Line Length and Fault Distance Considerations

in Traveling-Wave Protection and

Fault-Locating Applications

Bogdan Kasztenny and Venkat Mynam

Schweitzer Engineering Laboratories, Inc.

Presented at the
75th Annual Georgia Tech Protective Relaying Conference
Atlanta, Georgia
May 4–6, 2022

Original edition released May 2021

 1

Line Length and Fault Distance Considerations in

Traveling-Wave Protection and Fault-Locating
Applications
Bogdan Kasztenny and Venkat Mynam
Schweitzer Engineering Laboratories, Inc.

Abstract—This paper analyzes the impact of line length, fault both data exchange and time synchronization. In both
location, and locations of external and internal discontinuities on applications, the double-ended method does not require a
traveling-wave (TW) protection and fault-locating functions. The dedicated channel but shares the channel with protection
paper explains the underlying principles and derives a method to
calculate the minimum line length that yields an expected level of schemes. This avoids additional cost and complexity. The
accuracy and dependability. The paper serves as a tutorial on double-ended method allows a TW-based line monitoring
propagation and timing of TWs and is of interest to those function [6] for continuous line monitoring to detect, locate,
practitioners who evaluate, test, apply, and troubleshoot TW- and tabulate incipient or recurring line faults and fault
based devices. The paper is directly applicable to devices that use precursors.
a window-based method for detecting and time-stamping TWs; The single-ended TW-based fault-locating method (referred
however, the conclusions and findings can be extrapolated for
devices that use any signal-processing method for detecting TWs. to in this paper as the single-ended method) works with data
from the local line terminal and avoids the need for a digital
I. INTRODUCTION channel and time synchronization.
Our field experience with time-domain protection is
Following positive field experience with traveling-wave
excellent. Relays [4] and [5] have been installed to protect well
(TW) fault locators [1] [2], we have successfully introduced
over a hundred lines, have restrained for thousands of external
TW-based line protection [3] with field installations starting in
events, and have operated numerous times for internal faults.
early 2017 [4]. To date, the following TW-based line
These line protective relays have an excellent security record
protection, fault-locating, and line monitoring functions are
and a good dependability record. The observed trip times are on
available in protective relays and have been successfully
the order of 2–8 ms for the TD21 element, 1–2 ms for the TW87
deployed in the field [4] [5]:
scheme, and 1–2 ms plus the channel time for the permissive
• TW-based directional element, TW32. overreaching transfer trip scheme. The fault-locating accuracy
• TW-based differential scheme, TW87. is on the order of one tower span as demonstrated in the field
• Single-ended TW-based fault locating, SETWFL. since 2013 [2] and since 2016 [4].
• Double-ended TW-based fault locating, DETWFL. Measuring differences in TW arrival times and comparing
• TW-based line monitoring, LM. polarities of TWs are at the heart of any TW-based protection
Additionally, the ultra-high-speed line protection includes or fault-locating method. Indeed, a TW-based method can be
the following incremental quantity-based elements: defined as one that responds to differences in TW arrival times
• Incremental-quantity directional element, TD32. or relative polarities of TWs. TW arrival times and polarities
• Incremental-quantity distance element, TD21. are robust signal features. When a TW is properly detected in
In a typical application, the TD21 element is configured to signals acquired at high sampling rates, the arrival time and
trip directly without a protection channel, the TW32 and TD32 polarity of that TW are measured very accurately. The polarity
elements are used in a directional comparison pilot scheme, and and arrival time are principally independent from the signal
the TW87 scheme is used when a direct fiber channel is magnitude, fault resistance, and many properties of the power
available. A typical application uses phasor-based protection system. Therefore, interfering signals cannot easily alter and
elements and schemes for dependability in cases where the influence the TW polarity and arrival time.
time-domain protection restrains when the TW signals are too Reliable detection of TWs in a stream of signal samples
small or for other reasons [3]. Early microprocessor-based TW depends, however, on sufficient time separation between
line protective relays required a standalone backup relay. successive TWs. If two or more TWs arrive in quick succession,
Newer relays include phasor-based protection elements. a TW-based relay or a fault locator may have difficulties
The double-ended TW-based fault-locating method separating these TWs from one another. A blunt instrument of
(referred to in this paper as the double-ended method) can be faster sampling would not necessarily solve the problem
applied over a multiplexed channel (IEEE C37.94 encoding) for because the frequency response of instrument transformers and
data exchange and with IRIG-B-connected satellite clocks for secondary cables would become limiting factors [1]. A relay or
time synchronization, or it can use a direct fiber channel for a fault locator can detect and time-stamp two TWs only if the
 2

second TW arrives after a certain delay. To arrive separated by 98 percent of the speed of light in free space on overhead lines
a certain minimum time, the two TWs must travel two distances and at about 45 to 85 percent of the speed of light in free space
that differ by a certain minimum distance. on cable lines. Ethylene propylene rubber cable insulation
The following situations may lead to a train of TWs that results in a propagation velocity at the lower end of the range,
arrive in quick succession: oil filled paper insulation results in a propagation velocity at the
• Very short lines where the end-to-end TW travel time upper end of the range, and cross-linked polyethylene
is very short. insulation yields a propagation velocity in the middle of the
• Faults very close to either line terminal or close to any range. For simplicity, this paper uses 70 percent propagation
discontinuity on the line (such as a line tap). velocity when discussing cable lines. From the measurement
• Applications with very short lines connected to the and signal processing perspectives, a TW is a step change in
same bus as the protected line. current or voltage with transition times on the order of a few
The impact of TWs arriving in quick succession is different microseconds. Fig. 1 shows an ideal TW in the signal x (current
for different protection and fault-locating functions and TW- or voltage). The TW in Fig. 1 arrives at time t0, has a positive
detection methods. Moreover, the impact of insufficient TW polarity (the signal stepped up), and an instantaneous
separation is not necessarily a total loss of function but is rather magnitude of A0. The pre-step and post-step signal values
a gradual loss of protection dependability and fault-locating appear flat because the figure shows a very short span of time
accuracy. (microseconds), not allowing the curvature of the fundamental
This paper discusses the line length challenge and other frequency alternating current (ac) signal component to be
related issues as they apply to TW-based protection, fault visible.
locating, and line monitoring functions. The conclusions are
directly applicable to functions implemented in [4] and [5] but
may be extrapolated to other implementations. The paper is
organized as follows:
• Section II explains TW-detection and time-stamping
methods and focuses on the differentiator-smoother
filter used in [4] and [5] and used in a slightly
different form in [2].
Fig. 1. TW in the input signal x.
• Section III explains the proximity effect when a fault
is located too close to a discontinuity on the protected Protection and fault-locating functions in [4] and [5] use a
line, including line terminals and line taps, or too close differentiator-smoother (DS) filter [1] to detect TWs. A DS
to a discontinuity external to the protected line. filter is a finite-impulse response filter (FIR) with a data
• Section IV explains the issue of TWs aliasing, where window, as shown in Fig. 2. The DS filter is a least-square best-
multiple TWs arrive at the same time because they fit estimator for a step signal pattern. This is analogous to the
traveled the same distance after being reflected several Fourier filter being a least-square best-fit estimator for a sine
times. wave signal pattern. The DS filter detects step changes in the
• Section V discusses the case of a short line where TW input signal, the same way the Fourier filter detects sine waves
reflections from the opposite terminal arrive so early in the input signal. Based on the concept of a data window, the
that they blend with the TW from the fault. It also DS-based method for detecting TWs can be referred to as a
discusses long cable lines and the issue of TW window-based method. Other methods are possible [1] and
attenuation and dispersion. have been both applied in the field and proposed in literature.
This paper focuses on window-based TW-detection methods.
• Section VI introduces the concept of TW-based fault-
We denote the half-length of the DS filter window as
locating dependability as it applies to line length and
TWDSW (TW differentiator-smoother window). The gain
location of the fault.
coefficient for the filter is 1/TWDSW to ensure that the DS
• Section VII briefly discusses fault analysis and offline
filter output corresponds to the instantaneous TW magnitude,
fault-locating calculations as they relate to line length,
at least when the TW front is sharp. As with any FIR filter, the
fault location, TW aliasing, and proximity effects.
DS filter has a group delay equal to half its data window length,
• Section VIII discusses in detail the accuracy of TW-
i.e., the group delay is TWDSW.
based fault-locating methods and the dependability of
TW-based protection elements and schemes [4] [5].
We recommend that readers review the principles of
operation of the discussed protection and fault-locating
functions by reading [1], [3], [6], and [7].

II. TW DETECTION AND TIME-STAMPING

TWs are surges of electricity launched by a sudden change
in voltage, such as a line fault, that propagate at about Fig. 2. DS filter data window.
 3

In a practical TW-based device, the TWDSW parameter is Fig. 4 shows a case of two ideal TWs that arrive in quick
on the order of several microseconds. Long DS filter windows succession. The figure uses a TWDSW = 10 µs for detecting
allow better noise suppression. Short DS filter windows allow TWs. Fig. 4(a) shows the first TW that arrives at 0 µs (the step
detecting TWs that arrive separated by less time. Protection and change from 0.5 to 1.5 in the signal level) and the second TW
fault-locating functions in [4] and [5] use a common DS filter that arrives at 30 µs (the step change from 1.5 to 2 in the signal
for fault locating and TW-based protection. Therefore, their DS level). The two peaks in the DS filter output signal represent the
filter data windows are relatively long (TWDSW = 10 µs) TW magnitudes (1 and 0.5 respectively) and arrival times (DS
striking a good balance between protection, security, and time- filter output peaks are at 10 µs and 40 µs, respectively, and are
stamping resolution. Device [2] provides TW-based fault consistently shifted by the 10 µs group delay of the DS filter
locating only and it uses a shorter DS filter data window, on the with respect to the true arrival times of 0 µs and 30 µs).
order of 3 µs.
Fig. 3 shows the response of the DS filter to an ideal TW and
a dispersed TW (dispersion refers to the wavefront losing its
steepness as the TW travels over a long distance). The filter
output has a triangular shape when subjected to an ideal TW.
The peak of the output waveform represents the instantaneous
TW magnitude (including the TW polarity), and the time of the
peak represents the TW arrival time (with a constant group
delay of TWDSW).
TWs encounter dispersion when they travel on lossy lines.
Dispersion causes the TW front to lean rather than be an ideal
step (compare Fig. 3(b) and Fig. 3(a)). The DS filter responds
with a parabola-shaped output to dispersed TWs. TW time-
stamping algorithms in [2], [4], and [5] fit a parabola to the
samples near the DS filter output peak and calculate the time of
the peak (TW arrival time) as the time when the best-fit
parabola is at its extremum [1]. This approach results in Fig. 4. DS filter response to two TWs that arrive in quick
additional noise rejection and allows time-stamping resolution succession (the second TW is smaller than the first TW).
that is approximately five times better than the device sampling Fig. 4(a) shows a case where the second TW arrives after a
rate; for example, one can obtain an effective 0.2 µs time time longer than 2 ∙ TWDSW. In this case, the DS filter fully
resolution when sampling every 1 µs. separates (detects and correctly time-stamps) both TWs.
Fig. 4(b) shows a case where the second smaller TW arrives
when the DS filter output was halfway down (TWs are
separated by 0.75 ∙ 2 ∙ TWDSW). In this case, the second TW
starts exciting the filter before the filter has settled after the
previous TW. We can still see two separate peaks in the DS
filter output. The times of the two peaks correspond to the true
arrival times of the TWs. If the TW-detection algorithm (peak-
finding algorithm) is designed to select both peaks in the DS
filter output, the time stamps of the two TWs are correct.
Fig. 4(c) shows a case where the second smaller TW arrives
when the DS filter output was at its peak (TWs are separated by
0.5 ∙ 2 ∙ TWDSW). In this case, the two TWs blend in the DS
filter window. The DS filter output shows only one peak, and
the time of the peak corresponds to the first TW.
Fig. 4(d) shows a case where the second smaller TW arrives
when the DS filter output was halfway up (TWs are separated
Fig. 3. DS filter response to (a) an ideal TW and (b) a dispersed by 0.25 ∙ 2 ∙ TWDSW). The two TWs blend and the time stamp
TW. corresponds to the first TW.
Fig. 3(a) illustrates that the DS filter settles completely in a The second TW in Fig. 4 has a magnitude less than the first
time equal to 2 ∙ TWDSW. When the TW is dispersed, the DS TW. Consider, however, an opposite case when the second TW
filter settling time is slightly longer (2 ∙ TWDSW plus the time has a magnitude greater than the first TW (see Fig. 5). Fig. 5
of the TW transition from the pre-step to post-step levels, see teaches us that when the two TWs blend and the second TW is
Fig. 3(b)). When the filter settles, the step change in the input larger, the time stamp of the blended TW corresponds to the
signal is entirely removed from the filter data window – the second TW, not the first TW.
filter completely processed and “forgot” the previous TW and
is ready to process the next TW.
 4

correctly detected, or they may blend depending on

the ratio of the TW magnitudes.
• Two TWs that arrive separated by less than about
1.5 ∙ TWDSW will always blend. The TW-detection
algorithm will see them as a single TW. The time
stamp will be biased toward the TW with the greater
magnitude.
• If two TWs have similar magnitudes and they arrive in
quick succession and blend, the DS filter output may
have an ill-defined peak (flat top), challenging the
accuracy of the time-stamping algorithm.
• TW dispersion extends the DS filter window settling
time and makes detecting TWs that arrive in quick
succession slightly more difficult.
For example, a device with a TWDSW = 10 µs, such as [4]
and [5], correctly detects TWs that arrive separated by at least
Fig. 5. DS filter response to two TWs that arrive in quick
succession (the second TW is larger than the first TW).
20 µs, the device blends TWs that arrive separated by less than
about 15 µs, and it may – depending on the TW function – work
Fig. 6 shows two more ways in which two successive TWs with lower dependability and accuracy for TWs that arrive
can blend. If two TWs have similar magnitudes and arrive in separated by more than about 15 µs and less than 20 µs.
quick succession, the DS filter output can have an ill-defined Traveling for 20 µs on an overhead line, a TW traverses
peak (flat top), see Fig. 6 (a). The flat top challenges the peak- about 6 km (3.7 mi). Traveling for 20 µs on a cable line, a TW
finding and parabola-fitting algorithms, and it may result in traverses 4 km (2.5 mi). Therefore, a device that uses a 10 µs
poor accuracy of the time stamp. In Fig. 6(a), the time of the DS filter can reliably detect and time-stamp two TWs if they
DS filter output peak is between 10 and 15 µs, pointing to a TW travel distances that differ by about 6 km (3.7 mi) on overhead
arrival time of between 0 and 5 µs (the first TW arrived at 0 µs lines or 4 km (2.5 mi) on cable lines. In the next sections, we
and the second TW arrived at 5 µs). Fig. 6(b) shows a case discuss scenarios where the above requirement of the minimum
when the second TW has an opposite polarity relative to the distance difference is not met.
first TW. In this case, the first TW is time-stamped correctly,
but the second negative DS filter output peak occurs at 20 µs, III. PROXIMITY EFFECT
suggesting that the second TW arrived at 10 µs. In reality, the
second TW arrived at 5 µs; the time-stamping error for the By proximity effect, we mean a situation where two or more
second TW is therefore 5 µs. Two TWs that are of opposite TW discontinuities are located very close to one another. This
polarities and arrive in quick succession create an impulse, and includes the fault (first discontinuity) located close to a line
the DS filter output becomes an impulse response of the DS terminal, line tap, or an overhead-to-cable transition point. It
filter, i.e., it resembles the DS filter data window (compare also includes cases when two discontinuities are located close
Fig. 2 and Fig. 6(b). to one another irrespective of the fault location, such as a line
tap and a line terminal, two line taps, or a line terminal and a
terminal of a very short line connected to the terminal.
A. First Approximation of the TW Pattern
Consider a close-in fault on a two-terminal line, as shown in
the Bewley diagram in Fig. 7. The fault is located at the distance
M (mi or km) from the local terminal (L). The fault launches
two incident TWs that travel away from the fault and toward
the two line terminals. The first TW that arrives at the local
Fig. 6. DS filter response to two TWs that arrive in quick terminal (TWL1) partially reflects from the terminal and travels
succession: (a) ill-defined peak and (b) second TW time stamp is back to the fault. Part of that TW reflects from the fault and
inaccurate. travels back to the local terminal, arriving as the second local
Fig. 4, Fig. 5, and Fig. 6 teach us that a TW that follows the TW (TWL2). The process repeats several times until the TWs
first TW by less than 2 ∙ TWDSW may either skew the TW subside and dissipate. We can also expect that the TWs partially
arrival time or prevent the TW-detection algorithm from transmit through the fault and arrive at the remote terminal as a
finding either or both TWs in the input signal. similar series of TWs in quick succession (TWR1, TWR2, TWR3,
Expect the following when TWs follow in quick succession: and so on). TWs that arrive at the remote terminal reflect and
• Two TWs that arrive at least 2 ∙ TWDSW apart are travel back to the fault. When they reach the fault and the
correctly detected and time-stamped. nearby local terminal, they reflect and travel back to the remote
• Two TWs that arrive separated by more than terminal. TWs at both the local and remote terminals come in
1.5 ∙ TWDSW but less than 2 ∙ TWDSW may be bursts of several TWs in quick succession.
 5

this series propagates away and reflects off other discontinui-

ties, the TWs multiply and overlap.
The time difference between two consecutive TWs in Fig. 7
and Fig. 8 is proportional to the round-trip TW travel time
between the fault (F) and the terminal (L in Fig. 7) or between
the fault and the tap (T in Fig. 8).
According to this first approximation, the proximity
phenomenon does not impact devices that use a
TWDSW = 10 µs when the distance between the fault and the
discontinuity is longer than 3 km (1.8 mi) for overhead lines or
2 km (1.2 mi) for cable lines. When the distance is shorter, TWs
can overlap, and the TW-based functions may be impacted. The
first approximation, however, is a simplification that leads to
overly conservative observations.
B. Actual TW Pattern
A fault launches TWs only in the faulted phase(s). For
example, a Phase-A-to-ground fault only launches a TW in
Phase A. As the TW travels along a three-phase line, some
energy couples from the faulted phase to the healthy phases
Fig. 7. Fault close to a line terminal results in a series of TWs in (Phases B and C for a Phase-A-to-ground fault). However, for
quick succession (first approximation). this coupling to be effective, a TW must travel some distance
If the distance M is sufficiently long, the DS filter can detect [1], such as about 20 to 30 km (12 to 18 mi). If the fault is very
consecutive TWs (see Fig. 4(a) and Fig. 4(b)). If the distance is close to a line terminal, no TWs will develop in the healthy
too short, the TWs will blend (see Fig. 4(c) and Fig. 4(d)), phases. Therefore, the TW arrives and reflects from the line
preventing detection of individual TWs and potentially skewing terminal only in the faulted phase. This reflected TW arrives
the time stamp of the first TW. According to this first back at the fault, but because it traveled only a short distance, it
approximation, the problem is not limited to the local terminal. too arrives only in the faulted phase. When this TW encounters
Both the local and remote terminals receive a train of TWs in the fault, it reflects almost completely and travels back to the
quick succession, irrespective of if the fault is very close to the terminal. Only a very small portion continues to the remote
local or remote terminal. terminal on the faulted phase because the fault has a low
Fig. 8 shows a similar application challenge of a fault being characteristic (surge) impedance. TWs in the healthy phases
close to a line tap. A line tap is a reflection point for TWs. TWs would have passed through the fault location and continued to
travel back and forth between the fault (F) and the tap (T) the remote terminal, but the magnitude of these TWs is close to
causing a train of TWs to arrive at both line terminals. zero (the healthy phases have not coupled any energy from the
faulted phase because of the short distance traveled). Therefore,
no, or very small, TWs transmit through the fault and continue
toward the remote terminal.
Phase-to-phase faults do not create large TWs in the healthy
phase. The coupling from the two faulted phases on the third
healthy phase cancel. Conductor placement asymmetry is the
only reason for any TW energy to couple to the healthy phase
during phase-to-phase faults.
Three-phase faults do not allow TWs to propagate through
the fault because all three conductors include the fault (the fault
resistance is considerably less than the line characteristic
impedance, and therefore the fault has a very low characteristic
impedance in all three phases).
Considering all fault types, we can make the following
observations for faults close to a terminal:
• The local terminal measures TWs only in the faulted
phase(s).
Fig. 8. Fault close to a line tap results in a series of TWs in quick • Multiple TWs arrive at the local terminal in quick
succession (first approximation). succession, with magnitudes less than the first TW.
Fig. 7 and Fig. 8 show that two discontinuities that are close • The remote terminal measures only the initial TW.
to each other generate a series of TWs in quick succession. As This TW typically has the expected three-phase
 6

pattern (the faulted and healthy phases) because the

TW traveled a long distance.
• TW reflections from the local terminal do not
propagate through the fault and do not arrive at the
remote terminal.
Based on the above observations, Fig. 9 and Fig. 10 show
more realistic Bewley diagrams for the cases from Fig. 7 and
Fig. 8, respectively. These diagrams account for the effect of
TW propagation through the fault (a TW will not propagate
through the fault if it has not traveled a long enough distance
prior to arriving back at the fault).
Fig. 9 shows a series of TWs between the fault and the local
terminal. These TWs, however, do not propagate through the
fault. The remote terminal receives a single initial TW (TWR1).
This TW, when reflected from the terminal, travels back and
reflects from both the fault and the local terminal. As a result,
the first reflection from the fault (TWR2) is followed by a
reflection from the local bus (TWR3). The operating conditions Fig. 10. Fault close to a line tap (practical pattern).
for the single-ended method at the remote terminal are much
C. Field Case Example
more favorable compared with the local terminal.
Fig. 11 shows the local and remote currents and current TWs
for a Phase-C-to-ground fault on a 69 kV line [8]. The line is
only 8.49 mi long (end-to-end TW propagation time of
46.51 µs), and the line crew found the fault at 2.01 mi from the
local terminal.
Fig. 11 shows the initial TW at the local terminal
(−173 A primary) and the first reflection from the fault
(−70 A primary), separated by 22.318 µs. Two more reflections
from the fault are clearly visible in the local Phase C current.
The local current TWs are near zero in the healthy phases (A
and B) because no energy coupled to the healthy phases during
the 2 mi travel from the fault to the local terminal. The remote
current TWs in the healthy phases are near zero as well because
the travel distance to the remote terminal is only 6.5 mi. The
remote current TWs do not show any quick reflections resulting
from TWs oscillating between the fault and the local terminal
(the local current shows four clear reflections). The first TW at
the remote terminal is about –186 A primary. The next major
TW is the reflection from the fault after the round-trip time
from the remote terminal to the fault (–98 A primary). The two
Fig. 9. Fault close to a line terminal (practical pattern). TWs at the remote terminal are separated by 72.434 µs. The
Fig. 10 shows a series of TWs between the fault and the tap. first TWs at the local and remote terminals are separated by
These TWs, however, do not propagate through the fault. The 24.528 µs. The field case example in Fig. 11 confirms and
remote terminal receives a single initial TW (TWR1). This TW, illustrates the TW pattern in Fig. 9.
when reflected from the terminal, travels back and reflects from We calculate the TW-based fault location relative to the
both the fault and the tap. As a result, the first reflection from local terminal as follows.
the fault (TWR2) is followed by a reflection from the tap Double-ended method:
(TWR3). The operating conditions for the single-ended method 8.49 mi 24.528 μs
M= �1 − � = 2.006 mi (1a)
at the terminal that does not have a tap between the terminal 2 46.51 μs
and the fault (remote terminal in Fig. 10) are more favorable Single-ended method at the local terminal:
compared with the terminal that has a tap between the terminal
8.49 mi 22.318 μs
and the fault (local terminal in Fig. 10). M= ∙ = 2.037 mi (1b)
The cases in Fig. 9 and Fig. 10 are even more favorable than 2 46.51 μs
the Bewley diagrams imply because TWs attenuate each time Single-ended method at the remote terminal (the calculation
they reflect or transmit through a discontinuity. For example, shows the distance from the local terminal):
TWR2 has a magnitude greater than TWR3 in both figures. 8.49 mi 72.434 μs
M = 8.49 mi − ∙ = 1.879 mi (1c)
2 46.51 μs
 7

The 1.879 mi value differs by only 0.12 mi with respect to reflections from Bus B partially propagate through Bus L and
the true value of 2.01 mi. travel to the fault and the remote terminal. As a result, the
(a) reflection from the fault measured at the local terminal has a
form of several TWs in quick succession. Also, when the
reflection from the local terminal arrives at the remote terminal,
it has a form of several TWs in quick succession. The degree of
impact of the discontinuity (Bus B) depends on the termination
effect at Bus L. If the termination impedance is low (such as
when many lines are connected to the bus), then only a small
portion of the TW travels toward Bus B and an even smaller
portion re-enters the protected line after reflection from the
discontinuity (Bus B). Also, the termination impedance at Bus
B plays a role. The worst-case scenario is when the termination
impedance at Bus B is very low (such as when many lines are
connected to the bus) or very high (such as termination with
only a power transformer). Otherwise, only a fraction of the TW
that reached Bus B travels back toward Bus L.

(b)

Fig. 12. External discontinuity close to the line terminal results in a

series of TWs in quick succession.
An external discontinuity that is close to a terminal blurs the
TW reflected from the terminal and any other TW that
Fig. 11. Local (a) and remote (b) currents and current TWs for a originates from those reflections.
Phase-C-to-ground fault on an 8.49 mi line. In general, the case of Fig. 12 does not affect the double-
Fig. 11 and calculations (1) illustrate that TWs that have not ended method and the single-ended method at the terminal that
traveled for at least about 12 mi do not propagate through a fault is away from the external discontinuity. The single-ended
(the Bewley diagram in Fig. 9 is more accurate than the one in method at the local terminal may be affected by the challenge
Fig. 7). The field case also illustrates the scenario of a close-in to detect the reflection from the fault (loss of function) and by
fault when the second TW arrives just after 2 ∙ TWDSW the series of TWs reflected from Bus B, skewing the TW time
(arrival time difference of 22 µs compared with 20 µs); stamp (degraded accuracy). However, the impact is real only if
compare the local Phase C current TW in Fig. 11 to Fig. 4(a). the termination impedance at Bus L is not low; otherwise, the
Fig. 11 also illustrates the noise rejection capability of the DS reflections from Bus B are small and, therefore, inconsequen-
filter: the Phase C current contains high-frequency noise, yet tial.
the DS filter output is relatively undistorted, allowing correct
TW detection and time-stamping. IV. TRAVELING-WAVE ALIASING
D. Close-In Discontinuity External to the Protected Line We introduce the term TW aliasing to describe a scenario
Fig. 12 shows a Bewley diagram for a line fault in a system where two or more TWs arrive at the protection or fault-
with a discontinuity (Bus B) behind the local terminal. A locating device at the same time after reflecting multiple times
portion of the first TW at the local terminal propagates toward from various discontinuities in the network. The discontinuities
the discontinuity and oscillates between Buses B and L. The are not necessarily close to each other. Only the total travel
times are nearly identical.
 8

Fig. 13 shows a case when the fault is located at a similar

distance from the local terminal (L) as from an external bus (B).
The TWs reflected from the fault and from Bus B arrive at a
similar time (the blue oval shapes in the figure). The polarities
of the two overlapping TWs may be the same or opposite
depending on the termination impedance at Bus B. The two
overlapping TWs can add together (see Fig. 6(a)) or partially
cancel each other (see Fig. 6(b)), leading to an incorrect or
skewed time-stamp value.

Fig. 14. Fault near the line midpoint.

Fig. 15 shows a case when the fault is one-third of the line
length from the local terminal and two-thirds from the remote
terminal. The TW reflected from the remote terminal aliases
with the second reflection from the fault (TWL3).

Fig. 13. External discontinuity at a similar distance as the fault.

The operating conditions for the double-ended method are
good because the first TWs at both line terminals (TWL1 and
TWR1) are not affected by the aliasing phenomenon. Also, it is
unlikely that the aliasing would affect both terminals at the
same time (the remote terminal receives the first reflection from
the fault, TWR2, without aliasing). The TW aliasing in Fig. 13
may affect the single-ended method at the terminal with the
discontinuity behind it (L), but not the double-ended method,
and it is not likely to affect the single-ended method at the
terminal at the opposite end of the line (R).
Fig. 14 shows a case when the fault is located close to the Fig. 15. Fault near one-third of the line.
line midpoint. The TWs reflected from the fault and from the The cases in Fig. 13, Fig. 14, and Fig. 15 illustrate that the
remote terminal (R) arrive at the local terminal (L) at similar first TWs that arrive at both line terminals cannot alias and will
times. The polarity of the TW reflected from the remote always be measured correctly. Also, it is unlikely that TWs will
terminal is typically opposite of the one reflected from the fault. alias at both line terminals (the fault at the line midpoint
The two TWs can partially cancel, or the reflection from the notwithstanding). As a result, aliasing does not impact the
remote terminal can dominate the reflection from the fault. The double-ended method. It may impact the single-ended method
two TWs can also blend, leading to an incorrect or skewed time- but typically at one terminal only.
stamp value (see Fig. 6). When these TWs reflect from the Fig. 16 shows a case relevant for the TW87 scheme. A
terminal and travel back to the fault and the opposite terminal, discontinuity (Bus B) is located at such a distance that it reflects
they multiply, creating even more TWs that arrive again at the a TW that arrives at the local terminal at the same time as the
terminals. Again, the first TWs at both line terminals (TWL1 and expected exit TW (a TW at Terminal L if the external fault (E)
TWR1) are not affected, even though the subsequent TWs may were beyond Terminal R). Normally, a reflection from the
be blurred (may overlap). remote terminal arrives at the exit TW time. An external
discontinuity may reflect a TW that arrives at the same time and
it may impact the exit TW measurement in the TW87 scheme.
This TW aliasing may impact the dependability (but not the
security) of the TW87 scheme.
 9

Fig. 18 shows a Bewley diagram for an external fault behind

the remote terminal of a short line. The first TW (TWR1) is a
reverse TW (arrives from behind the relay). TWR2 is a reflection
from the local terminal and is a forward TW (arrives from the
direction of the line). If these two TWs arrive in quick
succession, then they may affect the security of the TW32
directional element. Voltage transformers will reproduce the
polarity and timing of the first TW (TWR1) but not necessarily
the subsequent TWs (TWR2) [1]. As a result, the presence of
TWR2 could cause the TW32 element to declare a forward fault
at Terminal R for the reverse fault in Fig. 18.

Fig. 16. TW aliasing potentially impacting dependability of the

TW87 scheme.
In this section, we showed examples of scenarios where
TWs may overlap irrespective of the line length and
irrespective of the distance to the fault or to a discontinuity.
These scenarios can impact the accuracy of the single-ended
Fig. 18. TWs for an external fault on a short line.
method at one terminal of the line as well as the dependability
of the TW87 scheme. The adverse effect is constrained to The TW32 element may use a data window, TW32WIN,
specific fault locations as they coincide with distances to such as 50 µs [4] [5]. You may be able to apply the TW32
discontinuities present in the system, irrespective of the line element only when the line propagation time is longer than that
length. Termination impedances reduce the magnitudes of the window (see Section VIII for more information).
aliasing TWs compared with the TWs that are expected and B. Long Cable Lines
used by the TW-based functions.
Cable lines are considerably more lossy than overhead lines
and cause TWs to attenuate and disperse. Attenuation reduces
V. LINE LENGTH EXTREMES
the magnitude of the TW as it propagates along the line.
A. Short Lines Dispersion makes the front of the TW lean over as it propagates
Fig. 17 shows a Bewley diagram for an internal fault on a along the line.
very short line (overhead or cable). TWs reflected from the The characteristic impedance of a cable line is
opposite line terminal arrive soon after the first TW arrives approximately five times less than that of an overhead line
from the fault and may overlap with the TW from the fault (approximately 70 Ω compared to approximately 350 Ω). This
given the DS filter window length. means that faults on cable lines launch current TWs with
magnitudes that are approximately five times greater than those
on overhead lines (assuming the same fault voltage). This
relative TW magnitude boost lessens the challenge of the
attenuation.
Dispersion creates another challenge on long cable lines.
Fig. 19 shows TWs with small dispersion (such as for an
overhead line) and large dispersion (such as for a long cable
line). To explain the dispersion challenge, the figure shows the
DS filter output for the two TWs by using the TWDSW
parameter of 3 µs and 10 µs.
Fig. 17. TWs overlap during an internal fault on a short line. When the DS filter window is sufficiently long compared
If the reflection from the opposite terminal (TWL2 for with the TW “ramp-up” time (both 3 µs and 10 µs DS filter
example) arrives quickly after the first TW from the fault windows in Fig. 19(a)), the DS filter output has a well-defined
(TWL1), then the second TW (TWL2) can either skew the time peak and it correctly captures the TW magnitude (full
stamp of the first TW or it can make it difficult to detect the sensitivity). When the DS filter window is short compared with
first TW (see Fig. 4). Note that the case of a short line is the TW ramp-up time (3 µs DS filter window in Fig. 19(b)), the
different than the close-in fault case in Fig. 9. In the case of a DS filter output does not have a well-defined peak (flat top),
short line, both terminals experience TWs that arrive in quick challenging the peak-detection and time-stamping algorithms,
succession. A close-in fault on a long line affects only the and it only captures a small portion of the TW magnitude
terminal that is closer to the fault. (reduced sensitivity).
 10

The diagram shows five TWs at the local and remote

terminals that are key for TW-based fault locating. The TW
arrival times for these five TWs are as follows:
t L1 = m ∙ TWLPT (2a)
t L2 = (2 − m) ∙ TWLPT (2b)
t L3 = 3 ∙ m ∙ TWLPT (2c)
Fig. 19. TWs with small (a) and large (b) dispersion; DS filter with t R1 = (1 − m) ∙ TWLPT (2d)
a 3 µs half-window (red) and 10 µs half-window (blue).
t R2 = 3 ∙ (1 − m) ∙ TWLPT (2e)
The DS filter window of 10 µs (default in implementations
[4] and [5]) is sufficiently long to address attenuation in cable We consider that two TWs can be reliably detected and time-
lines. The short DS filter window in [2] may be challenged in stamped if they arrive at least 2 ∙ TWDSW apart. Therefore,
applications to long cables. TWL2 and TWL1 can be reliably detected if:
Very long cables may cause significant attenuation and (2 − m) ∙ TWLPT − m ∙ TWLPT > 2 ∙ TWDSW (3a)
dispersion and they may challenge the dependability of TW- Solving (3a) we obtain:
based fault locators and the TW87 protection scheme. TWDSW
m<1− (3b)
TWLPT
VI. FAULT-LOCATING DEPENDABILITY
Condition (3b) teaches us that that individually detecting
In this section, we use observations from Sections II through TWL2 and TWL1 depends on the ratio of the TWDSW parameter
V and introduce a new concept of dependability contours for and TWLPT. Therefore, we introduce an auxiliary variable δ as
the double-ended and single-ended methods. A dependability follows:
contour is an area on a two-dimensional plane comprising the
TWDSW
per-unit fault location (m) and the ratio of the DS filter window 𝛿𝛿 = (4)
half-length (TWDSW) and the TW line propagation time TWLPT
(TWLPT). Inserting (4) into (3b), we can write that TWL2 and TWL1 are
Consider the following example. The TWDSW parameter of separated if:
the TW-based fault-locating device is 10 µs and the length of m < 1 − 𝛿𝛿 (5)
an overhead line is 100 mi (TWLPT for the line is about Similarly, TWL3 and TWL1 are separated if:
547 µs). Assume internal faults at 0.98 pu (98 mi) and 0.505 pu
3 ∙ m ∙ TWLPT − m ∙ TWLPT > 2 ∙ TWDSW (6a)
(50.5 mi). The first location is close to a terminal; the second
location is close to the line midpoint. Dependability contours Solving for m, we obtain:
determine whether these faults are within the dependability m > 𝛿𝛿 (6b)
limits of the double-ended and single-ended methods. Similarly, TWL3 and TWL2 are separated if:
A. TW Arrival Time Analysis 3 ∙ m ∙ TWLPT − (2 − m) ∙ TWLPT > 2 ∙ TWDSW (7a)
Consider an internal fault at location m > 0.5 pu, as shown Solving for m, we obtain:
in the Bewley diagram in Fig. 20. 1
m > (1 + 𝛿𝛿) (7b)
2
Finally, TWR2 and TWR1 are separated if:
3 ∙ (1 − m) ∙ TWLPT − (1 − m) ∙ TWLPT
(8a)
> 2 ∙ TWDSW
Solving for m, we obtain:
m < 1 − 𝛿𝛿 (8b)
The double-ended method works best if the first TWs at both
terminals are separated from the consecutive TWs, i.e., when
conditions (5), (6b), and (8b) are met. Because (5) and (8b) are
identical, we can state that the double-ended method works well
when:
m < 1 − 𝛿𝛿 𝐚𝐚𝐚𝐚𝐚𝐚 m > 𝛿𝛿 (9)
Fig. 20. Bewley diagram showing the five key TWs for an internal The single-ended method at the local terminal works best if
fault.
the first TW is separated from the first reflection from the fault
(6b) and from the reflection from the remote terminal (5).
Additionally, the reflection from the fault must be separated
from the reflection from the remote terminal (7b). We can state
 11

that the single-ended method at the local terminal works well

when:
1
m < 1 − 𝛿𝛿 𝐚𝐚𝐚𝐚𝐚𝐚 m > 𝛿𝛿 𝐚𝐚𝐚𝐚𝐚𝐚 m > (1 + 𝛿𝛿) (10a)
2
Condition (7b) is more restrictive than condition (6b);
therefore, we can simplify (10a) and state that the single-ended
method works well when:
1
m < 1 − 𝛿𝛿 𝐚𝐚𝐚𝐚𝐚𝐚 m > (1 + 𝛿𝛿) (10b)
2
Let us go back to our example of a TWDSW = 10 µs and a
TWLPT = 547 µs (δ = 10/547 = 0.0183). Condition (9) tells us
that the double-ended method is dependable for both fault
locations, m = 0.98 pu and m = 0.505 pu. Condition (10b) tells
us that the single-ended method is dependable for the fault Fig. 22. Dependability contour for the single-ended method –
location of 0.98 pu but not for the location of 0.505 pu. graphical representation of (10b).
B. Dependability Contours When the power line is long (small δ), the vertical line in
We can represent condition (9) for the double-ended method Fig. 21 and Fig. 22 moves to the left and the dependability
and (10b) for the single-ended method as contours, see Fig. 21 interval becomes larger. When the power line is short (large δ),
and Fig. 22. the vertical line in Fig. 21 and Fig. 22 moves to the right and
The shaded area in Fig. 21 is the dependability contour of the dependability interval becomes smaller. Fig. 23 shows the
the double-ended method. The vertical line in the figure percentage dependability of the double-ended and single-ended
represents a specific application (a TW-based fault-locating methods as a function of the parameter δ. Theoretically, the
device with the TWDSW parameter and a power line with the double-ended method does not work at all when δ > 0.5,
propagation time of TWLPT). The intersection points of the i.e., when TWLPT < 2 ∙ TWDSW. The single-ended method
vertical line with the shaded contour define the dependability does not work at all when δ > 0.33, i.e., when TWLPT <
limits for the double-ended method. 3 ∙ TWDSW.
The upper shaded area in Fig. 22 is the dependability
contour of the local single-ended method. The remote single-
ended method has a contour that is a mirror reflection (the
bottom shaded area). Assuming the application uses both the
local and remote fault-locating results, the effective depend-
ability contour includes both shaded areas. The potential
overlap of the TW reflected from the fault and reflected from
the opposite terminal (see Fig. 14) lowers the dependability for
faults near the line midpoint.

Fig. 23. Dependability of the double-ended and single-ended

methods as a function of δ.
You can use Fig. 23 to decide on the minimum line length
that justifies a TW-based fault-locating application. If
80 percent dependability is satisfactory, the corresponding δ is
0.100 for the double-ended method and 0.067 for the single-
ended method. For a device with a TWDSW = 10 µs, the
corresponding minimum line propagation times are therefore
100 µs and 150 µs, respectively. For an overhead line, these
propagation times correspond to the length of 29.4 km
(18.3 mi) and 44.1 km (27.4 mi) for the double-ended and
Fig. 21. Dependability contour for the double-ended method – single-ended methods, respectively. You can also use Fig. 23 to
graphical representation of (9). calculate the expected dependability for a given line length. For
example, for a 100 mi line and a device with a
TWDSW = 10 µs (δ = 0.0183), the double-ended method has a
 12

theoretical dependability of 96 percent, and the single-ended C. Refining Dependability Contours in Special Cases
method has a theoretical dependability of 94 percent. Let us consider a case when a certain fault location is near a
Fig. 24 and Fig. 25 plot the dependability as a function of discontinuity (such as a line tap) or when there is an external
the TWLPT and line length, respectively, for a device with a discontinuity that may reflect TWs and these TWs overlap with
TWDSW = 10 µs [4] [5]. The figures use a semilogarithmic the key TWs in Fig. 20. Ideally, we would like the fault location
scale for better readability. The figures clearly illustrate that the to be away, by at least a TWDSW time interval in terms of the
longer the line, the greater the benefits of the TW technology. TW travel time, from 1) the line discontinuity and 2) an
The dependability curves in Fig. 24 and Fig. 25 are overlapping TW reflected from an external discontinuity. The
conservative estimates. In applications to short lines, the TW- per-unit distance, ∆m, that a TW traveled during the TWDSW
based fault-locating methods do not abruptly lose dependability time is:
but gradually lose accuracy. TWDSW
∆m = =δ (11)
TWLPT
Equation (11) means that a discontinuity will remove a
portion of the dependability area in the shape of a triangle (the
slope between ∆m and δ is 1).
Fig. 26 shows an approximation of the dependability
contour for the double-ended method when a line tap is located
at 0.3 pu. The figure shows that dependability is impacted for
fault locations near the tap. Again, the contour in Fig. 26 is a
conservative estimation. Realistically, the tap will only skew
the time stamp and result in degraded accuracy rather than a
loss of function.

Fig. 24. Dependability of the double-ended and single-ended

methods as a function of the line propagation time.

Fig. 26. Dependability contour for the double-ended method when

a line tap is present at 0.3 pu.
Fig. 27 shows an approximation of the dependability
contour for the single-ended method when there is an external
discontinuity located 0.8 pu from the local terminal. The figure
shows that dependability of the single-ended method at the
remote terminal is impacted for fault locations near the 0.8 pu
location (see Fig. 16 for an explanation of the TW aliasing
issue). Again, the contour in Fig. 27 is a conservative
estimation. Realistically, the reflection from the external
discontinuity is reduced by the termination effect and will only
skew the time stamp and result in degraded accuracy rather than
a loss of function. Also, if the fault-locating results from both
line terminals are retrieved and used, the accuracy is improved.

Fig. 25. Dependability of the double-ended and single-ended

methods as a function of the line length for an overhead line (a) and a
cable line (b).
 13

Fig. 28. Dependability interval and tested accuracy of the double-

ended method applied to a 30 mi line.
Fig. 29 shows the theoretical dependability region and the
accuracy of the single-ended method as tested. The figure
shows that the method works for the entire line length except
faults about 1 mi away from the line midpoint (see Fig. 14 and
Fig. 27. Dependability contour for the single-ended method when Fig. 22). The method is accurate for faults within the
an external discontinuity is present at 0.8 pu from the local terminal.
dependability regions (0.1 mi error) and slightly less accurate
The method for evaluating the TW-based fault-locating outside of the dependability regions (0.15 mi error). The single-
dependability presented in this section allows us to draw the ended method works for faults located at least 2 mi from the
following conclusions: terminal (compare with Fig. 11 showing a field case of a fault
• Discontinuities (line terminals, line taps, and external located 2.1 mi from the terminal). Applying the single-ended
discontinuities if their reflected TWs propagate into method at the remote terminal covers the first 2 mi, and
the line) create “triangular holes” in the dependability applying the single-ended method at the local terminal covers
contour. the last 2 mi. Download and use the fault-locating results from
• The areas of impaired dependability grow in terms of both terminals for dependable single-ended fault locating. The
per-unit fault location as the line gets shorter. accuracy is slightly degraded for faults close to the opposite line
• The presented dependability contours are conservative terminal because of the skew in the time stamp of the TW
estimates; the field performance is typically better reflection from the fault (TWR3 skews the time stamp of TWR2
(reduced accuracy rather than loss of function). in Fig. 9).
• The “minimum line length” question for applicability
of TW-based fault locating is not a yes-or-no question.
Rather, for any given line length, one can expect a
certain guaranteed (minimum) dependability of TW-
based fault locating.
• The longer the line (the smaller the TWDSW/TWLPT
ratio), the higher the dependability, and the lesser the
impact of all discontinuities on the performance of
TW-based fault locating. Fig. 29. Dependability interval and tested accuracy of the single-
ended method applied to a 30 mi line.
D. Test Results for a Sample 30-Mile Line
If the single-ended TW-based method fails and the single-
Let us consider a sample 30 mi overhead line to illustrate the
ended impedance-based method reports the fault near the line
concepts of TW-based fault-locating dependability and
midpoint, assume the fault is indeed close to the line midpoint
accuracy. Devices [4] have been tested by using fault cases
and verify the fault location by analyzing the transient record
generated by using an electromagnetic transient program. This
(see Section VII).
application involves a TWDSW = 10 µs (the DS filter window
used in [4]) and a TWLPT = 164.3 µs (the TW propagation
VII. FAULT ANALYSIS AND OFFLINE FAULT LOCATING
time of the 30 mi line). According to (4), δ = 0.0609 and the
theoretical conservative dependability is 88 percent for the When performing fault analysis and offline fault locating by
double-ended method and 82 percent for the single-ended using ultra-high-resolution records, you have an option to
method. adjust the DS filter window length. This includes shortening the
Fig. 28 shows the theoretical dependability region and the window compared with the default length to allow detecting
accuracy of the double-ended method as tested. The figure and time-stamping TWs that arrive in quick succession and also
shows that the method works for the entire line length, lengthening the window to allow better noise suppression and
including fault locations as close to the terminals as 0.5 mi. The detecting and time-stamping of highly dispersed TWs. Some
method is very accurate for faults within the dependability event analysis software [9] allows modifying the TWDSW
region (0.01 mi error) and accurate outside of the dependability parameter for the offline-calculated TWs. When using [9] to
region (0.05 mi error). analyze transient records from [4] and [5], open the *.HDR
IEEE COMTRADE file with a text editor (see Fig. 30) and edit
 14

the TWDSW parameter before using [9] to open the record. apply a TWDSW = 20 µs to the record at the other
Remember to restore the TWDSW parameter to its default line terminal for better noise rejection and obtain the
value after analysis to avoid confusion when using the record peak time from the DS filter output as 211.5 µs. After
again in the future. correcting for the DS filter group delay, you can
compare the 211.5 – 20 = 191.5 µs time stamp with
the 117.2 µs time stamp.

VIII. REVIEW OF TW-BASED FUNCTIONS IN RELATION TO

LINE LENGTH AND FAULT LOCATION
Fig. 30. Modifying the TWDSW parameter in a text editor before This section reviews the impact of line length and fault
using [9] to analyze records from relays [4] and [5]. location on typical TW-based functions, assuming the DS filter
Follow these best practices with respect to the DS filter with the window half-length of 10 µs [4] [5].
window length when analyzing fault records and performing
A. Single-Ended TW-Based Fault Locating
offline fault locating:
• When using data from both line terminals, remember The following observations apply to the single-ended
to use the same TWDSW parameter for both records. method:
This ensures the same DS filter group delay. • The method is not effective on overhead lines shorter
Otherwise, with different TWDSW parameters, the than about 10 km (6 mi) and cable lines shorter than
TW time stamps at the two line terminals must not be about 6 km (4 mi). The method reaches 80 percent
compared because of the unequal group delay. In dependability for overhead lines longer than 50 km
advanced applications, you can correct for the unequal (30 mi) and cable lines longer than about 30 km
group delay as explained later in this section. (20 mi).
• If you encounter significant high-frequency ringing in • The method may be less effective on very long lines
the signals, consider adjusting the TWDSW to notch and cable lines because of attenuation and dispersion.
the ringing frequency out. Measure the period of the For example, if the fault is located 200 km (120 mi)
oscillatory signal component and select the TWDSW from the terminal, the method responds to a TW that
to be a multiple of the period. For example, to notch traveled 600 km (360 mi) after it was launched and
out a 3 µs ringing, select the TWDSW to be 6, 9, 12, before it arrived at the terminal as the first reflection
or 15 µs. from the fault.
• After you have obtained the first approximation of the • Offline applications of the method are possible on
time separation between two TWs, you can adjust the short lines by shortening the TWDSW when working
TWDSW to optimize noise rejection or sensitivity. with ultra-high-resolution transient records (see
The TWDSW parameter must not be more than half Section VII). Similarly, offline applications on long
the TW separation time. For example, if the two TWs lines can give better results by lengthening the
of interest arrive separated by 40 µs and you want to TWDSW.
reduce the high-frequency noise, you can select a • The method can perform poorly for fault locations that
TWDSW as long as 20 µs (double the default of alias with locations of line terminals, line taps, and
10 µs). However, if the two TWs of interest arrive discontinuities outside of the protected line.
separated by 12 µs, you need to select a TWDSW as • The method has a dependability gap for faults in the
short as 6 µs to see these TWs individually. middle of the line.
• Start with the default TWDSW of 10 µs to gain a • The method can be challenged by reflections from
general understanding of the fault location, the TWs network elements. Always inspect the impedance-
arriving at the terminal, and their polarities and arrival based fault-locating results as a verification of the
times. If needed, zoom in for selected TWs by TW-based fault location and when selecting the TW-
reducing the TWDSW parameter or zoom out by based fault location before dispatching the line crew.
increasing it. A shorter TWDSW reveals finer features • Typically, the local and remote line terminals have
but allows more noise. A longer TWDSW suppresses different operating conditions with respect to the
noise but potentially blends multiple TWs. single-ended method. Retrieve and inspect fault-
• You can obtain time stamps corrected for the DS filter locating results from both line terminals before
group delay by subtracting the TWDSW value from dispatching the line crew.
the time of the DS filter output peak. For example, if B. Double-Ended TW-Based Fault Locating
the DS filter output peak time is 123.2 µs when using
The following observations apply to the double-ended
a TWDSW = 6 µs, you can record the corrected time
method:
stamp as 123.2 – 6 = 117.2 µs. The 117.2 µs time
stamp is independent from the TWDSW parameter • The method may be less effective on overhead lines
and can be compared with time stamps obtained with shorter than about 6 km (4 mi) and cable lines shorter
different TWDSW values. For example, you may than about 4 km (3 mi). The method reaches
 15

80 percent dependability for overhead lines longer

than 30 km (20 mi) and cable lines longer than about
20 km (12 mi). These threshold values are
conservative estimates. The method works for shorter
lines but with slightly degraded accuracy.
• The method can be applied to relatively long overhead
lines and cable lines because it only uses the first
TWs, and the first TWs travel less than the line length.
For example, if the fault is located at 200 km (120 mi)
on a 300 km (200 mi) line, the method works with
TWs that travel 100 km (80 mi) and 200 km (120 mi)
before they arrive at the line terminals.
• The method works well for faults close to terminals
Fig. 31. Aliasing phenomenon explaining false positives in the line
and taps, but it may have slightly degraded accuracy monitoring function.
for such faults.
• Reflections from taps and network discontinuities do D. TW87 Scheme
not affect this method because the first TWs are The TW87 scheme includes several security conditions.
always launched by the fault and are never reflections. These conditions have yielded an excellent security record in
Subsequent TWs, if they arrive in quick succession, the field for relays [4] and [5]. However, these security
can skew the time stamps of the first TWs and erode conditions may impact the TW87 scheme dependability,
some accuracy, but they cannot make the method fail. depending on the line length and fault location:
• The method is reliable and accurate. If the method • The scheme must detect the first TWs at both line
should provide inaccurate results, inspect the settings terminals, and these TWs must be relatively similar in
and verify that the time synchronization between the shape to the ideal TW (step change). This requirement
local and remote devices was accurate when the fault can make the scheme slightly less dependable for
occurred. faults very close to the line terminals (see Fig. 4 and
• Offline applications of the method may be beneficial Fig. 5).
in special cases, especially on long cable lines when • The scheme verifies if the magnitudes of the first TWs
lengthening the TWDSW may allow detecting TWs at both line terminals as well as the sum of these first
with better sensitivity and time-stamping them with TWs are greater than certain minimum levels. This
better accuracy. requirement can make the scheme less dependable in
applications to long cable lines because of attenuation
C. TW-Based Line Monitoring and dispersion, especially for faults away from the
The TW-based line monitoring function leverages the middle section of the line.
double-ended method, and therefore it has the same line length • The scheme assumes an external fault and verifies the
limitations. However, unlike the fault-locating function, the presence of exit TWs to ensure security for external
line monitoring function may be configured to trigger on low- faults. If a discontinuity exists in the system and can
energy events with the intent to detect incipient faults and fault reflect a TW in such a way that it arrives at the
precursors. This high sensitivity can occasionally lead to false expected exit TW time for a specific internal fault
positives – external events detected and tabulated as internal (see Fig. 16), then the scheme may lose dependability
events. The line monitoring function includes the logic to for that internal fault.
prevent false positives. If they occur, however, these false In general, you can use the dependability of the double-
positives are typically tabulated at the line terminals (first and ended method (see Fig. 23 and Fig. 25 as well as the
last bins [6]) because external events often launch TWs that dependability contours) to approximate the TW87 scheme
pass through the monitored line. Occasionally, an external dependability in relation to the line length and fault location.
event can occur at a spot located at similar distances with This approach results in a base dependability of 80 percent or
respect to both terminals of the monitored line. If this event is higher. Other factors, such as excessive ringing in the
a false positive, it is tabulated at the location with the same secondary cables or faults that occurred when the instantaneous
distance difference to the line terminals as the external event voltage was low, may reduce that initial dependability estimate.
(see Fig. 31). If you suspect a false positive, search for spots
located at the same distance difference with respect to the line E. TW32 Directional Element
terminals. For example, if an event is tabulated at 15 km on a In reference to Fig. 18, the TW32 element design in [4] and
60 km line, and the event is a false positive, then the source of [5] requires the line to be long enough so that the TW reflected
the event is located at the spot that is 30 km closer to the local from the opposite terminal does not arrive within the TW32
terminal than to the remote terminal. data window, e.g., TW32WIN = 2 ∙ 50 µs. Implementations [4]
and [5] disable the TW32 logic for lines with a one-way TW
propagation time (a relay setting) shorter than 50 μs. You
 16

should refrain from using the TW32 element on tapped lines The operating conditions may be different for the double-
and hybrid overhead and cable lines if the discontinuity on the ended method and the local and remote single-ended methods
protected line is located at a distance shorter than 50 μs of travel depending on the fault location with respect to the line terminals
time from the line terminal. and discontinuities. For example, if the fault is located close to
the local terminal but the cable line attenuation prevents the
IX. CONCLUSIONS double-ended method from providing the result, the local
This paper reviews the impact of line length and fault single-ended method may work satisfactorily. Or, when a very
location on the dependability and accuracy of practical TW- close-in fault prevents the local single-ended method from
based protection and fault-locating functions. The paper providing the result, the remote single-ended method may work
conclusions and observations apply directly to [4] and [5] and satisfactorily. By retrieving and using all three results (double-
can be extended to [2] as well as any relay or fault locator that ended, local single-ended, and remote single-ended), you can
uses a window-based method for detecting and time-stamping overcome many of the line length constraints.
TWs. Devices [4] and [5] work with DS filter lengths that have
Line length is not the only factor that impacts the been selected for secure protection and accurate fault locating.
dependability and accuracy of TW-based functions. While it is When performing fault analysis and offline fault locating by
true that very short lines do not allow effective applications, one using relay records, you can adjust the DS filter window length
should keep in mind that in most cases, the impact of a short and balance the need for noise rejection and time-stamping
line length or close-in fault location is limited accuracy and resolution. Offline fault locating can be performed in multiple
dependability rather than a total loss of function. Also, other steps by adjusting the DS filter window length based on
factors beyond the line length and fault location play a role, information gained in previous steps. This allows you to resolve
such as line taps and other discontinuities internal or external to challenging and unusual cases that include tapped lines, lines
the protected line. with unusual terminations, and hybrid lines.
The double-ended method in [4] and [5] is highly
dependable for overhead lines longer than about 30 km (20 mi) X. REFERENCES
and it works well for overhead lines as short as 6 km (4 mi). [1] E. O. Schweitzer, III, A. Guzmán, M. V. Mynam, V. Skendzic,
B. Kasztenny, and S. Marx, “Locating Faults by the Traveling Waves
The single-ended method requires longer line lengths and
They Launch,” proceedings of the 40th Annual Western Protective
cannot be expected to perform for overhead lines shorter than Relay Conference, Spokane, WA, October 2013.
about 10 km (6 mi); it reaches 80 percent dependability for [2] SEL-411L Advanced Line Differential Protection, Automation, and
overhead lines 50 km (30 mi) in length and longer. These Control System Instruction Manual. Available: https://selinc.com.
numbers assume low-impedance line terminations, such as [3] E. O. Schweitzer, III, B. Kasztenny, A. Guzmán, V. Skendzic, and M.
when each terminal connects two or more lines in addition to V. Mynam, “Speed of Line Protection – Can We Break Free of Phasor
Limitations?” proceedings of the 41st Annual Western Protective
the line of interest. Relay Conference, Spokane, WA, October 2014.
The TW87 protection scheme and the line monitoring [4] SEL-T400L Time-Domain Line Protection Instruction Manual.
function incorporate the double-ended method in their logic. Available: https://selinc.com.
You can expect similar dependability and applicability limits [5] SEL-T401L Ultra-High-Speed Line Relay Instruction Manual.
for these functions as for the double-ended method. Available: https://selinc.com.
The TW32 directional element in [4] and [5] cannot be [6] B. Kasztenny, M. V. Mynam, T. Joshi, and D. Holmbo, “Preventing Line
applied on lines with a propagation time shorter than about Faults With Continuous Monitoring Based on Current Traveling
Waves,” proceedings of the 15th International Conference on
50 µs (about a 15 km (9 mi) overhead line) to ensure reflections Developments in Power System Protection, Liverpool, UK, March 2020.
from the opposite terminal do not arrive within the TW32 data [7] B. Kasztenny, A. Guzmán, N. Fischer, M. V. Mynam, and D. Taylor,
window. “Practical Setting Considerations for Protective Relays That Use
The paper proposes a methodology to calculate conservative Incremental Quantities and Traveling Waves,” proceedings of the 43rd
Annual Western Protective Relay Conference, Spokane, WA, October
dependability estimates for the single- and double-ended
2016.
methods. Use these dependability contours to estimate depend- [8] A. Sivesind, F. J. Sanchez, S. Cooper, F. Elhaj, “Traveling Wave Relay
ability for a particular line given the DS filter window length. Application, Commissioning, and Initial Experience,” presented at the
Faults located very close to line terminals and other 72nd Annual Conference for Protective Relay Engineers, College
discontinuities, such as taps or overhead-to-cable transition Station, TX, March 2019. Available: https://prorelay.tamu.edu/archive/.
points, challenge the single-ended method but not the double- [9] SYNCHROWAVE Event Software Instruction Manual. Available:
https://selinc.com.
ended method. At worst, the double-ended method may display
a slightly reduced accuracy for such faults.
XI. BIOGRAPHIES
Cable lines attenuate and disperse TWs much more than
Bogdan Kasztenny has over 30 years of experience in power system protection
overhead lines. TW-based functions are challenged when used and control. In his decade-long academic career (1989–99), Dr. Kasztenny
for cable lines that are too long to allow TWs to arrive at the taught power system and digital signal processing courses at several
line terminals with magnitudes great enough and wavefronts universities and conducted applied research for several relay manufacturers. In
sharp enough for reliable detection and time-stamping. The 1999, Bogdan left academia for relay manufacturers where he has since
designed, applied, and supported protection, control, and fault-locating
cable line length limitation depends on the voltage level. The products with their global installed base counted in thousands of installations.
greater the voltage, the longer the cable. Bogdan is an IEEE Fellow, a Senior Fulbright Fellow, a Distinguished CIGRE
 17

Member, and a registered professional engineer in the province of Ontario.

Bogdan has served as a Canadian representative of the CIGRE Study
Committee B5 (2013–2020) and on the Western Protective Relay Conference
Program Committee (2011–2020). In 2019, Bogdan received the IEEE Canada
P. D. Ziogas Electric Power Award. Bogdan earned both the Ph.D. (1992) and
D.Sc. (Dr. habil., 2019) degrees, has authored over 220 technical papers, and
holds over 50 U.S. patents.
Mangapathirao (Venkat) Mynam received his MSEE from the University of
Idaho in 2003 and his BE in electrical and electronics engineering from Andhra
University College of Engineering, India, in 2000. He joined Schweitzer
Engineering Laboratories, Inc. (SEL) in 2003 as an associate protection
engineer in the engineering services division. He is presently working as an
engineering director in SEL research and development. He was selected to
participate in the U. S. National Academy of Engineering (NAE) 15th Annual
U. S. Frontiers of Engineering Symposium. He is a senior member of IEEE and
holds patents in the areas of power system protection, control, and fault
location.

© 2021 by Schweitzer Engineering Laboratories, Inc.

All rights reserved.
20210519 • TP7007-01