Content uploaded by Lukas Ranzinger
Author content
All content in this area was uploaded by Lukas Ranzinger on Mar 15, 2023
Content may be subject to copyright.
1
DETECTING GROUND FAULTS IN SYNCHRONOUS
MACHINES USING SWEEP-FREQUENCY-
RESPONSE-ANALYSIS (SFRA)
Lukas Ranzinger1*, Prof. Dr.-Ing. Stephanie Uhrig1, Prof. Dr.-Ing. Stefan Tenbohlen2,
Tobias Rieder1
1Munich University of Applied Sciences, Munich, Germany
2University of Stuttgart, Stuttgart, Germany
*lukas.ranziger@hm.edu
Keywords: ROTATING MACHINES, DIAGNOSIS, GROUND FAULTS, SFRA
Abstract
Synchronous machines are important equipment in power plants and drive systems. Safe and trouble-free operation is essential,
but this can be interrupted by electrical or mechanical faults. Commonly occurring faults in synchronous machines are grounding
faults. This results in a conductive connection between parts of the winding and grounded elements. In this work, the Sweep-
Frequency-Response-Analysis (SFRA) method is used to detect these specific faults. The SFRA method is relatively simple and
fast to apply, is non-invasive and sensitive to various electrical and mechanical parameters of the machine. For the purpose of
ground-fault detection, multiple ground faults are generated at the stator and rotor winding and the changes in the frequency
response are observed. For example, artificial stator-to-ground faults were implemented on different phases and different
positions within one phase. It was found that the failure position can have an influence on the frequency response. Faults near
the middle of the winding show similar frequency responses, even with reverse measurement direction. However, faults near a
terminal result in specific changes after inverting the measurement direction. If a stator-to-ground fault was detected in a first
place, this characteristic might help to narrow down the fault location. Additionally, artificial rotor-to-ground faults were
implemented. Comparable behavior to the stator faults can be seen when measuring the field winding on the rotor. Beyond that,
the SFRA is sensitive enough to detect such failures by measuring purely the stator side without direct contact to the stator. The
fundamentals for this behavior are explained in detail using a previously developed model and verified by simulations.
1. Introduction
Condition diagnostics on synchronous machines is an
important component for operators in their operating strategy.
For this purpose, the machines are turned off at regular
intervals and many operating parts of the system are checked.
In order to prevent of electrical, magnetic or mechanical faults,
a variety of diagnostic methods are used. The aim is to observe
as many of the machine parts as possible because typical faults
can occur in a different machine areas. As Figure 1 shows, the
stator winding is affected by faults in about 60%. In more than
every tenth case, the rotor winding is damaged. Usually, a
different diagnostic method must be used for each position
shown. However, no differentiation has yet been made
between the types of faults. In the stator and rotor winding, for
example, short circuits, interruptions, or ground faults can be
found. The last mentioned is subject of this study. With the
SFRA, a new method for rotating machines is applied. It is
possible to detect ground faults in all machine areas with one
measurement.
The Sweep-Frequency-Response-Analysis (SFRA) offers
significant advantages compared to conventional methods. It
is sensitive to many magnetic and electrical parameters within
the machine at different locations. Therefore, an SFRA
measurement can initially be made as an overall picture. If
necessary, an individual fault at a specific location can then be
investigated with a specialized method. The SFRA is also
characterized by good reproducibility, as well as mobile,
simple, and safe application. The method is non-invasive and
the measuring voltage is usually 10 Vpp. Various own
publications describe very promising results for the SFRA also
in the application on rotating machines [1–3]. This confirms
the basic applicability for rotating machines.
Figure 1 Typical fault location for machines of larger power classes
[4].
As mentioned, the focus of this work is on ground faults of the
stator or rotor winding. In case of a fault, a conductive
connection is formed between parts of the winding and a
grounded part of the machine, e.g. the housing. Examples are
given in [5–8], e.g. ground faults are investigated on a removed
Stator
Winding
60%
Rotor
Winding
13%
Stator Core
1%
Bearing
13%
Others
13%
2
pole. Individual ground faults are clearly recognizable, a
transfer to the entire machine is not made. A 5 kVA
synchronous machine is measured in [9] with different ground
faults. However, only the affected stator phase is considered.
Furthermore, the explanations do not provide a verification by
a model of the fault expression, which is used in the present
work as a basis for determining the fault location. Ground
faults in the rotor of a complete machine are not investigated
in any of these papers.
2. SFRA Method
Frequency response analysis (FRA) can be performed as
sweep FRA (SFRA) or impulse FRA (IFRA). Due to the
higher accuracy over wide frequency ranges and the lower
interference sensitivity, the SFRA is preferred to the IFRA in
this work. As shown in Figure 2, the sinusoidal voltage U1 is
applied to an input terminal of the test object. The amplitude
is known (always10 Vpp in these studies), and the frequency
sweep is performed in predetermined limits (here 20 Hz –
2 MHz). The response signal U2 is measured at the output
terminal, which is then compared with the input signal U1. This
results in the frequency response consisting of amplitude
response and phase response. The present investigations are
mainly limited to the amplitude response D ([D] = dB), which
can be calculated according to the following formulas:
( 1 )
( 2 )
Figure 2 Schematic of SFRA measurements.
The SFRA can be executed differently for rotating machines,
depending on the measurement setup. In the case of a
capacitive coupling, there is no galvanic connection between
input and output terminal. This is the case, for example, when
the star point is opened, and the input signal is applied to a
terminal of phase U and the response is measured at phase V
or W. The measurements are then called "indirect"
measurements. Alternatively, a galvanic connection can be
measured, i.e., the input and output terminals are each on the
same phase and are termed "direct". These measurements can
be performed as "direct open circuit" or "direct short circuit".
The difference is that in the first case the field winding is open
circuit, whereas in the second case the field winding is short
circuited. All measurements shown here are "direct open
circuit" measurements with the star point opened.
The evaluation of SFRA results is usually done by comparing
actual measurements with reference measurements, so-called
fingerprints. These can be, for example, measurements at an
earlier time, a different phase, or a sister unit.
3. Characteristic trace and grey-box modelling
A grey-model is defined by using the most relevant machine
parameters but is nevertheless generally valid and can be easily
adapted to different machine types. Essential influences in
SFRA measurements can be explained clearly. The modelling
approach is therefore very well suited for this work and shows
some advantages over white box or black box modelling [10].
Figure 3 Single phase model for a synchronous machine with field
winding.
The single-phase model of a synchronous machine with field
winding is shown in Figure 3. The example shows the phase U
with the terminals U1 and U2. Accordingly, the left part of the
model with the indices "S" represents the stator-side three-
phase winding. The measurement setup of the SFRA
measurement is connected to the terminals U1 and U2, here
marked in red. Further parameters of the stator winding are:
RCu, LM: Copper resistance and main inductance
RFe: Magnetic circuit iron resistance
CP: Winding longitudinal capacitance
LL, CS: Leakage inductance with associated capacitance
RGnd,CGnd: Winding to ground path
Symmetrically modeled is the field winding with the terminals
F1 and F2 and the indices "F", which are contacted at the
machine by slip rings. Stator and rotor are connected through
inductive coupling (transformer principle) and the coupling
capacitor CCou.
The simulation of the model corresponds to a direct open
circuit measurement and results in four frequency ranges,
which are marked in colour in Figure 4. The amplitude
response in dB is plotted against the frequency as a logarithmic
scale. For low frequencies, the curve starts at 0 dB, where only
the copper resistor RCu,S has an effect. The range 1
characterizes a low pass behaviour due to the main inductance
LM,S. It results in a decreasing curve with -20 dB/decade
gradient, which is interrupted by the range 2, a double
resonance. Responsible for this is the rotor-side resonance of
the parameters LM,F and CP,F, which is also visible in stator-
side measurements due to inductive coupling. In range 3, the
curve flattens, and a local maximum is formed due to the
capacitive coupling to ground. The main resonance of the
parameters LM,S and CP,S can be seen in range 4.
3
Figure 4 Single phase model simulation for the direct open circuit
measurement.
4. Test object
The test object is a 3 kVA synchronous machine with a rated
speed of 1500 min-1 at 50 Hz and two pole pairs (p=2). For
SFRA measurements, the rotor angle must be specified during
measurements, the 0° position is selected freely. As an
example, the direct open circuit measurement at the rotor
position 60° for phase V is shown in Figure 5. The four
modelled frequency ranges from chapter 3 are also clearly
visible in the measurement. At higher frequencies above
100 kHz, additional smaller resonance points can be identified,
which result from further influences that are not considered in
the model. An example of this is the neighbouring phases,
which provide a further resonance point in frequency range 4.
Figure 5 Single phase and direct open circuit measurement of phase
V at 60°.
The reference measurement at one rotor angle is not sufficient
to characterize the machine. The rotor angle must be
considered. For this purpose, direct open circuit measurements
are repeated every 5°. In total, 72 measurements result for one
phase, which can all be seen summarized in Figure 6. The
result is a surface diagram in which the amplitude damping is
plotted against the frequency and the rotor angle. The
influence of the rotor angle is very clear, with repetitive
waveforms appearing at 360°/2p (here 90°) intervals. The
periodic course can be seen especially at the double resonance
and the main resonance.
Figure 6 Complete single phase and direct open circuit measurement
of phase V.
5. Artificial stator ground faults
The synchronous machine used as test object for this work has
a single wire winding. Individual wires of this winding were
contacted at various points of the phases V and W and a
conductive ohmic connection was built to grounded housing.
The results for phase V and W show very similar behaviour,
therefore only phase V is analysed here as an example.
In phase V, ground faults are implemented at two positions of
the winding. In Table 1, the resistance measurements between
the terminals V1 and V2 to the two different faults show their
position in the winding. Fault type 1 is located closer to
terminal V1 than to terminal V2. The resistance V1 to fault
type 1 is 0.571 Ω and smaller than the 2.626 Ω from V2 to fault
type 1. Fault type 2 is approximately centred in the winding,
and the resistances are approximately equal.
Table 1 Resistances between terminals and faults
Contact of:
V1
V2
Fault type 1
0,571 Ω
2,626 Ω
Fault type 2
1,491 Ω
1,343 Ω
In the grey box model, the ground fault can be represented by
splitting the main inductance LM,S, the copper resistor RCU,S
and the longitudinal capacitance CP,S, so that a connection to
ground can be made between them. In Figure 7 the
corresponding parameters are marked with the red superscripts
1 and 2. The fault is also drawn in red. A contact resistance
RContact results against ground, which varies depending on the
fault characteristic.
Figure 7 Single phase model for a synchronous machine with a stator
ground fault.
Various effects of this fault can be observed. As can be seen in
the modelling, a parallel path to the measuring resistor RM
(where voltage U2 is measured) is formed starting from the
fault location. If only the copper resistor and the main
inductance are considered, the influence of the fault for
frequency range 1 can be explained using the transfer function
in formula 3 and 4. The value of the amplitude ratio for low
frequencies must be higher in the fault case than in the fault-
free case. Equation 3 corresponds to the transfer function in
the fault case. For a finite contact resistance RContact, the ratio
U2/U1 for very low frequencies (20 Hz – 100 Hz) is always
smaller than in the fault-free case (formula 4). Formula 4
results for the case that RContact tends to infinity, i.e., there is no
conductive connection between winding and ground. This
4
corresponds to the original model from Figure 3 and the fault-
free case.
( 3 )
( 4 )
From the equations it results that the amplitude response in
case of a fault does not start at 0 dB as usual, but at higher
attenuation. The exact amount of the initial damping can be
calculated (formula 3). For this, however, the contact
resistance RContact and the ratio of Z1 and Z2 must be known.
The larger the contact resistance, the smaller the initial
amplitude damping. This can be shown with formula 3 and
results from considerations on the model in Figure 7. If the
contact resistance becomes larger, proportionally more current
flows through the measuring resistor RM and the voltage U2
becomes comparatively larger.
From the curve discussion of formula 3 with the input
parameter ratio Z1/Z2 it becomes obvious that the initial
damping D becomes smaller when Z1/Z2 becomes smaller.
Conversely, D becomes larger for larger Z1/Z2. Thereby it must
always be valid that Z1+Z2 = ZGes, with ZGes remaining
constant. In other words, the further away the fault is from the
input terminal, the higher the initial damping tends to be. If a
fault is detected, the measuring direction can be changed.
When comparing the two resulting traces, the fault location
can be localized, regardless of whether the contact resistance
RContact is known or not. It is sufficient to know for which
measuring direction the initial amplitude damping is larger. If
there is a fault and the traces of both measuring directions are
nevertheless overlapping, the ground fault is exactly in the
middle of the winding.
Two ground faults at different positions were implemented in
phase V (compare Table 1). Fault type 1 is located near
terminal V1, fault type 2 approximately in the middle of the
winding. The contact resistance is initially 0 Ω. In Figure 8,
the blue trace is the reference measurement. In the fault
measurements, a further differentiation is made according to
the measurement direction. For example, the abbreviation
V1V2 means that the input signal U1 is applied to terminal V1
and the response U2 is measured at V2. The red measurement
traces show a significant difference on where the fault is
located in the measurement circuit. As explained by formula
3, the damping at low frequencies depends on the fault
location. If the fault is close to the input terminal, the trace
starts at lower damping values than if the fault is further away.
Therefore, V2V1 has a higher damping for low frequencies
than V1V2. From about 10 kHz, the measurement traces
approach each other and finally run identically. In addition to
the initial amplitude damping, there is a further deviation from
the reference curve. The ground fault can be clearly identified.
A similar picture is seen for fault type 2. The fault is located in
the centre of the winding, minimally closer to terminal V2. The
measurements are almost identical, V2V1 is between 20 Hz
and 1 kHz minimally higher than the measurement V1V2.
Overall, the previous considerations can be confirmed, if the
fault is located in the centre of the winding, the different
measurement direction provides the same results.
Figure 8 Fault type 1 and 2 for direct single phase measurement of
phase V with RContact of 0 Ω.
As described, the contact resistance RContact also influences the
damping at low frequencies. This is shown as an example in
Figure 9 using the measurement of phase V and fault type 2.
The blue trace is the reference measurement, the remaining
traces are fault measurements at different contact resistances.
As expected, increasing contact resistance decreases the initial
amplitude damping. This behaviour can be observed for all
fault types. From a frequency of approximately 1 kHz, the fault
traces are identical again and the contact resistance no longer
has any influence on the measurement.
Figure 9 Fault type 2 for direct single phase measurement of
phase V with various RContact .
Finally, it is shown how fault types 1 and 2 impact the
neighbouring phases. The faults are still implemented in phase
V, the following measurements are from phase U. The
considerations are not necessarily relevant for fault detection,
because the fault can of course be detected clearly during the
measurement at phase V. Nevertheless, it is very helpful for
the general understanding. Figure 10 shows the measurement
U1U2 as reference and fault measurements. The trace is
changed by the ground fault only in the range of the main
resonance. This agrees with the considerations from the
general modelling, since the neighbouring phases also form
this range. The ground fault changes diverse inductances and
capacitances of phase V. The resulting resonances disappear
or change their frequency, and thus also change the
measurement trace of the neighbouring phases. Differences
caused by the contact resistance and the location of the fault
are difficult to determine. Measurements on phase V are far
better suited for this purpose.
5
Figure 10 Fault type 1 and 2 for direct single phase measurement of
phase U.
6. Artificial rotor ground faults
The synchronous machine is designed as an internal pole
machine with a non-salient rotor.
Figure 11 shows schematically the field winding with five
ground faults implemented. The position is defined by
resistance values. Fault type 1 and 5 as well as fault type 2 and
4 are symmetrical to the centre of the field winding. Fault type
3 is located in the centre.
Figure 11 Field winding with five ground fault types at different
locations.
In the grey box model, the ground fault can be represented by
splitting the main inductance LM,F, the copper resistance RCu,F
and the longitudinal capacitance CP,F, so that a connection to
ground can be made between them. The result is a similar
picture as shown in Figure 7, with the fault being installed on
the rotor-side. A contact resistance RContact results to ground. In
the following measurements, this is initially 1 Ω and
corresponds to the measured ground resistance between rotor
and stator.
Figure 12 shows direct open circuit measurements of phase V
at a rotor position of 60°. The reference measurement in blue
and fault measurements with fault types 1, 2 and 3 are shown.
It can be seen that a ground fault occurring in the rotor affects
the open circuit measurement only in frequency range 2 at the
double resonance. The position of the double resonance varies
depending on the fault location. A ground fault in the centre of
the field winding shows no change in the measurement trace
and therefore overlaps with the reference measurement. The
further the fault location moves from the centre of the field
winding in the direction of terminal F1, the smaller the
frequency of the double resonance becomes. As described, the
rotor-side parameters are split by the ground fault (see also
Figure 7). This results in a second double resonance, marked
with a red arrow, with significant smaller amplitude. Splitting
the main inductance LM,F also splits the amplitude of the
double resonance. The smaller a part of the main inductance
becomes, the smaller is its inductive coupling and thus the
amplitude.
Figure 12 Fault type 1, 2 and 3 for direct single phase measurement
of phase V with RContact of 1 Ω.
As described, the fault location affects the position of the
double resonance. Figure 13 shows that two ground faults
which are symmetrical with respect to the centre of the field
winding show the same measurement results. This can be seen
in the fault measurements of fault types 1 and 2 and their
respective symmetrical fault types 4 and 5. This means that it
is irrelevant for stator-side measurements if the ground fault
occurs at the beginning or at the end of the field winding. Only
the distance of the ground fault is to the centre of the field
winding matters. This implies that the direction of
measurement is irrelevant.
Figure 13 Fault type 1, 2, 4 and 5 for direct single phase measurement
of phase V with RContact of 1 Ω.
The fault impact results for Phase V are also applicable to the
other stator phases. Figure 14 shows this as an example using
phase U compared to phase V. The reference measurements
and the fault measurements for fault type 1 are shown in each
case. Based on the design of the test object and the relative
position between the field and three-phase winding, a rotor
position difference of 120° must be considered.
Figure 14 Fault type 1 for direct single phase measurement of phase
V and phase U with RContact of 1 Ω.
6
This 120° shift is the reason for showing phase V at rotor
position 60° and phase U at 300°. The double resonances are
equally defined in frequency and shape regardless of the
considered three-phase winding. The same is transferable for
phase W.
The contact resistance RContact influences the frequency and
shape of the double resonance only. Using the example of
phase U, the open circuit measurements with variable RContact
at a rotor position of 10° are shown in Figure 15. Above a
contact resistance RContact of 50 kΩ, the fault measurement no
longer shows differences compared to the reference
measurements. If the contact resistance becomes too high, this
again corresponds to the fault-free state of the machine and the
ground fault is no longer detected.
Figure 15 Fault type 1 for direct single phase measurement of phase
U with various RContact.
7. Conclusion
In sweep frequency response analysis (SFRA), the frequency
response of the stator three-phase windings of rotating
machines is recorded. The measurement is influenced by
various electrical and magnetic parameters in the direct and
indirect measuring circuit. Various faults change certain
parameters, and the measurement trace changes in its result.
SFRA can be used as a method for fault or condition diagnosis.
In the first part of this paper, the general characteristic
behaviour of SFRA measurements is explained using a model.
Four frequency ranges are defined.
One possible type of fault in rotating machines is the ground
fault. In this case, the insulation takes damage and a
conductive connection of the winding against grounded parts
of the machine occurs. This fault can appear both in the stator
and in the rotor. In a synchronous machine, different faults are
installed in different locations and the changes on the
frequency response are observed and explained. Both stator
and rotor faults can be detected clearly, only using the three-
phase winding on the stator side. The fault location can be
delimited via the characteristics of the trace of the fault
measurements. In addition, the influence of the contact
resistance from winding to ground is described for both cases.
8. References
[1] S. Uhrig, F. Öttl, N. Augeneder, and R. Hinterholzer,
“Reliable Diagnostics on Rotating Machines Using
FRA,” in Proceedings of the 21st International
Symposium on High Voltage Engineering: Volume 1, B.
Németh, Ed.: Springer, 2020, pp. 738–751. [Online].
Available: https://www.researchgate.net/publication/
337605417_Reliable_Diagnostics_on_Rotating_
Machines_Using_FRA
[2] L. Ranzinger, S. Uhrig, and F. Öttl, “Basic behaviour of
FRA measurements on rotating machines,” 22nd
International Symposium on High Voltage Engineering
(ISH).
[3] L. Ranzinger, S. Uhrig, R. Hinterholzer, and F. Öttl,
“Failure diagnosis in rotating machines using FRA
involving the rotation angle of the rotor,” 22nd
International Symposium on High Voltage Engineering
(ISH).
[4] M. Mostafaei and J. Faiz, “An overview of various faults
detection methods in synchronous generators,” IET
Electric Power Appl, vol. 15, no. 4, pp. 391–404, 2021,
doi: 10.1049/elp2.12031.
[5] A. ARANDA CARMONA, “Diagnostic method of
electrical rotors by applying the sweepfrequency
response analyzer (SFRA),” in Cigre A1-106.
[6] A. Mugarra, C. A. Platero, J. A. Martinez, and U.
Albizuri-Txurrka, “Large Salient Pole Synchronous
Machines Field Windings Diagnosis by Frequency
Response,” in 2018 XIII International Conference on
Electrical Machines (ICEM), Alexandroupoli, Sep. 2018
- Sep. 2018, pp. 1875–1880.
[7] C. Martin, J. Guerrero, P. Gomez Mourelo, and C.
Platero, “Ground Faults Location in Poles of
Synchronous Machines Through Frequency Response
Analysis,” IEEE Trans. on Ind. Applicat., vol. 58, no. 1,
pp. 113–122, 2022, doi: 10.1109/TIA.2021.3122412.
[8] H. Mayora, A. Mugarra, J. M. Guerrero, and C. A.
Platero, “Synchronous Salient Poles Fault Localization
by SFRA and Fault Diagram Method,” IEEE Trans.
Energy Convers., pp. 1–9, 2022, doi:
10.1109/TEC.2022.3182817.
[9] C. A. Platero, F. Blázquez, P. Frías, and D. Ramírez,
“Influence of Rotor Position in FRA Response for
Detection of Insulation Failures in Salient-Pole
Synchronous Machines,” IEEE Trans. Energy Convers.,
vol. 26, no. 2, pp. 671–676, 2011, doi:
10.1109/TEC.2011.2106214.
[10] M. Heindl, S. Tenbohlen, and R. Wimmer, Eds.,
Transformer modeling based on standard frequency
response measurements, Aug. 2011.
