23rd International Conference on Electricity Distribution

Lyon, 15-18 June 2015

Paper 0415

NEW DIGITAL METHOD FOR THE DIRECTIONAL DETECTION OF TRANSIENT

GROUND FAULTS
Stefan WERBEN
Siemens AG Germany
[email protected]

Ignaz HBL
KNG-Krnten Netz GmbH Austria
[email protected]

ABSTRACT
This paper describes a new digital sensitive method for
the directional detection of ground faults in resonantgrounded or isolated networks. The method evaluates the
transient response which occurs due to the state change
after the ground fault ignition.
In the past such methods were implemented in an analog
way, with disadvantages like limited sensitivity, no fault
recordings and others. The analog implementation was
caused by the fact that the frequency range of the
transient response is large, including frequencies up to
approximately 4 kHz (for the charging oscillation). Such
high signal frequencies require a respectively high
sampling frequency by the protection device which was
usually not available in the past. With constantly
increasing performance of modern digital protection
devices such high sampling frequencies are now possible,
allowing the implementation of fully digital algorithms
for the directional detection of transient ground faults.
The basis of the direction determination is the evaluation
of the discharging and charging oscillation of the
networks capacitance to ground, which occurs during the
first ms after the fault entry. However, due to the
frequency range and the energy content mainly the
charging oscillation of the sound phases is taken into
account.
The criterion for direction determination is the active
energy of the zero sequence system during the transient
response. By means of the sign of the active energy the
fault direction is determined. Especially for high
resistance ground faults the new method shows a
significant better sensitivity compared to analog methods
which depend only on comparing the signs of neutral
voltage and current.
The new method was tested and verified under different
conditions. Since static testing is not applicable due to
the required dynamic processes, different testing
approaches were chosen.
Testing was carried out a) by means of dynamic
simulations of the transient response, b) by use of ground
fault recordings from the field and c) finally by a
prototype application under real conditions in a meshed
110 kV network.
By the integration of this transient ground fault detection
method into a modern digital multifunctional protection
device all advantages of such a device (e.g. high sampled
fault recording) are provided, as well as the option for

CIRED 2015

Klaus BHME
Siemens AG Germany
[email protected]

the integration into an existing communication structure.

Due to the high reliability of this new method in
combination with the communication facilities a safe
fault location at the control station site is achieved.

INTRODUCTION
Various methods are being used for the detection of
ground faults in resonant-grounded or isolated networks.
The simplest method evaluates the zero sequence voltage
U0 (displacement voltage) or the zero sequence current
3I0. With these methods, only the detection of a ground
fault is possible. The localization of the fault via the
direction is not possible. Extension of the method to
include evaluation of the phase difference between U0
and 3I0 makes it possible to obtain the direction of the
ground fault; e.g. the watt-metric ground fault detection
falls into this category. With such methods it is only
possible to determine the ground fault direction under
stationary fault conditions, as the direction measurement
requires steady state values. In the event of nonstationary ground faults of short duration, it is not
possible to obtain the direction by evaluation of the phase
angle difference. Consequently, other methods based on
the transient ground fault signals are applied, e.g. the
transient ground fault method used in the device 7SN60.
During a ground fault in a system with compensated star
point grounding, the transient ground fault effect is
present during the initial ms of the fault. This is caused
by the discharge of the line capacitance in the faulted
phase, and the charging of the capacitance in the sound
phases [1][2]. These are then followed by the fault
extinction or the steady state fault current and voltage.
The transient ground fault effect is typically only of a
short duration (a few ms to a few network periods). The
duration is primarily dependent on the system
capacitance, the type of fault (high resistance, low
resistance) and the duration of the fault. Low resistance
faults are easier to detect than high resistance faults and
generally dont present any problems to the measuring
methods. The new directional transient ground fault
function evaluates this transient ground fault effect within
a short interval after the ground fault ignition and
determines the fault direction also in the event of highresistance faults.
The algorithm described here and tested in operation
significantly increases the sensitivity e.g. as compared

1/5

23rd International Conference on Electricity Distribution

Lyon, 15-18 June 2015

Paper 0415

with the device 7SN60. Ground faults with fault

resistances up to several k are detected securely. By
integrating the function within a multi-function protection
device (e.g. 7SJ85), a separate additional transient ground
fault protection device is not required. The function is
thereby also fully integrated into the existing
communication infrastructure. This function in
combination with the described measures facilitates
localizing the affected feeder section. The algorithm must
function very reliably for this purpose. Static test
methods are not suitable for the verification. The
reliability was therefore determined with tests in the
power system and simulation of various networks (e.g.
meshed networks) under varied conditions.

charge between the line capacitances is slower, as these

take place via the high fault resistance. These high
resistance faults only result in a small phase shift away
from the purely capacitive range towards the resistive
range. This small phase shift must be extracted and used
for the evaluation/classification of the direction.

MEASURING PRINCIPLE
The measuring principle evaluates the signals in a short
interval after the ground fault ignition. At this point, the
signal components resulting from charging and
discharging the ground capacitance of the network are of
particular interest. The discharge oscillation (frequency
range 500 Hz to 1000 kHz) in the faulted phase typically
ends within the first cycle of the charging oscillation. The
charging oscillation (frequency range 70 Hz to 4 kHz) in
the sound phases typically terminates within the first
power system frequency cycle [3]. The discharge
oscillation can therefore not be reliably detected, even
with relatively high sampling rates (e.g. 8 kHz). The
charging oscillation on the other hand can reliably be
measured with higher sampling rates (e.g. 8 kHz). An
important feature of these charging and discharging
cycles is that they are resistive and dependent on the fault
direction. This characteristic was already used by analog
transient ground fault relays to determine the fault
direction [2]. After detecting the fault ignition, the ground
current and displacement voltage are evaluated to
determine if they are in-phase or in other words have the
same sign or not. This is done in a small time window
following the instant of detecting the fault ignition. If the
signs are not equal the fault is in the forward direction
and alternatively the fault is in reverse direction.
This principle has been applied for a number of decades
and provides reliable results for faults with low resistance
(see Figure 1). The reason for this is that the resistive
current components (charging of the network
capacitance) dominate compared to the static ones in
systems with resonant grounded star point these are
shifted by approx. 90. Using this, the co-phase or
opposite phase condition of the zero sequence current and
zero sequence voltage will reliably determine the fault
direction. In the case of high resistance faults (see Figure
2) on the other hand, the static component is dominant.
As a result there is only a small phase shift between the
charging transient and the static component.
During high resistance ground faults the transfer of
CIRED 2015

Figure 1 Low resistance ground fault

Figure 2 High resistance ground fault

The active power in the zero sequence system has proven
to be a good criterion for determining the fault direction
of the transient ground fault (Equation 1 and 2). This
represents the resistive component of the signal. As the
charging takes place over a longer period of time, the
resistive component is present during this time. It
therefore makes sense to form an integral of this
component starting with the ground fault ignition. The
result is the active energy in the zero sequence system
(equation 3).

p0 (k ) = u0 (k )i0 (k )

(1)

2/5

23rd International Conference on Electricity Distribution

Lyon, 15-18 June 2015

Paper 0415

pa ( k ) =

compared with the sound feeders (fault in reverse

direction).

1
p0 (k ) dt
T k T

(2)

E0 (k ) = pa (k ) dt

(3)

The computed active energy must then only be compared

with set thresholds to determine the fault direction.
Exceeding the positive threshold indicates that the fault is
in reverse direction while dropping below the negative
threshold shows that the fault is in the forward direction.
The thresholds required for the direction decision must
still be determined. For this purpose a compromise
between sensitivity and a correct direction result must be
made. Three zones have proven to be effective as shown
in Figure 3. The energy thresholds are additionally
limited by a time Tmax after which a decision is no
longer made as the charge transfer process has been
largely completed by this time and the static fault
condition will influence the result. The threshold for
forward (-E0<) and reverse (+E0>) were determined
during the previously mentioned field trials and
simulations. An optimization process was applied for this
purpose so that all transient ground faults that were
available resulted in correct response. The results showed
that the resulting two thresholds were suitable for all the
transient ground faults. The user therefore does not have
to determine the thresholds or set them. Of interest is that
the reverse threshold is smaller than the forward
threshold (not shown true to scale in Figure 3).
E
reverse

+E0>

Tmax

unknown

-E0<

forward

Figure 3 Zones for direction

This is plausible as the transient oscillations from all
feeders are summated in the faulted feeder in the event of
a forward ground fault. This results in a greater energy
result in the faulted feeder (fault in forward direction) as

CIRED 2015

The ground fault detection methods have a problem with

circulating operational zero sequence currents that occur
in meshed or closed ring networks. These currents, that
result from a non-symmetry of the phase impedances or
systems connected in parallel, cause the phase of the fault
current to be shifted. They appear in the computed active
energy of the zero sequence system. The direction of
these currents does not have any correlation with the
actual fault current. As a consequence the direction
measurement may yield a wrong result. For correct
direction measurement the elimination of these currents is
necessary and this is done within the measuring method.
Initially, the simulation tool MATHLAB was used to
simulate the measuring method and test and optimize its
response to the available fault recordings. This field data
was used to create a reference data set for testing with an
OMICRON test set. With this setup the device was tested
to confirm that the code in the device was working
accordingly.

TEST OF THE MEASURING METHOD

Background
Contrary to many other protection functions, it is not
effective or even possible to test the transient ground
fault function with static test conditions. The Omicron
test set provides a simple simulation for testing the
ground fault function as well as the transient ground fault
function. In this case the test corresponds to a simple
radial system. This is not sufficient for the intensive
testing that is necessary for a newly developed measuring
method.
For testing the measuring method data from a number of
sources may be used. These are:
Artificial data
Data obtained with dynamic simulation
Field data from real ground faults
A further possibility is the application of prototypes in
the real power system.
Apart from the data required for testing the directional
ground fault function, the network topology also plays a
large part in determining the faulted feeder. The
following network topologies need to be considered:
Radial network
Meshed network or ring network
Most methods are reliable in radial networks. Experience
has shown that meshed or ring networks pose a greater
problem for the localization of the faulted feeder. For the
tests with data obtained from dynamic simulation,
amongst others, 2 topologies that are oriented on real

3/5

23rd International Conference on Electricity Distribution

Lyon, 15-18 June 2015

Paper 0415

systems were used. This also allowed the comparison of

simulation data with real fault recordings, which
validated the simulation results. This procedure allowed
further testing by modification of the test parameters.
These parameters were:
Fault location
Fault resistance
Fault duration
Point of ground fault ignition
Degree of compensation (over / under
compensated)

procedures had already been done there before [4],

which made accessible experience and infrastructure for
the detection and localization of ground faults possible.

Testing Methodology
The new measuring method was developed and tested
using approximately 200 fault recordings from a medium
voltage ring network N1 (Figure 4) and simulation data
from a radial network. The fault recordings have a high
sampling rate and were classified according to the fault
nature, so that they were suitable for the tests. The
implementation was done in a prototype, which had to
detect all the faults correctly. The prototype performance
was permanently compared with the MATHLAB
simulation during the entire testing process. The resulting
prototype was then installed in network N2.
110 kV
20 kV

Y
Y

Infeed
BB1

Y
Y

Infeed
BB2

BB=busbar

BB1

BB2

Coupling,
Normaly open

BB

BB

Closed ring

BB

Closed ring

Figure 4: Ring network N1

To collect experience, and to determine the suitability of
the new transient ground fault measuring method, a field
test was conducted, where four protection devices were
applied in a KELAG meshed 110 kV system (Figure 5).
In essence, this was a 110 kV parallel line in a high
alpine region with a high incidence of thunderstorm
activity. Furthermore pump storage generators were
connected to one of the parallel lines. The large number
of incidents within two years (transient ground faults,
switching operations, generation and pumping operation)
resulted in a comprehensive data set for further
evaluation.
The KELAG network proved to be very well suited for
this purpose, as tests for ground fault localization

CIRED 2015

Figure 5: 110kV network N2

The network N2 is schematically shown in Figure 6. The
protected parallel line has a length of approx. 40 km. The
one line has two T-Offs to the pump storage generators
(LG1, LG2additionally. Due to the superimposed ring
(lines Ln1, Ln2, Ln3) and the T-Offs LG1 and LG2 the
demands on protection functionality are challenging.

Field test of the prototype

The devices were applied to the parallel lines A1-A2 and
B1-B2, where they operated in parallel to the already
installed ground fault detection devices. The function was
applied with a high sensitivity to obtain fault recordings
from other influencing conditions, which were not ground
faults. All the ground faults on the observed parallel lines
were detected correctly. A large number of ground faults
in the remaining system were detected. The very sensitive
setting of the function resulted in a pick-up during
switching operations in the system. With the applied
SIPROTEC 5 devices it was possible to register fault
recordings with a sampling rate of up to 8 kHz. This wass
absolutely necessary for the analysis of the transient
ground fault effect. These fault recordings were then used
for further optimization of the transient ground fault
function and were added to the field record test data base.
The modified function was subsequently tested with all
the already available test cases and had to provide the
same results, as well as remain stable during switching
operations. The devices in the KELAG network N2 were
then updated with the improved measuring method and
remained in operation. Since the inception of the tests
with the improved measuring method, approx. 100 new
ground faults have occurred. All of these were correctly
detected. A pick-up during switching operations no
longer occurred.
The close cooperation with the system operator thereby
4/5

23rd International Conference on Electricity Distribution

Lyon, 15-18 June 2015

Paper 0415

resulted in an improvement and optimization of the

algorithm under real operating conditions. These tests are
ongoing and will be continued in the future.
Consequently the field data base will be increased
continuously.

Close to the relay location it is even possible to obtain

good results up to 10k. On the sound parallel line Lb
the direction reverses as expected. This fact however
does not result in a measurement where the line Lb is
indicated as the faulted line.

SUMMARY

Figure 6 Schematic representation of network N2

Simulation results
The network N2 that was used for testing is shown in
Figure 6. It was transferred into a simulation. This
allowed further investigation of the transient ground fault
function. Of particular interest is the boundary presented
by the detection of high resistance ground faults. For this
purpose the fault location and the fault resistance can be
varied.

Modern digital protection devices can provide high

sampling frequencies (e.g. 8 kHz) which allow measuring
the transient response after the ground fault ignition. This
was a precondition to develop a new digital and sensitive
transient ground fault detection function.
The methodology, whereby algorithms are developed in
MATHLAB and the target system in parallel, as well as
prototype testing in the field, has proven itself.
The new measuring method has proven to be very
dependable. The method is simple and robust and shows
significantly improved characteristics during high
resistance ground faults, when compared to e.g. the
proven 7SN600 transient ground fault relay. A secure
detection of ground faults up to 5 k is possible. For
testing, approx. 800 fault recordings are available at
present. The new measuring method is available in
protection devices of the SIPROTEC 5 family provided
by Siemens AG. The new function can be applied within
one device in parallel to other protection functions like
line or overcurrent protection which prevents from
applying two physical devices.

REFERENCES
[1] Gernot Druml, 2012, Innovative Methoden zur
Erdschlussortung und Petersen-Spulen Regelung,
Dissertation TU Graz, Graz, Austria, 19-27
[2] Ungrad, Winkler, Wiszniewski, 1991, Schutztechnik
in Elektroenergiesystemen, Springer-Verlag, Berlin
Heidelberg, Germany, 16-19, 165-167
[3] H.
Pundt,
1965,
"Untersuchung
der
Ausgleichsvorgnge
bei
Erdschlu
in
Energieversorgungsnetzen", Energietechnik 15. Jg.,
Heft 10, 469-477
[4] Tengg, Schoa, Schmaranz, Marketz, Druml, 2011,
"Neue Erdschluss-Ortungsverfahren im Praxistest",
ETG-Fachtagung Sternpunktbehandlung

Figure 7 Fault resistance versus fault location for the line

La
Figure 7 shows the result obtained with devices A1, A2,
B1 and B2 for ground faults on the line La (green =
forward, yellow = no decision, red = reverse). Up to 5 k
the forward direction is securely obtained. Meanwhile the
5 k sensitivity has been confirmed by customer testing.

CIRED 2015

5/5