184 IEEE Transactions on Power Delivery, Vol. 11, No.

1, January 1996

HIGH IMPEDANCE FAULT DETECTION DEVICE TESTER

V. L. Buchholz M. Nagpal J. B. Neilson R Parsi-Feraidoonian W.Zarecki
Member Non Member Member Non Member Non Member

Powertech Labs Inc.

Surrey, B.C. Canada V3W 7R7

Keywords - High Impedance Faults, Down Conductor Faults, attempt to distinguish the signatures of high impedance arcing
Protective Relays and Relay Testing. faults from normal loads. There are presently devices
coming on to the commercial market which implement these
Abstruct - High impedance or down conductor fault detection algorithms [2,31.
devices are now commerciallyavailable for evaluation. However, Although HIF detection devices are becoming available,
security and dependability of these devices can not be tested utilities are confronted with the fact that there are no com-
using conventional relay test apparatus and procedures. This monly accepted methods for testing these devices. Conven-
paper presents a test apparatus and procedures for testing high tional relay test procedures, using steady-state 60 Hz
impedance fault detection devices. The apparatus is capable of
playing back in real-time waveforms selected from a data
waveform injection, are not suitable. HIF detection devices
library, which includes seventy seven field recordings of high- require an extended time (sometimes up to a minute) to
impedance faults and feeder loads. Each recording is approxi- reliably differentiate an HIF fiom a normal load disturbance.
mately five minutes long and stored in the form of digitized data Field testing is the most acceptable solution, but a limited
sampled at 20 kHz. number of tests can be performed for every new HIF device.
Also utilities are reluctant to stage faults on feeders that are
I. INTRODUCTION supplying customers. The most economical and acceptable
alternative to field testing is laboratory testing using previous-
A distribution primary fault, which can not be detected by ly recorded high impedance fault and feeder load waveforms.
ordinary overcurrent or ground fault protection, is called a The purpose of this paper is to describe a method for
high impedance fault (HIF). These faults can occur when a evaluating and comparing HIF detection devices using
conductor comes in contact with an object such as a tree, or prerecorded data. The test method can combine digitally
falls on a surface of poor conductivity. The magnitude of the recorded staged HEs with load currents to simulate faulted
fault current generated is too low to harm electric apparatus, feeder waveforms. The method can evaluate both the
however undetected HIFs can cause fire or elecmc shock. dependability of the device to detect an HIF and the security
Considerable research work has been conducted to address of the device against false alarms on noisy loads.
the public safety concerns of HIFs. An excellent summary of
HIF research is given in Reference [l]. The majority of The paper is divided into descriptions of the following:
previous work has concentrated on devising detection
algorithms which examine feeder current and voltage the field tests which form the basis of the fault and normal
waveforms measured at the substation. The algorithms load data used in the €€IFdetection device testing,
adding fault current to load current waveforms to simulate
95 SM 427-5 PWRD A paper recommended and approved faulted feeder waveforms,
by the IEEE Power System Relaying Committee of the the method of digitizing and storing the recorded data,
IEEE Power Engineering Society f o r presentation at test devices and test protocol, and
the 1995 IEEEfpES summer Meeting, July 2 3 - 2 7 , 1995,
Portland, OR. Manuscript submittedDecember 2 2 , 1994;
the device for playing back the test data.
made available for printing April 2 7 , 1995.
The paper is based on two Canadian Electrical Association
projxts, co-funded by British Columbia Hydro, and per-
formed at Powertech Labs Inc. In the first project, completed
in June 1992 [4],recordings of more than one hundred high
impedance faults and normal feeder loads were made and
analyzed. The second project, which is still in progress,
includes building a test apparatus, developing test procedures
and testing three devices [2,3,5].

0885-8977/96/$05.000 1995 IEEE

 185

11. FIELD TESTS TABLE 1

RANGE OF FAULT CURRENT (A RMS) FOR CONDUCTORS
ON VARIOUS SURFACES AS A FUNCTION OF VOLTAGE.
The object of the field tests was to gather data on a large
number of high impedance faults and normal distribution
loads. The data were obtained at B.C. Hydro and at Sur- Surface 3.8 kV 7.2 kV 14.4 kV 20.2 kV
TransAlta Utilities, and on a test line connected to face Condition L-G L-G L-G L-G
Powertech’s High Power Lab. The faults were conducted on Tfle
a large number of varied surfaces at voltage levels from 3 kV
to U ) kV line-to-ground. Emphasis was given to obtaining a
broad cross section of test parameters.
All line currents were measured with standard substation
1 :2 Wet

Damp
1-3

1-2
7-25

1-11
40-120

5-70

current transformers. The secondaries of the main substation

ct’s were monitored with Pearson Model 101 instrumentation
ct’s. These have a flat response from 1Hz to 35 MHZ and an
output of 0.1 V/A. The number of turns from the secondary
of the main ct’s through the aperture of the Pearson ct’s was
varied to provide different current gains.
In order to check the overall frequency response of the
main ct’s and the instrumentation ct’s, a calibration current
signal was injected through the opening of the main ct and
monitored at the instrumentation ct. The calibration signal
was varied from 40 Hz to 5 kHz and the output of the
instrumentation ct’s remained flat to better than 2%.
All line voltages were measured with standard substation
potential transformers. The outputs of the vt’s were attenu-
ated with resistive dividers. The calibration was checked at Dry or
60 Hz only.
All signals were recorded on an Honeywell 101 analog
tape recorder/reproducer. The Honeywell 101 is an FM
recorder that accommodates 15 inch reels of 1 inch wide tape.
It has the capability of recording up to 15 data channels plus
one voice channel. The recorded signals could be the three-
phase and one neutral currents, filtered and unfiltered, the point of fault, current on a feeder experiencingan HIF can be
three phase voltages, and two timing signals. The recorder approximated by adding the current measured during an HIF
operated at 30 inches per second (IPS) in R I G Wideband test to the current measured on an unfaulted feeder. Simu-
Group I mode to yield a data bandwidth of 0 to 10 kHz lations were performed using the Electromagnetic Transient
within 1 dB at a signal to noise ratio (SNR) of 49 dl3. Program @MTP) [6] to investigate the extent of errors that
Table 1 gives the range of rms fault current as a function are likely to be caused by this approximation.
of voltage level and surface condition for energized A simple and accurate model for representing all kinds of
conductorslying on various surfaces. In Table 2 the rms fault HIFs is neither available nor easy to develop. If such a model
current is given for energized conductors touching various were possible, HIF detection would not be a problem. For
objects as a function of voltage. In addition to the fault tests, the purpose of this study, it was decided to use one of the
twenty three feeder loads were recorded at 12 and 25 kV HIF models already proposed in the literature. Emanual et al
(L-L). The circuits recorded varied from quiet residential to [7]modelled high impedance fault arcing on sandy soil. The
noisy industrial loads including an arc furnace and single- model is shown in Fig. 1. It includes two dc sources, Vp and
phase welder. V,, connected in antiparallel by means of two diodes. The
series impedance (R and X) controls the magnitude of arc
III. ADDING FAULT CURRENTS TO LOAD CURRENTS current. In the EMTP simulation, Vp was set to 1000 V, V, to
2500 V and R+jX to 123.4-kj12.3 0.When this model was
Most HIF current waveforms recorded either included the connected across 12.6 kV source, it generated arc current, i,
fault current alone or the fault current riding on small resistive of 50 A (rms), as shown in Fig. 1.
loads. A limited number of measurements were made on A typical 12.6 kV (L-G)distribution system was simulated
customer feeders experiencing HIFs. Because an HIF draws on EMIT. It consisted of a 10 km three-phase section,
very low current and distorts the line voltage very little at the
\ _-

186
TABLE 2
RANGE OF FAULT CURRENT (A RMS) FOR CONDUCTORS
TOUGHING VARIOUS OBJECTS AS A FUNCTION OF VOLTAGE

Object
TI3lClle.d
bY
Conductor

Deciduous
Tree

Deciduous
Tree

DeCidllOnS
Tree
Branch
I - Gmifemus
Tree
Coniferous
Tree
Branch

Wood
fencepost

Steel
fencepost

Concrete
slab un

Bias tires
on subsoil

Radial
tires 011

Fig. 1 HIFModel subsoil

CQIlCFek
followed by three single-phase laterals, 15 km on Phase A, curb
20 km on Phase B, and 15 km on Phase C . The loads were
represented as lumped star-connectedR-L circuits. Balanced
three-phase loads were connected at locations of 1 km, 5 km,
and lo km mO' the source and single-pb loads were 1. Compute an error signal, i, by performing point-by-point
placed at the end of the laterals such that typical currents, il,
were achieved with peak values of 292 A, 267 A and 238 A,
on Phases A, B and C, respectively. An HTF was applied on
Phase B of the feeder by using the arc model at location
ie = {air@)+ '$ - 4% (1)
30 km distance from source. Total current, $ which included where 0 is the phase shift applied to i, (mcurrent) with
both load and fault, was measured at the substation (or initial value equal to 0 and a is the scaling factor with
source) on Phase B. initial value equal to 1.
The following procedure was used to compare the addition
of the HH; arc current and the unfaulted feeder current to the 2. Calculate the ms enor, I,, as
faulted feeder current.
 187

IV.DATA PREPROCESSING
The analog tapes were in most cases multi-track recordings
with 10 k€hbandwidth. The first stage of processing was to
digitize five-minute sections of the analog recordings, for both
normal load and fault currents. The tapes were digitized with
12-bitresollution at 20 kHz sampling rate, and multiple tracks
where T is the period of 60 Hz waveform (16.67 ms). were simulltaneously digitized to maintain synchronization.
The digitizer was based on a personal computer, and the
3. Apply a phase shift and find emin,which gives the lowest results were written to a hard disk in binary format. The files
value of I,. were later moved across a local area network to a VAX
minicomputer, where they were converted to a standardized
4. Find the scaling factor, a,which will result in a value of data file brmat, sorted into directories and then backed up
rms error, I,, close to zero. onto 8 mm tape. It should be noted that each five-minute file
contains 6 million points, or 12 Mbytes of data. With 77 data
The reason for applying a phase angle correction is that sets containing between 2 and 6 files (up to 3 voltages and 3
there exists a phase angle difference between the two currents k a three-phase set) a total of about 4 Gbytes of data
voltages, at the substation where i, is measured and at the storage was required, and the sheer quantity of data presented
fault location. This phase angle difference is equal to emin, some difficulty. To keep track of the files, each data set was
which provides the minimum value of I,. A phase angle assigned a set number, and the suffixes IA, IB, IC, VA, VB,
correction alone can not reduce I, to zero. A scaling factor, VC were added to the data set number to uniquely identify
a less than 1, must multiply i, to bring I, as close to zero as each file in each data set.
possible. The reason for the scaling factor is that the voltage Early in the project it was noted that if fault and load
drop at the location where the fault is applied results in a currents are to be added as discussed in Section m,the proper
reduced arc current in the EMTP simulation providing $ for phase relaltionships must be maintained. The digitized data
use in Step 1 above, while i, used in point-by-point addition displayed small frequency errors, due to system frequency
is calculated assuming no-load voltage. variation, analog tape timing errors, and digitizer errors.
Fig. 2 shows the results of the analysis. This figure shows While these errors were small, in the order of O.Ol%, they
the fault current, 5 directly obtained from the EMTP were sufficiently large to cause phase shifts of several cycles
simulation, the computed fault current {il + a i e,,,)], and i,. over a fivle-minute data file. Before files could be compared
6
The values of e,,,, and a were -12' and 0.7 5. The I, will
never be zero due to the non-linear nature of an €IF.
or added, it was necessary to digitally synchronize each data
set to ex;wly 60 Hz. The synchronization was done by
However, the errors are negligible. selecting m e file of a data set (normally the A phase voltage)
as the key file. The first rising zero crossing was taken as the
start, and the file was synchronized from that point by
dividing ihe file into segments of 1000 points (50 ms or 3
cycles). At each segment boundary, if the rising zero crossing
did not fall on the boundary point a point was added or
deleted to synchronize the segment. The addition or deletion
was made at a point of zero slope to avoid introducing
discontiniiities. Once the key file was synchronized, each file
in that data set was synchronized to the key file, by adding or
deleting points in the same segments as in the key file. As
before, additions or deletions were made at the nearest point
of zero slope.
The next stage in processing was to trim the files to a
standard length. Long files were trimmed to 275 seconds,
starting with a rising zero crossing on A phase voltage, and
shorter files were extended to the same length. Recorded
faults with insufficient pre-fault data were extended by
Fig. 2 Result of analysis. duplicating the pre-fault section enough times to extend it to
the standard length. In addition, the files were scanned for
obvious irecording or digitizer errors, and when these were
found, segments of the file around the error were removed.
188
In all cases, cut and paste operations were performed changes in setting, the device should not operate when the
identically on all files of a data set, and were done at segment load current alone is applied.
boundaries to maintain synchronization.
When the f i a l results were examined, it was found that VI. TEST APPARATUS
some of the files with low signal levels had significant high
frequency components, apparently due to tape or digitizer Operating times of HIF detection devices are in the order of
noise. Fourier analysis showed that these components were minutes. The DFM uses high frequency currents up 750 Hz.
in the 2.5 to 10 kHz range. Since this band is not used by the An apparatus to be used for testing HIF detection devices
available HIF detection devices [2, 3,5], a digital filter was must be able to playback the data that represent minutes long
applied to the data sets to remove noise above 2 lrHz from the consinuous waveforms with a bandwidth of at least one kHi.
data. At the end of this process, all the data files were Relay test apparatus with the ability to playback field recorded
exactly synchronized to 60 Hz, starting and ending with a data are commercially available. The main limitation of these
rising zero crossing of the A phase voltage. apparatus is the number of samples per channel that can be
used. The E500 Doble test apparatus allows up to 32
V. TEST DEVICES AND PROTOCOL ksamples per channel. This apparatus can playback a data set
recorded at 2 kHz (bandwidth less than 1 E)for a
CEA has contracted to Powertech for testing and evaluation maximum of 16 seconds. This duration is not adequate for
of three HTF detection devices. Two devices, High testing H E detection devices.
Impedance Fault Alarm System (HIFAS) 121 and Digital A test apparatus was build at Powertech consisting of a
Feeder Monitor (DFM) [3], are now commercially available personal computer, a digital-to-analog converter, and voltage
for evaluation. A third HIF detector was developed at the and current amplifiers. An IBM-compatible Pentium personal
University of Manitoba [5]. The HIFAS uses changes in third computer is used as a host for the test apparatus. It runs at a
harmonic current magnitude and phase relation to the system 90 MHZ clock rate and includes 16 Mbyte RAh4,256 kbyte
voltage for detecting arcing in a down conductor. The DEM cache memory, and a 500 Mbyte hard disk. An ethernet
targets high impedance faults through expert system card is provided for access to a VAX disk (mass data storage)
integration of several detection and analysis algorithms. Two which is defined as a logical drive (i.e. drive e:). The digital-
algorithms, the Energy Algorithm and the Random Algorithm, to-analog card is a commercial product AT-AO-10 from
used in the DFM attempt to identify energy bursts in feeder National Instruments. It uses 12-bit digital-to-analog (D/A)
current at non-fundamental and non-harmonic frequencies (up converters for simultaneously generating ten outputs at rates
to 750 Hz) caused by the high impedance arcing and its up to 200 kHz. Each channel has a jumper selected option
random behaviour. The University of Manitoba relay looks for hntemal 10 V or extemal reference voltages. In this
at the positive and negative peaks of the feeder currents for apparatus design, the voltage channels have fixed D/A
finger prints of HIFs. converter resolution with reference to the intemal 10 V
A protocol has been prepared so that each device will be whereas current channels can have one of two resolutions
tested using a uniform test procedure and subjected to depending on an external reference voltage of either 5 V or
identical test waveforms. The protocol consists of three 1.5 V. The external reference voltage is used to improve the
stages of testing. Stage 1 tests are intended to evaluate the resolution of DlA conversions for signals of low magnitudes.
security of the devices, which will be tested with waveforms This feattne is especially useful for Stage 2 testing when HIF
that represent different load conditions and normal switching currents of low magnitudes are used. The board has a
events of the power system. Twenty three recordings of standard IBM AT bus interface which is directly plugged-in
various types of loads are available in Powertech’s HIF data to the host personal computer for its control and operation.
library. The device should not alarm when load currents are Three outputs of the D/A card are connected to buffer
applied to its input terminals. Stage 2 tests are dependability amplifiers and then to isolation transformers for generating
tests using the HIF currents. There are fifty four tests 120 or 69 V signals. The use of buffer amplifiers provides
representing various HTF conditions. In some cases, the HIF enough fan-out so that several detection devices can be
currents are riding on pure resistive loads and there are some connected in parallel without causing saturation of D/A
casea of HIF currents alone. Thus, the tests in this stage are outputs. Another three outputs are connected to three Doble
simple dependability checks and the device is suppose to F2100 amplifiers for generating currents up to 4.5 A (rms)
signal the presence of HIFs. Stage 3 tests are a combination when an external reference voltage of 5 V is used.
of dependability and security tests performed by The software for the test apparatus is written in ANSI C++
superimposing the HIF currents of Stage 2 on the load programming language for Windows. It runs on the host
currents of Stage 1. The device is expected to alarm when computer providing a user-interface to the test apparatus and
HIF current is superimposed on the load current. With no controlling the operation of the D/A card. Driver routines of
the D/A card are supplied by its manufacturer. The software
 . . 189

has three main modules: set-up, pre-processor, and start-up

modules. The set-up module allows the entry of hardware
settings which include ct and vt ratios, enabling and disabling
of output channels, and data rate. There are two pre-
processors, Re-processor 1 and Preprocessor 2. Pre-
processor 1 is used for Stage 1 and 3 testing when the feeder
currents include load cwrents. The D/A conversion resolution
of current channels is calculated for a 5 V external reference
voltage. Pre-processor 2 is used for Stage 2 testing when
feeder currents are low magnitude generally including HIF
currents alone. The resolution is calculated for a 1.5 V
external reference voltage. The pre-processor module
retrieves data values simultaneously from all selected data
files, one for each channel, residing on the VAX disk. These
data are processed by using appropriate scaling factors and
D/A converter resolutions. The processed data are written to
the output test file which resides on the local hard disk of the
host. The start-up module controls the operation of the D/A
card. It reads a test file created by the pre-processor module
and generates the desired analog outputs suitable for testing
the HIF detection devices. The start-up module imposes no
limitation on the number of data points per channel as it reads
the test file from the hard disk directly while the output
waveforms are being generated. The only limitation on the
number of data points is imposed by storage space available
on the hard disk.
The most important performance requirement of the test
apparatus is that it must reproduce field acquired signals
precisely. Several tests were performed to check the
apparatus. Results from one test are presented. This test 5
compares waveforms represented by the input data and analog
outputs from the test apparatus. For input and output
waveform comparison, data points at 20 kHz were generated
for a 60 Hz sinusoidal voltage and a 60 Hz triangular current Fig. 3. Input and output waveforms comparison.
waveforms. The current waveform was digitally low-pass
filtered with cut-off frequency 2 kHz to eliminate higher order available relay testers. It can continuously output six analog
harmonics. Using the two waveforms as inputs, the analog signals nepresenting input data recorded at rates up to
outputs from the apparatus were captured using a digital 20 kHz. With a 500 Mbyte hard disk, these outputs can run
scope. Fig. 3 shows the comparison between them. For up to 30 minutes .
comparison purpose, the outputs from the test apparatus are
converted to primary values. A good match is observed VII. REFERENCES
between the inputs and outputs.
[l] Detection of Downed Conductors on Utility
VII. CONCLUSIONS Distribution Systems, IEEE Tutorial Course,
9OEHO331O-PWR, 1989, NY, USA.
This paper demonstrated that the fault current in a feeder [2] High Impedance Fault Alarm System, A Users Guide
experiencing an W; can be approximated by adding currents to a Viable Solution to Distribution Down-Conductor
measured during HIF alone and on an unfaulted feeder. Detection, U.G. 33, February, 1994, Nordon
Before addition, appropriate phase angle correction and Technologies, FL, USA.
scaling factor must be applied to the HIF current so that the [3] Dligital Feeder Monitor, GEK-100643, General
error in the approximation is close to zero. Electric Company, PA, USA.
An HIF device tester has been built at Powertech Labs Inc. [4] The Characterization of High Impedance Faults,
This tester overcomes the drawbacks of commercially 038 D 721, June, 1992, Canadian Electrical
Amociation, Quebec, Canada.
190

PI A. F. Sultan, G. W. Swift and D. J. Fedirchuk, Dr.M. Nagpal obtained his M.Sc. and Ph.D. degrees from the
"Detecting Arcing Downed-Wires Using Fault University of Saskatchewan, Canada, in 1986 and 1990,
Current Flicker and Half-Cycle Asymmetry", IEEE respectively. He has been working in Powertech Labs Inc. for
Transactions on Power Delivery, Vol. 9, No. 1, the last three years. Before joining Powertech, he had worked
January, 1994, pp. 461-70. as a consulting engineer for approximately four years
[61 Electromagnetic Transient Program Reference designing industrial power systems. Dr. Nagpal is an adjunct
Manual (EMTP Theory Book), Prepared for professor at the University of British Columbia.
Bonneville Power Adminstration, Department of
Electrical Engineering, University of British Dr.Bruce Neilson received his Ph.D. in plasma physics from
Columbia, August, 1986. the University of British Columbia in 1981 and became a
[71 A. E. Emanuel and E. M. Gulachenski, "High professional engineer in 1987. He is an adjunct professor at
Impedance Fault Arcing on Sandy Soil in 15 kV UBC and a member of the IEEE. Since 1981 he has worked
DistributionFeeders: Contributionsto the Evaluation for B.C. Hydro R&D/Powertech Labs Inc.
of the Low Frequency Spectrum",IEEE Transactions
on Power Delivery, Vol. 5 , No. 2, April, 1990, pp. Mr. R. Parsi-Feraidoonian was bom in India. He received his
676-86. B.Sc. fiom Southem Illinois University at Carbondale and his
M.A.Sc. from the University of British Columbia. He is
1Mr. Buchholz, presently Director of the Transmission and currently employed at Powertech Labs Inc. as a research
Distribution Group at Powertech Labs, has 7 years experience engineer.
as a research engineer in geophysics, designing electronic
remote sensing equipment, and 14 years experience in Erlr. W. Zarecki obtained his diploma in electrical and
electrical engineering in the electric utilities. electronics technology from the British Columbia Institute of
His utility projects include work on grounding apparaius, Technology, Canada, in 1974. He is employed as a senior
electrical connectors, switchgear, power and instrument electronics/controltechnologist with Powertech.
transformers, batteries, and lighting equipment. He has been
the leader of two large Canadian Electrical Association
projects on high impedance faults in distribution systems.