Detection and Location of Faults in Underground

Cable using Matlab/Simulink/ANN and OrCad

Kunal Hasija Abhishek Kumar
Department of Electrical Engineering Department of Electrical Engineering
National Institute of Technology National Institute of Technology
Kurukshetra, India Kurukshetra, India
Kunalhasija.14@gmail.com Abhishek.nit0612@gmail.com

Shelly Vadhera Anurag Kishore

Department of Electrical Engineering Department of Electrical Engineering
National Institute of Technology National Institute of Technology
Kurukshetra, India Kurukshetra, India
Shelly_vadhera@rediffmail.com Akvs1991@gmail.com

Abstract—The paper deals with detection and location of the system, proper fault detecting and locating techniques are
faults in underground cable network. For precision a three step required. The inaccessibility of the underground cable makes
process is followed in this paper, with creation of transmission the location and detection of fault in the cable a challenging
system model using a Matlab/Simulink and followed by creation task. The fault detecting and locating techniques play a very
of faults in the system. In second step the fourier analyzed fault
important role in maintaining the system and thereby
voltages and currents obtained from the SIMULINK model are
fed to the training set of artificial neural network (ANN) in order increasing the reliability.
to detect the type of fault. In the last step, an independent
software OrCad is used to locate the fault distance from the Various fault locating methods like the sectionalizing
either ends using the principle of time domain reflectometry in a methods, acoustic detection method, Murray loop methods are
simulated practical underground distribution system. not used much because they suffer from many disadvantages.
The sectionalizing method can’t be employed because section
wise checking of underground cable is not possible. The
Index Terms— Underground cables, ANN, OrCad, Simulink, acoustic method may become disastrous at the time of rain and
Time domain reflectometry, Faults. it is a bit cumbersome method too. The Murray loop method is
Nomenclature—
based on the principle of Wheatstone and due to different
V1a = Sending voltage of phase a resistances of leads etc. this method is highly inaccurate. The
V2a = Receiving end voltage of phase a transmission line fault location done by [3] uses non-iterative
I1a = Sending end current of phase a method and doesn’t find the type of fault that has occurred in
I2a = Receiving end current of phase a the network. But this paper deals with both location of fault as
V1(1) = Positive sequence sending end voltage well as detection of type of fault using three different
V1(2) = Negative sequence sending end voltage softwares i.e., Simulink, ANN and OrCad.
V1(0) = Zero sequence sending end voltage
I1(1) = Positive sequence sending end current
I1(2) = Negative sequence sending end current
I1(0) = Zero sequence sending end current II. SIMULINK MODEL
Va = Fault voltage of phase a
Ia = Fault current of phase a A. Underground cable network
Simulink model adopted from [3] has been modified.
I. INTRODUCTION Basically distributed parameter line has been modeled as an
Use of underground power cable is expanding due to safety underground cable. A 1.1 kV, 25 sq. mm, unarmored copper
considerations and enhanced reliability in the distribution and cable parameters have been utilized [4]. The underground
transmission systems in recent times. Due to safety reasons network is of 10 km as shown in Fig.1. Fourier blocks and
and high power requirements in densely populated areas, use scope at sending, receiving and fault points are used to obtain
of underground cable has seen a sharp hike in recent times [1, the fundamental components of the voltages and currents at
2]. The increasing size of the power system has led to an these points respectively. All the faults namely LG, LL, LLG,
increase in the complexity of the system and our ever LLL, LLLG have been studied and comparison have been
increasing dependency on it has made the need to increase the made by shifting the fault points. Parameters used are R=0.727
reliability of the power system. To increase the reliability of Ÿ/km, L= 0.334 mH/km, C= 1.05 μF/km and length of the
line = 10 km.

978-1-4799-6042-2/14/$31.00 ©2014 IEEE

 B. Fault at mid-point C. Load end fault
From the model shown in Fig. 1 we can observe the nature Now the fault is moved from 5 km (i.e., from mid-point) to 8
of different types of fault. The waveforms of the fault voltage km (i.e., load end) from source end in order to observe the
and fault current for a line to ground fault (phase a to ground) variation in voltages and currents for different types of fault.
has been shown in Fig. 2 whereas the waveforms of the fault The comparison of voltages and currents of the faulted phase
voltage and fault current for a LLG fault (phase a and b to are shown in Fig. 4 and 5 respectively for either location.
ground) has been shown in Fig. 3.
Similarly we can not only obtain fault voltages and currents
but also sending and receiving end voltages and currents for
different types of faults.

Fig. 4 Comparison of fault voltage at mid-point and load

end during LG fault

Fig. 1 Simulink model for obtaining voltage and current

waveforms in an underground network

Fig. 5 Comparison of fault current at mid-point and load end

during LG fault

D. Simulation results
The run time of simulation is 0.08 second whereas the
transition times of circuit breakers have been taken as 1/50
Fig. 2 Fault voltage and current waveforms for LG fault seconds for opening and 2.5/50 seconds for closing. Using the
fourier block the fundamental values of voltages and currents
for each and every case can also be obtained and can further
be utilized for next step i.e. detection of the type of fault.
Table 1 shows the values of voltage and current phasors for
LG fault. Whereas, sequence voltage and currents obtained are
shown in Table 2.

III. DETECTION OF TYPE OF FAULT USING ANN

Artificial neural networks (ANN) are computational models
generally presented as systems of interconnected neurons
which can compute values from inputs. As per the simulation
Fig. 3 Fault voltage and current waveforms for LLG fault
results shown in Table 1 and 2, an array of fault voltages and
currents at different times can be found for various faults.
Time can be decided as per the transition times of circuit
breakers which is between 1/50 seconds and 2.5/50 seconds in
this case. A. Training set of ANN
In ANN firstly, a training set has to be provided in order to
Table 1 Voltage and current phasors during LG fault train the neural network so that it can develop a relationship
Quantities Values (p.u.) Values (p.u.) Values (p.u.) among the given data and show results for any input as per its
LG fault at LG fault at LG fault at unique ability of decision making [5]. Table 3 shows a set of
phase a phase b phase c values of fault voltages and currents obtained from the model
V1a 0.1381 – 0.7530 – 0.4883 +
j0.2237 j0.3556 0.0752
at 0.047 seconds.
V1b -0.1705 – -0.2434 – -0.6769 – Table 3 The data for training obtained from Simulink model
j0.4420 j0.0076 j0.4917
V1c -0.0758 + -0.3179 + 0.1053 + Case Va Vb Vc Ia Ib Ic Fault
j0.8480 j0.3670 j0.2315 type
I1a 3.3441 – 0.5636 – 0.4683 –
j7.4619 j1.5889 j1.6310 1 0.7224 0.7218 0.7234 1.801 1.8 1.8 No
I1b -1.6482 + -7.4934 + -1.6597 + fault
j0.4079 j0.8272 j0.3076
I1c 1.0960 + 1.1797 + 4.1485 + 2 0.000203 0.4526 0.8218 4.361 1.801 1.8 LG
j1.2824 j1.2225 j6.6288 (ag)
V2a 0.0930 – 0.7080 – 0.4433 +
j0.1241 j0.2569 j0.1749 3 0.806 0.0001 0.4678 1.8 3.974 1.8 LG
V2b -0.0615 – -0.1347 – -0.5680 – (bg)
j0.4529 j0.0184 j0.5025
V2c -0.1393 + -0.3816 – 0.0146 + 4 0.4608 0.8159 0.0002 1.8 1.8 4.408 LG
j0.7592 j0.2783 j0.1427 (cg)
I2a -2.3399 + 0.4422 – 0.5373 –
j4.0044 j1.8663 j1.8245 5 0.0002 0.0001 0.5979 5.309 2.403 1.8 LLG
I2b -1.8476 + 3.9969 + -1.8371 + (abg)
j0.4470 j0.0242 j0.5476
I2c 1.3942 + 1.3097 + -1.6603 – 6 0.0001 0.5884 0.0002 2.657 1.8 5.385 LLG
j1.3171 j1.3763 j4.0282 (acg)
. 7 0.5933 0.0002 0.0001 1.8 4.961 2.847 LLG
(bcg)
Table 2 Positive, negative and zero sequence voltages and
currents during LG fault 8 0.3623 0.3622 0.7245 2.297 2.532 1.8 LL
Quantities Values (p.u.) Values (p.u.) Values (p.u.) (ab)
LG fault at LG fault at LG fault at
phase a phase b phase c 9 0.7223 0.3612 0.3611 1.8 2.385 2.3 LL
(bc)
V1(1) 0.4595 – 0.4527 – 0.4667 – 10 0.3614 0.723 0.3615 2.744 1.8 2.71 LL
j0.1695 j0.1572 j0.1574
(ac)
V1(2) -0.2854 – 0.2364 – 0.0493 +
j0.1147 j0.2002 j0.2942 11 7.7e-5 6.9e-5 7.5e-5 2.876 2.392 2.788 LLL
V1(0) -0.0360 + 0.0637 + -0.0277 – (abc)
j0.0606 j0.0008 j0.0615
12 7.7e-5 6.9e-5 7.5e-5 2.876 2.39 2.788 LLLG
I1(1) 1.4592 – 1.3530 – 1.5659 –
j3.5616 j3.3759 j3.3765 (abcg)
I1(2) 0.9546 – 1.1274 + -2.0838 –
j1.9765 j1.6312 j0.0218
Similarly, we can formulate a large set of values at different
I1(0) 0.9307 – -1.9168 + 0.9860 +
j1.9236 j0.1542 j1.7687
times such as 0.03 seconds, 0.045 seconds etc. for an effective
V2(1) 0.4144 – 0.4077 – 0.4217 –
training of the ANN. The ANN has been trained with the help
j0.0699 j0.0576 j0.0578 of four variables i.e., phase a, b, c and g(ground) either being
V2(2) -0.2854 – 0.2364 – 0.0493 + in high state or low state, for example, for a fault between
j0.1147 j0.2002 j0.2942 phase a, b and ground the ANN is trained such that a, b &
V2(0) -0.0360 + 0.0637 + -0.0277 – ground are at high (1) state.
j0.0606 j0.0008 j0.0615
I2(1) -0.4528 + -0.3466 – -0.5600 –
j0.1053 j0.0800 j0.0797 B. Output of ANN
I2(2) -0.9553 + -1.1288 – 2.0838 + The output block of ANN is obtained by running the
j1.9761 j1.6303 j0.0233
training set. The values of fault voltages and currents obtained
I2(0) 0.9307 – 1.9167 – 0.9860 +
j1.9236 j0.1552 j1.7687 from Simulink model are entered as input and run the system.
The phases a, b, c and ground are resembled using four
colours i.e., yellow, pink, green and red respectively. The Amplitude of voltage pulse= -1V
logic being used is that high (or 1) is under fault, therefore, as Rise time = 0s
shown in Fig. 6 both yellow and red are near logic 1 whereas Fall time = 0s
rest are at 0, thus describing LG fault between phase a and Delay time = 0s
ground. This logic can be further extended for any type of Pulse width = 10μs
fault occurring at any time. Velocity of wave is given by equation (1)
1
v=
LC (1)
From Fig. 8 time t = 187 μs, and the distance of the fault is
given as the half of the distance travelled by the wave,
therefore,
d = vt/2 (2)
= 53398.8 x 187 x 10-6 / 2
= 4.9928 km

Fig. 6 Output of fault detection using ANN for LG fault

between phase a and ground

IV. FAULT LOCATION USING ORCAD

After detecting the type of fault using ANN, the distance of
fault point from the source end is calculated using the
principle of time domain reflectometry. In this paper, the
specimen used is simulated for different type of faults. Series Fig. 7 Circuit model of LG fault at midpoint of the line
inductance and shunt capacitance are taken as distributed
parameter and series resistance is considered as lumped. Also,
shunt conductance is neglected which is a considerable
approximation for underground lines [6]. Length and various
parameters of the cable are inserted as delay time and
characteristic impedance in the schematic tab of software. At
the time of fault, the fault impedance is considered zero and
the load is supposed to be removed as the circuit breaker or
isolator trips after fault inception. The open circuit at load end
is represented as a very large valued resistance of the order of
1019 ohms. A zero impedance fault is created somewhere on
the line (say at mid-point). A low voltage negative pulse
having pulse time much less than the period and having a zero
rise and decay time is sent along the cable from the source end
at t=0. The graph between voltage or current at source end
with respect to time is plotted and the time of return of the Fig. 8 Voltage waveform for LG fault at midpoint of line
pulse is noted and it is multiplied with the pulse velocity to get
the distance of fault point from the source end.
The OrCad model for the simulation of the LG fault is V. CONCLUSION
shown the Fig. 7 and the voltage-time graph at the source end This paper enables the researchers to detect and locate the
of the cable is drawn in Fig. 8. The values of the parameters faults in underground cable with the help of three robust
for the cable are given as software tools i.e., Simulink, ANN & OrCad. The ANN model
Length of cable = 10 km used is interconnected with the Simulink, which represents a
R= 0.727 Ÿ/km real time underground cable network, which in turn affects the
L= 334 μH/km working of OrCad model as per the type of fault detected in
C= 1.05 μF/km the preceding step. Hence, the method used in this paper
G=0 S/km operates in a sequential manner and proves to be useful in
Current limiter impedance=1kŸ/km detection and location of faults in underground cables.
Parameters of Voltage-pulse are under:
 REFERENCES
[1] C. K. Jung, K. H. Kim, J. B. Lee, B Klocklb, “Wavelet and neuro-fuzzy
based fault location for combined transmission systems”, Int J Electr
Power Energy Syst ; Vol 29: pp. 445–454, 2007.
[2] W Zhao, Y. H. Song, Y. Min. Wavelet analysis based scheme for fault
detection and classification in underground power cable system. Electr
Power Syst Res; Vol 53:pp 23–30,2000.
[3] Sumit and Shelly Vadhera, “Iterative and non-iterative methods for
transmission line fault-location without using line parameters,”
International Journal of Engineering and Innovative Technology
(IJEIT), vol 3, no. 1, July 2013.
[4] http://www.havells.com/Admin/Forms/Brochures%5CCable%20Catalog
ue%20Final-Aug -08.pdf
[5] J. S. R. Jang, C. T. Sun, and E. Mizutani, Neuro Fuzzy and Soft
Computing, Prentice Hall Inc, 1997.
[6] http://ipstconf.org/papers/Proc_IPST2001/01IPST009.pdf