You are on page 1of 10

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
1

Ground Fault Detection and Location for


Ungrounded DC Traction Power Systems
Jae-Do Park, Senior Member, IEEE

Abstract—A fault protection and location method for un-


grounded DC traction power systems is presented in this paper.
Many DC traction power systems have ungrounded power circuit
to increase the leakage path resistance. Although ungrounded sys-
tems can continue operating with a single ground contact unlike
solid- or low-resistance grounded systems because of the very low
fault current, a second ground fault in another pole will result in
a line-to-line fault that could cause a significant system damage.
However, it is difficult to detect the first ground contact due to
the low ground current and even harder to locate because it can
be seen by many detectors in the network. The proposed scheme
in this paper uses a probe unit to detect and locate the first single
ground fault in ungrounded traction power systems. The probe
unit applies probe voltage to detect the fault, and, once the fault
is detected, analyzes the response to DC or swept-frequency AC
probe voltage to locate the fault. The proposed concepts have been
verified with computer simulations and hardware experiments, Fig. 1. Schematic of an ungrounded double-track DC traction system.
and demonstrated a successful performance. A line-to-ground fault at the running rail is shown.

Index Terms—Fault diagnosis, Fault location, Grounding, Pro-


tection, DC power system.
i.e., floating, return rails significantly reduce ground leakage
current, as well as the corrosion level, compared to grounded
I. I NTRODUCTION systems [14]. It has been shown that the cumulative leakage
High-capacity public transportation systems such as rapid current in the solidly grounded system is approximately 30
transit and light-rail systems can be found in many urban times larger than the one in the ungrounded system at accel-
areas. Mostly DC or single-phase AC power systems are used. eration and deceleration regions [15]. However, ungrounded
For DC power systems, the standard voltages are 600/750V systems suffer from the possible high rail voltage that can
and 1.5kV for urban and regional systems, respectively [1]. be developed between platforms and earth ground when there
Although capacity and size are different by systems, typically is a rail-to-ground connection or a fault, which decreases the
they are running on steel rails. And the running rails are personal and equipment safety level [16], [17]. The grounding
usually used for return current path, while the positive pole is scheme can be actively altered between floating and solid
connected to the car through overhead contact system (OCS) ground using Over-Voltage Protection Device (OVPD) for
or the third conductor rail [2]. improved stray current reduction performance by monitoring
Power system faults can happen by occasional events such rail voltage [18].
as a broken OCS contact falling on metallic objects [3] caus- A diode-grounding can be a compromise between these
ing line-to-line and line-to-ground contact. Possible abnormal two schemes; the rail is solidly grounded when there is a
symptoms from a fault include high rate of rise (di/dt) ground current flowing. However, small voltage difference
of current, over-current, under- or over-voltage, excessive across the diode can happen even during normal operating
touch voltage and these could cause system outage, service conditions and the diode will conduct allowing higher stray
interruption, single-end feeding, hazard such as arc flash fire, current [13]. It has been reported that the diode-grounded
equipment damage, danger to passengers and workers [2], [4]– systems have more problems from stray currents and rail
[9]. It is well accepted that the line-to-ground fault is the most potential than ungrounded systems [19]. The ungrounded
common type of fault [8], [10]–[12]. systems can perform better by using an overvoltage protection
Typical grounding schemes for DC rail systems that use device and enhanced platform insulation, so that the risk of a
running rails as the return path are impedance-grounding, hazardous electric potential can be reduced [15], [20]. Modern
diode-grounding, and ungrounded [2], [13]. The DC power DC traction systems are not grounded at traction substations
grounding system needs to minimize DC stray current while (TSSs) [21].
maximizing safety, but these two goals are contradictory [13], A schematic of an ungrounded DC traction system and a
[14]. Solid grounding would be the safest because the unsafe negative-rail-to-ground fault is shown in Fig. 1. One of the
voltage between rail and ground can be suppressed, but it is advantages of the ungrounded system is being able to operate
not used in modern systems because of the corrosion caused normally even with a single phase-to-ground fault; however,
by ground current [13]. On the other hand, the ungrounded, it is critical to detect and correct the problem before a second

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
2

ground fault occurs on another pole because it will cause a


line-to-line fault [4], [10]. Although detecting the first ground
fault is readily possible using devices such as indicator lights,
voltmeters, and voltage-sensitive relays to detect the voltage
between rail and ground [10], the low fault current level makes
it a non-trivial task. Moreover, traditional detectors can only
indicate the existence of the ground fault; they cannot locate
the fault because it will be detected by any detectors in the
network [22], [23].
In this paper, a fault protection and location method for
ungrounded DC traction power systems is presented. The
proposed scheme uses a probe unit to detect and locate the
first ground fault while the traction system is in operation. The
probe unit applies probe voltage to detect the fault, and, once
the fault is detected, DC or swept-frequency AC probe current Fig. 2. A schematic of proposed protection system. Probe units at
analysis is used to locate the fault depending on the traction TSS are shown, which operates only one at a time. ZLP and ZLN
denote the impedance of positive and negative rail, respectively. A
power circuit’s operating condition. The proposed concept
single ground fault is also shown.
has been verified with computer simulations and hardware
experiments, and demonstrated a successful performance.
detection is not trivial because it is difficult to distinguish the
II. E XISTING T ECHNOLOGIES fault current and legitimate large load current, such as starting
current, especially if the fault happens far enough from the
A. Detection and Location
protective relay [9], [24]. Although it would be much safer
The detection of ground fault is relatively straightforward; if the first ground contact can be detected and located, such
it can be detected by devices such as overvoltage relay mon- techniques for preventive measure have not been investigated
itoring ground resistor voltage. However, consistent detection extensively.
and location of ground fault have been historically problematic
issues in DC power systems in general [5], [10], [24], [25].
B. Rail Impedance
Classical methods of ground fault localization include fault
isolation by network switching and circuit tracing using a The rail impedance has been investigated for traction system
signal injector and a hand-held detector. The circuit switching design, fault assessment, protection equipment selection, and
technique requires system shutdown, so it is difficult to use signaling subsystem design [31], [34]. A simplified traction
until there is a scheduled shutdown and loses the advantage of circuit model based on first-order linear RL equivalent circuit
ungrounded power system. Even if a faulted line is identified, and constant internal rail impedance have been used in most
the exact location would be searched by maintenance crew, reported rail traction simulation studies [34]. The equivalent
which is manpower intensive and time-consuming [10], [26]. circuit parameters consist of resistance and inductance of
Hand-held signal injector and receiver system, such as thumper feeder and return path. Typically, OCS or third rail and running
circuit, are used to search the ground fault [10]. rails are used for feeder and return path, respectively.
Common-mode-current-based techniques have been sug- It has been shown that the internal rail impedance is quite
gested to detect the faulted segment in multi-terminal DC constant for a given current, but it varies substantially as
systems such as ship board power systems [27], [28]. Although frequency changes due to the skin effect; resistance and
they can detect the AC- and DC-side ground fault, it is difficult inductance increases and decreases with increasing frequency,
to determine the detailed location of the fault. Advanced signal respectively [31], [35], [36]. The impedance of running rails
processing techniques such as traveling wave analysis [29] and has different characteristics than that of feeders because of its
noise pattern analysis [26] have also been suggested for ground material and non-circular shape [21]. However, rail impedance
fault location. However, the traveling wave methods is difficult has been investigated with various frequency and load current
to apply to traction power systems that have multiple TSSs values [31], [34]–[37], and the result showed a significant
connected to the bus in parallel. And analysis of noise pattern accordance in their characteristics even though materials and
from power electronics converters may not be appropriate for configurations were different [38]. Furthermore, it also has
traction power systems because 12-pulse passive rectifiers are been shown that the impedance/frequency characteristic for a
commonly used in TSSs [30]. particular track layout can be obtained [31].
Techniques to calculate the current/time profile for a remote
short circuit fault at a DC TSS using rail impedance was also III. P ROPOSED A LGORITHM
suggested [7], [31]. However, these techniques, as well as sig- A probe unit is proposed in this paper to detect and locate
nal injection, rate-of-rise detection [24], wavelet- and neural- the first single ground fault in ungrounded systems before a
network-based techniques [32], [33], are detecting line-to-line second ground fault happens on another pole, which could
short circuit fault in DC transit systems. Line-to-line fault cause a destructive line-to-line fault. It consists of a power
could cause a serious damage and service interruptions, but its supply that can generate voltages with arbitrary amplitude and

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
3

probe unit #2 and #3 as shown in Fig. 2. Once the faulted


section is detected, the faulted line can be easily identified
because the probe current that flows through the faulted line
will be much larger than other intact lines due to its smaller
impedance.
The probing frequency and duration will be design parame-
ters that need to be determined according to the system char-
acteristics, such as rail impedance, number of TSSs, and train
operating conditions. Furthermore, the detection sequence can
be coordinated with existing devices in TSSs. For example, a
line can be probed when the OVPD opens up to make the line
ungrounded, or when there is no train in a segment between
TSSs in order not to interfere the operation of OVPD.
Fig. 3. A schematic of probe unit using full-bridge inverter that can
generate voltages with arbitrary amplitude and frequency. The probe
unit also contains probe capacitor and current sensor. It is installed B. Fault Location
at TSSs the circuit is open when it is not injecting probing current,
and can have high output impedance when it is in probing mode. For a single ground connection at distance lF from the
TSS#2 as shown in Fig. 2, the normalized distance d to the
fault location can be defined with the line length between
frequency, and a probe capacitor to form an RLC circuit with adjacent TSSs lo .
the single ground contact, and it is installed at each TSS. lF
Although the probe unit is installed between the line and d= (1)
l0
ground, it does not turn a ungrounded system into a grounded
one because the probe unit is implemented with a switch- The overall fault circuit impedance can be derived as follows.
mode power supply; the circuit is open when it is not injecting A line impedance is defined as
probing current, and can have high output impedance when it
is in probing mode. The proposed system also contains current Zx = Rx + jωLx (2)
sensors to measure the probe current returning through ground.
The novelty of the proposed approach is the active probing where, x denotes the each line: up positive, down positive,
to detect and locate the first ground contact of the ungrounded up negative, and down negative. They are independently
power systems. This can be done while the system is in numbered from 1 to 4. Balanced current distribution between
operation unlike the existing techniques. Although the current parallel running rails is assumed. According to the position
state-of-the-art can detect the existence of the first ground of traction load, the fault can be located using the following
fault, locating the fault has been a non-trivial manpower- methods.
intensive task. The proposed method can automatically detect
and locate the fault with a simple probe unit without any 1) No Traction Load: Fig. 4(a) shows when there is no
human intervention, so substantial improvement on system traction load in and outside of faulted section. A DC probe
reliability and Mean Time Between Failure (MTBF) can voltage VP is superposed on the TSS power circuit and the
be expected. Fig. 2 shows the schematic of the proposed probe current flows only through the faulted side because the
protection scheme and a single ground fault. An electrical TSS on right is reverse biased. The probe unit can also be
schematic and DC/AC configurations of a probe unit are also connected to the positive pole to probe the positive lines.
shown in Figs. 3 and 4, respectively. The equivalent impedance becomes resistive because of the
DC excitation.

A. Fault Detection Zeq = {R3 + R4 − dR4 }//dR4 (3)


To detect the first ground connection, a probe unit applies
Then the fault location d can be found as follows using the
a probe voltage to the ungrounded power lines. The probe
measured currents from the probe unit and the faulted line.
current will not flow unless there is a ground connection
because the probe power supply is grounded. If some ground (R3 + R4 )(iP − iP 4 )
currents are detected, it could be either one of the four lines d= (4)
R4 i P
connected to the TSS assuming a typical double-track system
shown in Fig. 1: positive and negative pole of up and down If R3 = R4 , it can be further simplified as
lines. Only one probe unit probes the lines at a time and 2(iP − iP 4 )
it is followed by the one in the adjacent TSS. This probing d= (5)
iP
sequence can be readily implemented with the communication
between TSSs. During the sequence, the faulted segment can Because only the probe currents are used, the fault distance
be detected by checking the direction of the probing current. It can be calculated regardless of the fault resistance.
changes in TSSs at each side of the ground fault, for example,

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
4

then be given as
Zeq2 = Zeq1 + (1 − d)Z4 (8)
= (C − dR4 ) + j (D − dX4 ) (9)
and the total line impedance including faulted segment be-
comes
Zeq3 = Zeq2 //dZ4 (10)
2 2
= (dJ + d K) + j(dN − d O) (11)
Detailed coefficients can be seen in (12).
The overall impedance of the fault circuit will therefore be
given as
1
Zeq = Zeq3 − j + RF (13)
ωCP
1
= Req + j(ωLeq − ) (14)
ωCP
where,
Req = dJ + d2 K + RF (15)
2
dN − d O
Leq = (16)
ω
The fault circuit forms a second-order RLC circuit with Req ,
Leq , and CP as shown in Fig. 4(b). The stray capacitance
of the line is neglected because of the much larger probe
capacitor CP . The dynamics of the probe current iP for the
probe voltage VP can be obtained as
d2 iP (t) Req diP (t) 1 dVP (t)
Fig. 4. Fault circuit and probe unit. (a) when there is no traction load + + iP (t) = (17)
dt 2 Leq dt Leq CP dt
in and outside of faulted section. The TSS on right is reverse biased,
hence the probe current cannot flow through it. The probe unit is in Suppose the probe voltage VP (t) as
DC probing mode. (b) when TSSs supply traction power to a load.
If a TSS conducts, it is seen as a short circuit by the probe power VP (t) = M sin(ωt) (18)
supply because fault probe circuit is superposed on the TSS power
circuit. The probe unit is in AC probing mode. then the forcing function in (17) becomes
dVP (t)
= ωM cos(ωt) (19)
dt
2) TSSs Supply Traction Load: When both TSSs supply the ωM jωt ωM −jωt
traction load in inside or outside of the fault section as shown = e + e (20)
2 2
in Fig. 4(b), the probe current can flow through all four lines The steady-state solution for the probe current can be given
because the TSS rectifiers conduct for the traction current. as
However, it would be difficult to separate DC probe current
from traction current because the traction current changes as iP (t) = i1 (t) + i2 (t) (21)
the load moves. Hence, a swept-frequency AC voltage is used ωM ωM
= T1 ejωt + T2 e−jωt (22)
for fault location when the traction system is loaded instead 2 2
of the DC excitation. The probe current does not flow through where, T1 and T2 are complex numbers. Solving (17) for i1 (t)
the traction load, as shown in Appendix. yields
The total impedance of intact lines, line 1 to 3 in Fig. 4(b), ωM

Req 1

ωM jωt
2
can be calculated as (jω) + (jω) + T1 ejωt = e
2 Leq Leq CP 2
(23)
Zeq1 = Z1 //Z2 //Z3 (6) The coefficient T1 can be given as
= A + jB (7) 1
T1 = 1 Req
(24)
2
(−ω + Leq CP ) + j Leq ω
Because of the impedance of each line is a known parameter, Similarly,
the total impedance of intact lines Zeq1 is given as a complex
1
constant with coefficients A and B. T2 = 1 R
= T∗1 (25)
(−ω 2 + eq
) − j Leq ω
The impedance in parallel with the faulted segment lF will Leq CP

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
5

Hence, the overall solution for the probe current will be


ωM
T1 ejωt + T∗1 e−jωt

iP (t) = (26)
2
= ωM |T1 | cos(ωt + φ) (27)
where, φ = 6 T1 .
From (14), (24), and (27), it can be easily seen that the
magnitude of probe current is maximized at the resonance
condition. s
1 Fig. 6. Simulation: Probe current amplitude in power lines. A single
ω = ω0 = (28) ground connection is simulated in line 4. The current in the faulted
Leq CP
line has substantially larger amplitude compared to the one in intact
Once the faulted section is identified using the proposed de- line.
tection technique described in Section III-A, the probe power
supply will inject the swept-frequency signal to determine the
resonant frequency from the sampled probe current using Fast
Fourier Transform (FFT). With the identified resonant fre-
quency ω0 , the distance to the fault location can be estimated.
With the equivalent inductance expression (16), (28) becomes
1
Od2 − N d + =0 (29)
ω0 C P
and the distance d can be calculated as
( r ) Fig. 7. Simulation: Fault location error using DC probe current when
1 4O there is no traction load. The range of fault shown was 10-90% of
d= N ± N2 − (30)
2O ω0 CP the distance between TSSs.

It can be shown that (30) always have real solutions as


follows. The fault distant dLmax that has the maximum known parameter, the probe capacitance can be selected with
equivalent inductance Lmax can be found by differentiating a reference fault location and its resonance frequency.
the quadratic expression of the equivalent line inductance (16). 1
CP = (35)
N L̂eq ω̂02
dLmax = (31)
2O
where, L̂eq is a reference line equivalent inductance, e.g., 50%
The resonance frequency is minimized when the fault is fault distance, for a certain frequency ω̂0 . The frequency range
located at dLmax and it can be expressed as can then be determined using (33) and probing range, e.g.,
r
1 from d = 0.1 to 0.9. In case the line inductances are largely
ω0min = (32) different, the frequency range can be adjusted for each line.
Lmax CP
4O
= 2 (33)
N CP C. Computer Simulation
The discriminant of (30) will be minimum with ω0min , The proposed scheme has been verified using MATLAB and
PSCAD. A 6-km double-track system with four 12-pulse diode
4O
∆min = N 2 − =0 (34) rectifier TSSs, 1kV DC bus, copper cylindrical conductor for
ω0min CP overhead power lines and steel return rails has been simulated
hence, the discriminant is always greater than or equal to zero [1], [31], [34]. It was assumed that a single ground contact
and (30) will always have real solutions. It should be noted was detected by the detection method described in III-A. The
that all roots of (30) can be valid, i.e., less than 1, depending simulation parameters are shown in Table I.
on the line parameters and fault location and either one of A single ground connection has been simulated in line 4 for
them may not be easily ruled out. However, it can still reduce Fig. 4(a) case using MATLAB and the DC probe currents in
the location effort substantially. lines 3 and 4 in terms of fault distance are shown in Fig. 6. It
The probe capacitance and frequency sweep range needs to can be seen that the current in the faulted line has substantially
be determined so that the resonance frequency at all possible larger amplitude compared to the one in intact line, unless the
fault location can be detected. Because the line inductance is a fault location is at the far end of the line from the probe unit.

 
d(CE + DG) + d2 (CF − 2DH) + j d(CG − DE) − d2 (2CH + DF )
Zeq3 = (12)
C 2 + D2

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
6

Fig. 5. PSCAD simulation model. A double-track system with four TSSs, 12-pulse diode rectifiers, 2 km line length, copper cylindrical
conductor for overhead power lines and steel return rails has been implemented.

TABLE I
S IMULATION PARAMETERS .

Traction power Overhead power line Running rail Probe power unit
DC bus Vbus 1000 V Unit resistance Ru1 227 mΩ/km Unit resistance Ru2 200 mΩ/km Probe voltage VP 10 V
Line segment length l0 2 km Unit inductance Lu1 0.635 mH/km Unit inductance Lu2 0.4 mH/km Capacitor CP 3000 µF
Load resistance RL 0.01 Ω Fault resistance RF 0.01 Ω

Fig. 8. Simulation: Probe voltage and current for the case when the Fig. 9. Simulation: TSS supply currents and traction load current for
ground fault on a running rail at 50% distance and the traction load the case of 50% fault and 48% load distance from the probe unit in
is at 48% distance from the probe unit in TSS#2. TSS#2. It can be seen that the traction load current does not contain
the oscillating probe current.

The fault distance can be readily calculated using (4) and the
estimation result is shown in Fig. 7. was detected between TSS#2 and #3. The probe current flows
A PSCAD model was used to simulate Fig. 4(b) case that though the ground fault path and the magnitude changes as
rectifiers are conducting. The model is shown in Fig. 5. A excitation frequency increases. A ground fault at 50% distance
swept-frequency 10V-peak AC probe voltage was applied from from the probe unit was simulated and the current peak can be
the probe unit in TSS#2 when a low-resistance ground fault clearly seen. The probe voltage and current are shown in Fig.

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
7

Fig. 10. Simulation: FFT analysis on probe current for the case of
50% fault and 48% load distance from the probe unit in TSS#2. The
probe current peak happens at 179.58 Hz.
Fig. 12. Experiment setup. A double-track system with two diode
rectifier TSSs is simulated with RL circuit. A probe unit is imple-
mented with 3000µF capacitor bank, a function generator and an
audio amplifier. A single ground contact and traction load between
TSSs are also implemented.

TABLE II
FAULT LOCATION USING DC PROBE CURRENTS WHEN THERE IS NO
TRACTION LOAD . VARIOUS AMPLITUDES OF IP x WERE TESTED . IP x , IP ,
dest , AND derr DENOTE THE PROBE CURRENT MEASURED FROM THE LINE
AND GROUND , ESTIMATED FAULT LOCATION , AND ESTIMATION ERROR ,
RESPECTIVELY.

Fault location d = 0.33


IP 3 0.41 A 0.65 A 0.97 A 1.18 A 1.35 A
IP 4 2.04 A 3.25 A 4.90 A 5.72 A 6.54 A
IP 2.45 A 3.90 A 5.87 A 6.90 A 7.89 A
Fig. 11. Simulation: Fault location error in terms of ground fault and dest 0.3347 0.3333 0.3305 0.3420 0.3422
traction load location from the probe unit in TSS#2. The range of derr 0.41% 0% -0.85% 2.61% 2.66%
fault and load location tested was 10-90% of the distance between Fault location d = 0.67
TSSs.
IP 3 0.66 A 1.17 A 1.61 A 2.17 A 2.69 A
IP 4 1.29 A 2.30 A 3.12 A 4.22 A 5.15 A
8. A traction load at 48% distance from the probe unit was IP 1.95 A 3.47 A 4.73 A 6.39 A 7.84 A
also simulated to superpose the probe current onto the traction dest 0.6769 0.6744 0.6808 0.6792 0.6862
load current. It can be seen in Fig. 9 that the load current IL derr 1.54% 1.15% 2.11% 1.88% 2.93%
does not contain the oscillating probe current.
An FFT analysis was performed on sampled probe current.
For the 50% and 48% fault and load distance case, the current were used for TSSs, which are supplied by a step-down
peak occurred at 179.58Hz. Using the line parameters, (30) transformer to generate 13.5V DC bus. The probe unit consists
calculated the distance to the fault as 49.66%. The second of three 1000 µF bipolar capacitors, an audio amplifier, and a
root was 95.95% in this case. function generator. The experiment setup is shown in Fig. 12.
The performance of fault location was tested with the wide Ground fault detection when there is no traction load
range of fault and load location in 10% to 90% distance from between TSSs using DC probe current is tested. A ground
the probe unit. The result showed some error due to the DC fault at d = 0.33 and 0.67 is implemented and DC probe
ripple voltage as can be seen in Fig. 11, but it is fairly small. current is measured at the probe unit and each line. In this
In this simulation, the maximum error was 4.62% when d = experiment that a fault is implemented in the line 4, the probe
80%. It can be also seen that the location of the traction load current cannot flow through line 1 and 2 because the reverse
does not affect the fault location performance. biased TSS#2. The faulted line can be easily found because its
current IP 4 is significantly bigger than the intact line’s IP 3 .
The fault location can be estimated using (5), and the result is
IV. E XPERIMENTAL VALIDATION
shown in Table II, where dest and derr is the estimated fault
To verify the proposed ground fault detection scheme, a location and estimation error, respectively.
hardware testbed simulating a double-track system with two Also tested was the location of a ground fault when there is
diode-rectifier TSSs has been implemented. For total 3 km of a traction load between TSSs using swept-frequency voltage.
OCS line and running rail segments, three 470 µH and 670 µH The traction load was simulated with a 75 Ω resistor; A high
inductors (1468420C and 1447440C, Murata Power Solutions, resistance is chosen to easily see the effect of probe current on
Inc.) are used, respectively. The resistances in the traction traction load current. Ground faults located at d = 0.33 and
circuit were represented by the internal resistance of inductors 0.67 were tested with two load positions at dL = 0.33 and
(1 Ω and 0.75 Ω). Two 6-pulse three-phase diode rectifiers 0.67. The probe current in the faulted line was measured with a

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
8

Fig. 15. Experiment: Ground fault detection using swept-frequency


Fig. 13. Experiment: Function generator output voltage (top) and voltage when there is a traction load between TSSs. A ground fault
amplifier output voltage (bottom) when fault distance d = 0.33. It at d = 0.67 with traction load at dL = 0.33 was tested. FFT analysis
can be seen that the amplifier output voltage drops due to decreased showed that the frequency of peak current is 123.47 Hz.
impedance around resonance frequency.
TABLE III
FAULT LOCATION USING AC PROBE CURRENTS WHEN THERE IS A
TRACTION LOAD . Fpk , dest , AND derr DENOTE THE PROBE CURRENT
FREQUENCY PEAK , ESTIMATED FAULT LOCATION , AND ESTIMATION
ERROR , RESPECTIVELY.

Fault location d 0.33 0.67


Load location dL 0.33 0.67 0.33 0.67
Fpk 139.28 Hz 141.26 Hz 123.47 Hz 124.78 Hz
dest 0.3960 0.3791 0.7220 0.6275
derr 18.81% 13.74% 8.29% -5.88%

two tests d = 0.33 and dL = 0.33, and d = 0.67 and dL =


0.67. It can be shown that the traction load position did not
make a significant difference. The result of swept-frequency
experiments are shown in Table III. The estimation error was
Fig. 14. Experiment: Ground fault detection using swept-frequency higher compared to the DC probing, but it can be expected that
voltage when there is a traction load between TSSs. A ground fault precise circuit parameters and higher Q-factor can improve the
at d = 0.33 with traction load at dL = 0.67 was tested. FFT analysis
showed that the frequency of peak current is 141.26 Hz. accuracy. The inductors used in the experiments have ±10%
tolerance and the peak of the frequency response was not sharp
because of the amplifier power capacity. It can also be seen
current probe and digital oscilloscope’s (Tektronix DPO2014) that the effect of probe current on traction voltage and current
built-in FFT function was used for FFT analysis. is negligible as shown in Fig. 16.
The function generator sends a voltage command with
sweeping frequency from 0.1 to 275 Hz for 10 s and the am- V. C ONCLUSION
plifier generates output voltage about 2V with given frequency. A ground fault detection and location scheme for un-
The voltages are shown in Fig. 13, and it can be seen that the grounded DC rail power systems has been presented in
output voltage drops around resonance frequency due to the this paper. Although the ungrounded systems can continue
lack of power. Although it made the resonance not as sharp operating with a single ground contact, it is imperative to
as it would be, the resonant point of the fault circuit can be detect and locate the fault quickly because the second ground
readily identified. fault could cause a destructive line-to-line fault. The proposed
The probe current showed the resonant frequency of 141.26 protection scheme is capable of detecting the first ground
Hz and 123.47 Hz for the cases of fault position d = 0.33 fault by applying probe voltage through the ground connection
and load position dL = 0.67, and d = 0.67 and dL = 0.33, from a substation. Once a ground fault is identified, the probe
respectively. The probe currents and FFT results are shown in unit locates the fault by DC or swept-frequency AC response
Figs. 14 and 15. Eq. (30) estimated the fault location at dest = analysis on probe current. The proposed algorithm can be
0.3791 and 0.7220, which are reasonably close. As expected executed while the traction system is in operation without
in III-B2, the similar estimation result was obtained from other de-energizing the bus. The proposed method enables fast

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
9

Fig. 17. Mesh analysis for traction load path. dL denotes ratio of
traction load position and distance between TSSs and d0L is 1 − dL .

Fig. 16. Experiment: DC bus voltage and traction load current when
where,
probe current is applied for a ground fault at d = 0.33. It can be
seen that the effect of probe current on traction voltage and current V=[ V 0 0 0 V ]T (42)
is negligible. T
I = [ i1 i2 i3 i4 i5 ] (43)

detection and accurate location of the first ground contact, so The current vector can be obtained using the inverse
that proper preventive measure can be taken in a timely manner impedance matrix.
to improve the system reliability. It should be noted that if I = M−1
Z V (45)
multiple ground contacts exist when the lines are probed, the
estimated location would not be accurate although ground fault The individual mesh currents are given as
can still be detected. However, the probe unit can probe the V 2dL V
line in high frequency and the possibility for multiple ground i1 = − (46)
Z2 − 2dL Z2 Z2 − 2dL Z2
contacts to happen at the exactly same time is low. Therefore
i2 = 0 (47)
the first ground fault can be detected and located in most
cases. Successful performance of fault detection and location i3 = 0 (48)
have been shown using computer simulations and hardware V V (Z2 + Z3 − 2dL Z2 )
i4 = − (49)
experiments. Z2 − 2dL Z2 Z2 Z3 − 2dL Z2 Z3
V (Z2 Z3 + Z2 Z4 − 2dL Z2 Z3 − 2dL Z2 Z4 )
i5 =
Z4 (Z2 Z3 − 2dL Z2 Z3 )
A PPENDIX
V
M ESH A NALYSIS FOR T RACTION L OAD − (50)
Z2 − 2dL Z2
The probe current does not flow through the traction load The current in the traction load path is i2 − i3 , which is zero.
outside of the faulted section because the TSS is seen as a
short-circuit by the probe voltage. It can be shown using mesh
analysis technique that the probe current does not flow through R EFERENCES
the traction load inside of the faulted section either. The intact [1] R. Hill, “Electric railway traction: Part 3 traction power supplies,” Power
portion of the traction circuit, a traction load, and a single Engineering Journal, pp. 275–286, December 1994.
ground contact can be seen in Fig. 17 and the mesh currents [2] R. White, “DC electrification supply system design,” in 2007 3rd IET
Professional Development Course on Railway Electrification Infrastruc-
are shown as well. ture and Systems, 2007, pp. 35–62.
The mesh equations are given as follows for mesh 1 through [3] K. Pham, R. S. Thomas, and W. Stinger Jr, “Operational and safety
5, respectively, considerations for light rail DC traction electrification system design,”
in Transportation Research Circular E-C058: 9th National Light Rail
Transit Converence, 2003, pp. 650–668.
V = dL Z3 (i1 − i2 ) + d0L Z3 (i1 − i2 ) (36) [4] M. Valdes, T. Papallo, and B. Premerlani, “Finding fault - locating
a ground fault in low-voltage, high-resistance grounded systems via
0 = dL Z3 (i2 − i1 ) + dL Z2 (i2 − i4 ) + ZL (i2 − i3 ) (37) the single-processor concept for circuit protection,” IEEE Industry
0 = d0L Z3 (i3 − i1 ) + d0L Z2 (i3 − i4 ) + ZL (i3 − i2 ) (38) Applications Magazine, vol. 13, no. 5, pp. 24–30, 2007.
[5] X. Song, J. He, C. Dong, X. Yang, B. Kirby, and D. Writer, “Accurate
0 = dL Z2 (i4 − i2 ) + d0L Z2 (i4 − i3 ) + Z1 (i4 + i5 ) (39) track modelling with skin-effect for protection improvement in DC
railway system based on RTDS,” in 22nd International Conference and
V = Z1 (i4 + i5 ) (40) Exhibition on Electricity Distribution (CIRED 2013), June 2013, pp.
1–4.
where d0L denotes 1 − dL . It can be represented in a vector [6] M. Thong, A. Cheong, and H. Wijaya, “Overview of DC traction protec-
form tion scheme for Singapore rapid transit system,” in Power Engineering
Conference, 2005. IPEC 2005. The 7th International, Nov 2005, pp.
V = MZ I (41) 1–587.

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TVT.2015.2388785, IEEE Transactions on Vehicular Technology
10

dL Z3 + d0L Z3 −d0L Z3
 
−dL Z3 0 0
 −dL Z3 dL Z3 + dL Z2 + ZL −ZL −dL Z2 0 
d0L Z3 + d0L Z2 + ZL −d0L Z2
 
MZ =  −dL Z3 −ZL 0  (44)
−d0L Z2
 
 0 −dL Z2 Z2 + Z1 Z1 
0 0 0 Z1 Z1

[7] J. C. Brown, J. Allan, and B. Mellitt, “Calculation of remote short circuit Transactions on Industry Applications, vol. 47, no. 2, pp. 996–1002,
fault currents for DC railways,” IEE Proceedings B Electric Power 2011.
Applications, vol. 139, no. 4, pp. 289–294, 1992. [27] R. Cuzner, T. Sielicki, A. Archibald, and D. McFarlin, “Management
[8] Y. Pan, P. Silveira, M. Steurer, T. Baldwin, and P. Ribeiro, “A fault of ground faults in an ungrounded multi-terminal zonal DC distribution
location approach for high-impedance grounded DC shipboard power system with auctioneered loads,” in 2011 IEEE Electric Ship Technolo-
distribution systems,” in 2008 IEEE Power and Energy Society General gies Symposium (ESTS), 2011, pp. 300–305.
Meeting - Conversion and Delivery of Electrical Energy in the 21st [28] K. Karlsson and J. Svensson, “Fault detection and clearance in DC
Century, 2008, pp. 1–6. distributed systems,” in Proceeding of 2002 IEEE-NORPIE Workshop,
[9] B.-Y. Ku, “Simulation and emulation of rapid transit DC short circuit 2002, pp. CD–ROM pages 6.
current,” in Proceedings of the 2003 IEEE/ASME Joint Rail Conference, [29] F. Magnago and A. Abur, “Fault location using wavelets,” IEEE Trans-
2003, pp. 191–197. actions on Power Delivery, vol. 13, no. 4, pp. 1475–1480, 1998.
[10] T. Baldwin, J. Renovich, F., L. Saunders, and D. Lubkeman, “Fault [30] W. de Jager, “Buddy bidirectional supply for traction substations,” in
locating in ungrounded and high-resistance grounded systems,” IEEE Proceedings of the 2011-14th European Conference on Power Electron-
Transactions on Industry Applications, vol. 37, no. 4, pp. 1152–1159, ics and Applications (EPE 2011), 2011, pp. 1–10.
2001. [31] J. C. Brown, J. Allan, and B. Mellitt, “Calculation and measurement
[11] X. Wang, J. He, C. Dong, X. Yang, B. Kirby, and D. Writer, “Grounding of rail impedances applicable to remote circuit fault currents,” IEE
fault location in DC railway system,” in Electricity Distribution (CIRED Proceedings B-Electric Power Applications, vol. 139, no. 4, pp. 295–
2013), 22nd International Conference and Exhibition on, June 2013, pp. 302, 1992.
1–4. [32] C. Chang, S. Kumar, B. Liu, and A. Khambadkone, “Real-time detection
[12] C. Dong, J. He, X. Wang, J. Xu, L. Yu, and Z. Bo, “High-resistance using wavelet transform and neural network of short-circuit faults within
grounding fault detection and location in DC railway system,” in 11th a train in DC transit systems,” IEE Proceedings - Electric Power
International Conference on Developments in Power Systems Protection, Applications, vol. 148, no. 3, pp. 251–256, 2001.
2012., 2012, pp. 1–5. [33] C. Chang, Z. Xu, and A. Khambadkone, “Enhancement and laboratory
[13] D. Paul, “DC traction power system grounding,” IEEE Transactions on implementation of neural network detection of short circuit faults in DC
Industry Applications, vol. 38, no. 3, pp. 818–824, 2002. transit system,” IEE Proceedings - Electric Power Applications, vol. 150,
[14] I. Cotton, C. Charalambous, P. Aylott, and P. Ernst, “Stray current no. 3, pp. 344–350, 2003.
control in DC mass transit systems,” IEEE Transactions on Vehicular [34] R. Hill and D. Carpenter, “Determination of rail internal impedance for
Technology, vol. 54, no. 2, pp. 722–730, 2005. electric railway traction system simulation,” IEE Proceedings B Electric
Power Applications, vol. 138, no. 6, pp. 311–321, 1991.
[15] C.-H. Lee and H.-M. Wang, “Effects of grounding schemes on rail po-
[35] F. Filippone, A. Mariscotti, and P. Pozzobon, “The internal impedance of
tential and stray currents in Taipei rail transit systems,” IEE Proceedings-
traction rails for DC railways in the 1-100 kHz frequency range,” IEEE
Electric Power Applications, vol. 148, no. 2, pp. 148–154, 2001.
Transactions on Instrumentation and Measurement, vol. 55, no. 5, pp.
[16] J. Yu, “The effects of earthing strategies on rail potential and stray
1616–1619, 2006.
currents in DC transit railways,” in International Conference on Devel-
[36] A. Mariscotti and P. Pozzobon, “Measurement of the internal impedance
opments in Mass Transit Systems (Conf. Pub. No. 453), 1998, 1998, pp.
of traction rails at 50Hz,” IEEE Transactions on Instrumentation and
303–309.
Measurement, vol. 49, no. 2, pp. 294–299, 2000.
[17] S.-L. Chen, S.-C. Hsu, C.-T. Tseng, K.-H. Yan, H.-Y. Chou, and T.-M. [37] ——, “Measurement of the internal impedance of traction rails at au-
Too, “Analysis of rail potential and stray current for Taipei metro,” IEEE diofrequency,” IEEE Transactions on Instrumentation and Measurement,
Transactions on Vehicular Technology, vol. 55, no. 1, pp. 67–75, 2006. vol. 53, no. 3, pp. 792–797, 2004.
[18] S.-Y. Xu, W. Li, and Y.-Q. Wang, “Effects of vehicle running mode [38] ——, “Resistance and internal inductance of traction rails at power
on rail potential and stray current in DC mass transit systems,” IEEE frequency: a survey,” IEEE Transactions on Vehicular Technology,
Transactions on Vehicular Technology, vol. 62, no. 8, pp. 3569–3580, vol. 53, no. 4, pp. 1069–1075, 2004.
2013.
[19] Y.-S. Tzeng and C.-H. Lee, “Analysis of rail potential and stray currents
in a direct-current transit system,” IEEE Transactions on Power Delivery,
vol. 25, no. 3, pp. 1516–1525, 2010.
[20] C.-H. Lee and C.-J. Lu, “Assessment of grounding schemes on rail
potential and stray currents in a DC transit system,” IEEE Transactions
on Power Delivery, vol. 21, no. 4, pp. 1941–1947, 2006.
[21] “IEEE guide for rail transit traction power systems modeling,” IEEE Std
1653.3-2012, pp. 1–55, 2013. Jae-Do Park (M’07, SM’14) received his Ph.D. de-
[22] D. Love and N. Hashemi, “Considerations for ground fault protection gree from the Pennsylvania State University, Univer-
in medium-voltage industrial and cogeneration systems,” IEEE Trans- sity Park, in 2007. Dr. Park is currently an Assistant
actions on Industry Applications, vol. 24, no. 4, pp. 548–553, 1988. Professor of Electrical Engineering at the University
[23] J. Roberts, H. Altuve, and D. Hou, “Review of ground rault protec- of Colorado Denver. He is interested in various
tion methods for grounded, unqounded and compensated distribution energy and power system research and education in-
systems,” in 2008 IEEE Power and Energy Society General Meeting - cluding electric machines and drives, energy storage
Conversion and Delivery of Electrical Energy in the 21st Century, 2001, and harvesting systems, renewable energy sources,
pp. 1–8. grid-interactive distributed generation, microturbine
[24] J. Kohler and J. Li, “DC trolley fires-a new solution to an old problem,” control, and microgrid systems. Prior to his arrival at
IEEE Transactions on Industry Applications, vol. 31, no. 4, pp. 726–732, the University of Colorado Denver, Dr. Park worked
1995. for Pentadyne Power Corporation in California as Manager of Software and
Controls, where he took charge of control algorithm design and software
[25] J. Marrero, “Understand ground fault detection and isolation in DC
development for the high-speed flywheel energy storage system.
systems,” in IEEE Power Engineering Society Summer Meeting, vol. 3,
2000, pp. 1707–1711 vol. 3.
[26] Y. Pan, M. Steurer, and T. Baldwin, “Ground fault location testing of
a noise-pattern-based approach on an ungrounded DC system,” IEEE

0018-9545 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like