UTRA COLUMN

Frequency Response Analysis is a technique that has been suc-

cessfully applied on power transformers to detect mechanical
damage to windings since its development over 40 years ago

A review of transformer
FRA measurement and
diagnosis techniques
Column by the University Transformer Research Alliance (UTRA) www.university-transformer-research.com

cy responses of windings through

ABSTRACT day and requires a benchmark FRA
measurement for comparative diag- modelling and simulation.
Frequency Response Analysis (FRA) nostic methods. The use of numeri-
is a successful technique to de- cal indices to quantify differences in KEYWORDS:
tect mechanical damage in power two FRA signatures is described. It Frequency Response Analysis,
transformers with greater sensitiv- is important to understand the im- Sweep Frequency Response Anal-
ity than other measurements. The portance of understanding the phys- ysis, Windings, Numerical Indices,
SFRA method is commonly used to- ical meaning behind the frequen- FRA Interpretation

96 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023

 Y. YANG et al.

SFRA effectively measures the transfer admit-

tance of the winding using multiple frequency
points, generally in the range 2Hz to 2MHz

1. Introduction data (or any time series data with a suitable voltages even in noisy environments. Ini-
frequency content) into the frequency do- tial studies showed the capability of the
Frequency Response Analysis (FRA) is main, but it is generally considered easier technique to detect winding movement,
a technique that has been successfully to obtain good signal-to-noise ratios, good shorted turns and, in particular wind-
applied to power transformers to detect frequency definition and rejection of inter- ing partial axial collapse with far greater
mechanical damage to windings since its ference required for reliable site measure- sensitivity than impedance and capac-
development over 40 years ago [1]. One ments using a network analyser. A network itance measurements[1-2]. Measured
example of such damage is shown in Fig- analyser uses a swept frequency sine wave under power frequency, transformer
ure 1.1. FRA grew from low voltage im- as the input signal whilst measuring the impedance, winding-to-ground and
pulse test techniques, which record, for input and output signals through a narrow inter-winding capacitances do change
example, the neutral current when an im- band filter tracking the frequency. This is if the physical distances that determine
pulse is applied to a terminal. These tech- the technique (sometimes called SFRA) them change, and hence can defect ma-
niques had been successful in both detect- most commonly used today. jor winding deformations. However, the
ing winding damage and understanding wide frequency range employed by the
the impulse response for system transient With SFRA, it is relatively easy to choose SFRA measurement makes it sensitive to
studies. The raw impulse response, howev- the measurement bandwidth and fre- local movement or minor deformation
er, varies with cabling and impulse genera- quency range and can generally use low of the winding as well. SFRA effective-
tor parameters, and the frequency response
(the ratio of output voltage to input voltage
across a range of frequencies) eliminated SFRA effectively measures the transfer ad-
some of the uncertainties caused by mea-
surements, has better reproducibility and mittance of the winding using multiple fre-
was easier to interpret. Frequency response
can be obtained using a fast Fourier trans-
quency points, generally in the range 2Hz to
form algorithm (FFT) to convert impulse 2MHz
w w w . t ra n sfo r m e r s - m a g a z i n e . co m 97
UTRA COLUMN

Since SFRA is a low voltage wide frequency ly measures the transfer admittance of
the winding using multiple frequency
measurement technique, many factors can points, generally in the range 2Hz to
contribute to ‘false’ deviations 2MHz. Geometric dimension-depen-
dent electrical parameters, i.e., winding
inductances and capacitances, cause
resonance and anti-resonance points
across the frequency range, and local-
ised geometric deformation of winding
can therefore show itself as both fre-
quency shifts and amplitude changes
when being compared with a fingerprint
measurement on the intact winding.
FRA is normally used as a comparative
diagnostic method in which a reference
FRA trace is compared to the newly ob-
tained FRA trace, and the mechanical
condition of the winding is evaluated
based on the deviations between the two
traces. The fingerprint, i.e. a benchmark
FRA measurement for comparison, is
therefore vitally important and is ideal-
ly obtained when the winding is known
to be undamaged, but can be obtained
from another phase or another identical
transformer albeit with lower detection
sensitivity and more complicated inter-
pretation.

Since it is a low voltage wide frequency

measurement technique, many factors
can contribute to ‘false’ deviations. For
(a) instance, at higher frequencies, care
needs to be applied to earthing lead
arrangements, especially when a long
bushing is present. The tap-changer po-
sition causes the frequency response to
change, so that change must be consis-
tent between measurements. It is usual
to ensure that the whole tap winding is
included in the measurement as this is
often the winding most vulnerable to
mechanical damage.

The standardised mea-

surement methods
following IEC 60076-
18 help to ensure the
accuracy of the FRA
measurement results
and allow them to be
practically applied to
(b) detect and diagnose
Figure 1.1. (a) Damaged on-load tap changer and (b) damaged B phase of HV winding winding deformations
98 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023
This needs for a reliable benchmark
measurement taken when the trans-
The key electrical parameters that influ-
former is new, but which might not be ence frequency characteristics of winding
needed for several decades, drives stan-
dardisation in both the measurement are equivalent air core inductance, ground
device and the configuration of the capacitance and series capacitance, which
transformer when the measurement is
taken. We need to understand the sourc- are dependent upon the type of winding
es of variation first and then standardise
the measurement methods to reduce
them. This was achieved with CIGRE
working group A2.26 [3] and then the requirements, e.g., Sweep Frequency Through standardisation of measure-
development of IEC 60076-18, which Impedance (SFI) which measures the ments and connections, a considerable
introduced standard measurement con- short-circuit transformer impedance volume of FRA fingerprints for various
figurations and reporting [4]. versus the frequency [7]. In SFRA, the transformers at different voltage levels
end-to-end open circuit measurement and power ratings have been accumulat-
The standardised measurement meth- measures the transfer admittance of the ed by the electrical power industry over
ods following IEC 60076-18 greatly winding, while the secondary winding is the last decades. At present, the assess-
help to ensure the accuracy of the FRA open-circuited, whereas, in SFI, it mea- ment of FRA results demands examina-
measurement results and allow this sures the impedance of the winding, tion by skilled personnel as there is no
comparative technique to be practically when the secondary winding is short- reliable diagnostic or pass/fail criterion
applied to detect and diagnose wind- ed-circuited. Thus, the short-circuit available in IEC and IEEE standards
ing deformations. The “standard” FRA impedance at power frequency can be [4,6]. The lack of such criteria is due to
measurement connection method tends extracted from SFI. several complicating factors, including
to be the winding end-to-end open cir-
cuit measurement chosen for its sensi-
tivity to most deformations [5]. It has
to be recognised, however, that one of FRA plots of windings with low series ca-
the aims of the standard was to set out
a minimum set of benchmark measure- pacitance have more resonances and an-
ments, a very small subset of the possi- ti-resonances in high frequencies, in con-
ble combinations of winding connection
and tap-changer position, which could trast to those with high series capacitance
be used. A much wider set of measure-
ments might be useful for diagnosing
which tend to have a much smoother and
particular faults. rising trend
Direct comparison between the newly
obtained FRA trace with the fingerprint
will often yield the “normal – no devi-
ation” diagnostic result. However, de-
viations may sometimes appear, which
suggest an “abnormal” diagnostic result,
and require further detailed diagnosis,
for instance, end-to-end short-circuit
measurements may be applied when it
is necessary to eliminate the influence
of the transformer core, and inter-wind-
ing measurements can be considered if
the effect of inter-winding coupling on
FRA results is of concern. In parallel to
the development of IEC 60076-18, IEEE
standard C57.149 [6] also recommends
a minimum set of test connections for
different types of transformers, and the
end-to-end open circuit and end-to-
end short circuit measurements appear
to be the minimum set for each wind-
ing. In addition. Some countries and
utilities have also developed technical
guides with additional measurement Figure 2.1 Typical FRA traces of different winding types [8]

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 99
UTRA COLUMN

Numerous indices have been proposed, and caused by that damage to provide a pass/
fail criterion. In this context, the stan-
the performance of the numerical indices dard IEC 60076-18 is being revised to
can be characterised by five criteria, includ- incorporate the significant additional
site measurement experience accumu-
ing monotonicity, linearity, sensitivity, data lated since it was first published, as well
as university research-based progress on
size dependency, and index ratio modelling and simulations. This article
sets out to review the research work so
far carried out in the UTRA universities
the variety and sophistication of wind- active research, some of which are de- on the topic of FRA interpretation.
ing designs and the influence of core scribed in this article. The research aims
magnetisation at low frequencies and are twofold but linked: to understand
2. Winding design and how it
cables and earthing at high frequencies, and model in detail how winding fre-
preventing the use of simple algorithms. quency response is determined by wind- influences FRA traces
Interpretation of FRA results is, there- ing type, geometry and damage, and to Two types of winding design are com-
fore, still a challenge and the subject of be able to enumerate reliably deviations monly used in core-type transformers,

Figure 3.1 Application of a numerical index for FRA diagnosis [11]

100 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023

which are helical and disc windings. space coefficient , is also critical in de- current, low voltage applications such
The key electrical parameters of wind- termining the features of the winding’s as generator transformers where BIL is
ing are equivalent air core inductance FRA trace. lower than 325 kV. In a helical winding
(L), ground capacitance (Cg) and series structure, winding conductors are placed
Helical winding
capacitance (Cs), and Cs is most de- around the core in a way to form a he-
pendent upon the type of winding. The Helical windings are mainly used in high lix where conductors are parallel to each

Figure 3.2. Forced buckling of LV winding

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 101
UTRA COLUMN

A winding assessment factor called Stan- from 2 Hz to 2 kHz, winding interaction,

which means the inductive and capaci-
dard Deviation of Difference is proposed for tive coupling from the non-tested wind-
quantitatively identifying the degree of de- ing to the winding under test, dominates
the FRA feature in the region from 2 kHz
viation between two FRA signatures to 20 kHz, and the features dominated
by the winding itself will also be visible
in the frequency region from 20 kHz to
other in the radial direction. Each lay- a disc-type winding, turns of conductors 200 kHz [9]. In Figure 2.1, it can be seen
er of helical winding consists of a small are placed in a flat disc, and these discs clearly that FRA plots of windings with
number of winding turns, and typically a can be connected in series or parallel de- low series capacitance have more reso-
winding has a couple of layers. Typically, pending on the design considerations. nances, and anti-resonances in high fre-
the series capacitance of helical winding The individual conductors in the disc quencies, in contrast to those with high
is comparably smaller, and therefore, are insulated from each other. Press- series capacitance which tend to have a
its FRA trace usually has multiple reso- board spacers are mounted between much smoother and rising trend in the
nance points in high frequencies. Due to discs to acquire the required mechanical region from 20 kHz to 200 kHz.
LV voltage application, the FRA magni- strength while maintaining oil circula-
tude of helical winding also tends to be tion through windings. Typically, the
3. FRA numerical indices
higher than the other types of windings. series capacitance of plain disc winding
(also called continuous disc winding) is FRA diagnosis can be assisted by em-
If multiple layers are used in a high-volt- higher than helical winding, and inter- ploying numerical indices, which quan-
age winding design, they tend to have a leaved disc winding has an even higher tify the differences in two FRA signa-
high series capacitance. series capacitance. tures [10- 11]. A mathematical equation
is used to extract a single value from the
Disc-type winding
Typical FRA measurement results of reference and present FRA traces. The
Disc windings are commonly used in different winding types are presented in assessment of the transformer condition
power transformers due to their me- Figure 2.1, where the x-axis from 5 Hz to is then carried out based on the quanti-
chanical stability. In contrast to helical 1 MHz is in the logarithmic scale. Em- fied value of the indices, as depicted in
winding, disc-type windings are typi- pirically, the transformer core dominates Figure 3.1.
cally used with higher voltage levels. In the FRA feature in the frequency region
Some researchers used the FRA data
in the whole frequency range, whereas
others divide the whole frequency range
into several frequency regions to reach
reliable assessments. IEC and IEEE stan-
dards [4,6] stated that the division of fre-
quency regions is not standardised yet.

Numerous indices have been proposed,

and the performance of the indices can
be characterised by five criteria, including
monotonicity, linearity, sensitivity, data
size dependency, and index ratio [11-12].
To evaluate and compare the character-
istics of indices, different case studies
are considered through an experimental
setup which consists of a single-phase
transformer with HV and LV windings.
The windings correspond to a medium
voltage transformer of about 1 MVA. The
HV winding is a continuous disc winding,
and the LV winding is a helical winding.
In the experimental setup, three com-
monly occurring mechanical faults, axial
displacement (AD), radial deformation
(RD), and disk space variation (DSV) are
implemented in various steps, as shown in
Figure 3.2 and Figure 3.3.

These implemented mechanical faults

Figure 3.3 Axial collapse of the disk-type winding result in detectable resonant frequency

102 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023

Table 3.1. The equations of the winding assessment factor

shifts and amplitude changes in the FRA Understanding the physical meaning be-
measurement results [11].
hind the frequency responses of winding,
In this context, a winding assessment which is the relationship among winding
factor called Standard Deviation of Dif-
ference (SDD) is proposed for quanti- geometry, transformer equivalent electri-
tatively identifying the degree of devia-
tion between two FRA signatures [11].
cal components and corresponding FRA
In this method, the mean deviation of spectra, is key
difference is calculated to evaluate the
degree of deviation between two FRA
signatures, which is further standardised
with the total number of measurement
samples, thus, called Standard Deviation
of Difference (SDD). In this proposed
method, SDD is evaluated in a frequen-
cy window with a specific window size
(WS). The window slides from the start-
ing frequency to the ending frequency
of the FRA trace with a specific window
step (Wstep=1), as shown in Figure 3.4.
Consequently, the entire frequency
range of the FRA signature is scanned,
and SDD is evaluated in each window.
Accordingly, a vector is obtained that
characterises the deviation between two
FRA signatures as a function of frequen-
cy. Hence, the calculated SDD and the Figure 3.4 The basic principle of the winding assessment algorithm [13]
measured FRA signatures can be drawn
on the same scale in a graph, as shown
in Figure 3.5. The idea was first present-
ed in [14] and proved to be an effective
method for quantifying deviations be-
tween two FRA signatures in many cas-
es. The equations of the winding assess-
ment factor are given in Table 3.1.

Thus, the proposed method can detect

faults with high sensitivity, and it also
resolves the problem of fixed frequency
sub-bands. The minimum value of SDD
(MSDD) indicates the maximum devia-
tion between two FRA traces. It is worth
noting that MSDD has lower values for Figure 3.5 Application of the proposed method for identification of the degree of
the cases with deformations than those deviation between two FRA signatures of a three-phase transformer [13]

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 103
UTRA COLUMN

in normal cases. Thus, it is possible to

set a threshold to MSDD as a diagnostic
criterion, e.g., in [11] MSDD = -5 was
employed as a threshold for time-based
FRA comparisons; MSDD <- 5, indicates
abnormal condition whereas MSDD> -5
indicates normal condition [11]. Note
the diagnostic criterion is only valid for
these case studies.

The paper [7] was developed based on

SFI measurements, which were applied
to large power transformers of more
than 20 units of 110 kV and 5 units of
220 kV. The frequency range of SFI is
from 10 Hz to 1 MHz, with a frequency
resolution step of less than 2 Hz around
the power frequency. The paper [7] rec-
ommended that four parameters, in-
cluding the impedance value at power
frequency, the slope rate of the SFI trace
between 10Hz and 500Hz, the resonant
frequencies, as well as the correlation
(a) Lumped ladder network model [15]
coefficient between the measured trace
with the fingerprint, can be used for di-
agnosis. Appendix D of [7] gave thresh-
old values for the correlation coefficient
to determine the “normal – slight – obvi-
ous – severe deformed” winding status.

4. Interpretation of FRA
features
As a comparative technique, the difference
between FRA reference and measurement
results can indicate the existence of wind-
(b) Distributed multi-transmission line model [16] ing mechanical faults. Therefore, under-
standing the physical meaning behind the
frequency responses of winding, which is
the relationship among winding geome-
try, transformer equivalent electrical com-
ponents and corresponding FRA spectra,
is key. Numerous sensitivity studies, either
by experiment or simulation, have been
implemented to gain an understanding of
FRA physics, but a widely accepted gener-
al FRA interpretation guidance still needs
to be developed.

The challenge re-

mains to work back
from the FRA result to
determine the trans-
former winding char-
acteristics although
(c) Finite element model [17]
progress has been
Figure 4.1 Different white box modelling methods
made
104 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023
Table 4.1 Comparisons among white box models

Models Advantages Drawbacks

Physical meaning, accurate, wider fre-

Lumped element network model quency range up to 1-2 MHz. can repre- Complex, long time for calculation
sent the core, three phases.

Physical meaning, accurate, wider fre- Not sufficiently practical for the whole
Distribution multi-transmission line
quency range up to 10 MHz, specialised in transformer, for certain deformation simu-
model
the fast transient study lation

Physical meaning, accurate, flexible in Long computation time and large memory
Finite element model
performing practical deformation of computer required

4.1 Different modelling methods

Three types of transformer models have

been developed for the FRA study: the
white box model, the black box model and
the grey box model. White box models are
based on the laws of physics. They consist of
a large number of lumped circuit elements
whose parameters are calculated from
transformer design data. Three types of
white box models are used. The first model
is lumped element network model. As
shown in Figure 4.1 (a), it divides the wind-
ing into several sections, and each section
is represented by electrostatic and electro-
magnetic components, including series
and ground capacitance, magnetising in-
ductance, air core and mutual inductance.
This model is the most widely used white
box model. The second model is distrib-
uted parameters model. As shown in Figure 4.2 Reference FRA and the FRA reproduced by the estimated geometry [18]
Figure 4.1 (b), it is constructed based on
the multi-transmission line theory, and
each turn is represented by a transmission
line. Transmission line theory is applied to
The grey box model demonstrates the
calculate the voltage and current vectors at physical behaviour of the transformer with-
each sending and receiving end. The third
model is the finite element model. As il- out the design input from the manufacturer
lustrated in Figure 4.1 (c), it requires the
usage of FEM software, such as COMSOL
Multiphysics® or ANSYS. A 3D geometrical results when all transformer design details culated results match with measurement
model can be built in the design module of are available [9,15], and the challenge re- results with good accuracy. However, the
ANSYS, followed by a meshing process. mains to work back from the FRA result black box model circuit does not provide
to determine the transformer winding any information about internal voltages
The electrical potential of each small sec- characteristics although progress has been and currents. The parameters of black box
tion in the meshed grid is solved through made on this as described later in this models have no direct link with the trans-
stepwise physical equations. Table 4.1 sum- article. former’s geometrical dimensions.
marises the main advantages and disadvan-
tages of the three types of white box model- The black box model is constructed pure- The grey box model is in between the
ling methods. The white box models rely on ly based on FRA measurement data. The white box and black box models. It has
the assistance of the transformer manufac- model circuit is established according to the same physical-laws-based circuit
turer to provide design information. Signif- the number of resonances appearing on model as a white box model, but the
icant success has been achieved in building the FRA trace. Values of electrical param- model parameters are estimated using
‘white box’ models of transformers capable eters in the circuit are adjusted by algo- terminal measurement data without any
of accurately reproducing experimental rithms to reproduce the FRA until the cal- knowledge of the internal geometries.

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 105
UTRA COLUMN

Figure 4.3 Comparison of frequency response traces among measurement, GA and GA combined with IA [19]

agnose the winding deformation, i.e., the

With the combination of Genetic Algo- deformation position localisation and se-
rithms and Iterative Algorithms, the ac- verity evaluation.

curacy and efficiency of the identification 4.2 Understanding FRA features

process can be greatly enhanced through modelling and simulation

FRA analytic interpretation has been

developed for a single winding structure
The grey box model demonstrates the mated winding geometry parameters by by taking advantage of the lumped lad-
physical behaviour of the transformer using the GA. der network model and understanding
without the design input from the man- the transient behaviours of the voltage
ufacturer. Hence it is useful for the utili- The combination of the Genetic Algo- propagation along the winding. The FRA
ty to manage its in-service transformers. rithm (GA) and Iterative Algorithm (IA) responses for different winding struc-
The grey box model parameters are iden- was used to identify the ladder network tures have been categorised in terms of
tified using parameter estimation meth- parameters of the tested single winding the high- or low-series capacitance , as it
ods, and Artificial Intelligence algorithms from its FRA traces [18,20]. In the first controls the spacing coefficient , which is
are often employed to assist this optimis- step, GA searches for the suboptimal regarded as a key factor determining the
ation process. ANN, GA and PSO are the solutions of the network parameters degree of non-linear distribution of the
most popular meta-heuristic methods for among given ranges of network parame- impulse voltage along the winding [15].
optimisation problems. Using these algo- ters by comparing their FRA traces with The FRA responses of windings of high
rithms, the electrical parameters of trans- the measurement. In the second step, series capacitance exhibit an increasing
former equivalent circuits or even the IA computes and optimises the correc- trend of magnitude across the frequency
geometrical dimensions of windings can tion values by comparing the differences range with little resonances and anti-res-
be estimated. The differences between between two FRA traces at consecutive onances, as exemplified by interleaved
the FRA fingerprints generated from the iterative steps. Therefore, with the com- disc-type winding in Figure 4.4(a). The
estimated and reference values are used bination of GA and IA, the accuracy and windings of low series capacitance display
as a quantitative indicator to assess the efficiency of the identification process the steady magnitude trend in their FRA
optimisation accuracy. Figure 4.2 shows can be greatly enhanced. Several FRA responses with the features of resonances
a comparison between the reference FRA traces obtained in GA and IA are shown and quasi-antiresonances or anti-reso-
of a single winding and the FRA gener- in Figure 4.3. The final obtained ladder nances, as shown in Figure 4.4 (b) for a
ated by the grey box model, which esti- network parameters can be used to di- single layer or a continuous disc winding.

106 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023

FRA analytic interpre-
tation has been de-
veloped for a single
winding structure by
taking advantage of
the lumped ladder net-
work model and under-
standing the transient
behaviours of the volt-
age propagation along
the winding
(a) Interleaved disc winding (Cg=144 pF, Cs=278pF, α=0.72)

Based on the mathematic descriptions of

lumped element ladder network model,
the equations in Table 4.2.1 have been
derived from providing accurate predic-
tions of the pseudo-anti-resonant, reso-
nant and critical anti-resonant frequen-
cies [16]. These equations are only valid
for the simplified model of a single air
core winding discussed here.

From the formulas, it can be noticed that

α has the dominating effect on the shape
of frequency response. These equations
explain well why windings with low series
capacitance, such as plain disc windings
and single helical windings, always have
multiple resonances, whereas windings
with high series capacitance, such as in- (b) Continuous disc winding (Cg=168 pF, Cs=7pF, α=4.89)
terleaved disc type windings, always have Figure 4.4 Typical FRA responses of different winding structures [15]

Table 4.2.1. Equations for pseudo-anti-resonant, resonant and critical anti-resonant frequencies

Description Definition Equation Numbering

Local minimum between the adja-

cent resonances, where the fre-
quency response amplitude would
Pseudo-anti-resonance be concave without the sharp (2.1)
drop into the large negative dB
value as for the normal anti-reso-
nance, showing a U-shape.

A local maximum in amplitude and

sometimes resonant frequencies
Resonances (2.2)
are also called natural frequencies
of winding

Where the frequency response

‘Critical’ anti-resonance amplitude would drop sharply to a (2.3)
large negative dB value.

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 107
UTRA COLUMN

A double-peak feature is commonly ob- monly observed in medium frequency

regions for a transformer with more than
served in medium frequency regions for two windings, e.g., an autotransformer.
supergrid autotransformers in the UK Sensitivity studies have shown the first res-
onance of the double peak is mostly depen-
dent on the three-phase coupling through
the tertiary delta connection, while the
less or no resonance. All the formulas are When two windings are incorporated into second resonance of the double peak is
published in CIGRE TB812 [12], and the model, the influence of coupling be- produced as the series and the common
further extrapolations of analytical equa- tween windings becomes non-negligible. windings are strongly coupled due to their
tions for a single winding with an iron The mutual inductance and the interwind- relatively close turn ratio [9].
core with mutual inductive coupling con- ing capacitance between two windings are
sidered can be found in [12,16]. considered. A double-peak feature is com-
5. Conclusions
Significant progress has been made to-
wards understanding and modelling
FRA results in UTRA universities. It is a
Authors part of the pathway to developing reliable
Yaoxian Yang (M.Sc.–2016, B.Eng.–2015) is a Ph.D. pass/fail analysis and criteria for practical
candidate in High Voltage Technology in the Department use. Ultimately the goal of introducing
of Electrical and Electronic Engineering at the University of methods and criteria for FRA interpre-
Manchester, United Kingdom. He is also an associate of the tation into standards appears achievable.
Centre for Smart Grid at the University of Exeter. His main
research interests include the study of transformer condition Bibliography
monitoring techniques, modelling and simulation, [1] E. P. Dick, C. C. Erven, Transform-
frequency response analysis and transformer digital twin. er Diagnosis Testing by Frequency Re-
sponse Analysis, IEEE Trans. on Power
Shuhang Shen (Ph.D.–2020, M.Sc.–2016, B.Eng.–2013) Apparatus and Systems, Vol. PAS-97,
pp. 2144–2154, Nov/Dec 1978
is a Lecturer at the Centre for Smart Grid at the University
of Exeter, Exeter, U.K. His research interests include [2] P. T .M. Vaessen, E. Hanique,
condition monitoring of power equipment, electrical A New Frequency Response Analy-
modelling and operation of power networks in low- sis Method for Power Transformers,
carbon energy scenarios. IEEE Trans. on Power Delivery, Vol. 7,
pp. 384–390, Jan. 1992
Fan Zhang (Ph.D. –2019, B.Sc. –2014) worked as a joint
[3] CIGRE WG A2.26, CIGRE Techni-
Ph.D. student in the School of Information Technology and cal Brochure 342 - Mechanical Condi-
Electrical Engineering, at the University of Queensland, tion Assessment of Transformer Wind-
Brisbane, Australia from 2018 to 2019. He is currently an ings, 2008
Associate Professor at Xi’an Jiaotong University. His research
interests are multiphysics modelling, condition monitoring [4] IEC 60076-18:2012, Power trans-
of power equipment, and signal processing. formers – Part 18: Measurement of fre-
quency response
Shengchang Ji (Ph.D.–2003, B.Sc.–1998) is a Professor [5] J. A. S. B. Jayasinghe, Z. D. Wang,
at Xi’an Jiaotong University. His research work includes A. W. Darwin, P. N. Jarman, Wind-
investigations on vibro-acoustics characteristics and ing Movement in Power Transformers:
condition monitoring of power equipment. A Comparison of FRA Measurement
Connection Methods, IEEE Trans. on
Dielectrics and Electrical Insulation,
Chandima Ekanayake (SMIEEE, Ph.D.–2006, Tech.
Vol. 13, pp. 1342–1349, Dec. 2006
Lic–2003, B.Sc.–1999) is a Senior Lecturer in the School of
Information Technology and Electrical Engineering at the [6] IEEE Std. C57.149-2012, IEEE
University of Queensland, Australia. His research focuses Guide for the Application and Interpre-
on instrumentation for condition monitoring & protection tation of Frequency Response Analysis
of power system assets and developing ageing models based for Oil-Immersed Transformers
on laboratory and field measurements to quantify the ageing
of assets due to multi-factor ageing. [7] T/CEC 201-2019, Guidelines for
Detection and Diagnosis of Winding De-

108 TRANSFORMERS MAGAZINE | Volume 10, Issue 2 | 2023

formation of Power Transformers by the [17] M. F. M. Yousof, C. Ekanayake, [19] Y. X. Yang, S. Shen, Z. D. Wang, Dig-
Sweep Frequency Impedance Method T. K. Saha, H. Ma, A Study on Suitabil- ital Twinning of Helical Type Windings
ity of Different Transformer Wind- Based on FRA Data and GA Optimi-
[8] X. Mao, Z. D. Wang, P. Crossley et ing Models for Frequency Response sation, 2022 IEEE International Confer-
al., Transformer Winding Type Recog- Analysis, 2012 IEEE Power and En- ence on High Voltage Engineering and
nition Based on FRA Data and a Sup- ergy Society General Meeting, 2012, Applications (ICHVE), 2022, pp. 1-6
port Vector Machine Model, High Volt- pp. 1–8
age, no. 5, pp. 704–715, 2020 [20] F. Ren, Y. Xiao, C. Zhan, Ladder
[18] F. Ren, S. Ji, Y. Liu, Y. Shi, L. Zhu, Ap- Network Parameters Identification of
[9] D. M. Sofian, Z. D. Wang, J. Li, plication of Gauss-Newton Iteration an Isolated Winding by Combining the
Interpretation of Transformer FRA Algorithm on Winding Radial Defor- Intelligent Optimisation Algorithm
Responses— Part II: Influence of mation Diagnosis, in IEEE Transac- and GNIA, IET Generation, Trans-
Transformer Structure, IEEE Trans. tions on Power Delivery, vol. 34, no. 4, mission & Distribution, vol. 13, no. 2,
on Power Delivery, vol. 25, no. 4, pp. 1736–1746, Aug. 2019 pp. 296-304, Jan. 2019
pp. 2582–2589, Oct. 2010

[10] P. Picher, S. Tenbohlen, M. Lachman

et al., Current State of Transformer FRA Authors
Interpretation, Procedia Engineering, Tapan Kumar Saha (FIEEE, Ph.D.–1994, M. Tech–1985,
pp. 3–12, Vol. 202, 2017 B.Sc.–1982) is a Professor of Electrical Engineering at
the School of Information Technology and Electrical
[11] M. Tahir, S. Tenbohlen, S. Miyaza- Engineering at the University of Queensland, Australia.
ki, Analysis of Statistical Methods for
His research interests include condition monitoring of
Assessment of Power Transformer
Frequency Response Measurements, electrical plants, power systems and power quality.
IEEE Trans. on Power Delivery, Vol. 13,
pp. 618–626, Apr. 2020 Stefan Tenbohlen (Dr. Ing.–1997, diploma–1992) is a
Professor and Head of the Institute of Power Transmission
[12] CIGRE WG A2.53, CIGRE Tec­h­
nical Brochure 812 - Advances in the and High Voltage Technology at the University of Stuttgart,
interpretation of transformer Frequen- Germany. His main research fields are high voltage
cy Response Analysis (FRA), 2020 technique, power transmission and electromagnetic
compatibility (EMC).
[13] M. Tahir, S. Tenbohlen, Novel Cal-
culation Method for Power Trans-
former Winding Fault Detection using Paul Jarman joined the Central Electricity Generating
Frequency Response Analysis, 5th In- Board in the UK as a Research officer in 1984 where
ternational Colloquium “Transformer amongst other projects he worked on developing and
Research and Asset Management” (IC- introducing the FRA technique to the networks in the
TRAM’05 2019), Opatija, Croatia, Octo- UK. Following privatisation, he joined National Grid and
ber 09–12, 2019, pp. NM-10 worked on all aspects of transformer technology, including
testing, procurement, asset management and international
[14] M. Tahir, S. Tenbohlen, A Compre-
hensive Analysis of Windings Elec- standards. While with National Grid Paul worked with the
trical and Mechanical Faults Using University of Manchester on several research projects related to FRA. He was a
a High-Frequency Model, Energies, member of the IEC project team for IEC 60076-18, was chair of IEC TC14 and
vol. 13, no. 1, p. 105, Dec. 2019 is presently convenor of IEC MT 60076-1. In 2018 Paul left National Grid and
became a Professor at Manchester on a part-time basis in 2019.
[15] Z. D. Wang, J. Li, D. M. Sofian, In-
terpretation of Transformer FRA Re- Zhongdong Wang (FIEEE, FIET, Ph.D.–1999,
sponses— Part I: Influence of Winding M.Sc.–1993, B.Eng.–1991) is a Professor of Electrical
Structure, IEEE Trans. on Power Deliv-
Power Engineering and Head of Exeter Energy, Director of
ery, vol. 24, no. 2, pp. 703–710, April 2009
Centre for Smart Grid at the University of Exeter, United
[16] B. Cheng, Z. D. Wang, P. Crossley, Kingdom. Her research interests include smart grids
Using Lumped Element Equivalent and low carbon electrical energy networks, in particular
Network Model to Derive Analyti- condition monitoring techniques, thermal, electrical
cal Equations for Interpretation of transient and magnetic modelling techniques for power system networks,
Transformer Frequency Responses, dielectric insulation ageing mechanisms, alternative environmentally friendly
in IEEE Access, vol. 8, pp. 179486179496, liquid insulating materials, and transformer asset management.
2020

w w w . t ra n sfo r m e r s - m a g a z i n e . co m 109