1

Automatic Classification of Field Winding Faults in

Synchronous Motors based on Bicoherence Image
Segmentation and Higher Order Statistics of Stray
Flux Signals
Miguel E. Iglesias-Martínez, Jose Guerra Carmenate, Jose A. Antonino Daviu , Senior Member, IEEE,
Larisa Dunai , Member, IEEE, Carlos A. Platero, Senior Member, IEEE, J. Alberto Conejero, and Pedro
Fernández de Córdoba

Abstract— In this work, the application of the bicoherence (a as cage induction motors This is due to the inherent drawbacks
squared normalized version of the bispectrum) of the stray flux that WRSM have, namely: they need auxiliary elements
signal is proposed as a way of detecting faults in the field winding (variable speed drives, damper cages) or machines (pony
of synchronous motors. These signals are analyzed both under the
motors) for their starting, they have more complex constructive
starting and at steady state regime. Likewise, two quantitative
indicators are proposed, the first one based on the maximum characteristics, they have higher cost, etc… [3]. Despite these
values of the asymmetry and the kurtosis of the bicoherence problems, WFSM can be found in some applications involving
matrix obtained from the flux signals and the second one relying high output powers, since they show interesting features at
on an algorithm based on the bicoherence image segmentation of those power ranges, such as higher efficiencies and possibility
the obtained pattern for each analyzed state. The results are of power factor regulation [4-5]. Due to this, it is not so unusual
analyzed through a comparative study for the two considered
to find WFSM rated several MW that operate in certain plants,
motor regimes, obtaining satisfactory results that sustain the
potential application of the proposed methodology for the where they often are critical assets. An interesting case of a
automatic field winding fault detection in real applications. catastrophic failure in one of these machines, that implied
Index Terms— Bicoherence, Motors, Skewness-Kurtosis, Flux, losses of several million $ is reported in [6].
Winding Faults One of the failures that can happen in WFSM and that can
lead to very negative repercussions is the fault in the field
I. INTRODUCTION winding of these machines. This may happen due to the

C
ondition monitoring of electric motors is of crucial deterioration of the insulation between field winding turns
importance in the industry due to the primordial role of (caused by a diversity of reasons) which can progress to more
these machines that can drive a huge variety of loads in severe failures and even to the forced outage of the machine [7].
a vast range of different applications. Faults in the electric In this sense several works have been carried out using stray
motors can yield unplanned production downtimes, costly flux signals for this purpose [8-12], then specifically using
repairs and even user safety hazards. This is the reason why it wound field synchronous motors [13-14]. Despite the variety of
becomes crucial to develop techniques that are able to reliably works that have analyzed this problem, some issues still remain
detect possible failures or defects in the different components as the need of expertness of the user that interprets the results
of the electric motor when these failures are still at their early of some of the proposed methods, a fact that complicates their
stages of development in the machine [1]-[2]. implementation in autonomous condition monitoring devices
Wound field synchronous motors (WFSM) are far less that can be applied in the field.
widespread in industry than other types of electric motors, such This work proposes the analysis of stray flux signals both

Miguel E. Iglesias Martínez’s work was supported by the postdoctoral Jose Guerra Carmenate is with the Departamento de Telecomunicaciones,
research scholarship "Ayudas para la recualificación del sistema universitario Universidad de Pinar del Río Hermanos Saíz Montes de Oca, Pinar del Río
español 2021-2023. Modalidad: Margarita Salas", UPV, Ministerio de 20100, Cuba ([email protected]).
Universidades, Plan de Recuperación, Transformación y Resiliencia, Spain. Jose A. Antonino-Daviu is with the Instituto Tecnológico de la Energía,
Funded by the European Union-Next Generation EU. This work is also Universitat Politècnica de València, E-46022 Valencia, Spain (e-mail:
supported by Generalitat Valenciana (reference CIAICO/2021/020) [email protected]).
Miguel Enrique Iglesias Martínez is with InterTech Interdisciplinary Larisa Dunai is with the Centro de Investigación en Tecnologías Gráficas
Modeling Group from the Instituto Universitario de Matemática Pura y Universitat Politècnica deValència, E-46022Valencia, Spain (e-mail:
Aplicada, Universitat Politècnica de València (UPV), E-46022 Valencia, Spain [email protected]).
(e-mail: [email protected]) and with Grupo de Ingeniería Física, Escuela de Carlos Platero is with Universidad Politécnica de Madrid, Calle de José
Ingeniería Aeronáutica y del Espacio, Universidad de Vigo, Edif. Manuel Gutierrez Abascal, nº 2, 28006 Madrid, Spain. ([email protected])
Martínez Risco, 32004 Ourense. Spain. (email: Pedro Fernández de Córdoba is with the Instituto Universitario de
[email protected]) Matemática Pura y Aplicada, Universitat Politècnica de València, E-46022
Valencia, Spain (e-mail: [email protected]).
 2

under starting and at steady-state for the detection of field NxN matrix that contains the frequency values of the amplitude
winding faults in WFSM. To this end, two high-order statistical bispectrum matrix of the analysed stray flux signal [15].
analysis using bicoherence are evaluated through the combined
(𝐵𝑥𝑁 (𝑓)) = |𝐵𝑥𝑁 (𝑓1 , 𝑓2 )|𝑖 ∀𝑖 = 1, … , 𝑁 ()
application of the skewness and kurtosis indicator. While the
bicoherence enables to obtain a visual qualitative representation On the other hand, the bicoherence can be interpreted as an
of the frequency content of the analyzed signals. The second automatic index that can be calculated from a single signal. It
analysis, based on skewness and kurtosis indicator, enables the takes values between 0 and 1, which makes it a convenient
quantitative discrimination between different fault conditions. measure to quantify the degree of phase coupling in a signal.
Moreover, the methodology proposes in this paper enables Then, the Bicoherence, according to equation (2), is defined by
an automatic application of the methodology which is crucial (3), where the numerator contains the square magnitude of the
for its possible implementation in autonomous devices. The bispectrum in all segments of the time series and the
results included in the paper, which represent an incremental denominator is the factor that normalizes the bispectrum.
advance versus those presented in [15], show the potential of Hence, 0 ≤ 𝐵(𝑓1 , 𝑓2 ) ≤ 1.
the approach and its suitability for making the diagnostic user- |𝐵(𝑓1 , 𝑓2 )|2
independent and facilitating its possible future extrapolation to 𝑏𝑖𝑐 2 (𝑓1 , 𝑓2 ) = (3)
the field. 𝐸{|𝑋(𝑓1 )𝑋(𝑓2 )|2 }𝐸{|𝑋 ∗ (𝑓1 + 𝑓2 )|2 }
In synthesis, the main contributions of this work can be listed To verify the validity of the theoretical proposal, several
as follows: experiments are carried out, the results of which are shown
• Proposal of a new methodology for the detection of field below in section IV.
winding faults based on stray flux data analysis; the method
relies on high order spectral analysis for the calculation of the III. EXPERIMENTS
bicoherence. The method enables an easy qualitative detection Different tests were developed using a 4-pole, 400 V WFSM
of the fault, since it is based on visual patterns. rated 3.4 kVA. The motor was started through a 3-phase variable
• Proposal of indicators to quantify the severity of the motor voltage source. The motor was started thanks to the rotor that
fault using the combination of the kurtosis and the skewness of assumed the function of the damper, enabling an asynchronous
the flux signal. The indicator is verified in different regimes, mode starting. The load was an induction motor that acted as
proving its usefulness for the determination of the fault severity. generator. A picture of the experimental test bench is shown in
• Proposal of an automatic fault detection scheme based on the Fig.1.
segmentation of the image contour obtained from the The inter-turn failure was forced by using a variable resistance
bicoherence calculation. This scheme makes the method RF that could be inserted in parallel at the connections between the
suitable for potential incorporation of the method in automatic poles that can be externally accessed (see Fig.2). The three tested
condition monitoring systems. cases were: healthy machine (no resistor connected, i.e., RF = ∞),
The paper is organized as follows: Section II shows the basic partial interturn fault in one pole (connected resistor RF=2.7 Ω) and
fundamentals of the analysis based on bicoherence. In Section complete short-circuited pole (connected resistor RF=0 Ω).
III the experimental details are shown. In Sections IV and V, The stray-flux signals both under the starting and at steady-state
respectively, the obtained results using bicoherence and the were captured using a coil sensor with air core with N1=1000 turns
proposed algorithm based on the image segmentation are and round shape. It had an external diameter of 80mm and an
illustrated. Finally, the conclusions of the work are detailed in internal diameter of 39mm (see Fig. 2).
Section VI.

II. THEORETICAL FOUNDATIONS: BICOHERENCE ANALYSIS

Higher order spectral analysis is based on the Fourier
Transform 𝑋(𝑓) of the higher order cumulants of a discrete
sequence {𝑥(𝑛)}𝑁−1
𝑛=0 and it can be expressed as follows (where
𝑋(𝑓) is the Fourier Transform of the sequence {𝑥(𝑛)}𝑁−1 𝑛=0 )
[16]-[17]:
𝑁 𝑁
𝐵(𝑓1 , 𝑓2 ) = ∑ ∑ 𝐶3𝑥 (𝜏1 , 𝜏2 ) ∙ 𝑒 −2𝜋𝑓1 𝜏1 ∙ 𝑒 −2𝜋𝑓2𝜏2
𝜏1 =1 𝜏2 =1

1
= 𝑋(𝑓1 , 𝑓2 ) ∙ 𝑋(𝑓1 ) ∙ 𝑋(𝑓2 ) ()
𝑁2

In this work, the bispectrum absolute value of the stray flux

signal is employed. The discrete theoretical description is given Fig. 1. Experimental testbench.
by (2), where N is the number of rows of the square matrix
(NxN) obtained from the bispectrum. The obtained result is a
 3

lower corners either at steady state or under starting. Note also

Stray flux sensor that at steady state (Fig. 4 (b)) a slight circular pattern appears
Added in the center, around the central frequencies, which is not
resistances observed under starting (Fig. 4(a)).
(a)

Fig. 2. Inserted resistor to simulate the field winding fault and stray flux sensor
used.

IV. RESULTS USING BICOHERENCE ANALYSIS

A. Bicoherence plots.
Sometimes the use of classic signal processing techniques
such as the FFT does not constitute an accurate indicator to
visually identify a pattern related to the failure in the motor,
since the harmonics can be found overlapping with other non- (b)
fault related harmonics or even with the spectral noise. An
example is depicted in Fig. 3, in which the spectra of the stray-
flux signals corresponding to both motor in healthy state, motor
with moderate field winding failure and motor with severe field
winding failure are shown. Note that a change in the spectral
frequencies pattern linked with the winding failure is not clearly
observable. Due to this, more sophisticated tools as the
bicoherence are proposed in this paper, as a tool to visually
enhance the detection of the failure.

Fig. 4. Bicoherence for healthy condition: (a) under starting, (b) at steady-state

On the other hand, in Fig. 5 (moderate fault level) the

bicoherence patterns are different compared to the healthy case.
The patterns on the corner are not so concentrated and they
spread showing a triangular pattern that expands from the
corner, both for the starting (Fig.5 (a)) and for the steady-state
(Fig. 5(b)) regimes. Moreover, there is a clear circular pattern
in the middle of the contour plot at steady-state regime (Fig. 5
(b)). Finally, Fig. 6 (severe field winding fault level) shows that,
as the level of fault worsens, the triangular patterns are even
Fig. 3. Comparison of the frequency spectrum of the stray flux signals for a wider for both regimes, while the circular pattern in the middle
motor in healthy state (black), moderate field winding failure (red) and severe is clearly noticeable.
field winding failure (blue)
The triangular patterns appearing in the bicoherence plots of
The following figures show the contour plots corresponding Figs. 4 and 5 are clear signatures of the fault and are explained
to the bicoherence of the stray flux signals captured under by the interactions between the fault frequency components that
starting and at steady-state operation. In Fig. 4 (healthy appear when the field winding failure is present. Moreover, the
condition), no clear pattern can be observed in the upper and
 4

width of the pattern is linked with the fault severity, as observed (a)
in these figures. The detection of these patterns is rather simple
and can be carried out by non-expert users. Moreover, they can
be detected by automatic pattern recognition algorithms
programmed to this end.
(a)

(b)

(b)

Fig. 6. Bicoherence for the severe field winding fault condition: (a) under
starting, (b) at steady-state.

In summary, the proposed method is based, first, on

Fig. 5. Bicoherence for the moderate field winding fault condition: (a) under obtaining the bicoherence plots of the healthy machine both
starting, (b) at steady-state.
during the start-up and at steady-state. In addition, quantitative
More specifically, Fig.4 shows a star-shaped pattern that values of the indicators based on skewness and kurtosis must be
corresponds to the interactions between the frequencies of the computed. Afterwards, during the motor life, the bispectrum
harmonics that are already present in healthy condition over the plots must be periodically obtained and the fault indicators
spectrum at steady-state; moreover, a circular pattern can be computed and compared with the healthy ones, detecting
observed around of 0.1 which indicates the phase coupling with deviations that are indicative of the appearance of the fault.
respect to the fundamental frequency. Additionally, there are
also replicas of the interactions between multiple harmonics On the other hand, Fig. 7 shows the analysis of the
and the fundamental component that yield a circular pattern computational cost for evaluating the bicoherence for a data
about 0.2. window ranging from 256 samples to 4096 samples. The
simulation was carried out on a computer with a W10 12GB of
In the subsequent figures corresponding to the faulty field RAM and a third-generation i7 processor. Note that the
winding (Figs. 5 and 6), it can be observed that, at steady state, computational cost rises exponentially as the number of
the external pattern around 0.2 disappears and only the 0.1 samples to be processed increases. However, also note that the
pattern around the fundamental frequency remains. On the other analyses displayed in this paper were performed with only 256
hand, there is a residual noise that is only visible when the samples of the stray flux signals of the motor in each state,
motor has severe damage. The resulting plots clearly differ which would imply a computational cost of only 1.3 seconds.
from that for healthy condition and allow the detection of the
failure.
 5

informational sources for the quantification of the fault severity.

The difference in the values of the method based on kurtosis
reveals a greater range of fault indication, both during start-up
and at steady-state, although in the latter the trend is more
erratic.

TABLE I. TABLE I. VALUE OF THE SKEWNESS INDICATOR FOR

DIFFERENT FAULT CONDITIONS

Under starting
Healthy Moderate fault Severe fault
8.7227 7.1297 7.0359
At steady-state
Healthy Moderate fault Severe fault
9.6385 8.2746 9.1977

TABLE II. TABLE II. VALUE OF THE KURTOSIS INDICATOR FOR

Fig. 7. Computational cost for evaluating the bicoherence for a data window DIFFERENT FAULT CONDITIONS
ranging from 256 samples to 4096 samples .
Under starting
Healthy Moderate fault Severe fault
In the following sections, two quantitative indicators are 88.8614 61.8200 54.7955
proposed to automatically detect the failure and determine the At steady-state
fault severity. The analyzes were performed both during startup Healthy Moderate fault Severe fault
and at steady state. 102.6651 75.1721 94.1826

Other experiments can be carried out using other types of

B. Fault Indicator Based on Skewness and Kurtosis of
quantities such as vibrations or acoustic signals of the motor to
Bicoherence.
extend the results obtained. The generalization of the method
In this work, the use of two fault severity indicators based on only implies obtaining a pattern of the healthy state of the motor
high-order statistical analysis, such as asymmetry (S) and at the beginning, during start-up and at steady-state, which can
kurtosis (k) (see (4)), is proposed. The proposed indicators are be carried out during motor commissioning.
based on the maximum values of the asymmetry and the
kurtosis of the bicoherence matrix both when the motor is at V. BICOHERENCE IMAGE SEGMENTATION
steady state as well as during startup:
This section is intended to deepen in the automatic fault
𝐸(𝑥 − 𝑢)3 𝐸(𝑥 − 𝑢)4 diagnosis, by applying an image processing methodology to the
𝑆= 𝑘=
𝜎3 𝜎4 bicoherence plots obtained in previous sections for each fault
(4) condition and considered regime (see Figs. 4 to 6). Based on
this method, a second fault indicator is proposed.
Then the respective indicators are given by (5), where 𝑢 is the In the bicoherence plots of Figs. 4 to 6, a triangular pattern
mean of the bicoherence, σ is its standard deviation, and E(t) that enabled to detect the fault condition was identified. This
represents the expected value of the t quantity: pattern varies as the failure increases, but there is a certain
graphic similarity in all the triangular obtained patterns.
𝐸( 𝑏𝑖𝑐 2 (𝑓1 , 𝑓2 ) − 𝑢)3 In order to obtain a function that defines the degree of motor
𝐼𝑛𝑑𝑆 = max ( ) fault as a function of the relative changes of the bicoherence
𝜎3
pattern, an algorithm based primarily on binary segmentation
(5) using the Otsu method [18] to filter the relevant information is
proposed. Once a clean pattern is obtained, the information is
𝐸( 𝑏𝑖𝑐 2 (𝑓1 , 𝑓2 ) − 𝑢)4 processed. Otsu's method is a variance-based technique to find
𝐼𝑛𝑑𝑘 = max ( ) the threshold value where the weighted variance between
𝜎4
foreground and background pixels is minimal. The Otsu method
Table I and II show the value of the skewness and kurtosis- [19-20], can be described as follows:
based indicators, respectively, revealing a progressive
decrement under starting as the winding fault worsens. The 𝜎 2𝐵
trend of both indicators during startup follows a common 𝐾𝑖 = (6)
𝜎 2𝐺
decreasing pattern, then in steady state it follows an erratic
trend. In any case, both fault indicators clearly differ between where 𝐾𝑖 is the threshold value, 𝜎 2 𝐵 is a global variance of
healthy and faulty conditions, becoming interesting the entire image and 𝜎 2 𝐺 is the variance between classes. Once
 6

every image is segmented, to process the information, three

fixed-size quadrants are extracted from all the images; hence,
there is a total of nine quadrants, corresponding to the
bicoherence during startup and at steady-state for each
condition (healthy, moderate and severe failure). Fig. 7 and 8
show, respectively, the results of this segmentation process.

Fig. 9. Bicoherence image segmentation at steady state. a) Healthy b)

Moderate Fault c) Severe Fault.

In the previous figures it can be noticed that, unlike what

happens at steady state, the circular pattern of harmonics does
not appear during the startup. Once the segmentation is
obtained, an analysis of the pixels "on/off" is carried out in each
quadrant. The relationship is represented as follows:

Fig. 8. Bicoherence image segmentation during startup. a) Healthy b) 𝐹(𝑒, 𝑎) = 𝑃(𝑥, 𝑦) ∇ 𝑥, 𝑦 ∈ ℝ ()
Moderate Fault c) Severe Fault.

The three selected regions for each picture are: the upper where 𝑃(𝑥, 𝑦) is the point that represents the relation in a
right and lower left corners, representing the variation of the plane.
triangular patterns, and the center region that is the one that A. Function to evaluate the fault severity
differentiates between the starting and the steady state. Once the segmentation is obtained, a fault indicator function
is proposed. It is based on spiraling through a matrix of fixed
size that contains the relevant information of each one of the
selected quadrants.
If the patterns obtained in Figs. 7 and 8 are analyzed, it can
be noticed that, as the fault becomes more severe, a triangular-
shaped area of "off" pixels in the center of the pattern (figures
 7

7b, 7c, 8b, 8c) is generated. When the fault increases, the "on" The updating process of each loop is executed until reaching
pixels are very close to the edges, due to the increase in the the matrix dimensions, represented by the 𝑚𝑎𝑥𝑥 y 𝑚𝑎𝑥𝑦
triangular area (Figs.7c and 8c respectively). On the contrary, variables respectively. Before the completion of the first
in the healthy case, there is a pattern of "on" pixels concentrated iteration of the general loop, all auxiliary variables represented
close to the quadrant center. The idea is to weight with a greater by 𝑎𝑢𝑥𝑥 , 𝑎𝑢𝑥𝑦 are updated. These variables correspond to the
value those pixels that are closer to the edges, since the iteration index for the X and Y quadrant dimension,
underlying hypothesis is that the edges are related to the fault respectively.
while the center to the healthy state. Finally, the cumulative sum of the weights represented by
Based on these ideas, three quadrants are taken in each image each pixel, which in this case corresponds to the variable
for the three considered conditions (see Figs 7 and 8); the 𝑤𝑒𝑖𝑔𝑡ℎ, is updated. This cumulative sum corresponds to the
central quadrant is taken to discriminate between the START- weights of the pixels of the matrix of each quadrant, for each
UP regime and the STEADY-STATE, while the UPPER motor condition.
RIGHT and LOWER LEFT quadrants are used to assess the The whole process is repeated from step 1 to step 3 until the
severity of the motor failure. analysis of all graphical information provided by the
The main idea of the proposed algorithm is to run through segmentation of the bicoherence contour image is covered.
each quadrant following a spiral of the 𝑁𝑥𝑁 quadrant matrix;
in this way, the farthest layer from the center is run in each
iteration (Fig. 10).
A weight with value 100 is initialized and it will decrease
10% of its current value in each iteration, in such a way that the
weight is maximum in iteration 1 (closer to the edge) and
minimum in the last iteration (closer to the center). The matrix
iteration process is shown in Fig. 10.

Fig. 11. Ilustration of: Step 1 of the flowchart description of the functionality
Fig. 10. Spiral matrix algorithm for fault indicator function.
of the proposed method based on bicoherence image segmentation.

The function takes an array as an input, that corresponds to

the quadrant to be analyzed. Starting from a null value,
𝐹𝑎𝑖𝑙𝑢𝑟𝑒 = 0 and using the following operation variables:
• 𝑚𝑎𝑥𝑥 → X matrix Dimension (quadrant)
• 𝑚𝑎𝑥𝑦 → Y matrix Dimension (quadrant)
• 𝑎𝑢𝑥𝑥 → iteration of the matrix by the X dimension
• 𝑎𝑢𝑥𝑦 → iteration of the matrix by the Y dimension
• 𝑤𝑒𝑖𝑔𝑡ℎ → Pixel Weight

A flowchart description of the functionality of the proposed

method that yields an indicator of the severity of motor fault is
shown in Fig. 11, 12 and 13 respectively.
The first step of the iteration process is to calculate the upper
quadrant region of the matrix obtained from the contour
representation of the bicoherence (see Figs. 8 and 9,
respectively).
Subsequently, the regions of the right and lower quadrant are
calculated, and finally the region of the left quadrant. Each
iteration is performed taking into account the variables listed Fig. 12. Ilustration of : Step 2 of the flowchart description of the functionality
above. of the proposed method based on bicoherence image segmentation.
 8

is also proposed. The method is based on segmenting the area

of relevant information and developing a spiral matrix
processing algorithm to obtain a quantitative indicator of the
fault severity.
Under both regimes, it is possible to discern, by using the
bicoherence image pattern whether the motor is healthy, or it
has a certain level of field winding failure. Likewise, it is shown
that the indicators calculated based on higher order central
cumulants as skewness and kurtosis are also a good alternative
to determine the fault severity, showing better results under
starting.
Finally, the results of this work can be generalized and
incorporated into real time motor condition monitoring
systems. The use of high order statistics offers the possibility of
detecting other types of faults given the noise immunity of high
order spectral analysis techniques. This can be an interesting
tool to explore the detection of other types of faults in
synchronous motors as well as to apply the method to other
machines, as induction motors.

Fig. 13. Ilustration of: Step 3 of the flowchart description of the functionality REFERENCES
of the proposed method based on bicoherence image segmentation. [1] Abid, A., Khan, M.T. & Iqbal, J. A review on fault detection and diagnosis
techniques: basics and beyond. Artif Intell Rev 54, 3639–3664 (2021).
[2] Gundewar, S.K., Kane, P.V. Condition Monitoring and Fault Diagnosis
As a result of the application of the proposed algorithm, a of Induction Motor. J. Vib. Eng. Technol. 9, 643–674 (2021).
non-dimensional indicator is obtained that facilitates to [3] Zamudio-Ramirez, R. A. A. Osornio-Rios, J. A. Antonino-Daviu, H.
Razik and R. d. J. Romero-Troncoso, "Magnetic Flux Analysis for the
determine the fault severity; this indicator increases its value as Condition Monitoring of Electric Machines: A Review," in IEEE
the severity of the failure gets worse. The results can be Transactions on Industrial Informatics, doi: 10.1109/TII.2021.3070581.
observed in Fig. 14, in which an upward trend can be observed [4] J. Shin, Y. Park and S. B. Lee, "Flux-Based Detection and Classification
of Induction Motor Eccentricity, Rotor Cage, and Load Defects," in IEEE
for both considered regimes (i.e. the startup and the steady Transactions on Industry Applications, vol. 57, no. 3, pp. 2471-2480,
state), as the fault severity increases. May-June 2021.
[5] M. E. Iglesias Martínez, J. A. Antonino-Daviu, P. F. de Córdoba, J. A.
Conejero and L. Dunai, "Automatic Classification of Winding
Asymmetries in Wound Rotor Induction Motors Based on Bicoherence
and Fuzzy C-Means Algorithms of Stray Flux Signals," in IEEE
Transactions on Industry Applications, vol. 57, no. 6, pp. 5876-5886,
Nov.-Dec. 2021.
[6] J. Antonino-Daviu. V. Fuster-Roig, S. Park, Y. Park, H. Choi, J. Park and
S.B. Lee, "Electrical Monitoring of Damper Bar Condition in Salient-Pole
Synchronous Motors Without Motor Disassembly," in IEEE Transactions
on Industry Applications, vol. 56, no. 2, pp. 1423-1431, Mar-Apr 2020.
[7] J. Yun, S. B. Lee, M. Šašić and G. C. Stone, "Reliable Flux-Based
Detection of Field Winding Failures for Salient Pole Synchronous
Generators," in IEEE Transactions on Energy Conversion, vol. 34, no. 3,
pp. 1715-1718, Sept. 2019.
[8] L. Frosini, “Novel diagnostic techniques for rotating electrical machines
-A review”, in Energies, 2020, 13(19), 5066,
https://doi.org/10.3390/en13195066
Fig. 14. Output and trend of the fault indicator function based on bicoherence
image segmentation. [9] K. N. Gyftakis, P. A. Panagiotou and S. B. Lee, "Generation of
Mechanical Frequency Related Harmonics in the Stray Flux Spectra of
Induction Motors Suffering From Rotor Electrical Faults," in IEEE
VI. CONCLUSION Transactions on Industry Applications, vol. 56, no. 5, pp. 4796-4803,
Sept.-Oct. 2020, doi: 10.1109/TIA.2020.3002975.
This work has proposed a field winding fault detection
[10] X. Wang et al., "Stray Flux-Based Rotation Angle Measurement for
method for WFSM based on the bicoherence analysis of stray Bearing Fault Diagnosis in Variable-Speed BLDC Motors," in IEEE
flux signals. In the paper, the potential of the use of high-order Transactions on Energy Conversion, vol. 36, no. 4, pp. 3156-3166, Dec.
spectral analysis and its combination with statistical indicators 2021, doi: 10.1109/TEC.2021.3079444.
based on cumulants such as asymmetry and kurtosis is assessed. [11] Garcia-Calva T, Morinigo-Sotelo D, Fernandez-Cavero V, Romero-
Troncoso R. Early Detection of Faults in Induction Motors A
Two methodologies for different regimes (starting and steady- Review. Energies. 2022; 15(21):7855.
state) are applied. https://doi.org/10.3390/en15217855
On the other hand, a fault indicator function based on the [12] F. J. M. Dias, G. G. Sotelo and R. d. A. Júnior, "Performance Comparison
of Superconducting Machines With Induction Motors," in IEEE
segmentation of the image patterns obtained from bicoherence
 9

Transactions on Applied Superconductivity, vol. 32, no. 7, pp. 1-5, Oct.

2022, Art no. 5202805, doi: 10.1109/TASC.2022.3188461. Jose Antonino-Daviu (S’04, M’08, SM’12) received
[13] M. F. Shaikh, J. Park, Y. Park, S. B. Lee and J. A. Antonino-Daviu, his M.S. and Ph. D. degrees in Electrical Engineering,
"Electrical Testing for Detection and Classification of Open Damper Bar both from the Universitat Politècnica de València, in
and Shorted Field Winding Failures in Wound Field Synchronous 2000 and 2006, respectively. He also received his Bs.
Motors," in IEEE Transactions on Industry Applications, vol. 58, no. 4, in Business Administration from Universitat de
pp. 4532-4541, July-Aug. 2022, doi: 10.1109/TIA.2022.3167017. Valencia in 2012. He was working for IBM during 2
[14] K. N. Gyftakis, C. A. Platero and J. K. Nøland, "Multi-Parametric years, being involved in several international projects.
Monitoring of Medium-Power Generators with Brushless Exciters under Currently, he is Full Professor in the Department of
Mechanical Faults," 2022 International Conference on Electrical Electrical Engineering of the mentioned University,
Machines (ICEM), Valencia, Spain, 2022, pp. 1524-1529, doi: where he develops his docent and research work. He
10.1109/ICEM51905.2022.9910891. has been invited professor in Helsinki University of Technology (Finland) in
2005 and 2007, Michigan State University (USA) in 2010, Korea University
[15] J. G. Carmenate et al., "Bicoherence and Skewness-Kurtosis Analysis for
(Korea) in 2014 and Université Claude Bernard Lyon 1 (France) in 2015. He is
the Detection of Field Winding Faults in Synchronous Motors using stray
IEEE Senior Member since 2012 and he has published over 300 contributions,
flux signals," 2022 IEEE Energy Conversion Congress and Exposition
including international journals, conferences and books. He is also Associate
(ECCE), 2022, pp. 1-5
Editor of IEEE transactions on Industrial Informatics, IEEE Industrial
[16] M. E. Iglesias-Martínez, P. Fernández de Córdoba, J. A. Antonino-Daviu Electronics Magazine, IEEE transactions on Sustainable Energy, IEEE
and J. Alberto Conejero, "Bispectrum Analysis of Stray Flux Signals for transactions on Cybernetics and IEEE Journal of Emerging and Selected Topics
the Robust Detection of Winding Asymmetries in Wound Rotor Induction in Industrial Electronics. He is also IEEE IAS Prominent Lecturer for 2022-23.
Motors," 2020 IEEE Energy Conversion Congress and Exposition He was General Co-Chair of IEEE SDEMPED 2013 and ICEM 2022. He
(ECCE), 2020, pp. 4485-4490. received the Nagamori Award from Nagamori Fundation in Kyoto, Japan in
[17] M. E. I. Martínez, J. A. Antonino-Daviu, C. Platero, L. Dunai, J. A. 2018, for his contributions in electric motors transient analysis area. In 2019,
Conejero and P. F. de Cordoba, "Multifractal Spectrum and Higher Order he was awarded with the SDEMPED Diagnostic Achievement Award in
Statistics for the Detection of Field Winding Faults in Wound Field Toulouse, France.
Synchronous Motors," 2021 IEEE 19th International Conference on
Industrial Informatics (INDIN), 2021, pp. 1-7. Larisa Dunai (M´19), Associate Professor at UPV,
[18] Otsu, N., "A Threshold Selection Method from Gray-Level Histograms." obtained her MSc degree in Electronic Engineering in
IEEE Transactions on Systems, Man, and Cybernetics. Vol. 9, No. 1, 2003 from Technical University of Moldova and a
1979, pp. 62–66. Master degree in Electronic Engineering in 2004 at the
[19] W. A. Mustafa, W. Khairunizam, Z. Ibrahim, S. A. B. and Z. M. Razlan, same university. After obtaining the MSc degree joined
"A Review of Different Segmentation Approach on Non Uniform the Technical University of Moldova as Assistant
Images," 2018 International Conference on Computational Approach in professor at the Radio electronics and
Smart Systems Design and Applications (ICASSDA), 2018, pp. 1-6, Telecommunications Department. In 2007 she started
[20] Jaiswal, S., Pandey, M.K. (2021). A Review on Image Segmentation. In: working as a researcher in the Research Center in
Rathore, V.S., Dey, N., Piuri, V., Babo, R., Polkowski, Z., Tavares, Graphic Technology (CITG) of the Universidad Politecnica de Valencia. In
J.M.R.S. (eds) Rising Threats in Expert Applications and Solutions. November 2008 she joined the UPV as an Assistant professor of Graphic
Advances in Intelligent Systems and Computing, vol 1187. Springer, Design Department. In 2010 obtained her PhD at Universitat Politècnica de
Singapore. València. In 2013 she received the MIT Innovators Award for Spain and in
2014 the Michael Richey Medal from Royal Institute of Navigation.

VI. BIOGRAPHIES Carlos A. Platero (M’10-SM’20) was born in Madrid,

Spain, in 1972. He obtained his Dipl. degree and Ph.D.
Miguel Enrique Iglesias Martínez received a degree degree in Electrical Engineering from the Universidad
in Telecommunications and Electronics Engineering Politécnica de Madrid, Spain, in 1996 and 2007
from the University of Pinar del Río (UPR) in 2008 respectively. From 1996 to 2008 he worked in ABB
and a Master's Degree in Digital Systems from the Generación S.A., Alstom Power S.A. and ENDESA
Technological University of Havana, Cuba, in 2011. Generación SA, and was involved in the design and
In 2020 he received his doctorate in Mathematics at commissioning of power plants. In 2002 he started
Universitat Politècnica de València, obtaining best teaching at the Electrical Engineering Department of the
dissertation award in the area of science. He is Universidad Politécnica de Madrid and joined an energy research group. In
currently a postdoctoral researcher of the "Margarita 2008 he became a full-time Associate Professor.
Salas" Program, of the Ministry of Universities of the
Government of Spain, assigned to the School of J. Alberto Conejero received the B.Sc. and
Aeronautical and Space Engineering of the University of Vigo. He is also an M.Sc.degree in mathematics from the Universitat de
associate researcher in the Interdisciplinary Modeling Group, InterTech València, Valencia, Spain, in 1998, the M.Sc. degree in
(www.intertech.upv.es) of the University Institute of Pure and Applied bioinformatics and biostatistics from the Universitat
Mathematics at the Universitat Politècnica de València. His areas of research Oberta de Catalunya and the Universitat de Barcelona,
interest are signal processing, noise analysis and information extraction, as well Barcelona, Spain, in 2020, and the Ph.D. degree in
as pattern recognition systems. applied mathematics from the Universitat Politècnica de
València (UPV), Valencia, Spain, in 2004, receiving the
Outstanding Dissertation Award. Since 2020, he has
José Guerra Carmenate was born in Pinar del Río Cuba been a Full Professor with UPV. He is currently the Director of the Department
in 1995. He received the B.Sc. degree in of Applied Mathematics, UPV, and responsible for the M.Sc. program on
telecommunications and electronics engineering from the Mathematics Research with UPV. He has been Visiting Scholar with Bowling
University of Pinar del Río, Pinar del Río, Cuba, in 2019. Green St. University, USA; Kent St. University, USA; Università del Salento,
Is currently working toward the Ph.D. degree in Italy; Universität Tübingen, Germany; and Czech Academy of Sciences, Czech
mathematics at the Universitat Politècnica de València, Republic. His research interests include dynamical systems, partial differential
Valencia, Spain. His research interests include equations, graph theory, network science, and the multidisciplinary applications
acquisition and analysis of environmental data, design of mathematics to computer science, engineering, and biotechnology. Dr.
and implementation of IoT sensors, as well as digital signal processing. Conejero was the recipient of the Teaching Excellence Prize of UPV in 2014,
and he won the XPRIZE $500k Pandemic Response Challenge with
VALENCIA IA4COVID team in 2021.
 10

Pedro Fernández de Córdoba was born in Valencia

in October 1965. He received the B.Sc., M.Sc., and
Ph.D. degrees in physics from the Universitat de
València (UV), Valencia, Spain, in 1988, 1990, and
1992, respectively. He also received the Ph.D. degree in
mathematics from the Universidad Politécnica de
Valencia (UPV), Valencia, in 1997. His research work
was performed at UV, UPV, the Joint Institute for
Nuclear Research (Russia), the University of Tübingen
(Germany), and the Istituto Nazionale di Fisica
Nucleare, Torino, Italy, among others. He is currently a Professor with the
Department of Applied Mathematics, UPV. His research interests include the
area of modeling and numerical simulation of physical and engineering
problems, mainly focusing on the nu- merical treatment of heat and mass
transfer problems. Dr. de Córdoba is Doctor Honoris Causa from the University
of Pinar del Rio (Cuba), a member of the Colombian Academy of Exact,
Physical and Natural Sciences, a member of the Académie Nationale des
Sciences, Arts et Lettres du Bénin, Profesor Invitado of the University of Pinar
del Rio, and Profesor Visitante “Ad Honorem” of the Universidad del
Magdalena (Colombia). Furthermore, since its establishment on September 30,
2011, he has been a member of the Board of the Spanish Mathematics-Industry
network (www.math-in.net).