFA'03, pp. 537–544.
0.06
[7] L. Prikler and H. Hoildalen, 1998, ATPDraw users'
Pd3 of section BC
0.04
manual, SINTEF TR A4790.
0.02
[8] H. Darwish and N. Elkalashy, 2005, “Universal Arc
Pd3 of section BD
0 Representation Using EMTP,” IEEE Trans. on Power
-0.02 Delivery, Vol. 2, no. 2, pp 774-779.
-0.04 [9] Wavelet Toolbox for MATLAB, Math Works 2005.
-0.06 [10] J. Haung, C. Shen, S. Phoong and H. Chen, 2006,
-0.08 "Robust Measure of Image Focus in the Wavelet
Pd3 of section BE
-0.1
Domain" International Symposium on Intelligent
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time(s) Signal Processing and Communication Systems,
b- The discriminator Pd3 at node B to determine the faulty section.
0.4 ISPACS2005.
0.3 Pd3 of section EK
0.2
ir(Feeder 5)
0.1 Feeder 5
0
-0.1 ir(Feeder 4)
-0.2
66/20kV Feeder 4
Transformer
U A
-0.3
Feeder 3 ir(Feeder 3)
-0.4
Pd3 of section EF
-0.5
-0.6 ir(Feeder 2)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time(s) Feeder 2 J
c- The discriminator Pd3 at node E to determine the faulty section.
C I
Figure 8 The discriminator Pd3 when the fault occurred in section EF. Feeder 1
ur(A)
B D H
APPENDIX E K
ir(AB)
Figure 9 illustrates the ATPDraw network. It contains the F
MV network, the universal arc representation and the
residual current and voltage waveforms computation. The ur(B)
ir(BE)
feeders are represented using a frequency dependent JMarti TREE1
model. Their configuration is shown in Figure 10. R(t)
ur(D) ir(EF)
REFERENCES Rtree
[1] N. Elkalashy, M. Lehtonen, H. Darwish, M. Izzularab Universal Arc =1.0
I
and A. Taalab, 2007, "Modeling and Experimental Representation
Verification of a High Impedance Arcing Fault in MV Varc
CTR
Networks" Accepted at IEEE Trans. Dielectric and RES
Electrical Insulation.
[2] Report of PSRC Working Group D15, 1996, “High τ
Impedance Fault Detection Technology”.
[3] G. Druml, A. Kugi and O. Seifert, 2003, "A New Figure 9 The ATPDraw network.
Directional Transient Relay for High Ohmic Earth 1.1m
Faults" 17th International Conference and Exhibition 1.1m
1.1m
on Electricity Distribution, CIRED. Raven
d=10.11mm
[4] N. Elkalashy, M. Lehtonen, H. Darwish, A. Taalab and r=0.536 /km
8.1m soil resistivity
M. Izzularab, 2007, “DWT-Based Extraction of 250 m
Residual Currents throughout Unearthed MV Networks
for Detecting High Impedance Faults due to leaning
Figure 10 The feeder configuration.
Trees” Accepted at European Transaction on
Electrical Power ETEP.
CIRED2007 Session 3 Paper No 0174 Page 4 / 4
Related docs
