See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261714882

Modeling and detection of high impedance faults

Article  in  International Journal of Electrical Power & Energy Systems · October 2014

DOI: 10.1016/j.ijepes.2014.03.046

CITATIONS READS

22 735

4 authors, including:

Vicente Torres J.L. Guardado

Instituto Tecnológico de Morelia (ITM) Instituto Tecnológico de Morelia (ITM)
9 PUBLICATIONS   57 CITATIONS    84 PUBLICATIONS   1,052 CITATIONS   

SEE PROFILE SEE PROFILE

Serguei Maximov
Instituto Tecnológico de Morelia (ITM)
59 PUBLICATIONS   441 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

New magnetic shunt technology for power transformers View project

All content following this page was uploaded by Serguei Maximov on 12 October 2017.

The user has requested enhancement of the downloaded file.

 Electrical Power and Energy Systems 61 (2014) 163–172

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Modeling and detection of high impedance faults

V. Torres ⇑, J.L. Guardado, H.F. Ruiz, S. Maximov
Instituto Tecnologico de Morelia, Morelia 58130, Mexico

a r t i c l e i n f o a b s t r a c t

Article history: High impedance faults are difficult to detect by conventional overcurrent relays mainly because of their
Received 21 February 2013 low current magnitudes. This paper describes a model for representing high impedance faults in electrical
Received in revised form 23 January 2014 distribution systems. The model is based in a non-linear resistance representing the high impedance path
Accepted 21 March 2014
during this kind of faults. Based on this model, the performance of several electric variables associated to
high impedance faults is analyzed and an algorithm for high impedance fault detection in electrical dis-
tribution systems is presented. Field measurement and computer simulations validate the high imped-
Keywords:
ance fault model and the proposed fault detection algorithm.
Arc modeling
Electrical distribution system
Ó 2014 Elsevier Ltd. All rights reserved.
High impedance faults
Harmonic analysis

Introduction
High impedance fault (HIF) detection represents a real chal-
lenge to protection engineers in electric utilities because of the
complexity and variability of the phenomena involved. HIFs have
the characteristic of producing low current magnitudes compared
to nominal load currents, which makes difficult its detection by
conventional overcurrent relays in a reliable way [1,2]. Some addi-
tional features that influence HIF detection are weather conditions,
fault conditions, fault distance, short-circuit rate, conductor type,
etc. There exist many causes leading to HIFs in Electrical Distribu-
tion Systems (EDS). For example, a HIF may occur when an ener-
gized conductor makes contact with the soil surface without a
solid grounding, or when the conductor get in touch with a branch
tree or any other object representing a high impedance path to
ground. Under these circumstances, the relatively high voltages
close to human beings and the electric arc associated to HIFs rep-
resents a serious risk to public safety and a major concern on
industrial installations.
HIFs are closely related to the grounding method used in EDS. A
description of these methods can be found in [3]. In México, North
America and Latin American countries the common practice is to
use a solidly grounding system. This approach facilitates the
setting of conventional overcurrent relays for detecting low imped-
ance faults. Typical fault levels at the distribution substations bus
are 5–6 kA, 1.5 kA at the end of the feeder and load currents in
the range 300–500 A. Since typically a HIF produce currents
⇑ Corresponding author. Tel.: +52 443 3171870.
E-mail address: vtg_1982@hotmail.com (V. Torres).
smaller than 75 A, these events are difficult to detect even for con-
ventional ground relay with settings in the range 200–300 A. A
complete description of the current profile along a typical distribu-
tion feeder is presented in [4].
The problem of HIF detection has long been recognized by the
industry and several techniques have been proposed in the litera-
ture. In 1988, Huang et al. evaluated the performance of four differ-
ent algorithms for HIF detection using a staged fault test [5]. The
evaluations were done off-line and the results demonstrated that,
under certain circumstances, some algorithms perform better than
others. Latter, in the 90s, Emmanuel et al. carried out extensive
measurements of harmonic currents at a ‘‘staged ground imped-
ance fault in sandy soil’’. The aim was to assess to what extent har-
monic currents can be used for HIF detection [6]. In 1997, Zori et al.
[7] developed an algorithm for arcing fault detection using the bus
voltages and the odd harmonics.
The above developments were followed in 2004 by Stoupis et al.
[8], who presented an algorithm based on the Discrete Wavelet
Transform (DWT), higher order statistics and an Artificial Neural
Network (ANN) in order to detect HIFs in medium voltage net-
works. Latter, Hou presented a HIF detection algorithm based on
the sum of current differences using a one cycle difference filter
[9]. In 2006, Adamiak et al. [10] applied expert systems and pattern
recognition techniques to calculate harmonic energy levels during
HIFs.
In 2007, Elkalashy et al. used the DWT to analyze arc reignitions
associated to high impedance faults caused by leaning trees [11].
The low current magnitudes observed and the periodicity of arc
reignitions makes the DWT a suitable tool to analyze this kind
of phenomena with good accuracy. In 2011, Baqui et al. also used

http://dx.doi.org/10.1016/j.ijepes.2014.03.046
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
164 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172

the DWT and ANNs for HIF detection in distribution feeders. In this
15
approach, changes in current waveforms caused by HIFs and nor-
mal switching events were also considered [12]. The obtained re- 10
sults validate the effectiveness of the proposed methodology to
discriminate HIFs from switching transients. 5

Current (A)
In 2012, Samantaray proposed an ensemble decision tree which
0
provides effective decision on HIF detection [13]. This approach
also used extended Kalman filters to analyze the harmonic content
-5
in the current signals and feed a classifier based on random forest
techniques. Good results for HIF detection in large distribution -10
systems are reported by the author. Recently, Bakar et al. [14]
proposed a method to detect and locate HIFs based on the DWT, -15
where the fault characteristics are extracted from voltage 0.2 0.4 0.6 0.8 1 1.2
measurements and then compared with a database obtained from Time (s)
computer simulations They also apply rank analysis to locate the
likely faulted section. Fig. 1. HIF current for a bare conductor.
A key aspect in the development of a HIF detection algorithm
is the need of accurate models for representing HIFs. One of the
first models for representing HIFs in EDS was proposed by 1
Emmanuel et al. in [6]. The HIF was represented by means of 0.8
two DC sources interconnected by two diodes in order to obtain 0.6
the non-linear characteristics of voltage and current (v–i). In 0.4

Current (A)
[15], this model was improved by adding two resistances and 0.2
inductances between the diodes. Latter, Nam et al. [16] proposed 0
a new model to simulate HIFs by means of two time dependent -0.2
series resistances. The combined action of both resistances allows -0.4
representing HIFs with good accuracy. Latter, Elkalashy et al. [17] -0.6
proposed a HIF model based in two components: an arc model -0.8
and a resistance. -1
The main contribution of this paper is the development of a
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
general purpose model for representing HIFs of varied characteris-
tics in EDS based in a single non-linear resistance. A second contri- Time (s)
bution is the development of an algorithm for HIF detection in EDS.
Fig. 2. HIF current for a cover conductor.
In order to achieve these objectives, the paper has been organized
as follows: in the second section, the results of several field tests
are presented. Based on these measurements, in the third section Figs. 1 and 2 show typical current measurements for bare and
a model for representing HIFs is proposed. The model is based in cover conductors during HIFs respectively. For all the tests carried
calculating a non-linear resistance representing the high imped- out in bare conductors, there was a variability in the peak current
ance path associated to the fault. This section also includes a meth- magnitudes in the range 7–15 Apeak, the mean value is 9.8366 and
odology for calculating the model parameters. In the fourth the standard deviation is 0.1440.
section, an algorithm for HIF detection in EDS is presented and For cover conductor the variability in all the tests is in the range
tested. Finally, the paper conclusions are drawn. It should be men- 0.25–1 Apeak, the mean value is 0.675 A. and the standard deviation
tion that the HIF model and the HIF detection algorithm were is 0.0721. As expected, cover conductors represents a higher
implemented in the MODELS section of the Alternative Transient impedance to fault currents because of the insulation sheat, mean-
Program (ATP) [18] for validation purposes. ing smaller currents magnitudes during HIFs. The oscillograms pre-
sented in these figures share many HIF features also reported by
other authors [6,16,17,19]. It should be mention that during the
Measurements of high impedance faults

The first step in the development of a model for representing x 10

4

HIFs in EDS is to measure currents and voltages during this kind 1.5
Current x 500
Voltage
of events. Field measurements provide reliable information about
HIF behavior, which would be useful for understanding the dy- 1
namic of the phenomena involved and also laid the basis for fur-
ther model development. 0.5
Current (A)
Voltage (V)

Twenty field tests were carried out in a 13.8 kV, 8 miles

(12.8 km) long distribution feeder owned by the national utility 0

of México (CFE). All the measurements were carried out during a

sunny day in the dry season using a sampling frequency of -0.5

20 kHz. The HIFs were simulated at the end of the feeder by means
of a fallen conductor over the soil surface covered by dry grass. -1

Bare and cover conductors representative of those used in Mexico

(3/0) were used during testing. Current and voltages were mea- -1.5
0.18 0.2 0.22 0.24 0.26 0.28 0.3
sured on site and the different data files were recorded using the Time (s)
Common Transient Data Exchange (COMTRADE) protocol, in order
to be available for further processing in ATP. Fig. 3. Voltage and current during a HIF for bare conductors.
 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172 165

different tests conventional overcurrent relay protecting the feeder x 10

4

did not operate. 1.5

Figs. 3 and 4 show voltages and currents measured during the

1
firsts few cycles after fault inception. Observe that HIF currents in-
crease steadily up to a peak value of 11 and 0.9 Apeak for bare
0.5
and cover conductors respectively. On the other hand, voltage

Voltage (V)
waveform remains practically unchanged after fault inception. This
0
voltage behavior is explained by the fact that the low current levels
during HIFs do not represent an important load to the feeder, and
-0.5
consequently no significant change in voltage waveform is ob-
served. Measured currents also show significant harmonic distor-
-1
tion in both cases. Another important feature is that current and
voltage measured are practically in phase, suggesting a strong
-1.5
resistive component typical of HIFs. Figs. 5 and 6 show the non-lin- -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
ear characteristics v–i during HIFs. Current (A)
The harmonic content in the different current waveforms was
analyzed using the Discrete Fourier Transform (DFT). Figs. 7 and Fig. 6. Characteristics v–i for cover conductor.
8 show the evolution in the time domain from the first to the fifth
harmonic for bare and cover conductors, respectively. The analysis
12
was carried out for the waveforms shown in Figs. 3 and 4. On the I1
I2 x 3
other hand, Table 1 show the harmonic content of current wave- I3 x 2
forms as a function of the first harmonic for both, bare and cover 10 I4 x 3
I5 x 3
conductors. The results presented in Table 1 are an average of
the different tests carried out. 8

Current (A)
From the above set of measurements, the general characteris-
tics observed during HIF are low current magnitudes due to the 6
high impedance path, random variability on peak values combined
with some asymmetry, a non-linear v–i characteristics and low
4

4
2
x 10
1.5 Voltage
Current x 10000
0
1 0.15 0.2 0.25 0.3 0.35 0.4
Time (s)
0.5
Current (A)

Fig. 7. Harmonic currents magnitudes for bare conductors.

Voltage (V)

0
0.7
I1
-0.5 I2 x 4
0.6 I3 x 2
I4 x 3
-1 I5 x 3
0.5
Current (A)

-1.5
0.94 0.96 0.98 1 1.02 1.04 1.06 0.4
Time (s)
0.3
Fig. 4. Voltage and current during a HIF for cover conductors.
0.2

0.1
4
x 10
1.5 0
0.9 0.95 1 1.05 1.1 1.15
1 Time (s)

Fig. 8. Harmonics currents magnitudes for cover conductors.

0.5
Voltage (V)

0 Table 1
Average harmonic content.
-0.5
Harmonic Bare conductor % fundamental Cover conductor % fundamental

-1 2nd 4.214 11.59

3rd 29.22 23.14
4th 1.527 2.756
-1.5 5th 0.930 4.814
-15 -10 -5 0 5 10 15
6th 0.098 0.239
Current (A) 7th 2.560 2.528
8th 0.580 0.715
Fig. 5. Characteristics v–i for bare conductor.
166 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172

frequency harmonic components, mainly the third harmonic. All
these characteristics also depend on the conductor type: bare or
cover conductor.
High impedance fault model
Then, by substitution of (5) into (2), the arc conductance g(t) can
be represented like the sum of the ‘‘slow’’ component ĝ(t), which
does not change significantly during the period T, and the rapidly
oscillating component dG(t):
gðtÞ ¼ g^ðtÞ þ dgðtÞ

Model description The representative equation for the slow part of the arc conduc-
tance can be obtained by applying the averaging operation (6) to
The development of an algorithm for HIF detection requires a both sides in Eq. (2). Then, due to the linearity of the operator (6):
suitable representation for the high impedance existing during this dg^ G0  g^
kind of events. The set of measurements presented in the previous ¼ ð7Þ
dt s
section suggest a strong non-linear resistive component for repre-
senting HIFs. In particular, this paper proposes a model based in a and the solution to (7) in the time domain is:
single non-linear resistance. This component would represent the
g^ðtÞ ¼ G0 ð1  et=s Þ ð8Þ
high impedance path due to the electric arc associated to HIFs
and also the return path through ground and any other contact ob- From (8), the average electric arc resistance can be calculated as
ject like a branch tree, for example. For a given fault, the electrical the inverse of ĝ(t).
parameters for this non-linear resistance can be calculated from
measurements. Since each measurement is unique, then the model Parameters calculation
parameters will vary in a certain range. The model development
and the calculation of these parameters are now addressed. For bare and cover conductors, the parameters G0 and s in (8)
The non-linear resistance proposed for representing HIFs can be can be calculated from Fig. 3 by using the least square method.
modelled like an electric arc of varied characteristics. Latter, the Let us denote ĝf(tk), if(tk) and uf(tk) as the fault conductance, current
comparison between measured and calculated results will justify and voltage respectively, taken from measurements at discrete
this assumption. In this sense, Cassie and Mayr [20,21] introduced time instants t1, t2,. . ., tN. The averaged (‘‘slow’’) arc conductance
their first descriptions of the arc conductivity in the form of a first can be calculated as follows:
order differential equation. These arc dynamic equations have been vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
,
u n
improved and modified along the time in order to increase the uX 2 X n
g^f ðtk Þ ¼ t if ðtkþi Þ u2f ðtkþi Þ
model validity range and reduce the associated computational bur-
i¼n i¼n
den. Most of these equations are of the form [22]:
  where the number of samples n is taken large enough to smooth the
d ln g 1 ui
¼ 1 ð1Þ conductance curve. ĝ(tk) is the theoretical value of the fault conduc-
dt sðu; iÞ Pðu; iÞ
tance at the discrete time tk predicted by Eq. (8). Then, the sum of
where u and i are the arc voltage and current respectively, g = i/u is squared deviations of the theoretical values from the experimental
the arc conductance, s(u,i) is the time constant and P(u,i) is the cool- data for a given s is:
ing power. Kizilcay proposed a modified arc model which can be ob-
X
N
tained from the general Eq. (1) by considering s as a constant and SðG0 ; sÞ ¼ ðg^ðt k Þ  g^f ðt k ÞÞ
2

the heat dissipation power as P = P0 + u0|i| [23,24]. As a result the k¼1

arc equation takes the form:
The minimum of the sum of squares is found by solving the fol-
dg GðiÞ  g lowing system of equations:
¼ ð2Þ
dt s @ðSðG0 ; sÞÞ @ðSðG0 ; sÞÞ
where the steady state conductance is a function of the arc current ¼ 0; ¼0
@G0 @s
[22]:
This system of equations can be solved numerically by any
2
i existing method like the descendent gradient [22]. The results
GðiÞ ¼ ð3Þ
P0 þ u0 jij for a HIF in bare conductor, Fig. 3, are G0 = 0.000932 f and
s = 0.0298 s. Applying the same procedure to current and voltage
In order to calculate G(i), in this paper is considered that if the
measurements during a HIF in cover conductor, Fig. 4, the results
electric arc current in (3) has a sinusoidal (or quasi-sinusoidal)
are G0 = 0.000065 f and s = 0.051 s.
form with frequency x = 2p/T, then, according to (3), the steady
state conductance is a periodic function with frequency 2x, and
the steady state conductance can be expanded in Fourier series HIF model validation
with the even harmonics:
Comparison with measurements
X
1
GðtÞ ¼ Gn ej2nxt ð4Þ
n¼1 In order to validate the HIF model described in the above sec-
tion, a model for simulating the interaction between the EDS and
If the period of (4) (T/2)  s, then the steady state conductance the non-linear resistance during HIFs was developed. The model
is the superposition of a constant component G0 and the rapidly was built up in the MODELS section of ATP. Fig. 9 shows a
oscillating component dG(t): schematic view of the EDS and the HIF model representing the
GðtÞ ¼ G0 þ dGðtÞ ð5Þ non-linear resistance due to the high impedance path. Therefore,
once that G0 and s are calculated from measurements for a given
where the rapidly oscillating part can be filtered by the operation: fault event, these parameters are used like inputs to the HIF model
Z T
in ATP [25]. Then, the model simulates the dynamic of the fault
 ¼1
GðtÞ GðtÞdt ¼ G0 ð6Þ resistance during the HIF by calculating the solution to Eq. (8).
T 0 For simulating a HIF with different characteristics, G0 and s can
 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172 167

where gk is the calculated conductance and g M k is the measured con-

ductance. In general, the total error in magnitude for bare conductor
is smaller 4% and for cover conductor about 9%. These results show
that the assumptions made for the development of the HIF model
are correct, and that the model is capable of calculating the arc con-
ductance during a HIF.
The comparison was extended to current and voltage wave-
forms during a fault event. Fig. 12 shows the theoretical current
Fig. 9. Distribution system and HIF model.
and voltage waveforms during a HIF calculated in ATP for a bare
conductor. In fact, Fig. 12 is quite similar to the measured results
be calculated from measurements or adjusted to fit to the new
shown in Fig. 3. In general, current and voltage waveforms shown
fault characteristics. The computer code for implementing the
in both figures have a good correlation in magnitude, waveform
HIF model in the MODELS section of ATP is given in the Appendix
and timing.
A1. On the other hand, the feeder characteristics are given in the
Current waveforms are of particular interest because of its har-
Appendix A2.
monic content. Fig. 13 shows in detail a comparison between the
For comparative purposes, Figs. 10 and 11 show the perfor-
current waveforms obtained from measurements and the results
mance of the HIF conductance after fault inception for bare and
calculated using the model proposed in this paper, Figs. 3 and 12
cover conductors. The measured results in both figures were ob-
respectively. The superposition of both waveforms shows a good
tained using voltage and current phasors measured on each time
correlation between measured and calculated results for bare
instant. On the other hand, the calculated results were obtained
conductors, justifying the assumptions made during the model
by using the HIF model shown in Fig. 9, Eq. (8), for the bare and
development. The total error in Fig. 13 was calculated by means
cover conductors respectively. In general, the results show good
of (9) and it is 4%. The variability for the measured results is
agreement between the measured and calculated results. The total
10.08–12.45 Apeak and for the calculated results the variability is
error was calculated by means of the following expression:
7.792–12.73 Apeak.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn ffi
M 2 Table 2 shows a comparison of the measured and calculated
k¼1 jg k  g k j
error ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi ð9Þ harmonic currents existing in the waveforms shown in Fig. 13.
Pn 2 Observe that the harmonic content for the fundamental and the
k¼1 jg k j
third harmonic, measured and calculated, are quite similar, they

-3
x 10
1 4
x 10
0.9 1.5
Current x 500
Voltage
0.8
1
Conductance (S)

0.7
0.6 0.5
Voltage (V)
Current (A)

0.5
0.4 0

0.3
-0.5
0.2
0.1 Measured
Calculated -1
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
-1.5
Time (s) 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Time (s)
Fig. 10. Measured and calculated conductance during a HIF for bare conductor.
Fig. 12. Calculated voltage and current using the proposed model.

-5
x 10
8
15
7 Measured
Calculated

6 10
Conductance (S)

5 5
Current (A)

4
0
3
-5
2

Measured
-10
1
Calculated

0 -15
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3
Time (s) Time (s)

Fig. 11. Measured and calculated conductance during a HIF with cover conductor. Fig. 13. Measured and calculated current during a HIF for bare conductor.
168 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172

Table 2
Harmonic content analysis for Fig. 13.

Bare conductor
Harmonics Measured Calculated
Mean (A) Standard deviation Min. (A) Max. (A) Mean (A) Standard deviation Min. (A) Max. (A)
1st 9.7405 0.1925 7.6975 10.0619 9.8244 0.165 7.5187 10.2415
2nd 0.3063 0.0741 0.1309 0.4349 0.0124 0.2942 0.00037 0.188
3rd 2.7041 0.0743 2.120 3.160 2.5211 0.1893 1.577 2.660
4th 0.0754 0.03 0.0037 0.2066 0.0049 0.0707 0.0057 0.1486
5th 0.1163 0.032 0.0107 0.3266 0.4849 0.369 0.4298 0.5735
6th 0.0324 0.0179 0.00059 0.1155 0.0200 0.0171 0.00029 0.076
7th 0.1835 0.0557 0.0716 0.3032 0.142 0.0431 0.1056 0.2077
8th 0.0512 0.0209 0.00036 0.1219 0.0142 0.0372 0.0021 0.0535

are the main sources of harmonic distortion in the current wave- x 10

4

forms, see Table 1. There exist some differences between measured 1.5

and calculated results for the remaining harmonics. Some of these

1
differences are due to small measurement errors and noise. It
should be mention that these harmonics contribute in a small pro-
0.5
portion to the overall harmonic content in the current waveform.

Voltage (V)
On the other hand, observe that the harmonic content is higher
0
for the measured results. This is explained by the fact that the mea-
sured results are slightly higher that the calculated results, see
-0.5
Fig. 13.
Fig. 14 shows the evolution of the harmonic content in the cur-
-1
rent waveform calculated as a function of time, using the proposed
model. Observe that Fig. 14 is very similar to Fig. 7, obtained from
-1.5
measurements. Finally, Fig. 15 shows the characteristic v–i during -15 -10 -5 0 5 10 15
the HIF using the proposed model, which is very similar to the one Current (A)
obtained from field test and shown in Fig. 5.
The analysis was extended to cover conductors. Fig. 16 shows a Fig. 15. Characteristics v–i for bare conductor using the proposed model.
comparison of the current waveforms measured and calculated
after fault inception. The superposition of both waveforms shows
1
a good correlation between measured and calculated results with Measured
0.8 Calculated
a total error of 9%. The variability for measured results is
0.27–0.97 A and for calculated results is 0.31–0.85 A. Table 3 shows 0.6
a comparison of the measured and calculated harmonic content 0.4
existing in the current waveforms shown in Fig. 16. Again the fun-
Current (A)

0.2
damental and the third harmonic are the dominant frequencies
0
and the remaining harmonic contribute in a small proportion to
the overall harmonic distortion. -0.2
-0.4

Comparison with other authors -0.6

-0.8
An advantage of the proposed model is the fact that HIFs can be -1
represented by means of a single non-linear resistance, which can 0 0.05 0.1 0.15 0.2 0.25 0.3
Time (s)

Fig. 16. Measured and calculated current during a HIF for cover conductor.
12
I1
I2x3
10 I3x2
be calculated from measurements. This is a qualitative improve-
I4x3 ment since previous models include several components (diodes,
I5x3
8
resistances, inductances and non-linear resistances) for represent-
ing HIFs and the calculation of all these parameters is not an easy
Current (A)

task [6,15–17].
6
An additional benefit of the proposed model is that can also be
used for representing HIFs of varied characteristics. The measured
4
HIFs reported by the authors in this paper have magnitudes in the
range of 0.25–13 Apeak and they were carried out in dry grass. How-
2
ever, measurements carried out by Nam on crushed and robust
pebbles have peak magnitudes up to 80 Apeak with longer periods
0 of time to stabilize the high impedance [16]. These characteristics
0.15 0.2 0.25 0.3 0.35 0.4
Time (s) can be simulated in the proposed model by modifying the magni-
tudes of G0 and s. Figs. 17 and 18 show a reproduction of the
Fig. 14. Harmonics currents for bare conductor using the proposed model. current waveforms reported by Nam, using the model developed
 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172 169

Table 3
Harmonic content analysis for Fig. 16.

Cover Conductor
Harmonics Measured Calculated
Mean (A) Standard deviation Min. (A) Max. (A) Mean (A) Standard deviation Min. (A) Max. (A)
1st 0.6815 0.0559 0.2099 0.7931 0.5926 0.0937 0.2325 0.5992
2nd 0.0237 0.0096 0.00098 0.0839 0.0038 0.0202 0.00007 0.0424
3rd 0.172 0.0163 0.0282 0.1982 0.1975 0.0773 0.0814 0.2002
4th 0.015 0.0074 0.0002 0.0479 0.0031 0.0116 0.0018 0.0183
5th 0.019 0.0077 0.002 0.0404 0.0017 0.0192 0.00061 0.0219
6th 0.0102 0.0044 0.00023 0.025 0.0012 0.0092 0.00026 0.0235
7th 0.02 0.005 0.0026 0.0314 0.000915 0.0192 0.00017 0.0147
8th 0.0064 0.004 0.00017 0.0261 0.000757 0.0057 0.00008 0.0088

100 condition exists. The algorithm can be summarized in the follow-

80 ing steps for each phase:
60
Step 1. Measure current waveforms at the beginning of the
feeder.
40
Step 2. Calculate even harmonic components from the 2nd to
Current (A)

20 the 8th.
0 Step 3. Calculate odd harmonic components from the 3nd to the
-20 9th.
-40 Step 4. Form three signals groups: sum of odd harmonics, sum
of even harmonics and 3rd harmonic.
-60
Step 5. Apply the following logic:
-80
!
-100 X
H X
H
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 IF Iað2nþ1Þ > Ia3 AND Ia3 > Iað2nÞ
Time (s) n¼1 n¼1
AND T d > T  T s
Fig. 17. HIF on crushed pebbles using the proposed model.
Step 6. IF the above logic is negative, return to step 1, ELSE.
Step 7. Calculate the average harmonic distortion (THDavg) in
100 the three phases.
80 Step 8. If a THDa is greater than kTHDavg THEN ? TRIP the line
60 ELSE return to step 1.
In the above algorithm, H is the number of harmonics consid-
40
ered, typically H = 4, Ia is the current in phase a, Td = 200 ms is
Current (A)

20
the delay time to trip the feeder and Ts is the time of fault incep-
0 tion. For example, if during a given event the sum of odd harmonics
-20 is higher than the third harmonic and the third harmonic is greater
-40 than even harmonics, and these relationships are maintained in a
-60
single phase for a period of time greater than Td, then a high
impedance fault is suspected in the feeder and the logic described
-80
in steps 7 and 8 is applied. The delay time Td is adjustable and aids
-100 to discriminate high frequency events in the system, k is a coeffi-
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (s) cient helpful to discriminate HIF in a phase conductor with
non-linear loads in the feeder, typically k = 1.05 and THDavg is the
Fig. 18. HIF on robust pebbles using the proposed model. average THD for the three phases.
In order to assess the performance of the proposed algorithm, a
in this paper. A comparison of these current waveforms with those model was built up in the MODELS section of ATP, with all the algo-
presented by Nam in [16] show good agreement in magnitude, rithm features already mentioned. The model representing the HIF
timing and waveforms. This comparison illustrates the model ver- detection algorithm is connected at the beginning of the distribu-
satility to simulate HIFs of varied characteristics in distribution tion feeder, emulating the operation of a distribution relay, see
systems. Fig. 19. Several scenarios were analyzed in order to evaluate the
operational characteristics of the proposed algorithm.
HIF fault detection algorithm
Cases of study
Algorithm description
Fig. 19 shows the representation in ATP for the distribution fee-
The developments presented in the previous sections allow rep- der used to carry out the set of measurements shown in ‘measure-
resenting HIFs with an acceptable accuracy, as shown in Figs. 13 ments of high impedance faults’. The feeder characteristics are
and 14. Thus, after a careful analysis of HIFs behavior under varied given in the Appendix A2. Observe that the MODEL with the HIF
circumstances, a new algorithm for HIF detection in EDS is pro- detection algorithm is located at the beginning of the feeder. On
posed. The algorithm is based in analyzing the current harmonic the other hand, the MODEL used for representing HIFs is located
content in the feeder in order to determine whether a HIF at the end of the feeder. The basic idea is to simulate along the
170 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172

Fig. 19. Distribution feeder representation in ATP.

feeder several system events and to assess the algorithm perfor- 170
mance under varied scenarios. These events are switching capaci- 160
tor banks, motor starter, transformer energization and of course 150 Ib

high impedance faults. It should be mentioned that load currents 140

considered in phases a, b, and c were 119, 123 and 117 A respec-

Current (A)
130
Ia
tively, with an average THD = 4.5%. These values were measured 120
at the substation. 110
Table 4 shows several events simulated along the feeder and its 100
Ic
time of occurrence. On the other hand, Fig. 20 shows the phasor 90
current magnitudes, phases b and c have an offset of ±25 A for clar- 80
ity of presentation. From Fig. 20 it is evident that detecting a HIF 70
A B C D
based only in the fundamental components is not an easy task. 60
The low current magnitudes during HIFs are difficult to identify 0.2 0.4 0.6 0.8 1 1.2 1.4
in the background of relatively high magnitudes of load currents. Time (s)
In fact, the HIF at t = 0.8 s in phase a can be consider like a small
Fig. 20. Currents in the feeder for several events shown in Table 1.
increase on load current.
Fig. 21 shows the performance in phase a, at the beginning of
the feeder, of the third, sum of even and sum of odd harmonics 50
with an offset of ±25 A for the events shown in Table 4. Some of
40
these events produce a wide range of harmonics, but the HIF Odd Harmonics
detection algorithm requires only the low frequency components. 30

For example, during capacitors bank switching (300 kVAr) at 20

Current (A)

t = 200 ms, event A, odd harmonics are greater than the 3rd and 10 3th Harmonic
even harmonics but the algorithm does not trip the feeder during 0
the event because they decay rapidly and last only for a few cycles
-10
(<100 ms). However, observe that on two occasions at very low Even Harmonics
-20
harmonic magnitudes, the even harmonics are greater than the
3rd and odd harmonics for a few moments, alarming the algorithm. -30 A B C D
Trip
These situations do not trip the feeder because they last for less -40
than 200 ms. Thus, there are two alarms but not a trip signal. -50
Motor starters (50 hp) were simulated at t = 400 ms, event B. In 0.2 0.4 0.6 0.8 1 1.2 1.4

this case, smaller magnitudes of low frequency harmonics were Time (s)
found in the waveforms and after a short period of time they prac-
Fig. 21. Harmonic components in current waveform.
tically disappear. These characteristics are taken into account by
the HIF detection algorithm for discriminating motor starter from
HIFs. Observe that the algorithm does not trip the feeder but there in odd harmonics. Therefore, the increase of even harmonics com-
is an alarm due low harmonic magnitudes. pared to odd harmonic is used by the algorithm to discriminate
Transformer energization (300 kVA) was simulated at transformer energization from HIFs. Observe that the alarm signal
t = 600 ms, event C. In this case inrush currents produce a signifi- for this event is activated but does not last for long.
cant amount of even harmonic components and smaller increases Finally, a HIF was simulated in the feeder at t = 800 ms, event D.
In this case the 3rd and odd harmonics component remains practi-
cally constant for periods of time greater than 200 ms, then the
Table 4
algorithm identify that the event is a HIF and send a trip signal
Events simulated in the distribution feeder.
to the circuit breaker. From the above results, the conclusion of this
Event Time Description study is that the proposed algorithm discriminate properly
A 0.20 s Capacitor bank switching different events in the EDS from HIFs. The algorithm can still be im-
B 0.40 s Motor starter proved further by considering a minimum harmonic current to trip
C 0.60 s Transformer energization
the circuit breaker and avoid alarming the model of HIF detection
D 0.80 s High impedance fault
algorithm.
 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172 171

Conclusions Table A3.2

Transformer parameters.

This paper has proposed a new model for representing HIFs in Nominal voltage H 13,800 V, delta
EDS, including a methodology for calculating the model parame- Nominal voltage X 440 V, Wye
Nominal power 300 kVA
ters from measurements. Another contribution is the development
Nominal freq. 60 Hz
of an algorithm for HIF detection in distribution feeders. The HIF Impedance 11%
model is based in modeling the high impedance path associated Efficiency 99.80%
to faults. On the other hand, the algorithm for HIF detection in dis- PcuH/PcuX 50%
tribution feeders is based in the harmonic analysis of current
waveforms and certain logic aimed to discriminate HIFs from other
phenomena existing in the EDS.
Table A3.3
Several field tests were carried in order to obtain information Capacitor bank parameters.
about the performance of HIFs under varied circumstances. Based
Nominal voltage 13.8 kV
in the analysis of these measurements, a representation for HIFs
Reactive power 300 kVAr
is proposed. Some practical assumptions facilitate the solution of Nom. current 12.55 A
the representative model equations. The results provided by the Xc 634.8 Ohm
HIF model were compared with those obtained during measure- Capacitance 4.17 lF
ments. The obtained results show that the HIF model can repre-
sents accurately the measured high impedance faults. The time
constant and the average conductance during the fault also facili- Appendix A
tate the representation of HIFs of varied characteristics reported
by other authors. On the other hand, the analysis of field test show See (Tables A2.1–A3.3).
some features typical of HIFs such as arcing, low current magni-
tudes, random behavior, non-linear v–i characteristics, asymmetri- MODEL HIF
cal magnitudes and harmonics, which suggest a strong non-linear CONST gmin {val:1.0E-6}
resistive component. v0 {val:11267}
The DFT was used to obtain the harmonic content of current DATA g0
waveforms. The results show that HIFs have a high content of tau
3rd harmonic. This characteristic is used for the development of tinit
a HIF detector for distribution systems solidly grounded. Several INPUT vt
tests, some of them presented in the paper, show the effectiveness OUTPUT rt
of the proposed algorithm. VAR gs
Finally, HIFs require complex models because of its own gd
non-linear dynamics leading to high harmonic content on phase gt
currents during this kind of events. However, this complex phe- sg
nomenon was successfully simulates in the MODELS section of rt
ATP by means of a single non-linear resistor. EXEC
gs: = g0(vt2/(v02))
gs: = gs(1exp((ttinit)/tau))
Table A2.1 IF (t 6 tinit) THEN
Distribution feeder geometry. gd: = gs
ELSE
Cond. No. Phase X (m) Y tower (m) Y min (m)
LAPLACE(gd/gs): = 1|/(1| + 5e4|s)
1 1 1.28 10.05 8.6
ENDIF
2 2 0 10.05 8.6
3 3 0.64 10.05 8.6 sg: = (gd 6 gmin) OR (prevval(gd) 6 gmin)
gt: = (gminsg + gdnot(sg))
rt: = 1/gt
ENDEXEC
ENDMODEL
Table A2.2
Conductor characteristic.

Conductor Diam. Diam. Rcd (20 °C) Radius Radius

type (cm) steel (cm) (ohm/km) (cm) steel (cm)
References

3/0 AWG 1.275 0.425 0.3367 0.6375 0.2125 [1] IEEE power system relaying committee working group D15 report. High
impedance fault detection technology, <http://grouper.ieee.org/groups/td/
dist/documents/highz.pdf>; March 1, 1996.
[2] Detection of downed conductors on utility distribution systems. IEEE PES
tutorial course 90EH0310-3-PWR, 1989.
[3] Fischer N, Hou D. Methods for detecting ground fault in medium voltage
Table A3.1 distribution power systems. White paper SEL, <https://www.selinc.com/
Motor parameters. WorkArea/DownloadAsset.aspx?id=2385>; 2006. p. 1–19.
[4] IEEE power system relaying committee working group report, downed power
Nominal voltage 440 V lines: why they cannot always be detected. IEEE Powers Engineering Society,
Rating 50 HP February 1989.
poles 4 [5] Huang C, Chu H, Chen M. Algorithm comparison for high impedance fault
Nominal freq. 60 Hz detection based on staged fault test. IEEE Trans Power Delivery October
Nominal P.F. 0.892 1988;3(4):1427–35. http://dx.doi.org/10.1109/61.193941.
Nominal speed 1788 Rpm [6] Emmanuel A, Cyganski D, Orr JA. High impedance fault arcing on sandy soil in
15 kV distribution feeders: contribution to the evaluation of the low frequency
172 V. Torres et al. / Electrical Power and Energy Systems 61 (2014) 163–172
spectrum. IEEE Trans Power Delivery April 1990;5(2):673–86. http://
dx.doi.org/10.1109/61.53070.
[7] Zori KJ, Duric MB, Terzija V. Arcing fault detection on overhead lines from
voltage signals. Int J Electr Power Energy Syst 1997;19(5):299–303. http://
dx.doi.org/10.1016/S0142-0615(96)00053-1.
[8] Stoupis J, Maharsi M, Nuqui R. Reliable detection of high-impedance faults
caused by downed conductors. ABB Rev 2004:28–32.
[9] Hou D. Detection of high-impedance faults in power distribution systems. In:
Paper presented at power systems conference: advanced metering, protection,
control, communication, and distributed resources, 2007. PSC 2007 13–16
March 2007, p. 85–95. doi: 10.1109/PSAMP.2007.4740902.
[10] Adamiak M, Wester C, Thakur M. High impedance fault detection on
distribution feeders. Protection and control Journal GE April 2006; p. 25–31.
<http://www.gedigitalenergy.com/multilin/journals/issues/pcj_2006-04.pdf>.
[11] Elkalashy NI, Lehtonen M, Darwish HA, Taalab A, Izzularab M. A novel
selectivity technique for high impedance arcing fault detection in
compensated MV networks. Eur Trans Electr Power ETEP May 2008:344–63.
http://dx.doi.org/10.1002/etep.179.
[12] Baqui I, Zamora I, Mazón J, Buigues G. High impedance fault detection
methodology using wavelet transform and artificial neural networks. Electr
Power System Res July 2011;81(7):1325–33. http://dx.doi.org/10.1016/
j.epsr.2011.01.022.
[13] Samantaray SR. Ensemble decision trees for high impedance fault detection in
power distribution network. Int J Electr Power Energy Syst December
2012;43(1):1048–55. http://dx.doi.org/10.1016/j.ijepes.2012.06.006.
[14] Bakar A, Ali M, Tan C, Mokhlis H, Arof H, Illias HA. High impedance fault
location in 11 kV underground distribution systems using wavelet transforms.
Int J Electr Power Energy Syst February 2014;55:723–30. http://dx.doi.org/
10.1016/j.ijepes.2013.10.003.
[15] Zamanan N, Sykulski J. Modeling an arcing high impedances fault based on the
physical process involved in the arc. In: Proceedings of the 6th WSEAS, Lisbon,
View publication stats
Portugal; September 22–24, 2006. p. 28–33 <http://www.labplan.ufsc.br/
congressos/wseas/papers/517-413.pdf>.
[16] Nam SR, Park JK, Kang YC, Kim TH. A modeling method of high impedance fault
in distribution system using two series time-varying resistances in EMTP, IEEE
Power Engineering Society summer meeting, vol. 2. Vancouver, BC; 15–19-
July, 2001. p. 1175–80, doi: 10.1109/PESS.2001.970231.
[17] Elkalashy NI, Lehtonen M, Darwish HA, Izzularab MA, Taalab AMI. Modeling
and experimental verification of high impedance arcing fault in medium
voltage networks. IEEE Trans Dielectr Electr Insul April 2007;14(2):375–83.
10.1109/TDEI.2007.344617.
[18] Dommel HW. ATP/EMTP rule book. Canadian American EMTP User Group,
1997.
[19] Torres V, Ruiz HF. High impedance fault detection using discrete wavelet
transform. IEEE Electr Robot Automot Mech Conf 2011:325–9. http://
dx.doi.org/10.1109/CERMA.2011.60.
[20] Cassie AM. Theorie Nouvelle des Arcs de Rupture et de la Rigidité des Circuits.
Cigre, Report 1939; 102: p. 588–608.
[21] Mayr O. Beiträge zur theorie des statischen und des dynamischen lichtbogens.
Arch Elektrotech 1943;37(12):588–608.
[22] Guardado JL, Maximov S, Melgoza E, Naredo JL. An improved Mayr-type arc
model based on current-zero measurements. IEEE Trans Power Delivery
January 2005;20(1):138–42. http://dx.doi.org/10.1109/TPWRD.2004.837814.
[23] Kizilcay M. Dynamic arc modelling in EMTP/TACS. In: 10th European EMTP
users group meeting, Copenhagen, Denmark, May 1986. p. 1–15.
[24] Prikler L, Kilzicay M, Ban G, Handl P. Modeling secondary arc based on
identification of arc parameters from staged fault test records. Int J Electr
Power Energy Syst 2003;25:581–9. http://dx.doi.org/10.1016/S0142-
0615(03)00018-8.
[25] Laurent Dube. MODELS RULE BOOK (New Version), April 1996.