The Bearing Faults Detection Methods For Electrica

energies
Review
The Bearing Faults Detection Methods for Electrical
Machines—The State of the Art
Muhammad Amir Khan 1 , Bilal Asad 1,2, * , Karolina Kudelina 2 , Toomas Vaimann 2 and Ants Kallaste 2
1 Department of Electrical Power Engineering, The Islamia University of Bahawalpur,

Bahawalpur 63100, Pakistan
2 Department of Electrical Power Engineering and Mechatronics, Tallinn University of Technology,
12616 Tallinn, Estonia
* Correspondence: [email protected] or [email protected]
Abstract: Electrical machines are prone to faults and failures and demand incessant monitoring
for their confined and reliable operations. A failure in electrical machines may cause unexpected
interruptions and require a timely inspection of abnormal conditions in rotating electric machines.
This article aims to summarize an up-to-date overview of all types of bearing faults diagnostic
techniques by subdividing them into different categories. Different fault detection and diagnosis
(FDD) techniques are discussed briefly for prognosis of numerous bearing faults that frequently occur
in rotating machines. Conventional approaches, statistical approaches, and artificial intelligence-
based architectures such as machine learning and deep learning are discussed summarily for the
diagnosis of bearing faults that frequently arise in revolving electrical machines. The most advanced
trends for diagnoses of frequent bearing faults based on intelligence and novel applications are
reviewed. Future research directions that are helpful to enhance the performance of conventional,
statistical, and artificial intelligence (machine learning, deep learning) and novel approaches are well
addressed and provide hints for future work.
Keywords: bearing fault diagnosis; condition monitoring; feature extraction; fault detection and
diagnoses; principal component analysis; neural networks; vibration signals; spectral analysis; genetic
algorithm; support vector machines; power spectral density
Citation: Khan, M.A.; Asad, B.;
Kudelina, K.; Vaimann, T.; Kallaste,
A. The Bearing Faults Detection
Methods for Electrical 1. Introduction
Machines—The State of the Art. Electrical machines are considered the backbone of the industry and play a major
Energies 2023, 16, 296. https:// role in industrial, commercial, and domestic applications. These electric machines have
doi.org/10.3390/en16010296 to work under different circumstances such as extreme ambient temperature, frequently
Academic Editor: Abu-Siada Ahmed varying load conditions, fluctuations in voltages and currents, high moisture, and overloads
causing faults and failures [1]. Bearing faults are considered the most frequent types and
Received: 29 November 2022
are accountable for 30% to 40% of failures in rotating electric machines. Previous studies
Revised: 16 December 2022
suggest that 80% of failures in bearings occur due to a lack of proper lubrication [2]. Bearings
Accepted: 19 December 2022
are subjected to rotation, which is mostly because of the mechanical stress endured during
Published: 27 December 2022
rotational movement and bearing currents. The failures in the bearings mostly happen
due to poor installation, poor assembling, improper use, and improper maintenance. The
movement of current in the bearing depends upon shaft voltages and capacitive currents
Copyright: © 2022 by the authors. caused by frequency and power supply control inverters. Another cause for the failures
Licensee MDPI, Basel, Switzerland. of bearings is contamination, mostly caused by foreign substances from external sources
This article is an open access article in the bearing lubricant. The foreign substances contain dirt (sand) and water entering
distributed under the terms and through the seal, causing the failures in bearings [3]. Electrical and mechanical failures in
conditions of the Creative Commons machines have some unique features to diagnose them, and most techniques employing
Attribution (CC BY) license (https:// motor current signature analysis (MCSA) are implemented [4]. Figure 1 shows the ball
creativecommons.org/licenses/by/ bearing with broken cage, material fatigue and moisture-based corrosion.
4.0/).
Energies 2023, 16, 296. https://doi.org/10.3390/en16010296 https://www.mdpi.com/journal/energies

Energies 2023, 16, x FOR PEER REVIEW 2 of 54

employing motor current signature analysis (MCSA) are implemented [4]. Figure 1 shows
Energies 2023, 16, 296 the ball bearing with broken cage, material fatigue and moisture-based corrosion. 2 of 54
employing motor current signature analysis (MCSA) are implemented [4]. Figure 1 shows
the ball bearing with broken cage, material fatigue and moisture‐based corrosion.
Figure 1. (a) Ball bearing structure. (b) Outer raceway fault. (c) Inner raceway fault.
Mostly, faults and failures occur in bearings due to microscopic sub-surfaces, metal
surfaces,
Mostly, andfaults
cyclicand loading stress
failures on the
occur bearing.due
in bearings Theseto variable and sub-surfaces,
microscopic
Mostly, faults and failures occur in bearings due to microscopic sub‐surfaces, metal cyclic loads producemetal
separation
surfaces, in the bearing surfaces, which further leads to
and cyclic loading stress on the bearing. These variable and cyclic a sapling
surfaces, and cyclic loading stress on the bearing. These variable and cyclic loads produce in the rolling element
loads produce
of the bearings.
separation in the These
bearing developments
surfaces, which willfurther
increase noise
leads to aand vibrations,
sapling
separation in the bearing surfaces, which further leads to a sapling in the rolling element which element
in the rolling alter the
internal
of the dimension
bearings. Theseof bearings.
developments The parameters
will increaseresembling
noise and noise, vibration,
vibrations,
of the bearings. These developments will increase noise and vibrations, which alter the which and temper-
alter the
ature
internal cause more
dimension saplings
of in
bearings.the bearing
The and
parameters damage it
resembling entirely.
noise,
internal dimension of bearings. The parameters resembling noise, vibration, and temper‐ Thermal
vibration, stress
and comprises
tempera-
50%cause
ture of failures
more in bearings
saplings in and also wipes
the bearing andout the lubricant
damage andThermal
it entirely.
ature cause more saplings in the bearing and damage it entirely. Thermal stress comprises creates conditions unsafe
stress comprises
for
50% consistent
of failures operations.
in bearings Frequent
and also failures
wipes in
out bearings
the are
lubricant also
and due to
creates
50% of failures in bearings and also wipes out the lubricant and creates conditions unsafe flaking
conditionsand pitting,
unsafe
unusual
for wearoperations.
consistent patterns, saplings,
Frequentrust, corrosion,
failures creeping,
in bearings and
are also usually
due
for consistent operations. Frequent failures in bearings are also due to flaking and pitting, someand
to flaking other rela-
pitting,
tively
unusual small
wearcombination
patterns, saplings,of causes
rust,that are
corrosion, mostly correctable
creeping, and
and usually
unusual wear patterns, saplings, rust, corrosion, creeping, and usually some other rela‐ predictable.
some Another
other relatively
reason
small for bearingoffailure
combination causesisthat
fluting, which correctable
are mostly occurs when and alternating
predictable.
tively small combination of causes that are mostly correctable and predictable. Another current
Another (AC) or direct
reason for
current
bearing (DC)
failure passes
is fluting,through
whichbearings
occurs when during operation
alternating and
current harms
(AC) orthe
reason for bearing failure is fluting, which occurs when alternating current (AC) or direct symmetry
direct currentof (DC)the
bearings
passes due
through to circulating
bearings currents
during [5].
operation When an
and harms unfixable failure
the symmetry
current (DC) passes through bearings during operation and harms the symmetry of the is found
of in the
the bearings bearings,
due to
it is necessary
circulating to explore
currents the causes
[5]. When of the defects
an unfixable failuretoisprevent
found in future failures in
the bearings,
bearings due to circulating currents [5]. When an unfixable failure is found in the bearings,
it these precise
is necessary
parts
to of
explore thethemachines.
causes ofFigure
the 2 shows
defects to the major
prevent types
future of classifications
failures in these
it is necessary to explore the causes of the defects to prevent future failures in these precise established
precise parts in
of roll-
the
ing bearings with their specific types.
machines. Figure 2 shows the major types of classifications established
parts of the machines. Figure 2 shows the major types of classifications established in roll‐
in rolling bearings
with their specific types.
ing bearings with their specific types.
Figure 2. The major classifications of rolling bearings.

Traditionally, vibration analyses are mostly employed for the extraction of essential
Traditionally,
features vibration
through sensors analyses are
for diagnostics andmostly employed
can also forthe
be used for thegeneration
extraction of
of datasets,
essential
features
which arethrough
further sensors
minimized for diagnostics
by dimension and can also
reduction be used
techniques for the
such generation
as principal
Traditionally, vibration analyses are mostly employed for the extraction of essential of
com-da-
tasets, which are further minimized by dimension reduction techniques such as principal
ponent analysis (PCA) and linear discriminant analysis (LDA) for accurate measurements
features through sensors for diagnostics and can also be used for the generation of da‐
component
of distinctiveanalysis (PCA)
values for and linearofdiscriminant
the diagnosis analysis To
a defective bearing. (LDA) forprecise
extract accurate
tasets, which are further minimized by dimension reduction techniques such as principal
andmeasure-
useful
ments of distinctive
information values signals,
from vibration for the diagnosis of a defective
signal processing bearing.
techniques areTo extractthat
adopted
component analysis (PCA) and linear discriminant analysis (LDA) for accurate measure‐ precise and
change
useful information
information from thefrom
time vibration
domain tosignals, signal
the frequency processing techniques
domain through are adopted
fast Fourier that
transform
ments of distinctive values for the diagnosis of a defective bearing. To extract precise and
(FFT). The timely inspection of these abnormal conditions on rotating machines is very
useful information from vibration signals, signal processing techniques are adopted that
important for their secure and consistent operations. In the last few years, many researchers
are paying attention and preceding their research with the employment of advanced tech-
niques based on a hybrid algorithm to diagnose abnormal conditions in rotating machines.

change information from the time domain to the frequency domain through fast Fourier
transform (FFT). The timely inspection of these abnormal conditions on rotating machines
Energies 2023, 16, 296 is very important for their secure and consistent operations. In the last few years, many 3 of 54
researchers are paying attention and preceding their research with the employment of
advanced techniques based on a hybrid algorithm to diagnose abnormal conditions in
rotating machines. Many diagnostic techniques/algorithms are implemented to diagnose
Many diagnostic techniques/algorithms are implemented to diagnose the bearing faults
the bearing faults and are addressed by a variety of available signals, which consist of
and are addressed by a variety of available signals, which consist of thermal imaging [6],
thermal imaging [6], vibration [7], acoustic noise [8], stator current spectrum [9], and many
vibration [7], acoustic noise [8], stator current spectrum [9], and many more. Figure 3 shows
more. Figure 3 shows the different percentage contributions of faults and failures in roll-
the different percentage contributions of faults and failures in rolling bearings, the detail
ing bearings, the detail can be studied in [10].
can be studied in [10].
Figure
Figure3.3.Causes
Causesofoffailures
failuresininrolling
rollingbearings.
bearings.
Severaltypes
Several typesofofdiagnostic
diagnosticapproaches
approachesare areavailable
availablefor
fordifferent
differenttypes
typesofoffaults
faultsthat
that
occur on rotating machines, including conventional approaches, which
occur on rotating machines, including conventional approaches, which mostly use vibra-mostly use vibration
signals
tion to diagnose
signals the the
to diagnose faults, statistical
faults, approaches,
statistical approaches,model-based
model-based approaches,
approaches,hard-ware-
hard-
based approaches,
ware-based history-based
approaches, approaches,
history-based and artificial
approaches, intelligence-based
and artificial approaches
intelligence-based ap-
(machine learning and deep learning). These approaches employ different signals that
proaches (machine learning and deep learning). These approaches employ different sig-
follow patterns such as motor current signature analysis (MCSA), vibration signals, acoustic
nals that follow patterns such as motor current signature analysis (MCSA), vibration sig-
signals, electromagnetic signals, voltage, and current signals to diagnose numerous types
nals, acoustic signals, electromagnetic signals, voltage, and current signals to diagnose
of faults in rotating machines [10]. Currently, condition monitoring techniques are utilized
numerous types of faults in rotating machines [10]. Currently, condition monitoring tech-
on a variety of equipment such as revolving machines and auxiliary equipment to prevent
niques are utilized on a variety of equipment such as revolving machines and auxiliary
serious issues that may result from their damage. Condition monitoring of electrical ma-
equipment to prevent serious issues that may result from their damage. Condition moni-
chines implements multiple types of sensors that are used to sense any abnormal condition
toring of electrical machines implements multiple types of sensors that are used to sense
that happened during operation and collect useful information for long-term prediction.
any abnormal condition that happened during operation and collect useful information
In electrical motors, condition monitoring is implemented individually on different parts,
for long-term prediction. In electrical motors, condition monitoring is implemented indi-
and measurements are taken by employing vibration-based techniques, signal processing
vidually on different parts, and measurements are taken by employing vibration-based
techniques, and temperature differences for their diagnostics. The importance of bearing
techniques, signal in
failure diagnosis processing
revolvingtechniques, and temperature
electric machines differences
is clear because for theirthat
it is estimated diagnos-
half of
tics.
the failures that occur in revolving electrical machines are bearing faults [11]. Aisvariety
The importance of bearing failure diagnosis in revolving electric machines clear
because it is estimated that half of the failures that occur in revolving electrical
of different faults investigated in the literature that influence the performance of rotating machines
are bearing
electric faults [11].
machines A varietyinofFigure
is displayed different
4. faults investigated in the literature that influ-
ence the performance of rotating electric machines is displayed in Figure 4.
Energies 2023,
Energies 16,296
2023,16, x FOR PEER REVIEW 44 of 54
of54
54
Energies 2023, 16, x FOR PEER REVIEW 4 of

Figure
Figure 4. A
4. A variety
variety of
of different
different faults
faults occurring
occurring in
in rotating
rotating electrical
electrical machines.
machines.
Figure 4. A variety of different faults occurring in rotating electrical machines.
There are
There are three
are three targets of
three targets
targets of failure
of failure analysis:
failure analysis: detection,
analysis: detection, isolation,
isolation,and
andidentification.
identification.
Condition
Condition monitoring
monitoring of of revolving electric machines
revolving electric machines ensures
ensuresthe
thechances
chancesof takingthe
oftaking
taking the
the
minimum
minimum risk for
risk for operations
operations and
and reduces
reduces the
the minimum
minimum effective
effective cost
cost of
of maintenance.
minimum risk for operations and reduces the minimum effective cost of maintenance. In maintenance. In In
online
online condition monitoring,
condition monitoring, thethe operating
operating system
system will
will generate
generate real-time
real-time data
data at
online condition monitoring, the operating system will generate real‐time data at critical at critical
critical
moments
moments of
of the
themachine
machineduring
duringrunning
runningconditions.
conditions.Periodic
Periodiccondition
conditionmonitoring
moments of the machine during running conditions. Periodic condition monitoring inves‐ monitoringinves-
in-
tigates monitoring during a fixed interval of time and reduces the chances of fault in ma-
vestigates monitoring during a fixed interval of time and reduces the chances
tigates monitoring during a fixed interval of time and reduces the chances of fault in ma‐ of fault in
chines [12]. The classification of proposed approaches discussed briefly in this article is
machines [12]. The classification of proposed approaches discussed briefly in this
chines [12]. The classification of proposed approaches discussed briefly in this article is article is
shown
shown in
in Figure
Figure 5.
shown in Figure 5. 5.

Figure 5. Different bearing fault diagnostic approaches.

Energies 2023, 16, 296 5 of 54
2.2.Conventional
ConventionalTechniques
Techniquesfor forDiagnosis
DiagnosisofofBearingBearingFaults
Faults
Faultdiagnostic
Fault diagnostictechniques
techniquessuchsuchas asconventional
conventionalapproaches
approachesare arethe
thetypical
typicalmethod
method
for the detection and diagnosis of bearing faults by employing vibration,
for the detection and diagnosis of bearing faults by employing vibration, current, voltage, current, voltage,
strayflux,
stray flux,temperature,
temperature,acoustic
acousticmeasurements,
measurements,etc. etc.These
Theseconventional
conventionalapproaches
approachesare are
utilized to extract useful attributes from the vibration spectrum, waveforms,
utilized to extract useful attributes from the vibration spectrum, waveforms, and signals and signals
totoaccurately
accuratelyclassify
classifyabnormal
abnormalconditions.
conditions.In In[13,14],
[13,14],wavelet
waveletpacket
packetdecomposition
decompositionwas was
employedtotoobserve
employed observethethestator
statorcurrent
currentandanddetect
detectthe
theabnormal
abnormalconditions
conditionsininthe
thebearing
bearing
ofofIM.
IM.TheTheWPT
WPTcancananalyze
analyzedifferent
differentbearing
bearingfaults
faultsunder
undervarying
varyingload
loadconditions
conditionsandand
detect the non-stationary nature of faults in diverse frequency modes. In
detect the non-stationary nature of faults in diverse frequency modes. In [15], the authors [15], the authors
investigatedthe
investigated the
ballball bearings’
bearings’ faultsfaults by utilizing
by utilizing a modified
a modified winding-based
winding-based model (MWFM)model
(MWFM)
with with thecircuit
the coupled coupled circuit
model modelvariations
through through variations
in the air in theofair
gap IM.gap of IM.
Stator Stator
current
current spectrum
spectrum analysis isanalysis is employed
employed to observe to the
observe the amplitudes
amplitudes of faulty of faultysignatures
current current signa-
for
tures
the for the diagnosis
diagnosis of bearingoffaults.
bearing faults.6 Figure
Figure shows 6the shows the development
development of harmonics
of harmonics in statorin
stator current
current due bearing.
due to bad to bad bearing.
Figure6.6.The
Figure Thecomparison
comparisonofofharmonics
harmonicsininstator
statorcurrent
currenttotodiagnose
diagnosehealthy
healthyand
andouter
outerrace
racebearing
bear-
ing conditions
conditions [15]. [15].
In
In[16],
[16],the
theauthors
authorsproposed
proposed a strategy to to
a strategy diagnose
diagnose thethe
bearing
bearingandand
eccentricity fault
eccentricity in
fault
permanent magnet synchronous machines (PMSM) by utilizing the motor
in permanent magnet synchronous machines (PMSM) by utilizing the motor current sig-current signature
analysis (MCSA)(MCSA)
nature analysis and stator
and current harmonics.
stator current The stator
harmonics. current
The stator spectrum
current is used
spectrum as
is used
aashealthy indicator of the reference spectrum at different speed tests with the vibratory
a healthy indicator of the reference spectrum at different speed tests with the vibratory
indicator
indicatortotodiagnose
diagnosebearing faults.
bearing Figure
faults. 7 shows
Figure 7 showsthe the
stator current
stator harmonics
current spectrum
harmonics spec-
totrum
diagnose bearing (redline) and eccentricity faults (blue-line) in electric machines.
to diagnose bearing (redline) and eccentricity faults (blue-line) in electric machines.
Energies 2023, 16,Energies 2023, 16, x FOR PEER REVIEW

296 6 of 54 6

Figure 7. Stator current spectrum to diagnose bearing faults (redline) and eccentricity faults (bluel-
Figure Stator
7. in
ine) PMSMcurrent
[16].spectrum to diagnose bearing faults (redline) and eccentricity faults (blueline)
Figure 7. Stator current spectrum to diagnose bearing faults (redline) and eccentricity faults (b
in PMSM [16].
ine) in PMSM [16].
In [17], the authors proposed a technique to diagnose linear bearing faults in linear
In [17], the authors proposed a technique to diagnose linear bearing faults in linear
brushless ACIn [17], the authors proposed a technique to diagnose linear bearing faults in li
machines by utilizing the frequency spectrum of vibration signals. In [18],
brushless AC machines by utilizing the frequency spectrum of vibration signals. In [18], the
the authors recommended bearing fault-finding of PMSM under dynamic conditions
brushless AC machines by utilizing the frequency spectrum of vibration signals. In
authors recommended bearing fault-finding of PMSM under dynamic conditions through
through the
discrete wavelet
authors decomposition
recommended andfault‐finding
bearing finite elementof analysis
PMSM (FEA). dynamic
From stator
discrete wavelet decomposition and finite element analysis (FEA). Fromunder
stator current condit
current harmonics, DWT is utilized to extract a significant attribute from vibration
through discrete wavelet decomposition and finite element analysis (FEA). From st signals
harmonics, DWT is utilized to extract a significant attribute from vibration signals for
for bearing fault diagnosis. In [19–22], the authors investigated the park’s vector approach
bearing faultcurrent harmonics, DWT is utilized to extract a significant attribute from vibration sig
diagnosis. In [19–22], the authors investigated the park’s vector approach
to diagnosing bearing faults in IM through motor supply current and bearings ball pass
for bearing fault diagnosis. In [19–22], the authors investigated the park’s vector appro
to diagnosing bearing faults in IM through motor supply current and bearings ball pass
frequencies (BPFs). Figure 8 points out the comparison of healthy bearings frequency with
to diagnosing bearing faults in IM through motor supply current and bearings ball
frequencies (BPFs). Figure 8 points out the comparison of healthy bearings frequency
failure infrequencies (BPFs). Figure 8 points out the comparison of healthy bearings frequency
the outer race of bearings by implementing the park’s vector approach via cur-
with failure in the outer race of bearings by implementing the park’s vector approach via
rent signals.
failure in the outer race of bearings by implementing the park’s vector approach via
current signals.
rent signals.
Figure 8. The8.impact
Figure of outer
The impact raceway
of outer fault fault
raceway on (a)on
phase a (b) aphase
(a) phase b, (c) b,
(b) phase phase c, (d)c,overall
(c) phase phasor
(d) overall phasor
diagram,
diagram, (e) bearing
(e) bearing fault fault Park’s
Park’s vectorvector
(BFPV) (BFPV)
[19]. [19].
Figure 8. The impact of outer raceway fault on (a) phase a (b) phase b, (c) phase c, (d) overall ph
In [23], diagram, (e) bearing fault Park’s vector (BFPV) [19].
the authors suggested the procedure to identify inner raceway and outer
raceway bearing defects of induction motors via spectral analysis. Discrete wavelets
In [23], the authors suggested the procedure to identify inner raceway and outer r
transform (DWT) is used for extracting useful information from the faulty bearing of
way bearing defects of induction motors via spectral analysis. Discrete wavelets transf
induction motors and comparing characteristic frequencies of healthy and faulty indicators

In [23], the authors suggested the procedure to identify inner raceway and outer race-
way bearing defects of induction motors via spectral analysis. Discrete wavelets transform
In [23], the authors suggested the procedure to identify inner raceway and outer race-
(DWT) is used for extracting useful information from the faulty bearing of induction7 mo-
Energies 2023, 16, 296
way bearing defects of induction motors via spectral analysis. Discrete wavelets transform of 54
tors and comparing characteristic frequencies of healthy and faulty indicators for fault
(DWT) is used for extracting useful information from the faulty bearing of induction mo-
investigation. In [24–26], the authors anticipated the noninvasive bearing faults of IM by
tors and comparing characteristic frequencies of healthy and faulty indicators for fault
employing stray flux measurements and flux probes for the analysis of different kinds of
investigation. In [24–26],
for fault investigation. Inthe authors
[24–26], theanticipated the noninvasive
authors anticipated bearing faults
the noninvasive of IM
bearing by
faults
bearing faults. Figure 9 shows the spectrum of healthy bearings (black spectral lines) and
employing stray fluxstray
of IM by employing measurements and flux probes
flux measurements and fluxfor the analysis
probes for the of different
analysis kinds of
of different
faulty of
bearings (red spectral lines) obtained by stray flux measurements and power spec-
bearing
kinds faults.
bearingFigure 9 shows
faults. Figure the spectrum
9 shows of healthyof
the spectrum bearings
healthy(black spectral
bearings lines)
(black and
spectral
tral density for bearing fault diagnosis.
faulty bearings (red spectral lines) obtained by stray flux measurements and power spec-
lines) and faulty bearings (red spectral lines) obtained by stray flux measurements and
tral density for bearing fault diagnosis.
power spectral density for bearing fault diagnosis.
Figure 9. Magnetic stray flux spectrum for healthy bearing (black spectral lines) and inner raceway
fault (red
Figure spectral lines)
9. Magnetic strayto diagnose
flux bearing
spectrum faults. bearing
for healthy Reproduced
(blackwith permission
spectral of [26].
lines) and inner raceway
Figure 9. Magnetic stray flux spectrum for healthy bearing (black spectral lines) and inner raceway
fault (red spectral lines) to diagnose bearing faults. Reproduced with permission of [26].
fault (red spectral lines) to diagnose bearing faults. Reproduced with permission of [26].
In [27,28],the authors suggested a methodology to evaluate the normal operating
conditions and faulty
In [27,28],the conditions
authors of bearings
suggested for IM depending
a methodology on the
to evaluate theinstantaneous value
normal operating
In [27,28],the authors suggested a methodology to evaluate theinstantaneous
normal operating
of the motor voltage. Figure 10 shows the results of bearing healthy conditions and value
conditions and faulty conditions of bearings for IM depending on the outer
conditions and faulty Figure
conditions of bearings for IMofdepending on theconditions
instantaneous
and value
raceway
of defects
the motor obtained from
voltage. 10 applying
shows thethe instantaneous
results bearingfrequency
healthy of motor voltage.
outer
of the motor
raceway voltage.
defects Figure
obtained 10 applying
from shows thethe
results of bearingfrequency
instantaneous healthy conditions and outer
of motor voltage.
raceway defects obtained from applying the instantaneous frequency of motor voltage.
Figure 10. PSD of the stator current for bearings: (a) Healthy case; (b) outer raceway defect obtained
from the instantaneous frequency of motor voltage [27].
In [29], the authors suggested the technique of identifying the generalized roughness
during abnormal conditions in bearings through spectral kurtosis energy utilizing vibration
Figure 10. PSD of the stator current for bearings: (a) Healthy case; (b) outer raceway defect obtained
from the instantaneous frequency of motor voltage [27].
Energies 2023, 16, 296 8 of 54
In [29], the authors suggested the technique of identifying the generalized roughness
during abnormal conditions in bearings through spectral kurtosis energy utilizing vibra-
tion
or or current
current signals.
signals. In [30],
In [30], thethe authors
authors suggesteda amethodology
suggested methodologytotodiagnose
diagnose bearing
bearing
faults of
faults of variable-speed
variable-speedwindwindturbines
turbinesthrough
throughpower
powerspectral
spectral density
density (PSD)
(PSD) forfor extract-
extracting
ing features
features of faults
of faults fromfrom stator
stator current
current measurements.
measurements. Characteristic
Characteristic features,
features, stator
stator cur-
current,
rent,fault
and and signatures
fault signatures of bearings
of bearings are utilized
are utilized for thefor the diagnosis
diagnosis of windof turbine
wind turbine
bearingbearing
faults.
faults. Figure
Figure 11 shows 11 the
shows the of
results results of the bearing
the bearing cage
cage fault fault
and andrace
outer outer race
fault byfault by imple-
implementing
menting threshold impulse values and power spectral density.
threshold impulse values and power spectral density.
Figure 11.
Figure 11. Power
Power spectral
spectral density (PSD) results
density (PSD) results for
for the
the diagnosis
diagnosis of
of bearing
bearing fault
fault in
in the
the wind
wind turbine
turbine
bearing cage
for (a) bearing cage fault
fault and
and (b)
(b) outer
outer race
race fault.
fault. Reproduced
Reproduced with
with permission
permission of
of [30].
[30].
In
In [31],
[31], the
the authors
authors proposed
proposed thethe prognostic
prognostic methodology
methodology to to diagnose
diagnose thethe bearing
bearing
failures in electric railway traction motors by placing a high-frequency inductance
failures in electric railway traction motors by placing a high-frequency inductance coil near
coil
the
nearrailway traction
the railway bearing
traction to extract
bearing useful
to extract information
useful for for
information thethe
classification of bearing
classification of bear-
faults. In [32,33],
ing faults. thethe
In [32,33], authors
authorsrecommended
recommendedbearingbearingfaults
faultsof
ofvariable-speed converter-
variable-speed converter-
fed
fed electrical machines by applying systematic approaches to investigate bearing wear
electrical machines by applying systematic approaches to investigate bearing wear
conditions
conditions andandincipient
incipientbearing
bearingfaults
faultsthrough
through mutual analysis
mutual of of
analysis electrical current
electrical signals
current sig-
and vibration signals. In [34], the authors suggested a technique to point out the ball bearing
nals and vibration signals. In [34], the authors suggested a technique to point out the ball
failures by utilizing bearing-rated frequency and vibration indicators. The fundamental
bearing failures by utilizing bearing-rated frequency and vibration indicators. The funda-
frequency is compared with the frequency of the inner raceway and outer raceway of the
mental frequency is compared with the frequency of the inner raceway and outer raceway
bearing to diagnose abnormal conditions. In [35], the authors investigated the detection
of the bearing to diagnose abnormal conditions. In [35], the authors investigated the de-
of bearing faults in PMSM by employing the instantaneous power factor and the torque
tection of bearing faults in PMSM by employing the instantaneous power factor and the
oscillations in damaged bearings. The diagnostic index was developed to validate healthy
torque oscillations in damaged bearings. The diagnostic index was developed to validate
and faulty bearings conditions at variable loads. Figure 12 shows the results of the fault
healthy and faulty bearings conditions at variable loads. Figure 12 shows the results of the
index for healthy and outer raceway bearings faults by employing instantaneous power
fault index for healthy and outer raceway bearings faults by employing instantaneous
factor and the torque oscillations in PSPM.
power factor and the torque oscillations in PSPM.
In [28], the authors implemented angular contact for evaluating the bearing faults
by employing the envelope technique to identify inner race, outer race, and ball bearing
failures. In [36,37], the authors applied vibration analysis to remove the nonbearing fault
component (RNFC) filter to detect healthy, inner race, outer race, and double holes in the
outer race bearings. Frequency-domain features are compared with time-domain features
to analyze the abnormal conditions. Figure 13 shows the outcomes obtained by the RNFC
filter to detect healthy, inner race, outer race, and double holes in the outer race bearings.
Energies 2023, 16, 296 9 of 54
Figure 12. Power spectrum density based on frequency spectrum for diagnostics of bearing healthy
condition and outer raceway fault [35].
In [28], the authors implemented angular contact for evaluating the bearing faults by
employing the envelope technique to identify inner race, outer race, and ball bearing fail-
ures. In [36,37], the authors applied vibration analysis to remove the nonbearing fault
filter to
Figure 12.detect
Powerhealthy,
spectruminner race,
density outer
based on race, and spectrum
frequency double holes in the outer
for diagnostics race bearings.
of bearing healthy
Figure 12. Power
condition spectrum
and outer density
raceway based on frequency spectrum for diagnostics of bearing healthy
fault [35].
condition and outer raceway fault [35].
In [28], the authors implemented angular contact for evaluating the bearing faults by
employing the envelope technique to identify inner race, outer race, and ball bearing fail-
ures. In [36,37], the authors applied vibration analysis to remove the nonbearing fault
filter to detect healthy, inner race, outer race, and double holes in the outer race bearings.
Figure 13.
Figure 13. RNFC
RNFC filter
filter results
results for
for the
the diagnosis
diagnosis of
of (a)
(a) healthy
healthy and
and inner
inner race
race defects
defects and
and (b)
(b) outer
outer
race defects
race defects and
and double
double holes
holes in
in outer
outer race
racedefects
defects[36].
[36].
In [38], the authors suggested the wireless sensor-based network to observe the sit-
uation of bearings and to be able to explore the characteristics of vibration signals using
accelerometers. The measurements of accelerometers are compared with a piezoelectric
transducer, which is set as a reference signal to evaluate the results of the bearing faults.
In [39,40], the authors proposed to diagnose the bearing faults of a brushless direct current
motor (BLDC) at different speed variations through its phase current examination. Multiple
signal processing techniques were applied, such as zero-phase filtering, Hilbert transform,
Figure 13. RNFC
and accurate filter results
rotating anglefor the diagnosis
curves, to obtainoffrequency
(a) healthycharacteristics
and inner race defects and (b)
of the noisy outer
current
race defects and double holes in outer race defects [36].
signal for faulty bearings assessment. In [41], the authors proposed an application for the
accelerometers. The measurements of accelerometers are compared with a piezoelectric
transducer, which is set as a reference signal to evaluate the results of the bearing faults.
In [39,40], the authors proposed to diagnose the bearing faults of a brushless direct current
motor (BLDC) at different speed variations through its phase current examination. Multi-
ple signal processing techniques were applied, such as zero-phase filtering, Hilbert trans-
Energies 2023, 16, 296 10 of 54
form, and accurate rotating angle curves, to obtain frequency characteristics of the noisy
current signal for faulty bearings assessment. In [41], the authors proposed an application
for the diagnosis of bearing gearbox faults in high-speed locomotives employing temper-
ature sensor
diagnosis of signals.
bearingAxle box monitoring
gearbox for bearing
faults in high-speed health is applied
locomotives basedtemperature
employing on temper-
ature signals that are obtained through sensors from the wireless transmission. In [42], the
sensor signals. Axle box monitoring for bearing health is applied based on temperature
authors proposed
signals that a wavelet-based
are obtained through approach to find
sensors from theout the bearing
wireless faults in the
transmission. rotating
In [42], the
authors proposed a wavelet-based approach to find out the bearing faults
electrical machines using vibration analysis via wavelet transform. Multiple techniquesin the rotating
electrical
such machinesanalysis,
as temporal using vibration
spectralanalysis
analysis,via wavelet analysis,
envelope transform. Multiple
and wavelettechniques
transfor-
such as temporal analysis, spectral analysis,
mation are applied as fault diagnostic approaches. envelope analysis, and wavelet transformation
are applied
In [43], as
thefault diagnostic
authors approaches.
suggested an improved variational mode decomposition (IVMD)
In [43], the authors suggested
to trace the incipient faults in rolling an improved variational
bearings using mode decomposition
a combination of traditional(IVMD)
varia-
to trace the incipient faults in rolling bearings using a
tional mode decomposition and empirical mode decomposition (EMD). combination of traditional varia-
The intrinsic
tional mode decomposition and empirical mode decomposition (EMD). The intrinsic mode
mode function (IMF) set the evaluation index for the reconstruction of signals employing
function (IMF) set the evaluation index for the reconstruction of signals employing the
the transient impulse monitoring index and Hilbert envelope analysis to evaluate bearing
transient impulse monitoring index and Hilbert envelope analysis to evaluate bearing
failures. In [44], the authors proposed to diagnose the inner race and outer race faults of
failures. In [44], the authors proposed to diagnose the inner race and outer race faults
roller bearings by IM, utilizing high-resolution spectral analysis of the stator current. Fou-
of roller bearings by IM, utilizing high-resolution spectral analysis of the stator current.
rier transform was implemented to visualize the stator current and observe the frequen-
Fourier transform was implemented to visualize the stator current and observe the frequen-
cies of small magnitudes via high resolution. In [45], the authors suggested an approach
cies of small magnitudes via high resolution. In [45], the authors suggested an approach
to finding out the inner raceway, outer raceway, and ball bearings failures through the
to finding out the inner raceway, outer raceway, and ball bearings failures through the
probabilistic model of fault vibrations obtained from accelerometers. Figure 14 shows the
probabilistic model of fault vibrations obtained from accelerometers. Figure 14 shows the
results of receiver operating characteristic (ROC) curves for evaluating the performance
results of receiver operating characteristic (ROC) curves for evaluating the performance
of the probabilistic model to diagnose ball bearing faults, inner race faults, and outer race
of the probabilistic model to diagnose ball bearing faults, inner race faults, and outer race
faults.
faults.
Figure
Figure 14.
14. ROC
ROC curves
curves for
for evaluating
evaluating the
the performance
performance ofof (a)
(a) ball
ball bearing
bearing fault,
fault, (b)
(b) outer
outer race
race fault,
fault,
and
and (c)
(c) inner
inner race
race fault.
fault. Reproduced
Reproduced with
with permission
permission of
of [45].
[45].
In
In [46],
[46], the
the authors
authors proposed
proposed the
the phase-locked
phase-locked loop-based
loop-based approach
approach toto find
find out
out the
the
bearing
bearing faults
faults of
of doubly
doubly fed induction generators
fed induction generators (DFIG) used in
(DFIG) used in windmills.
windmills. Vibration
Vibration
signals
signals were
were obtained
obtained via
via aa resampling
resampling technique
technique based
based on
on the
the angle
angleofofthe
therotor
rotorposition,
position,
and experimental results were validated with simulation results. Figure 15 shows the
and experimental results were validated with simulation results. Figure 15 shows the
comparison results of the auto-power spectrum for the diagnosis of bearing faults in direct
comparison results of the auto-power spectrum for the diagnosis of bearing faults in direct
wind turbines
wind turbines through
through thethe current
current demodulated
demodulated signals.
signals.
In [47], the authors suggested an intelligent-based strategy for bearing fault diagnosis
in a brushless direct current motor (BLDC) and permanent magnet synchronous generators
(PMSG) employing synchro-squeezing wavelet transform (SWT) with touch-less order
tracking (TOT). SWT can extract instantaneous frequencies of fault signals and also re-
structure the harmonic components correctly to diagnose abnormal bearing conditions.
In [48], the authors recommended the diagnostic techniques for multiple kinds of bearing
failures (outer race, inner race, lack of lubrication, and healthy status) in IM by utilizing
noninvasive, contactless thermal infrared imaging. DWT was employed to extract data
in the form of two dimensional and converted it into thermal images for evaluation of
bearing faults in electric machines. In [49,50], authors proposed electrostatics-based sensors
Energies 2023, 16, 296 11 of 54
for the diagnosis of roller bearing fault performance degradation by implementing spec-
tral regression. The constructive attribute was extracted through the time and frequency
Figure
domain 15.from
(a) DFIG windoperating
normal turbine topology. (b) Comparison
conditions of the
of bearings auto-power
through spectrum for
the Gaussian healthy
model to
and faulty
predict thebearing faultsuseful
remaining in wind
lifeturbines
(RUL) for doubly
of the fed induction
bearings. Figure generators.
16 shows theReproduced
results ofwith
the
Energies 2023, 16, x FOR PEER REVIEW
permission of sensor
[46]. for evaluating the degradation assessment depending on the fusion 11 of 54
electrostatic of
various attributes and shows the assessment of the bearing whole life with time.
in a brushless direct current motor (BLDC) and permanent magnet synchronous genera-
tors (PMSG) employing synchro-squeezing wavelet transform (SWT) with touch-less or-
der tracking (TOT). SWT can extract instantaneous frequencies of fault signals and also
restructure the harmonic components correctly to diagnose abnormal bearing conditions.
noninvasive, contactless thermal infrared imaging. DWT was employed to extract data in
the form of two dimensional and converted it into thermal images for evaluation of bear-
ing faults in electric machines. In [49,50], authors proposed electrostatics-based sensors
domain from normal operating conditions of bearings through the Gaussian model to pre-
dict
Figurethe remaining useful life topology.
(RUL) of the bearings. Figure 16 shows the results of the
Figure 15.
15. (a)
(a) DFIG
DFIG wind
wind turbine
turbine topology. (b)
(b) Comparison
Comparison of
of the
the auto-power
auto-power spectrum
spectrum for
for healthy
healthy
electrostatic sensor
and faulty bearing for evaluating
faults the degradation
in wind turbines assessment
for doubly fed induction depending on the fusion
generators. Reproduced of
with
and faulty bearing faults in wind turbines for doubly fed induction generators. Reproduced with
various
permissionattributes
of [46]. and shows the assessment of the bearing whole life with time.
permission of [46].
in a brushless direct current motor (BLDC) and permanent magnet synchronous genera-
tors (PMSG) employing synchro-squeezing wavelet transform (SWT) with touch-less or-
der tracking (TOT). SWT can extract instantaneous frequencies of fault signals and also
restructure the harmonic components correctly to diagnose abnormal bearing conditions.
noninvasive, contactless thermal infrared imaging. DWT was employed to extract data in
the form of two dimensional and converted it into thermal images for evaluation of bear-
ing faults in electric machines. In [49,50], authors proposed electrostatics-based sensors
domain from normal operating conditions of bearings through the Gaussian model to pre-
dict the remaining useful life (RUL) of the bearings. Figure 16 shows the results of the
electrostatic sensor for evaluating the degradation assessment depending on the fusion of
various attributes and shows the assessment of the bearing whole life with time.
Figure
Figure 16.
16. Performance
Performance evaluation
evaluation of
of complete
complete life
life of
of aa bearing
bearing through time [47].
through time [47].
In [51], the authors proposed a scheme to diagnose corrosion failures in bearings

of a doubly fed induction generator in windmills based on a bi-spectrum modulation
signal. Outcomes obtained from the bi-spectrum are compared with fundamental fre-
quency and characteristic frequency to identify faulty harmonics in the current spectrum.
Figure 17 shows the stator current waveform in the time domain to classify healthy and
faulty bearings.
In [51], the authors proposed a scheme to diagnose corrosion failures in bearings of
a doubly fed induction generator in windmills based on a bi-spectrum modulation signal.
Outcomes obtained from the bi-spectrum are compared with fundamental frequency and
characteristic frequency to identify faulty harmonics in the current spectrum. Figure 17
Energies 2023, 16, 296 shows the stator current waveform in the time domain to classify healthy and faulty12bear-
of 54
ings.
Figure 17. The

Figure 17. Thewaveform
waveformofofthe
thestator
statorcurrent
currentinin
thethe
frequency domain
frequency forfor
domain healthy andand
healthy faulty bearings.
faulty bear-
ings. Reproduced with permission of [51].
Reproduced with permission of [51].
In
In [52,53],
[52,53], the
the authors
authors proposed
proposed conditional
conditional monitoring
monitoring of of journal
journal bearings
bearings byby imple-
imple-
menting motor current signature analysis (MCSA) of induction machines.
menting motor current signature analysis (MCSA) of induction machines. The experi- The experimental
setup was created for low voltage rotating machines for evaluation of journal bearing faults
mental setup was created for low voltage rotating machines for evaluation of journal bear-
by comparing the observations of different designs. In [54], the authors suggested an
ing faults by comparing the observations of different designs. In [54], the authors sug-
online fault diagnostic methodology for the evaluation of bearing failures in brushless DC
gested an online fault diagnostic methodology for the evaluation of bearing failures in
machines and brushed DC motors by employing a stochastic resonance-based adaptive
brushless DC machines and brushed DC motors by employing a stochastic resonance-
filter. The stochastic resonance-based adaptive filter was utilized to obtain the phase of
based adaptive filter. The stochastic resonance-based adaptive filter was utilized to obtain
purified signal for the analysis of diverse bearing faults in rotating machines.
the phase of purified signal for the analysis of diverse bearing faults in rotating machines.
In [55], the authors proposed an embedded system for the diagnosis of bearing failures
In [55], the authors proposed an embedded system for the diagnosis of bearing fail-
in permanent magnet synchronous motors (PMSM) using tachometer-based fast and online
ures in permanent magnet synchronous motors (PMSM) using tachometer-based fast and
order analysis (FOOA) composed of two algorithms for obtaining useful information from
online order analysis (FOOA) composed of two algorithms for obtaining useful infor-
sinusoidal currents and calculating the envelope order spectrum for the classification of
mation from sinusoidal currents and calculating the envelope order spectrum for the clas-
bearing faults. In [56], the authors suggested the diagnosis of inner race creep bearing faults
sification
of of bearing
wind turbine faults. In
generators by[56], the authors
utilizing suggested
vibration the diagnosis
and temperature of innerFailures
analysis. race creep
of
bearing faults of wind turbine generators by utilizing vibration and temperature
bearing creep were achieved by continuous monitoring through vibration speed harmonic analysis.
Failures
and of bearing
absolute creep values.
temperature were achieved
Figure 18 byshows
continuous monitoring
the recognition through vibration
of generator bearing
speed harmonic and absolute temperature values. Figure 18 shows the recognition
inner race creep by utilizing vibration and temperature investigation and their replacement of gen-
erator bearing inner race creep by utilizing vibration and
by showing slowly developing trends in the behavior of the bearing. temperature investigation and
theirThe
replacement by showing slowly developing trends in the behavior of the bearing.
authors of [57] suggested a strategy for the recognition of outer race-bearing faults
in IM by employing the homogeneity algorithm (HA). The homogeneity algorithms (HO)
can identify the changes in normal regime from vibration signals and alarm the presence
of bearing faults in IM efficiently. In [58], the authors proposed bearing fault diagnostic
indicators by employing MCSA based on normalized triple covariance for IM. In [59], the
authors paid attention to diagnosing the faulty bearings through the envelope harmonic
product spectrum and the adaptive second order cyclo-stationarity blind deconvolution
algorithm for the recognition of incipient bearing faults in locomotives. A comparison of
different conventional diagnostic techniques is presented in Table 1.
Energies2023,
Energies 16,296
2023,16, x FOR PEER REVIEW 13 of
13 of5454
Figure18.
Figure 18. The
The behavior
behavior of
ofthe
theweighted
weightedaverage
averagecondition indication
condition forfor
indication bearing replacement.
bearing Re-
replacement.
produced with permission of [56].
Reproduced with permission of [56].
The
Table 1. authors of
A summary of different
[57] suggested a strategy
conventional for for
techniques thebearing
recognition of outer race-bearing
fault diagnosis.
faults in IM by employing the homogeneity algorithm (HA). The homogeneity algorithms
Technique/Assisting (HO) can identify the Cons
Pros changes in normal regime
Bearing from vibration
Fault signals and alarm
Applications the pres-
References
Technique ence of bearing faults in IM efficiently. In [58], the authors proposed bearing fault diag-
Type
Wavelet packet nostic indicators
Efficient technique for by employing MCSA based on normalized triple covariance for IM. In
transform extracting[59], the authors
features to paid computing
Higher attention to diagnosing the faulty bearings through the envelope
Ball bearings Induction motors [13,14]
+ diagnose abnormal cost
harmonic product spectrum and the adaptive second order cyclo-stationarity blind de-
MCSA conditions
convolution algorithm for the recognition of incipient bearing faults in locomotives. A
Modified windings
Accurate comparison
for faulty of different conventional diagnostic techniques is presented in Table 1.
approach Need inductance
signals having Ball bearings Induction motors [15]
+ as a parameter
harmonicsTable 1. order
in their A summary of different conventional techniques for bearing fault diagnosis.
MCSA
Technique/Assisting
Stator current Bearing Fault
Pros
Works well in noisy Cons Applications
Permanent magnet References
Technique environments to extract
harmonics Not applicable to
Ball bearingsType synchronous [16,52,54]
Wavelet+ packet trans- stationary signals
features machines
MCSA Efficient technique for ex-
form Higher computing
Finite element tracting features to diag-High Ball bearings Induction motors [13,14]
+ Able to handle cost Permanent magnet
analysis nose abnormal conditions
computational cost
+ MCSA incredibly complex
demands more
Ball bearings synchronous [17,18]
faults at the same time machines
Modified
Wavelet windings ap-
transform memory
Accurate for faulty signals
proach Needto inductance as Ball bearings Induction motors
Park vector having
Easy to harmonics
implement for in their
Difficult [15]
approach + implement undera parameter Three-phase
+MCSA
order
recognizing multiclass
varying load
Ball bearings
induction motors
[19–22]
problems in machines
StatorMCSA
current harmonics conditions
Permanent mag-
Works well in noisy envi- Not applicable to
+ Ball bearings net synchronous [16,52,54]
ronments to extract features stationary signals
MCSA machines
Finite element analysis Able to handle incredibly High computational Permanent mag-
+ complex faults at the same cost demands more Ball bearings net synchronous [17,18]
Wavelet transform time memory machines
Energies 2023, 16, 296 14 of 54
Table 1. Cont.
Technique/Assisting Bearing Fault

Pros Cons Applications References
Technique Type
Discrete wavelet
Able to analyze the Poor directionality
transform Inner raceway and
faulty signal in the time and a lack of phase Induction motors [18]
+ outer raceway
and frequency domain information
FFT
Stray flux
For same flux
measurements Less sensitive and
measurements, shows
+ demands precise Ball bearings Induction motors [24–26]
greater values of faulty
Power spectral calibration
signals
density
Instantaneous Fluctuations in
Provide an
frequency measuring the
understanding of where Ball bearings and
+ exact numeric Induction motors [27,28,30]
the useful information is roller bearings
Power spectral value of faulty
lying in the faulty area
density signals
Spectral kurtosis
Fast approximation Demands prior
energy Outer race, inner
ability for fault detection information and Induction
+ race faults of ball [29]
and classification rates hasa high machines
Vibration and bearings
in electrical machines computational cost
current signals
High-frequency
Ability to detect
inductance coil Hasa larger
abnormal conditions in
+ decrement in the
machines under high Ball bearings Induction motors [31]
Vibration and detection of faulty
voltages and current
acceleration signals
circumstances
signals
Main bearing
Provides excellent
fundamental Unable to
localization of
frequency reconstruct the Ball bearings Wind turbines [34]
transientvalues of
+ signal coefficients
bearing faults
Vibration signals
Mutual
Measures the value in
information on
multiple predictable Hasa greater risk
current and Converter-fed
means and shows high of manipulation of Ball bearings [32,33]
voltage electrical machines
efficiency to isolate data
+
bearing faults
FFT
Measures the value in
Impact of load on
multiple predictable Has a greater risk Permanent magnet
stator current Cylindrical roller
means and shows high of manipulation synchronous [35]
+ bearings
efficiency to isolate for faulty data machines
MCSA
bearing faults
Envelope analysis Distortion will
Aggressive and has the
+ occur if the faulty Angular contact
ability to work under Induction motors [28]
Airborne noise signal is bearings
varying loads conditions
signals over-modulated
WPT/RNFC filter Has the ability to detect Demands incessant
+ bearing faults at early data for fault Roller bearings Induction motors [36,37,42]
Vibration signals stages analysis
Rational energy
harvester Precise to detect the
Poor results for
+ fault under varying Roller bearings Washing machines [38]
noisy signals
Piezoelectric loads conditions
sensors
Energies 2023, 16, 296 15 of 54
Table 1. Cont.

Technique Type
Hilbert transform Noninvasive, suitable
Does not provide
+ for constant load Variable speed
perfect information BLDC motors [39,40]
Zero phase conditions to classify bearings
on faulty signal
filtering multiple faults
Local outlier factor
Easy to implement,
algorithm Limited
ability to work in a strict
+ bandwidth for Axle-box bearings High-speed trains [41]
environment to detect
Abnormality index faulty signals
the fault
model
Improved
variational mode Noninvasive, suitable to Affected by
decomposition classify multiple faults environmental Ball bearings Rotating machines [43]
+ at the same time changes
EMD
High-resolution
spectral analysis of Accurate and can Expensive and has
Inner race and Induction
stator current measure faults where a nonlinear [44]
outer race faults machines
+ segregation is difficult response
Phase modulation
Probabilistic model Has the ability to
+ decompose a signal into Poor performance
Ball bearings Induction motors [45,56]
Vibration multiresolution to for noisy signals
disturbances segregate bearing faults
Adaptive
Aggressive and has the The complex
stochastic
ability to implement calculation for Brushed/brushless
resonance filter Wind turbines [46,54]
under low slip nonperiodic faulty DC motors
+
conditions signals
Vibration signals
Synchro-
Offer simultaneous
squeezing wavelet Inner race and
localization of faulty
transform Not applicable to outer race
signals in the FT domain BLDC motors [47]
+ stationary signals Ball
to detect abnormal
Tacholess order bearing
conditions
tracking
Thermal infrared
Hasthe ability to
imaging Expensive and Outer race, inner
perform in low visibility
+ hasa nonlinear race faults of ball Induction motors [48]
situations for accurate
Discrete Wavelet response bearings
detection of faults
transform
Spectral regression Cost-effective and High cost and
+ provides a wide range affected by
Roller bearings Induction motors [49,50]
Electrostatic for fault-complicated environmental
sensors environments conditions
Modulation/bispectrum
Cost-effective and High cost and
analysis
provides a wide range affected by
+ Roller bearings Wind turbines [51,55]
for fault-complicated environmental
Fundamental
environments conditions
frequency
Homogeneity The best algorithm for
algorithm comparative analysis of Time-consuming Outer race ball
Induction motors [57]
+ the multiple faults and expensive bearing
Vibration signals occurred in machines
Energies 2023, 16, 296 16 of 54
Table 1. Cont.

Technique Type
Normalized triple Proves as a valuable tool The chances of
covariance for the selection of faulty errors are very
Ball bearings Induction motors [58]
+ signals from the matrix high due to
MCSA of random variables complex values
Envelope The most common Vibrations also
harmonic product technique for analysis of
Energies 2023, 16, x FOR PEER REVIEW contribute to Locomotive
Roller bearings [59]16 of 54
+ faulty conditions in unwanted parts of bearing fault
CYCBD spectrum bearings the signals
3. Statistical Techniques for Diagnosis of Bearing Faults
3. Statistical Techniques for Diagnosis of Bearing Faults
Statistical approaches are generally employed in research to collect and explore a
Statistical approaches are generally employed in research to collect and explore a
large amount of data and find out hidden trends. These techniques are implemented to
large amount of data and find out hidden trends. These techniques are implemented to
explore the trends, the relationship between variables, and approximation analysis by us‐
explore the trends, the relationship between variables, and approximation analysis by using
ing quantitative data. Statistical parameters commonly implement root mean square val‐
quantitative data. Statistical parameters commonly implement root mean square values,
ues, mean values, variance, skewness, kurtosis, and crest factor attributes for measuring
mean values, variance, skewness, kurtosis, and crest factor attributes for measuring the
the consequences of bearing faults. In [60], the authors proposed an approach to diagnos‐
consequences of bearing faults. In [60], the authors proposed an approach to diagnosing
ing the outer race and inner race bearing faults of IM using statistical time features with
the outer race and inner race bearing faults of IM using statistical time features with
neural networks. Statistical time features for bearing failures were obtained from vibra‐
neural networks. Statistical time features for bearing failures were obtained from vibration
tion signals and compressed by using curvilinear component analysis for visualization
signals and compressed by using curvilinear component analysis for visualization behavior.
behavior. Feature‐extracted signals were employed to identify a crack in the outer race,
Feature-extracted signals were employed to identify a crack in the outer race, deformation
deformation of the seal, and a hole in the outer race. Figure 19 shows the frequency plot
of the seal, and a hole in the outer race. Figure 19 shows the frequency plot for the
for the damaged bearing seal condition and hole in the outer race fault by implementing
damaged bearing seal condition and hole in the outer race fault by implementing stray
stray flux measurements.
flux measurements.

Figure 19. (a) Applied probes for amplification and filtering (b) frequency plot for the damaged
Figure 19. (a) Applied probes for amplification and filtering (b) frequency plot for the damaged
bearing seal condition and (c) hole in outer race fault [60].
bearing seal condition and (c) hole in outer race fault [60].
In [61], the authors implemented empirical mode decomposition (EMD) of acoustic

In [61], the authors implemented empirical mode decomposition (EMD) of acoustic
signals obtained through kurtosis and crest factors of the time domain for checking the
signals obtained through kurtosis and crest factors of the time domain for checking the
status of healthy and faulty roller bearings. In [62], the authors suggested a step-varying
status of healthy and faulty roller bearings. In [62], the authors suggested a step‐varying
vibrational resonance (SVVR) algorithm to investigate the faulty status of a bearing by
vibrational resonance (SVVR) algorithm to investigate the faulty status of a bearing by
regulating the diverse parameters. Extraction features by SVVR were compared with simu-
regulating the diverse parameters. Extraction features by SVVR were compared with sim‐
ulated signals and practical signals, and results from SVVR depicted incredible precision
in extracting and enhancing the weak information status of bearing fault detection in the
time domain. In [63], the authors pointed out the bearing faults by employing the empir‐
ical cumulative distribution function to explore hidden patterns collected from extracted
Energies 2023, 16, 296 17 of 54
lated signals and practical signals, and results from SVVR depicted incredible precision in
extracting and enhancing the weak information status of bearing fault detection in the time
domain. In [63], the authors pointed out the bearing faults by employing the empirical cu-
mulative distribution function to explore hidden patterns collected from extracted features
through the signal-to-noise ratio (SNR). Extracted features were employed to diagnose the
statistical spectral images and classify the multiple bearing faults that frequently take place
in revolving electrical machines.
In [64], the authors proposed a methodology for bearing fault diagnosis under steady
and
inconsistent speed operations. Absolute value principal component analysis (AVPCA) 17 of 54
was employed for plotting images of vibration signals through the time–frequency domain,
fast Fourier transform, and probability plot. By applying AVPCA, the identification of
bearing faults was made possible and could also be used to diagnose gearbox faults.
shows 20
Figure theshows
resultthe
of result
the outer race,
of the outerinner race,
race, and
inner ball
race, andbearing fault detection
ball bearing and their
fault detection and
classification by implementing the sum of square error (SSE) distance evaluation.
their classification by implementing the sum of square error (SSE) distance evaluation.
Figure 20.
Figure 20. Sum
Sum of square error
of square error (SSE)
(SSE) distance
distance variation
variation for
for outer
outer race
race inner
inner race
race and
and ball
ball bearing
bearing
fault. Reproduced
fault. with permission
Reproduced with permission of
of [64].
[64].
In [65],
In [65], the
the authors
authors suggested
suggested aa variety
variety of
of research trends for
research trends for the
the diagnosis
diagnosis of roller
of roller
bearing
bearing faults
faults depending
dependingon onthe
thetime-varying
time-varying Kalman
Kalman filter forfor
filter accurate prediction
accurate of the
prediction of
remaining
the remaininguseful life (RUL).
useful The Kalman
life (RUL). The Kalmanfilterfilter
is employed
is employed on linear and quadratic
on linear func-
and quadratic
tions to forecast
functions the RUL
to forecast the in roller
RUL in bearings. Figure 21
roller bearings. shows
Figure 21the bearing
shows the test rig diagram
bearing test rig
for the prediction
diagram of RUL inofroller
for the prediction RUL bearings.
in roller bearings.
In [66], the authors suggested the identification of incipient bearing faults in rotating
machinery through single value decomposition (SVD) and the squared envelope spectrum
(SES) by employing vibration signals. The optimal singular component (SC) was utilized
through the kurtosis of SES for information gain and to demonstrate the results of incipient
faults. Figure 22 shows the setup for detecting bearing faults and their related results using
the kurtosis variation for different stages.
In [67], the authors proposed a methodology to identify and distinguish the multiple
kinds of ball bearing failures extracted by empirical mode decomposition (EMD) of non-
stationary time series signals. The Case Western Reserve University (CWRU) dataset was
applied to check the validity of the work to diagnose multiple kinds of bearing failures at
different loading conditions. Figure 23 shows the results of ball bearing faults at 0% loading
conditions and 100% loading conditions by implementing Kullback–Leibler divergence
based on empirical mode decomposition (EMD).
fault. Reproduced with permission of [64].
In [65], the authors suggested a variety of research trends for the diagnosis of roller
bearing faults depending on the time-varying Kalman filter for accurate prediction of the
remaining useful life (RUL). The Kalman filter is employed on linear and quadratic func-
Energies 2023, 16, 296 tions to forecast the RUL in roller bearings. Figure 21 shows the bearing test rig18diagram
of 54
for the prediction of RUL in roller bearings.

In [66], the authors suggested the identification of incipient bearing faults in rotating
machinery through single value decomposition (SVD) and the squared envelope spec‐
trum (SES) by employing vibration signals. The optimal singular component (SC) was
utilized through the kurtosis of SES for information gain and to demonstrate the results
of incipient faults. Figure 22 shows the setup for detecting bearing faults and their related
Figure 21. Degradation
Figure tracking
21. Degradation for the
tracking for RUL of roller
the RUL bearings
of roller
results using the kurtosis variation for different stages. [65].[65].
bearings

(a) (b)
Figure 22. (a) Bearing test rig and (b) kurtosis variation of incipient bearing faults at different stages
Figure 22. (a) Bearing test rig and (b) kurtosis variation of incipient bearing faults at different
[66].
stages [66].
kinds of ball bearing failures extracted by empirical mode decomposition (EMD) of non‐
different loading conditions. Figure 23 shows the results of ball bearing faults at 0% load‐
ing conditions and 100% loading conditions by implementing Kullback–Leibler diver‐
gence based on empirical mode decomposition (EMD).
kinds of ball bearing failures extracted by empirical mode decomposition (EMD) of non-
different loading conditions. Figure 23 shows the results of ball bearing faults at 0% load-
Energies 2023, 16, 296 19 of 54
ing conditions and 100% loading conditions by implementing Kullback–Leibler diver-
gence based on empirical mode decomposition (EMD).

Figure 23. Ball bearing results based on EMD at (a) 0% loading conditions and (b) 100%19loading
of 54
Figure 23. Ball bearing results based on EMD at (a) 0% loading conditions and (b) 100% loading
conditions. Reproduced with permission of [67].
conditions. Reproduced with permission of [67].
In
In [68],
[68], the
the authors
authors proposed
proposed 1.5-dimensional
1.5-dimensional symmetric
symmetric difference
difference analytical
analytical energy
energy
operators
operators to to enhance
enhance thetheextracted
extractedfeatures
featuresforforbearing
bearingfault
faultdiagnosis.
diagnosis.AA statistical
statistical filter-
filtering
ing channel was employed to reduce the background noise and improve the signal-to-
channel was employed to reduce the background noise and improve the signal-to-noise
noise ratiofor
ratio SNR SNR thefor the faulty
faulty bearing’s
bearing’s characteristic
characteristic frequency.
frequency. Figure Figure
24 shows24 shows the re-
the results for
sults for the diagnosis
the diagnosis of abnormalof abnormal
conditionsconditions in faultythrough
in faulty bearings bearingsdifferent
throughhigh-pass
different filters
high-
pass filters and
and matches matches
their effectstheir
witheffects withfiltering.
statistical statistical filtering.
Figure
Figure 24.
24. (a)
(a) Time
Time domain
domain of
of three
three different
different statistical
statistical filtering
filtering algorithms
algorithms using
using high
high pass
pass filter
filter
(HPF, green), Empirical mode decomposition (EMD, blue) and statistical filtering (SF, red). (b)
(HPF, green), Empirical mode decomposition (EMD, blue) and statistical filtering (SF, red). (b) Feature Fea-
ture extraction spectrum result for the diagnosis of bearing fault characteristic frequency. Repro-
extraction spectrum result for the diagnosis of bearing fault characteristic frequency. Reproduced
duced with permission of [68].
with permission of [68].
In
In [69],the
[69],the authors
authors suggested
suggested an an approach
approach forfor the
the investigation
investigation ofof roller
roller bearing
bearing faults
faults
detection and classification using radial internal clearance (RIC). Statistical
detection and classification using radial internal clearance (RIC). Statistical condition indi-condition in-
dicators, fast Fourier transform (FFT), and continuous wavelet transform
cators, fast Fourier transform (FFT), and continuous wavelet transform (CWT) techniques (CWT) tech-
niques were utilized
were utilized to evaluate
to evaluate the results.
the results. An experimental
An experimental setupsetup
was was created
created to validate
to validate the
the RIC measurements to predict the lifetime of roller bearings accurately.
RIC measurements to predict the lifetime of roller bearings accurately. In [70], the authorsIn [70], the au-
thors proposed
proposed the bearing
the bearing fault diagnosis
fault diagnosis of hybrid
of hybrid stepperstepper
motorsmotors by employing
by employing the
the revolv-
revolving angle estimation
ing angle estimation of motor
of motor current.
current. Variational
Variational nonlinear
nonlinear chirpchirp
modemode decompo-
decomposition
sition was applied
was applied for thefor the extraction
extraction of features
of features from from the vibration
the vibration signalssignals spectrum
spectrum orderorder
and
and resembledthe rotating angle curve to diagnose bearing faults. In [71], the
resembledthe rotating angle curve to diagnose bearing faults. In [71], the writers suggested writers sug-
gested
a hybrid a hybrid approach
approach utilizing
utilizing the kurtosis
the kurtosis spectrum
spectrum and and envelope
envelope spectrum
spectrum analysis
analysis of
of the stator current for the prognosis of roller bearing faults detection and classification
in aerospace and industrial applications. An estimation index was set with different bear-
ings experiments performed under variable load conditions. Figure 25 shows the results
of the stator current spectrum with healthy and damaged bearings for different bearing
Energies 2023, 16, 296 20 of 54
the stator current for the prognosis of roller bearing faults detection and classification in
aerospace and industrial applications. An estimation index was set with different bearings
experiments performed under variable load conditions. Figure 25 shows the results of the
stator current spectrum with healthy and damaged bearings for different bearing faults
by employing a fast kurtogram and a wavelet kurtogram and enhancing the results of
envelope analysis for bearing fault diagnosis.
Figure 25.
Figure 25.Comparison
Comparison ofof the
the stator
stator current
current spectrum
spectrum for
for (a)
(a) healthy
healthy bearing
bearing and
and (b)
(b) faulty
faulty bearing,
bearing,
where significant change in the amplitude of certain harmonics is highlighted by circles
where significant change in the amplitude of certain harmonics is highlighted by circles [71]. [71].
In [72],
In [72], the
the authors
authors suggested
suggested aa methodology
methodology that that depends
depends on on the
the improved
improved version
version
of singular
of singular value
valuedecomposition
decomposition(SVD) (SVD)for forthe
theselection
selection of of
thethe
kurtosis
kurtosisenvelope andand
envelope op-
timized byby
optimized ananenhanced
enhanced version
versionofofwavelet
waveletpacket
packettransform
transform(WPT)
(WPT) forfor the
the diagnosis
diagnosis ofof
roller bearing
roller bearing faults
faults detection
detection and and classification.
classification. An An indicator-based
indicator-based filter
filter was
was designed
designed
to extract
to extract frequency
frequency characteristics
characteristics fromfrom anan envelope
envelope spectrum
spectrum analysis
analysis forfor the
the bearing
bearing
fault diagnosis.
fault diagnosis. In In[73],
[73],thetheauthors
authors proposed
proposed thethe
rectangular
rectangularresampling
resampling methodology
methodology for
the diagnosis
for the diagnosisof bearing faultsfaults
of bearing in electric motorsmotors
in electric under variable speeds using
under variable speedstheusing
Kanade–the
Lucas–Tomasi (KLT) algorithm.
Kanade–Lucas–Tomasi Faulty indications
(KLT) algorithm. were obtained
Faulty indications throughthrough
were obtained the micro-
the
phone and recorded
microphone by a high-speed
and recorded camera
by a high-speed in aninangular
camera domain-based
an angular domain-based angle curve
angle for
curve
obtaining
for characteristic
obtaining characteristicfrequency
frequency in in
brushless
brushlessdirect
directcurrent
currentmotors
motors(BLDC).
(BLDC). Figure
Figure 2626
shows the instantaneous rotating angle (IRA) and instantaneous rotating frequency (IRF)
for the
for the diagnosis
diagnosis of of the
the inner
inner race
race and
and outer
outer race
race bearing
bearing faults.
faults.
In [74], the authors suggested failures of roller bearings in inverter-fed revolving
machines during transient bearing current patterns and estimate the remaining useful
life (RUL). In [75], the authors proposed an efficient technique to find out the unknown
abnormal conditions of bearings employing time-varying speed characteristics and strong
background noise. Stochastic resonance was utilized on the input signal to evaluate the non-
stationary features of the faulty signal, and an imaginary index was set up to convert the
angular domain into a spectrum for bearing fault diagnosis. In [76], the authors proposed
a speed-based methodology to evaluate the main bearing fault of wind turbines (MBWT)
by implementing the signals of shaft speed under a steady-state load. Various kinds of
ball/roller bearing faults, such as outer-race faults were diagnosed through tower signals
and empirical mode decomposition. Figure 27 shows the geometry of the wind turbines
main bearing.
Figure 26.Instantaneous rotating angle (IRA) and instantaneous rotating frequency (IRF) curve for
diagnosis of (a) outer race bearing fault and (b) inner race bearing fault. Reproduced with permis-
fault diagnosis. In [73], the authors proposed the rectangular resampling methodology for
the diagnosis of bearing faults in electric motors under variable speeds using the Kanade–
Lucas–Tomasi (KLT) algorithm. Faulty indications were obtained through the micro-
phone and recorded by a high-speed camera in an angular domain-based angle curve for
obtaining characteristic frequency in brushless direct current motors (BLDC). Figure 26
Energies 2023, 16, 296 21 of 54
for the diagnosis of the inner race and outer race bearing faults.
In [74], the authors suggested failures of roller bearings in inverter-fed revolving ma-
chines during transient bearing current patterns and estimate the remaining useful life
(RUL). In [75], the authors proposed an efficient technique to find out the unknown ab-
normal conditions of bearings employing time-varying speed characteristics and strong
background noise. Stochastic resonance was utilized on the input signal to evaluate the
non-stationary features of the faulty signal, and an imaginary index was set up to convert
the angular domain into a spectrum for bearing fault diagnosis. In [76], the authors pro-
posed a speed-based methodology to evaluate the main bearing fault of wind turbines
(MBWT) by implementing the signals of shaft speed under a steady-state load. Various
kinds of ball/roller
Figure bearingrotating
26. Instantaneous faults,angle
such (IRA)
as outer-race faults were
and instantaneous diagnosed
rotating through
frequency (IRF) tower
curve for
signals
Figure and empirical
26.Instantaneous mode decomposition.
rotating angle (IRA) Figure
and 27 shows
instantaneous the geometry
rotating frequency
diagnosis of (a) outer race bearing fault and (b) inner race bearing fault. Reproduced withof the
(IRF) wind for
curve
permission
turbines
of [73], main
diagnosis of bearing.
(a)
the initial
outer race bearing fault and (b) inner race bearing fault.
value of IRF at t = 0 is shown by red circle and arrow.
Reproduced with permis-
sion of [73], the initial value of IRF at t=0 is shown by red circle and arrow.
Figure 27.27.
Figure Wind turbine
Wind main
turbine bearing
main (WTMB)
bearing geometry
(WTMB) [76].
geometry [76].
Figure 28 28
Figure shows
shows thethe
results of of
results thethe
main wind
main windturbine bearing
turbine bearingfault byby
fault employing
employing
absolute value principal component analysis (AVPCA) via the shaft speed signal under
constant
constantloads. Ball-free
loads. faultfault
Ball-free (BFF),(BFF),
ball outer race fault
ball outer race(BOF), ball inner
fault (BOF), ballrace fault
inner (BIF),
race fault
and ball bearing fault (BBF) are evaluated by different input schemes to diagnose bearing
(BIF), and ball bearing fault (BBF) are evaluated by different input schemes to diagnose
faults.
bearing faults.
In [77], the authors proposed the identification of roller bearing faults by implementing
multivariable statistical process control methods (MSPC) such as higher-order cumulants
analysis (HCA), independent composite analysis (ICA), and principal component anal-
ysis (PCA). Two bearings’ life datasets were evaluated, and the outcomes provided the
suggested approaches and showed the paramount precision for the diagnosis of bearing
faults. Figure 29 shows the results of higher order cumulants analysis (HCA), independent
component analysis (ICA), and principal component analysis (PCA) through multivariate
statistical process control (MSPC) to diagnose the rolling elements bearing faults; the red
horizontal lines show dynamic control limits, and the black dashed lines demonstrate the
incipient bearing fault.
Figure 28 shows the results of the main wind turbine bearing fault by employing
constant loads. Ball-free fault (BFF), ball outer race fault (BOF), ball inner race fault (BIF),
Energies 2023, 16, 296 and ball bearing fault (BBF) are evaluated by different input schemes to diagnose bearing 22 of 54
faults.
Figure 28. (a) IMF order of faulty generator bearing vibration signal (b) Sum of square error (SSE)
distance variation to diagnose wind turbine main bearing fault (WTMB) detection at different inputs
(red arrow shows the outer race fault order and its harmonics) [76].
In [77], the authors proposed the identification of roller bearing faults by implement-
ing multivariable statistical process control methods (MSPC) such as higher-order cumu-
lants analysis (HCA), independent composite analysis (ICA), and principal component
analysis (PCA). Two bearings’ life datasets were evaluated, and the outcomes provided
the suggested approaches and showed the paramount precision for the diagnosis of bear-
ing faults. Figure 29 shows the results of higher order cumulants analysis (HCA), inde-
pendent component analysis (ICA), and principal component analysis (PCA) through
multivariate statistical process control (MSPC) to diagnose the rolling elements bearing
Figure 28. (a) IMF order of faulty generator bearing vibration signal (b) Sum of square error (SSE)
faults; the red horizontal lines show dynamic control limits, and the black dashed lines
distance variation to diagnose wind turbine main bearing fault (WTMB) detection at different inputs
demonstrate the incipient bearing fault.
(red arrow shows the outer race fault order and its harmonics) [76].
Figure 29.
Figure 29. Dynamic
Dynamic monitoring
monitoring ofof bearing
bearing faults
faults by
by (a)
(a) PCA,
PCA, (b)
(b) ICA,
ICA, and
and (c)
(c) HCA
HCA approaches.
approaches.
Reproduced with permission of [77] (The X label shows the total number of sample points,
Reproduced with permission of [77] (The X label shows the total number of sample points, and and the
the
Y-label shows the number of training samples).
Y-label shows the number of training samples).
In [78], the authors proposed a hybrid strategy to investigate wind turbine generator
bearing faults by employing sparse representation and shift-invariant dictionary learning
for the extraction of features in the frequency domain. In [79], the authors suggested linear
discriminant analysis (LDA) and principal component analysis (PCA) for the judgment of
ball bearing faults focused on global spectrum analysis. Envelope spectrums were applied
for the extraction of features from vibration signals and to obtain useful information from
Energies 2023, 16, 296 23 of 54
In [78], the authors proposed a hybrid strategy to investigate wind turbine generator
bearing faults by employing sparse representation and shift-invariant dictionary learning
for the extraction of features in the frequency domain. In [79], the authors suggested linear
discriminant analysis (LDA) and principal component analysis (PCA) for the judgment of
ball bearing faults focused on global spectrum analysis. Envelope spectrums were applied
Energies2023,
Energies 16,xxFOR
2023,16, FORPEER
PEERREVIEW
REVIEW
for the extraction of features from vibration signals and to obtain useful information 23from
23 ofof54
54
bearing-specific characteristics and frequencies. Figure 30 shows the results of ball bearing
failures depending on the global pattern of vibration signals by utilizing PCA.
Figure30.
Figure
Figure 30.(a)
30. (a)Discrimination
(a) Discriminationof
Discrimination ofbearing
of bearingfaults
bearing faultsby
faults byPCA;
by PCA;(b)
PCA; (b)contour
(b) contourof
contour ofPDF
of PDFfor
PDF forestimated
for estimatedclass
estimated classof
class ofball
of ball
ball
bearingfault
bearing
bearing fault[79].
fault [79].
[79].
Similarly,Figure
Similarly,
Similarly, Figure31
Figure 31
31 shows
shows
shows the
the
the results
results
results aof
of of aafaulty
faulty
faulty ballbearing
ballball bearing
bearing based based
based on
on the on theglobal
the
global global
range
range
range
of ofofvibration
vibration
vibration signals
signalssignals byutilizing
by utilizing
by utilizing LDA for LDA
LDA for
thefor thediagnosis
the diagnosis
diagnosis ofball
of ballof ballbearing
bearing bearing faults.
faults.faults.
Figure31.
Figure
Figure 31.(a)
31. (a)Discrimination
(a) Discriminationof
Discrimination ofbearing
of bearingfaults
bearing faultsby
faults byLDA;
by LDA;(b)
LDA; (b)contour
(b) contourof
contour ofPDF
of PDFfor
PDF forestimated
for estimatedclass
estimated classof
class ofball
of ball
ball
bearing
bearing fault
fault [79].
[79].
bearing fault [79].
InIn[80],
In [80],the
[80], theauthors
the authorssuggested
authors suggestedaaastatistical-based
suggested statistical-basedindicator
statistical-based indicatorfor
indicator forthe
for thediagnosis
the diagnosisof
diagnosis ofbearing
of bearing
bearing
faultsin
faults
faults ininPSPM
PSPMby
PSPM byemploying
by employing
employing the
the
the stator
stator
stator currents
currents
currents forthe
forfor
the thereplacement
replacement
replacement ofoffaulty
of faulty faulty bearings
bearings
bearings with
withhealthy
with
healthy healthy bearings.
bearings.
bearings. Figure Figure
Figure 3232shows
32 shows shows thereplacement
the replacement
the replacement ofoffaulty
of faulty faulty bearings
bearings
bearings with withwithhealthy
healthy
healthy ones
onesutilizing
ones utilizing
utilizing aavibration
vibration
a vibration indicator.
indicator.
indicator. N-change
N-change
N-change stands
standsstands
for when forawhen
for when aabearing
bearing bearing isisreplaced
is replaced replaced with
with
with a faulty
afaulty
faulty
aone, andone,one, andN-default
and
N-default N-default stands
stands
stands for when for
for whenaavibration
a when
vibration vibration indicator
indicatorindicator startstotorespond.
starts tostarts
respond. respond.

Energies 2023, 16, 296 24 of 54
Figure 32. Replacement of healthy bearings with faulty ones by using statistical indicators [80].
In
In[81],
[81],thetheauthors
authors proposed
proposed a prognostic
a prognostic approach
approach to forecasting
to forecastingthe remaining
the remaininguse-
ful life (RUL) of a bearing by utilizing the bidirectional strategy based on multiple statis-
useful Inlife
[81],(RUL)
the authors proposed
of a bearing by autilizing
prognostic theapproach to forecasting
bidirectional the remaining
strategy based use-
on multiple
tics of vibration signals. An enhanced Kalman filterfilter
was implemented as a as
maximization
ful life (RUL)
statistics of a bearing
of vibration by An
signals. utilizing
enhanced the bidirectional
Kalman strategy based on multiple
was implemented statis-
a maximiza-
algorithm to estimate the remaining usefulusefullife (RUL) of bearings. In [82], authors pro-
tics of
tion vibrationtosignals.
algorithm estimate Anthe
enhanced
remaining Kalman filter lifewas
(RUL)implemented
of bearings. as In
a maximization
[82], authors
posed to estimate the remaining useful life (RUL) and identification of bearing faults by
algorithm to
proposed toestimate
estimatethe theremaining
remaininguseful
usefullife life(RUL)
(RUL)and of identification
bearings. In [82], authorsfaults
of bearing pro-
introducing the the
time–frequency characteristics for extracting features fromfrom
vibration sig-
posed
by to estimate
introducing the remaining useful
time–frequency life (RUL) and
characteristics identification
for extracting of bearing
features faults by
vibration
nals. Curve fitting and extended Kalman filtering approaches were employed for the as-
introducing
signals. Curve thefitting
time–frequency
and extended characteristics
Kalman filtering for extracting
approaches features
werefrom vibration
employed for sig-
the
sessment of bearing RUL. Figure 33 shows the predicted results for the RUL of the bearing
nals. Curveof
assessment fitting
bearing andRUL.
extended
FigureKalman
33 shows filtering
the predictedapproaches
resultswere employed
for the RUL of theforbearing
the as-
by employing an
an extended Kalman filter
filter with aa curve-fitting approach.
sessment
by employing of bearing RUL. Figure
extended Kalman33 shows the predicted
with results
curve-fitting for the RUL of the bearing
approach.
by employing an extended Kalman filter with a curve-fitting approach.
Figure
Figure 33.
33. RUL
RULestimation
estimationof
ofbearings
bearings(RUL
(RULprediction
predictionversus
versusactual
actualprediction)
prediction)Reproduced
Reproducedwith
with
permission
permission of
of [82].
[82].
Figure 33. RUL estimation of bearings (RUL prediction versus actual prediction) Reproduced with
permission of [82].
Energies 2023, 16, 296 25 of 54
In [83], the authors proposed a signal processing strategy to recognize rolling bearing
failures in revolving electric machines by employing an adaptive morphological update
lifting wavelet (AMULW). In [84], the authors suggested diagnosing the bearing faults
by utilizing stator current noise cancellation and statistical control to suppress dominant
components and bearing fault frequencies. Sigma limits were applied to detect the variation
in the magnitude of noise cancellation stator current and statistical process control technique
for accurate diagnosis of bearing faults. In [85,86], the authors proposed a statistical
approach to find out the multiple kinds of abnormal conditions in roller bearings by
implementing vibration energy on the spectra Quest machinery fault simulator. Table 2
shows a comparative analysis of several statistical diagnostic techniques.
Table 2. A summary of different statistical techniques for bearing fault diagnosis.

Pros Cons Applications Reference
Technique Type
Curvilinear
component Dominant to identify Not capable of
analysis bearing faults and not processing the data
+ affected by surrounding based on complex
Stray flux noise values
measurements
Empirical mode
decomposition Ability to decompose a
Limited to detect Gears and roller
+ faulty signal into Electric machines [61]
IR and OR faults bearings
Kurtosis and crest multiresolution
factors
Step-varying
Noninvasive and can
vibrational Demand high
segregate the faults
resonance computational Ball bearings Induction motors [62]
under low slip
+ efficiency
conditions
Duffing oscillator
Statistical spectral
Influential results for Calculation and
analysis Incipient ball
classification of bearing interruption are AC machines [63]
+ bearing faults
faults difficult
ECDF
Absolute value
principal
Reduce overfitting of Loss of useful
component 200 HP induction
faulty signal and information about Ball bearings [64]
analysis motors
improves visualization the faulty signals
+
RSB-BFD
Box-Cox
Estimate
transformation Noninvasive and can Demand complex
remaining useful Induction motors [65]
+ segregate faults processing
life
Kalman filter
Single value
decomposition Efficient and stable Results have worse Incipient bearing Aerospace and
+ algorithm removes noise quality and are faults of ball industrial [66,71]
Squared envelope from faulty signals limited bearings applications
spectrum
Kullback–
Leiblerdivergence Demand more
Ability to decompose Inner and outer
(KLD) computational
nonlinear faulty signals race ball bearing Induction motors [67]
+ power faulty
to isolate faults faults
Empirical mode signals
decomposition
Energies 2023, 16, 296 26 of 54
Table 2. Cont.

Technique Type
Recurrence Sensitive for
quantification Aggressive, able to complete
analysis diagnose faults under visualization of Roller bearings Induction motors [68]
+ steady load conditions faulty data in a
Statistical filter single view
Statistical
Complex to
condition Ability to find out the
implement and
indicators shortest way for the Roller bearings Induction motors [69]
demand large
+ segregation of faults
memory
FFT, CWT
Variational
Noninvasive and
nonlinear chirp
accurate under
mode Poor visualization Hybrid stepper
steady-state conditions Ball bearings [70]
decomposition of faulty bearing motors
to isolate abnormal
+
conditions
Order analysis
Improved singular
value Noninvasive and has
Lack of real-time
decomposition high performance to Roller bearings Induction motors [72]
response
+ classify multiple faults
WPT
Contactless
angular Ability to perform
resampling excellently under Expensive, subject
Roller bearings Induction motors [73]
method varying load conditions to sensors failures
+ to detect the faults
Order Analysis
Redundant
second-generation
wavelet packet
Easy to calculate, high Haspoor Estimation of RUL
transform Rotating machines [74,82]
signal-to-noise ratio directionality in roller bearings
+
Extended Kalman
filter
Stochastic
resonance Adaptive to load
Very sensitive to
+ changes and noise Ball bearings Induction motors [75]
harmonics
Characteristics resilient
frequency analysis
AVPCA Bad results due to
Adaptive and able to
+ harmonics and Main wind turbine Inverter-fed
predict accurately faulty [76]
Shaft speed unwanted bearing rotating machines
signal
measurements vibrations
Multivariable
statistical process Simple to implement,
Interruption of
control shows exceptional Ball/roller
results is Wind turbines [77]
+ results to segregate bearings
impossible
HCA, ICA, and faults
PCA
Sparse
Able to remove Inner race and
representation
correlated features from Information loss outer race Wind turbines [78]
+
faulty signals ball/roller bearing
Laplace transform
Energies 2023, 16, 296 27 of 54
Table 2. Cont.

Technique Type
Statistical spectral
indicators The powerful Difficult for more
Ball/roller Low rotational
+ dimensions reduction than two-class [79,80,84]
bearings speed machines
LDA, PCA, and algorithm classification
MCSA
Clustering and
change point Demands prior
Easily handles problems
detection domain to extract
for more than two faults Roller bearings Induction motors [81]
algorithm (KPDA) faulty signals
at the same time
+ features
K-means clustering
Adaptive
morphological
update lifting Powerful technique and
Picks unwanted
wavelet coupling artificial Roller bearings Induction motors [83]
faulty signals
+ intelligence
Linear
decomposition
Maximum kurtosis
denoising Self-adapted data
Unable to identify
algorithm mining technique and
failures in a noisy Roller bearings Induction motors [85,86]
+ better visualization of
environment
Minimum entropy bearing faults
deconvolution
4. Artificial Intelligence (AI) Techniques for Diagnosis of Bearing Faults

In the past few years, researchers are paying attention to artificial intelligence-based
approaches (ML and DL) for the diagnosis of bearings’ abnormal conditions in electric
machines. These techniques follow hidden patterns to diagnose bearing faults and offer
sufficient diagnostic results. In [87], the authors implemented artificial bee colony archi-
tecture to identify bearing faults in three-phase induction motors. Faulty data was taken
out by characteristics of the joint in the sequence between the stator current signals in
the time domain. In [88,89], the authors suggested the efficient net optimizer for bearing
fault classification and their degradation level. Acoustic signals were employed to gain
faulty knowledge from spectrograms in the time domain. In [90], the authors proposed a
technique, a defective signature wavelet image (DWSI), with deep convolution networks to
trace the bearing faults by utilizing acoustic emission signals.
In [91,92], the authors diagnosed multiple roller bearing faults by experimental tests
and obtained the images by spectrograms to train CNN for bearing fault diagnosis in an
IM. In [93], the authors suggested an influential ML algorithm support vector machine
(SVM) with adaptive cascade fault observer and fuzzy orthonormal regressive to diagnose
the bearing faults of rotary machines. In [94], authors proposed support vector machines
(SVM) and decision trees (DT) to sort out the bearing faults in electric machines. Attribute
withdrawal methodology was employed to extract features from bearings that were opti-
mized by employing particle swarm optimization to recognize bearing faults. In [95],the
authors proposed a recurrent neural network (RNN) to detect faulty bearings in a noisy
environment and create a dataset by adding random noise for training the RNN. Figure 34
shows the confusion matrix plot for the RNN-WDCNN on the CWRU dataset to evalu-
ate bearing faults. Diagonal elements in the confusion matrix are accurately classified,
and non-diagonal elements are unclassified (normal, ball bearing, inner race, outer race)
bearing faults.
mized by employing particle swarm optimization to recognize bearing faults. In [95],the
authors proposed a recurrent neural network (RNN) to detect faulty bearings in a noisy
environment and create a dataset by adding random noise for training the RNN. Figure
34 shows the confusion matrix plot for the RNN-WDCNN on the CWRU dataset to eval-
uate bearing faults. Diagonal elements in the confusion matrix are accurately classified,
Energies 2023, 16, 296
and non-diagonal elements are unclassified (normal, ball bearing, inner race, outer race)
28 of 54
bearing faults.
Figure 34. Confusion matrix plot for proposed algorithm RNN-WDCNN to diagnose ball bearing,
Figure 34. Confusion matrix plot for proposed algorithm RNN-WDCNN to diagnose ball bearing,
inner raceway, and outer raceway bearing faults [95].
inner raceway, and outer raceway bearing faults [95].
In [96],
In [96], the
the authors
authors utilized
utilized aa support
support vector
vector machine
machine (SVM),
(SVM), K-nearest
K-nearest neighbor
neighbor
(KNN),and
(KNN), anddecision
decisiontrees
trees(DT)
(DT)for
forthe
the classification
classificationofof incipient
incipientbearing
bearingfaults
faultsby
by utilizing
utilizing
the CWRU
the CWRU dataset.
dataset. Features
Features ofof the
the dataset
dataset were
were tuned
tuned by by EMD-KLD
EMD-KLD for for bearing
bearing fault
fault
classification.Figure
classification. Figure3535shows
showsthe
the results
results of the
of the direct
direct acyclic
acyclic graph
graph support
support vector
vector ma-
machine
chine (DAG-SVM), KNN, and DT for evaluating
(DAG-SVM), KNN, and DT for evaluating bearing faults. bearing faults.
Figure35.
Figure 35. Analysis
Analysis results
results of
of bearing
bearing faults:
faults: (a)
(a) training
training accuracy;
accuracy; (b)
(b) testing
testing accuracy.
accuracy. Reproduced
Reproduced
withpermission
with permissionofof[96].
[96].
In[97],
In [97],the
theauthors
authorsrecommended
recommendedthe thediagnosis
diagnosisof ofbearing
bearingfaults
faultsby
byemploying
employingdiscrete
discrete
wavelet transform (DWT) and the ensemble ML algorithm. Current signals were
wavelet transform (DWT) and the ensemble ML algorithm. Current signals used
were to ex-
used to
tract features using discrete wavelet transform (DWT) for creating a dataset for training the
extract features using discrete wavelet transform (DWT) for creating a dataset for training
the random
random forest
forest (RF) (RF) and extreme
and extreme gradient
gradient boostingboosting for bearing
for bearing fault classification.
fault classification. Figure 36
Figure
shows 36theshows
ROC the
curveROCforcurve for evaluating
evaluating the classification
the classification performance
performance of RF and of extreme
RF and
extreme gradient boosting for the diagnosis of bearing
gradient boosting for the diagnosis of bearing faults. faults.
In [97], the authors recommended the diagnosis of bearing faults by employing discrete
wavelet transform (DWT) and the ensemble ML algorithm. Current signals were used to ex-
tract features using discrete wavelet transform (DWT) for creating a dataset for training the
random forest (RF) and extreme gradient boosting for bearing fault classification. Figure 36
Energies 2023, 16, 296 shows the ROC curve for evaluating the classification performance of RF and extreme 29 of 54
gradient boosting for the diagnosis of bearing faults.
Figure 36.
Figure 36. ROC
ROC curves
curves for
for (a)
(a) RF
RF and
and (b)
(b) XG-Boost
XG-Boost for
for bearing
bearing fault
fault classification
classification[97].
[97].
In [98],
In [98],the
theauthors
authorssuggested
suggesteddeepdeepbelief
belief network
network andand
thethe Dempster–Shafer
Dempster–Shafer theory
theory for
for diagnosis
the the diagnosis of bearings
of bearings in rotating
in rotating AC machines.
AC machines. CWRUCWRU was putwasintoput into practice
practice to
to validate
validate
the the performance
performance of the proposed
of the proposed algorithm.algorithm. In [99],
In [99], the the proposed
authors authors proposed a feed-
a feed-forward
forward
neural neural(FFNN)
network networkto(FFNN)
identify to
theidentify thebearing
outer-race outer-race bearing
faults for onefaults for one HP
HP induction in-
motor
duction motor at its early stages. Linear discriminant analysis (LDA) was utilized for ex-
at its early stages. Linear discriminant analysis (LDA) was utilized for extracting features
tracting
to train thefeatures
FFNN.toFigure
train the FFNN.the
37 shows Figure 37 shows
outcomes the outcomes
of projection regionsoffor
projection regions
the classification
formultiple
of the classification of multiple
bearing faults bearing
by using faults
an FFNN atby using an
different FFNN at different stages.
stages.
Figure 37.
Figure 37. Projection
Projection of
of decisions
decisions region
region for
for multiple
multiple faults
faults classification
classificationobtained
obtainedby
byFFNN
FFNN[99].
[99].
In [100], authors suggested an approach to diagnose the roller bearing faults in electri-
In [100], authors suggested an approach to diagnose the roller bearing faults in elec-
cal machines utilizing ensemble local x-tics-scale decomposition (ELCD) and an extreme
trical machines utilizing ensemble local x-tics-scale decomposition (ELCD) and an ex-
learning machine (ELM). Figure 38 shows the results for the prediction of fault diagnosis in
treme learning machine (ELM). Figure 38 shows the results for the prediction of fault di-
rolling bearing elements by implementing ELCD and ELM.
agnosis in rolling bearing elements by implementing ELCD and ELM.
Figure 37. Projection of decisions region for multiple faults classification obtained by FFNN [99].
In [100], authors suggested an approach to diagnose the roller bearing faults in elec-
trical machines utilizing ensemble local x-tics-scale decomposition (ELCD) and an ex-
Energies 2023, 16, 296 treme learning machine (ELM). Figure 38 shows the results for the prediction of fault 30 ofdi-
54
agnosis in rolling bearing elements by implementing ELCD and ELM.
Figure
Figure 38.Test
38. Testclassification
classificationresults
resultsfor
forthe
thediagnosis
diagnosisof
of rolling
rolling bearing
bearing element
element faults
faults in
in rotating
rotating
machines [100].
machines [100].
In [101], the authors suggested an intelligent neuro-wavelet neural network in combi-

nation with variational mode decomposition (VMD) to diagnose bearing faults. Vibration
signals were employed to extract useful information and remove the redundant features to
construct a dataset for training the intelligent neuro-wavelet neural network and estimating
bearing faults. In [102], the authors suggested an improved shuffled frog leaping algorithm
(ISFLA) for incipient fault detection of roller bearings. Particle swarm optimization was
utilized to tune the parameters of ISFLA for the detection of roller bearing faults. In [103],
the authors proposed a stack sparse auto-encoder (SSAE) with a deep neural network
(DNN) for the diagnosis of bearing faults by employing envelope spectrum analysis. The
complex envelope spectrum had silent features to extract features of each fault type present
in bearing at variable speeds. In [104], the authors proposed the Wasserstein generative
adversarial network and convolutional neural network (WG-CNN) for the diagnosis of
roller bearing faults. A high-quality bearing fault dataset was produced, and outcomes
were compared with the CWRU dataset for evaluation. In [105], the authors suggested
an artificial neural network (ANN) with a multilayer perception artificial neural network
(MLP-ANN) for roller bearing fault diagnosis. In [106], the authors suggested an artificial
neural network incorporated with Laplace transform to diagnose the inner race, outer
race, and rolling element faults. Time domain vibration signals were further processed
with Laplace transform for extracting useful information from bearings. Figure 39 shows
the Laplace wavelet kurtosis for the diagnosis of multiple bearing faults from simulated
vibration signals.
In [107], the authors proposed discriminant feature distribution analysis for the diag-
nosis of IM bearing faults with a genetic algorithm (GA) filter to optimize the attributes and
train KNN to classify bearing faults. In [108], the authors suggested compressive sampling
(CS) on the numerous measurements vector (MMV) and feature ranking framework for
the categorization of roller bearing faults obtained through pulsation indication. Support
vector machines (SVM), artificial neural networks (ANN), and logistic regression (LR) were
employed to validate the outcomes. In [109], the authors proposed K-means clustering op-
timized by a genetic algorithm (GA). The genetic algorithm can overcome the drawbacks of
K-means clustering for extracting useful information to point out faulty bearings. Figure 40
shows the 3D plot of clustering results for testing data to diagnose bearing faults for the
healthy, inner race, outer race, and ball damage conditions.
suggested an artificial neural network (ANN) with a multilayer perception artificial neu-
ral network (MLP-ANN) for roller bearing fault diagnosis. In [106], the authors suggested
an artificial neural network incorporated with Laplace transform to diagnose the inner
race, outer race, and rolling element faults. Time domain vibration signals were further
processed with Laplace transform for extracting useful information from bearings. Figure
Energies 2023, 16, 296
39 shows the Laplace wavelet kurtosis for the diagnosis of multiple bearing faults31from
of 54
simulated vibration signals.
for the categorization of roller bearing faults obtained through pulsation indication. Support
vector machines (SVM), artificial neural networks (ANN), and logistic regression (LR) were
employed to validate the outcomes. In [109], the authors proposed K-means clustering op-
timized by a genetic algorithm (GA). The genetic algorithm can overcome the drawbacks of
K-means clustering for extracting useful information to point out faulty bearings. Figure 40
shows the 3D plot of clustering results for testing data to diagnose bearing faults for the
healthy, inner race, outer race, and ball damage conditions.
Figure 39. Laplace wavelet kurtosis for the diagnosis of multiple bearing faults [106].
Figure 39. Laplace wavelet kurtosis for the diagnosis of multiple bearing faults [106].
In [107], the authors proposed discriminant feature distribution analysis for the di-
agnosis of IM bearing faults with a genetic algorithm (GA) filter to optimize the attributes
and train KNN to classify bearing faults. In [108], the authors suggested compressive sam-
pling (CS) on the numerous measurements vector (MMV) and feature ranking framework
Figure 40.
Figure 40. The
The3D
3Dplot
plotofof
the
theclustering algorithm
clustering for for
algorithm the the
diagnosis of different
diagnosis bearing
of different faults.faults.
bearing Re-
produced with
Reproduced permission
with permission of of
[109].
[109].
In [110], the
In [110], the authors
authorsproposed
proposedananFGPA-based
FGPA-based online
online multi-core
multi-core system
system to diag-
to diagnose
nose and monitor in real-time the bearing faults signature extraction by employing
and monitor in real-time the bearing faults signature extraction by employing time–fre- time–
frequency analysis of acoustic signals through multiclass support vector machines.
quency analysis of acoustic signals through multiclass support vector machines. In [111], In [111],
the
the authors
authors suggested
suggested the the convolutional
convolutional neural
neural network
network (CNN)
(CNN) and
and envelope
envelope spectrum
spectrum
(ES) for the diagnosis of roller bearing faults utilizing acoustic emission analysis
(ES) for the diagnosis of roller bearing faults utilizing acoustic emission analysis (AE) under
(AE) un-
der variable speeds operations. The envelope spectrum was used to extract raw signals
and provide useful information about frequency peaks to create datasets and train the
CNN for the diagnosis of faults in roller bearings. In [112], the authors suggested the dis-
crete orthogonal Stock-well transform (DOST) with transfer learning to diagnose the
Energies 2023, 16, 296 32 of 54
variable speeds operations. The envelope spectrum was used to extract raw signals and
provide useful information about frequency peaks to create datasets and train the CNN
for the diagnosis of faults in roller bearings. In [112], the authors suggested the discrete
orthogonal Stock-well transform (DOST) with transfer learning to diagnose the healthy
Energies 2023, 16, xx FOR
2023, 16, FOR PEER condition, outer race, and inner race bearing faults through vibration imaging. Figure
PEER REVIEW
REVIEW of 41
32 of 54
Energies 32 54
shows vibration-based images obtained from the DOST to diagnose the healthy condition,
inner raceway, and outer raceway faults.
Figure
Figure 41.
41. Vibration-based
Vibration-basedimages from
images discrete
from orthogonal
discrete Stock-well
orthogonal transform
Stock-well (DOST)
transform for a bear-
(DOST) for a
ing’s healthy conditions (HC),inner race fault (IRF), and ball fault (BF) [112].
bearing’s healthy conditions (HC),inner race fault (IRF), and ball fault (BF) [112].
In
In [113],
[113], the
theauthors
authorsdiagnosed
diagnosedthe abnormal
the abnormal conditions in bearing
conditions andand
in bearing gears by uti-
gears by
lizing feature
utilizing featuresparse representation
sparse through
representation throughslow
slowfeature
featureanalysis
analysis(SFA)
(SFA)and
andlong
longshort-
short-
term
term memory
memory (LSTM).
(LSTM). Figure
Figure 42
42 shows
shows the
the monitoring
monitoring chart
chart for
for the
the diagnosis
diagnosis of
of bearing
bearing
fault
fault using
using sparse
sparse three-way
three-way data-based
data-based dynamic SFA (STBDSFA).
dynamic SFA (STBDSFA).
Figure 42. (a)

Figure 42. (a) Monitoring
Monitoring chart
chart for
for the
the diagnosis
diagnosis of
of bearing
bearing fault
fault by
by employing
employing STBDSFA
STBDSFA to detect
to detect
fault at the 121st point in the total of 500 samples. (b) Confusion matrix plot for bearing faults
fault at the 121st point in the total of 500 samples. (b) Confusion matrix plot for bearing faults clas-
sification [113].
classification [113].
In
In [114],
[114], the
the authors
authors proposed
proposed aa fuzzy
fuzzy sliding
sliding mode
mode observer
observer with
with decision
decision trees
trees (DT)
(DT)
for the diagnosis of roller bearing faults in industrial applications. In [115], the authors
for the diagnosis of roller bearing faults in industrial applications. In [115], the authors pro-
proposed a hybrid
posed a hybrid algorithm
algorithm by employing
by employing empirical
empirical waveletwavelet transform
transform (EWT), (EWT), fuzzy
fuzzy entropy,
entropy,
and SVMand SVM (EWTFSFD)
(EWTFSFD) to bearing
to diagnose diagnose bearing
faults. Thefaults. Thealgorithm,
proposed proposed EWTFSFD,
algorithm,
EWTFSFD,was utilized to decompose vibration signals into frequency and amplitude
modulation to achieve effective outcomes for bearing fault diagnosis. In [116], the authors
proposed the fault diagnosis of rolling element bearings by employing detrended fluctu-
ation analysis (DFA) and improved detrended fluctuation analysis (IDFA) incorporated
with linear discriminant analysis to trace out the bearing healthy state, inner race, outer
Energies 2023, 16, 296 33 of 54
was utilized to decompose vibration signals into frequency and amplitude modulation to
achieve effective outcomes for bearing fault diagnosis. In [116], the authors proposed the
fault diagnosis of rolling element bearings by employing detrended fluctuation analysis
(DFA) and improved detrended fluctuation analysis (IDFA) incorporated with linear dis-
criminant analysis to trace out the bearing healthy state, inner race, outer race, and
Energies 2023, 16, x FOR PEER REVIEW cage
33 of 54

fault in feature vectors. Figure 43 shows the testing results of DFA and IDFA for tracing the
multiple kinds of bearing faults.

Figure 43. Testing results for multiple kinds of bearing faults by implementing (a) DFA and
Figure 43.Testing results for multiple kinds of bearing faults by implementing (a) DFA and (b) IDFA
[116].
(b) IDFA [116].
In [117],
In [117], the
the authors
authors recommended
recommended K-nearest neighbor
K‐nearest (KNN)
neighbor andand
(KNN) random forest
random (RF)
forest
and decision trees (DT) to classify the bearing faults of IM. The autocorrelation technique
(RF) and decision trees (DT) to classify the bearing faults of IM. The autocorrelation tech‐
was employed for the extraction of the informative attribute from bearings and generated a
nique was employed for the extraction of the informative attribute from bearings and gen‐
dataset by using the accelerometer. Figure 44 shows the confusion matrix plot to diagnose
erated a dataset by using the accelerometer. Figure 44 shows the confusion matrix plot to
the healthy condition, inner raceway, and outer raceway faults of IM bearings through
diagnose the healthy condition, inner raceway, and outer raceway faults of IM bearings
KNN and RF.
through KNN and RF.
In [118], the authors proposed the diagnosis of incipient roller bearing faults through
wavelet transform (WT) optimized by particle swarm optimization (PSO) from vibration
signals at diverse speeds. In [119], the authors proposed an artificial neural network (ANN)
and convolutional neural network (CNN) for the diagnosis of bearing faults in rotating
machines using thermal images. Six kinds of thermal images of the rolling element were
diagnosed (one healthy and five faulty). Figure 45 shows the results of thermal images to
diagnose bearing faults for a healthy state, outer race, inner race, ball defect, cage defect,
and lack of lubrication.
In [120], the authors proposed a sensor-driven technique to diagnose the bearing
faults using a convolutional neural network with s-transform (ST-CNN). To enhance the
performance of the proposed algorithm, data from the sensor were converted into a 2D
time–frequency matrix, and the coefficient matrix was fed to ST-CNN to diagnose the
bearing faults. Figure 46 shows the block diagram of ST-CNN for training and testing
procedures and also shows the confusion matrix plot for diagnosing IRF and ORF.

In [117], the authors recommended K-nearest neighbor (KNN) and random forest
(RF) and decision trees (DT) to classify the bearing faults of IM. The autocorrelation tech-
nique was employed for the extraction of the informative attribute from bearings and gen-
erated a dataset by using the accelerometer. Figure 44 shows the confusion matrix plot to
Energies 2023, 16, 296 diagnose the healthy condition, inner raceway, and outer raceway faults of IM bearings
34 of 54
through KNN and RF.
In [118], the authors proposed the diagnosis of incipient roller bearing faults through
wavelet transform (WT) optimized by particle swarm optimization (PSO) from vibration
signals at diverse speeds. In [119], the authors proposed an artificial neural network
(ANN) and convolutional neural network (CNN) for the diagnosis of bearing faults in
rotating machines using thermal images. Six kinds of thermal images of the rolling ele-
ment were diagnosed (one healthy and five faulty). Figure 45 shows the results of thermal
images to diagnose bearing faults for a healthy state, outer race, inner race, ball defect,
cage defect,
Figure and lackROC
44. Combined of lubrication.
curve for proposed machine learning classifiers [117].
Figure 44. Combined ROC curve for proposed machine learning classifiers [117].
Figure
Figure45.Thermal
45. Thermalimages
imagesresults
resultsfor
for(a)
(a)healthy
healthystate,
state,(b)
(b)outer
outerrace,
race,(c)
(c)inner
innerrace,
race,(d)
(d)ball
balldefect,
defect,
(e)
(e)cage
cagedefect,
defect,and
and(f)
(f)lack
lackof
oflubrication.
lubrication.Reproduced
Reproducedwithwithpermission
permission ofof [119].
[119].
In [120], the authors proposed a sensor-driven technique to diagnose the bearing

faults using a convolutional neural network with s-transform (ST-CNN). To enhance the
performance of the proposed algorithm, data from the sensor were converted into a 2D
time–frequency matrix, and the coefficient matrix was fed to ST-CNN to diagnose the
bearing faults. Figure 46 shows the block diagram of ST-CNN for training and testing
procedures and also shows the confusion matrix plot for diagnosing IRF and ORF.
Energies
Energies 2023, 16, 296
2023, 16, x FOR PEER REVIEW 3535of
of 54
54
Figure 46. (a) ST-CNN block diagram for training and testing. (b) Confusion matrix plot for IRF and
Figure
Figure 46.
46. (a)
(a) ST-CNN
ST-CNN block
block diagram
diagram for
for training
training and
and testing.
testing. (b)
(b) Confusion
Confusion matrix
matrix plot
plot for
for IRF
IRF and
and
ORF [120].
ORF
ORF [120].
[120].
In
In [121],
[121],the
In [121], the authors
theauthors
authors suggested
suggested
suggested aa physics-based
physics-based
a physics-based convolutional
convolutional
neural network
neural network network
(PCNN)
(PCNN)
(PCNN)
algorithm algorithm
algorithm utilizing
utilizingutilizing kurtosis
kurtosiskurtosis and envelope
and envelope
and envelope analysis
analysis
analysis for rolling
for rolling
for rolling bearing
fault fault
bearing
bearing fault diag-
diag-
diagnosis.
nosis.
nosis. Figure
FigureFigure 47, shows
47, shows
47, shows the results
the results
the results of rolling
of rolling
of rolling bearing
bearingbearing faults detection
faults detection
faults detection and
and theirand their localization
their localization
localization through
through
through
PCNN and PCNN
PCNNtheirand
and their
their comparison
comparison
comparison with SVM with
andSVM
with SVM
ANN. and
and ANN.
ANN.
Figure
Figure 47.
47. (a)
(a) Speed
Speed tests
tests for
for the
the diagnosis
diagnosis of
of bearing
bearing faults
faults using
using PCNN.
PCNN. (b)
(b) Comparison
Comparison ofof fault
Figure 47. (a) Speed tests for the diagnosis of bearing faults using PCNN. (b) Comparison of fault
fault
detection and localization results with SVM and ANN. Reproduced with permission of [121].
detection
detection and
and localization
localization results
results with
with SVM
SVM and ANN. Reproduced
and ANN. Reproduced with
with permission
permission of
of [121].
[121].
In [122], the authors recommended Gramian angular field (GAF) integrated with
Dense-Net for feature extraction and convert them into 1D time series images to diagnose
the bearing faults. Figure 48 shows the patterns of GAF with Dense-Net for diagnosing
inner race and outer race faults bearing faults in electric machines.
In [122], the authors recommended Gramian angular field (GAF) integrated with
Dense-Net for feature extraction and convert them into 1D time series images to diagnose
Energies 2023, 16, 296 36 of 54
the bearing faults. Figure 48 shows the patterns of GAF with Dense-Net for diagnosing
inner race and outer race faults bearing faults in electric machines.
Figure 48.
Figure 48. Gramian
Gramianangular
angularfield
field(GAF)
(GAF)with
withDense-Net
Dense-Netpatterns
patternsfor
forrecognition
recognitionofofthe
the(a)(a)
inner race
inner race
and (b) outer race bearing faults expressed by using [122].
and (b) outer race bearing faults expressed by using [122].
In
In [123],
[123], the
the authors
authorsproposed
proposedaaconvolutional
convolutionalneural neuralnetwork
networkthrough
throughFault-Net
Fault-Netfor
for
the classification
the classification of of multiple
multiple types
types of
of bearing
bearing faults
faults employing
employingvibration
vibrationsignals.
signals.Mean
Mean
and median
and medianchannels
channelswere wereapplied
appliedto to
extract information
extract acquire
information fromfrom
acquire raw signals to cre-to
raw signals
ate a dataset and train CNN for bearing fault diagnosis. In [124], the
create a dataset and train CNN for bearing fault diagnosis. In [124], the authors proposed authors proposed
support vector
support vectormachines
machinesforfor bearing fault
bearing diagnosis
fault diagnosisusing radial
using basis basis
radial function (RBF) ker-
function (RBF)
nel in rotating machines. A hybrid feature pool was obtained through
kernel in rotating machines. A hybrid feature pool was obtained through time-domain time-domain fea-
tures, envelope power spectrum, and wavelet energy optimized by
features, envelope power spectrum, and wavelet energy optimized by the kernel functionthe kernel function for
providing
for providing input to the
input SVM
to the SVMforfor
bearing
bearing fault diagnosis.
fault In In
diagnosis. [125], the
[125], authors
the authorssuggested
suggested
the finite
the finite element
element method
method (FEM)
(FEM)to toovercome
overcomethe theshortage
shortageofofdata
datathat
thatarearea amuch-needed
much-needed
part of
part of the
the field
field of
ofmachine
machinelearning
learning(ML)
(ML)for fordiagnosis
diagnosisofofbearing
bearingfaults.
faults.Simulation-based
Simulation-based
finite element methods were employed to enlarge the datasets and applied totoML
finite element methods were employed to enlarge the datasets and applied MLclassifiers
classi-
to sort out bearing faults. Figure 49 shows the simulation results of the inner raceand
fiers to sort out bearing faults. Figure 49 shows the simulation results of the inner race and
outer race
outer race bearing
bearingfaults
faultsinincomparison
comparisonwith with experimental
experimental data
data andand synthetic
synthetic data
data 37 for
for of 54
their
their diagnosis.
diagnosis.
Figure 49.
Figure 49. Inner
Innerrace
raceand
andouter
outerrace
racebearing
bearingresults were
results obtained
were byby
obtained (a)(a)
experimental data
experimental andand
data (b)
synthetic data. Reproduced with permission of [125].
(b) synthetic data. Reproduced with permission of [125].
In [126,127], the authors proposed support vector machines (SVM) to diagnose the
bearing faults, employing complex signals in the rotating machines. Composite multiscale
weighted permutation entropy (CMWPE) was utilized to extract features and convert
them into the high dimensional feature set to input SVM for fault diagnosis. Figure 50
Energies 2023, 16, 296 37 of 54
Figure 49. Inner race and outer race bearing results were obtained by (a) experimental data and (b)
synthetic data. Reproduced with permission of [125].
In [126,127], the authors proposed support vector machines (SVM) to diagnose the
bearingIn [126,127],
faults, the authors
employing proposed
complex support
signals in thevector machines
rotating (SVM)Composite
machines. to diagnosemultiscale
the
bearing faults, employing complex signals in the rotating machines. Composite
weighted permutation entropy (CMWPE) was utilized to extract features and convert them multiscale
intoweighted
the highpermutation
dimensional entropy (CMWPE) was utilized to extract features and convert
feature set to input SVM for fault diagnosis. Figure 50 shows
them into the high dimensional feature set to input SVM for fault diagnosis. Figure 50
the results of bearing faults using CMWPE incorporated with SVM to diagnose the normal
shows the results of bearing faults using CMWPE incorporated with SVM to diagnose the
state (NOR),
normal inner
state race inner
(NOR), fault (IRF), outer
race fault raceouter
(IRF), fault race
(ORF), and
fault ball bearing
(ORF), and ballfaults(BBF)
bearing in
rolling element
faults(BBF) in bearings.
rolling element bearings.
Figure
Figure 50.50. CMWPE entropy
CMWPE entropy curves
curvestoto
diagnose
diagnose(a) normal, (b) inner
(a) normal, (b) raceway, (c) outer (c)
inner raceway, raceway,
outer (d)
raceway,
ball bearing faults, and (e) mean values for all states of fault faults. Reproduced with permission of
(d) ball bearing faults, and (e) mean values for all states of fault faults. Reproduced with permission
[126].
of [126].
In [128], the authors proposed the characteristics of time-domain zero-crossing fea-
In [128],
tures the authors
and defected proposed
frequencies the characteristics
obtained of time-domain
by vibration signals zero-crossing
to classify bearing features
faults. The
anddefective
defected frequencies
frequencies obtained by
are represented by vibration signals
spectral analysis, andtoclassification
classify bearing faults. The
is completed
by using
defective
Energies 2023, 16, x FOR PEER REVIEW FFNN for
frequencies bearing fault diagnosis.
are represented Figure
by spectral 51 shows
analysis, the results
and of multiple
classification bearing
is completed
38 of 54
failuresFFNN
by using by employing timefault
for bearing domain zero-crossing
diagnosis. Figurewith FFNN. the results of multiple bearing
51 shows
failures by employing time domain zero-crossing with FFNN.
Figure 51.
Figure 51.Zero-crossing
Zero-crossing features
features characteristics
characteristics for
for the
the diagnosis
diagnosis of
of multiple
multiple bearing
bearing faults.
faults. Repro-
Repro-
In [129], the authors suggested the incorporation of a multivariable feedback extreme

learning machine (MFELM) with convolutional neural networks (CNN) for the diagnosis
of IM faulty bearings. The MFELM provided the ultimate decision about layers of neural
networks from vibration signals through sensors to generate datasets through a spectro-
Energies 2023, 16, 296 38 of 54
Figure 51.Zero-crossing features characteristics for the diagnosis of multiple bearing faults. Repro-
In [129],
In [129], the
the authors
authors suggested
suggested the
the incorporation
incorporation ofof aa multivariable
multivariable feedback
feedback extreme
extreme
learning machine (MFELM) with convolutional neural networks (CNN) for the diagnosis
of IM faulty bearings. The MFELM provided the ultimate decision about layers of neural
networks from
networks from vibration
vibration signals
signals through
through sensors
sensors to
to generate
generate datasets
datasets through
through aa spectro-
spectro-
gram. Figure 52 shows the results of roller bearing faults and the prediction of their useful
gram. Figure 52 shows the results of roller bearing faults and the prediction of their useful
remaining life
remaining life using
using MFELM
MFELM and and relative
relative root
root mean
mean square
square value
value (RRMS).
(RRMS).
Figure 52.
Figure 52. (a)
(a) Short-term
Short-term degradation
degradation results
results for
for bearing
bearing faults
faults prediction.
prediction. (b)
(b) Performance
Performance of
of the
the
proposed diagnostic methodology for forecasting of remaining useful life (RUL) [129].
proposed diagnostic methodology for forecasting of remaining useful life (RUL) [129].
In [130],
In [130], the
the authors
authors proposed
proposed incipient
incipient fault
fault detection
detection and
and health
health monitoring
monitoring of of
bearings through complementary ensemble empirical mode decomposition
bearings through complementary ensemble empirical mode decomposition (CEEMD) (CEEMD) in-
corporated with
incorporated withkernel-based
kernel-basedsupport
supportvector
vectormachines.
machines.Figure
Figure53
53shows
shows the
the results
results ofof
CEEMD
Energies 2023, 16, x FOR PEER REVIEW incorporated with kernel-based SVM for estimating the lifetime bearings
CEEMD incorporated with kernel-based SVM for estimating the lifetime bearings and and
39 their
of 54
their degradation
degradation stagesstages
basedbased on experimental
on experimental data. data.
Figure 53.
Figure 53. (a)
(a) Test
Test data
data of
of bearing
bearing lifetime.
lifetime. (b)
(b) Degradation
Degradation stages
stages based
based on
on experimental
experimental data
data [130].
[130].
In [131],
In [131], the
the authors
authors suggested
suggested support
support vector
vector machines
machines and
and decision
decision trees
trees for
for the
the
diagnosis of bearing faults in rotating machines through noncurrent defects
diagnosis of bearing faults in rotating machines through noncurrent defects obtained byobtained by
statistical features for dataset generation. In [132], the authors recommended data-driven
probabilistic E-support
probabilistic E-support vector
vector regression
regression for
for estimating
estimating the
the remaining
remaining useful
useful life
life of
of roller
roller
bearings. Wavelet
bearings. Wavelettransform
transformwas
wasutilized
utilized for
for extracting
extracting informative
informative features
featuresfrom
fromthe the time
time
domain, frequency domain, and time scale domain through accelerometers. Figure 54
shows the actual and computed RUL for provided data.
Figure 53. (a) Test data of bearing lifetime. (b) Degradation stages based on experimental data [130].
In [131], the authors suggested support vector machines and decision trees for the
diagnosis of bearing faults in rotating machines through noncurrent defects obtained by
Energies 2023, 16, 296 39 of 54
probabilistic E-support vector regression for estimating the remaining useful life of roller
bearings. Wavelet transform was utilized for extracting informative features from the time
domain, frequency
domain, frequency domain,
domain, and
and time
time scale
scale domain
domain through
throughaccelerometers.
accelerometers. Figure
Figure 5454
shows the
shows the actual
actual and
and computed
computedRUL
RULforforprovided
provideddata.
data.
Figure54.
Figure 54.(a)
(a) Prognosis
Prognosis result for prediction
result for prediction of
of RUL.
RUL. (b)
(b) Zoom
Zoom view
view [132].
[132].
In [133], the
In thewriters
writersproposed
proposeda deep
a deepneural
neuralnetwork
network(DNN)
(DNN)for the
for diagnosis of bear-
the diagnosis of
ing faults.
bearing Vibration
faults. signals
Vibration were employed
signals were employedfor feature extraction
for feature to convert
extraction multi-domain
to convert multi-
images and
domain fed and
images to thefedDNN
to thefor
DNNthe classification of bearing
for the classification faults. In
of bearing [134],Inthe
faults. authors
[134], the
authors
proposed proposed health monitoring
health monitoring of ballof ball bearings
bearings usingusing Hilbert–Huang,
Hilbert–Huang, supportsupport vector
vector ma-
machines
chines SVM, SVM,andand support
support vector
vector regression
regression SVR.SVR. Hilbert–Huang
Hilbert–Huang was was implemented
implemented to
to haul
haul out attributes from stationary vibration signals. SVM and
out attributes from stationary vibration signals. SVM and SVR were utilized SVR were utilized to classify
classify
and
and estimate
estimate thethe RUL
RUL of of bearings,
bearings,respectively.
respectively.
In
In [135],
[135], the
the authors
authors suggested
suggested the the sparse
sparse representation
representation theory
theory incorporated
incorporated with with
kernel
kernelfunctions
functionsofofsupport
support vector
vector machines
machines (SVM)
(SVM) for the
for diagnosis of bearing
the diagnosis faults.faults.
of bearing Root
mean squaresquare
Root mean error (RMSE) and mean
error (RMSE) andsquare correlation
mean square coefficient
correlation (MSCC)were
coefficient utilized uti-
(MSCC)were for
feature extraction optimized by PSO to obtain significant information for training
lized for feature extraction optimized by PSO to obtain significant information for training purposes.
In [136], theInauthors
purposes. recommended
[136], the a Bayes classifier
authors recommended a Bayesfor the classification
classifier for theofclassification
bearing faults of
in PMSM employing linear discriminant analysis (LDA) of the stator currents. In [137],
the authors suggested a stacked pruning sparse denoising auto encoder (sPSDAE) for
evaluating the rolling bearing faults. The auto encoder was used for extracting features
from subsequent layers to feed sPSDAE for learning. In [138], the authors proposed the
classification of bearing faults by implementing support vector machines (SVM), which
uses bi-spectrum analysis to extract features for generating datasets. Bi-spectrum analyses
were further subjected by PCA to reduce dimensions and input to SVM for classification of
bearing faults as shown in Figure 55.
In [139], the authors investigated the methodology for evaluating the bearing fault
diagnosis of metallic, ceramic, and hybrid bearings based on an adaptive stacked auto en-
coder. In [140], the authors suggested a complete ensemble empirical mode decomposition
with adaptive noise (CEEMDAN) with piecewise aggregate approximation (PAA) for the
decomposition of long signals to enhance the demodulation and filtering for the diagnosis
of the bearing’s abnormal conditions accurately. In [141], the authors suggested continuous
wavelet transform CWT and convolutional neural networks to predict the bearing fault size
diagnosis based on deep transfer learning algorithms (DTL) and time–frequency images.
Vibrations signals were employed to obtain useful information to create datasets using
continuous wavelet transform (CWT) and scalogram images. Figure 56 shows the confusion
matrix plot by employing the proposed methodology for the true class and predicted class.
(sPSDAE) for evaluating the rolling bearing faults. The auto encoder was used for extract-
ing features from subsequent layers to feed sPSDAE for learning. In [138], the authors
proposed the classification of bearing faults by implementing support vector machines
(SVM), which uses bi-spectrum analysis to extract features for generating datasets. Bi-
Energies 2023, 16, 296 spectrum analyses were further subjected by PCA to reduce dimensions and input to 40 SVM
of 54
for classification of bearing faults as shown in Figure 55.
Figure55.
Figure 55.Bi-spectra
Bi-spectraresults
resultsfor
forthe
thediagnosis
diagnosisofof(a)
(a)outer
outerrace
racefault
faultand
and(b)
(b)ball
ballbearing
bearingfault
fault[139].
[139].
In [139], the authors investigated the methodology for evaluating the bearing fault
diagnosis of metallic, ceramic, and hybrid bearings based on an adaptive stacked auto
encoder. In [140], the authors suggested a complete ensemble empirical mode decompo-
sition with adaptive noise (CEEMDAN) with piecewise aggregate approximation (PAA)
for the decomposition of long signals to enhance the demodulation and filtering for the
diagnosis of the bearing’s abnormal conditions accurately. In [141], the authors suggested
continuous wavelet transform CWT and convolutional neural networks to predict the
bearing fault size diagnosis based on deep transfer learning algorithms (DTL) and time–
frequency images. Vibrations signals were employed to obtain useful information to cre-
ate datasets using continuous wavelet transform (CWT) and scalogram images. Figure 56
shows the confusion matrix plot by employing the proposed methodology for the true
class and predicted class.
Figure
Figure 56.
56. Confusion
Confusion matrix
matrix plots
plots and
and their
their accuracies
accuracies for
for prediction
prediction of
of bearing
bearing fault
fault size
size using
using aa
convolutional
neural network
network with
with continuous
continuous wavelet transform [141].
wavelet transform [141].
In
In [142],
[142], the authors proposed
the authors proposed multichannel
multichanneland andmultilevel
multilevelone-dimensional
one-dimensionalconvolu-
convo-
lutional neural networks (1D-CNN) for the detection of enhanced bearing
tional neural networks (1D-CNN) for the detection of enhanced bearing faults
faults inin electric
electric
motors
motors bybyemploying
employingvibration
vibrationsignals
signalsobtained
obtained through
through accelerometers.
accelerometers.In [143], the au-
In [143], the
thors suggested
authors suggested support
supportvector
vectormachines
machines(SVM)
(SVM)andandfiber
fiberBragg
Bragggrating
grating(FBG)
(FBG) toto detect
detect
bearing
bearing failures
failures inin three-phase
three-phase induction
induction motors.
motors. Outer
Outer race
race faults
faults were
were diagnosed
diagnosed in in the
the
proposed work at 47 different frequencies, and principal component analysis (PCA) was
employed as
employed as aa feature
feature reduction
reduction technique
technique for
for the
the dimension
dimension reduction
reduction of
of the
the dataset
dataset forfor
gaining excellent results. Figure 57 shows the power spectral density (PSD)
gaining excellent results. Figure 57 shows the power spectral density (PSD) results for the results for
the recognition
recognition of healthy
of healthy bearings
bearings and outer
and outer raceway
raceway abnormal
abnormal conditions
conditions at different
at different load-
loading conditions.
ing conditions.
thors suggested support vector machines (SVM) and fiber Bragg grating (FBG) to detect
bearing failures in three-phase induction motors. Outer race faults were diagnosed in the
employed as a feature reduction technique for the dimension reduction of the dataset for
gaining excellent results. Figure 57 shows the power spectral density (PSD) results for the
Energies 2023, 16, 296 recognition of healthy bearings and outer raceway abnormal conditions at different41 load-
of 54
ing conditions.
Figure 57. Power spectral density PSD of dynamic strain signal for (a) healthy bearings and (b) outer
raceway faults. Reproduced with permission of [143].
In [144], the authors investigated abnormal conditions of ball bearings by employing

maximal overlap discrete wavelet transform (MODWT) and image edge detection for
grid-fed three-phase induction machines. In [145], the authors recommended the method-
ology to diagnose bearing faults in electric machines by employing a convolutional neural
network with wireless sensor network data fusion. In [146], the authors suggested the
progressive detection of bearing faults in three-phase induction motors by employing
the matrix pencil method and MUSIC algorithm via pre-fault frequency cancellation. A
very comprehensive review article on sound and acoustic emission-based fault diagnostic
techniques is presented in [147]. Some AI based fault detection techniques are summarized
in Table 3.
Table 3. A summary of different artificial intelligence techniques for bearing fault diagnosis.

Technique Type
Intelligent and
Artificial bee
self-organizing Local search ability
colony
algorithm for is poor for Ball bearings Induction motors [87]
+
segregation of bearing collecting results
Decision trees
faults
Network structure
Efficient-Net, CNN Noise-resilient and does
is complex and
+ not demand labeled data Ball bearings Induction motors [88,89]
requires pre
CWT for faults classification
training
Defective signature Inexpensive,
wavelet image to find the shortest path Demands large
+ to diagnose bearing amounts of data
DCNN faults
Convolutional
Ability to detect
neural network Demands very
significant features Roller bearings Induction motors [91,92]
+ large database
without supervision
Transfer learning
Energies 2023, 16, 296 42 of 54
Table 3. Cont.

Technique Type
Adaptive cascade
observer Easily handle complex
Training rate is
+ tasks in noisy Roller bearings Induction motors [93]
very slow
Fuzzy orthonormal environment
regressive
Support vector
Best classifier as
machines The kernel
compared to other
+ selection process is Ball bearings Rotating machines [94]
algorithms and fast to
Particle swarm difficult
train
optimization
Recurrent neural Capable of memorizing Demands high
network sequential events and processing time
+ receiving input from and overfitting to
WD-CNN multiple variables problems
EMD-KLD
Excellent classification Difficult to
+ Roller bearings Induction motors [96]
algorithms understand
SVM, KNN, DT
Ensemble RF,
XG-Boost Excellent predictive Hard to tune the Roller/ball
Induction motors [97]
+ capability parameters bearings
DWT
Deep belief
Does not provide
network Has excellent learning
accurate results
+ strategy and greedy for Ball bearings Rotating machines [98,133]
among dependable
Dumpster Shafer complex problems
variables
theory
Feed-forward
neural network Computationally
High
+ powerful algorithms Outer raceway Small HP
computational cost [99]
Linear demand less training faults induction motors
and robust to noise
discriminant time
analysis
Ensemble local
x-tics Easy to implement and
Demands large
decomposition deal with noisy faulty Ball bearings Rotating machines [100]
searching time
+ data
SVM
Neuro wavelet
approach Demands low Sensitive and
Outer raceway Small HP
+ computational power frequent learning [101]
faults induction motors
Variational mode and less memory issues
decomposition
Improved shuffled
frog leaping High-speed convergence
Demand
algorithm rate and ability to
pre-training phase Roller bearings Rotating machines [102]
+ predict accurate fault
for accurate results
Back propagation size
neural network
Energies 2023, 16, 296 43 of 54
Table 3. Cont.

Technique Type
Stack pruning
sparse auto
Powerful hybrid Time-consuming
encoder
algorithm and can learn and has a complex Roller bearings Induction motors [103,138]
+
independently structure
Deep neural
network
Wasserstein
generative Approximation model Hard to
adversarial and implemented where implement, and
network loss function exists in training time is
+ training very unstable
CNN
Artificial neural
network Requires few neurons to Time-consuming
+ establish excellent and has a complex Roller bearings Induction motors [105]
Deep neural capability structure
network
Artificial neural
Approximation model Hard to
network
and implemented where implement, and
loss function exists in training time is
Laplace wavelet
training very unstable
transform
Discriminant
feature
distribution No need for linearly Sensitive to noisy
Roller bearings Induction motors [107,117]
analysis separable classes faulty data
+
GA, KNN
Multiple
measurement
vector (MMV) Exhibits less error and is Demands mutually
+ susceptible to overfitting exclusive classes
Compressing
sampling
K-means clustering Suitable to implement Poor results for
+ on any size of the small datasets and Ball bearings Rotating machines [109]
GA dataset overfitting
Has the ability to solve Its fault detection
FPGA, SVM
complex problems by and classification
implementing kernel accuracy depend
Acoustic emission
functions on the value of k
Provides sparse Training time is
CNN
representation and slow and lack of
extracts only relevant transparency in the
envelope spectrum
features from the dataset results
A large number of
Stock-well
Demands less training samples are
transform Inner and outer
time and the ability to needed for Induction motors [112]
+ race faults
handle nonlinear data excellent
Transfer learning
performance
STBDSFA Simple, easy to
Demand high Low-speed
+ implement, and training Roller bearings [113]
storage machines
LSTM time is less
Energies 2023, 16, 296 44 of 54
Table 3. Cont.

Technique Type
Fuzzy sliding
mode observer Easily deals with noisy
Overfitting Roller bearings Rotating machines [114]
+ data for accurate results
Decision trees
Integrated High
Fast, accurate, and
empirical wavelet computational cost
requires less memory;
transform and requires a Ball bearings Induction motors [115]
also, its interference
+ high-memory
speed is high
SVM system
Improved
detrended Lazy algorithm
Easy to understand and
fluctuation and consumes too
works accurately for Ball bearings Induction motors [116]
analysis much time for
classification problems
+ training
SVM
Self-adaptive
Deals with the complex
wavelet transform High
relationships between Roller bearings Rotating machines [118]
+ computational cost
nonlinear functions
PSO
Convolutional Demands a lot of
Accurate predictive
neural network power and is very
models for nonlinear Ball bearings Induction motors [119]
+ slow due to many
and complex problems
Thermal Images layers
Convolutional
Extremely accurate and Demands more
neural network
much faster than other data for precise Ball bearings Rotating machines [120]
+
algorithms results
S-transform
Physics-based
neural network
Fast approximation Needs large
+ Ball bearings Rotating machines [121]
capability memory
Kurtosis and
envelope analysis
Gramian angular
Powerful algorithms
filed (GAF) Limited to the
and better feature Ball bearings Induction motors [122]
+ number of inputs
extraction capability
Dense-Net
Convolutional
neural network Excellent algorithm for Difficult to
+ nonlinear relationships understand
Fault-Net
Hierarchical radial
Powerful and can Results are not
basis function
handle very complex transparent due to Ball bearings Rotating machines [124]
+
problems overfitting
SVM
Generative
adversarial
Ability to handle The possibility of
network
categorical features and error is high due to Roller bearings Induction motors [125]
+
faulty datasets large datasets
Finite element
analysis
Energies 2023, 16, 296 45 of 54
Table 3. Cont.

Technique Type
Modified
multiscale
weighted Difficult to
Excellent algorithm for
permutation understand the Ball bearings Induction motors [126,127]
nonlinear relationships
entropy discrete faulty data
+
SVM
Feed-forward
Multiclass
neural network Handles complex
problems are not
+ problems; also handles Ball bearings Induction motors [128]
addressed
Time domain overfitting problems
properly
zero-crossing
Multivariate
feedback extreme Convolutional filters to The computational
6-pole induction
learning transform 2D data into training cost is Ball bearings [129]
motors
+ 3D for high performance expensive
RRMS
CEEMD
Capable of receiving Demands labeled
+ Roller bearings Induction motors [130,141]
high accuracy data
SVM
One vs. all Powerful architectures, Training time is
+ easy to implement, and slow due to a large Ball bearings Rotating machines [131]
SVM, DT hard to overfit number of datasets
Reliability density Best classification results
function and good for High memory RUL of roller
PMSM [132]
+ high-dimensional data, requirements bearings
SVR less risk of overfitting
Hilbert–Huang Ability to solve
transform multiclass problems for Kernel selection is
Ball bearings PMSM [134]
+ both continuous and difficult
SVM, SVR discrete data
SVM, DT, KNN,
Simple to implement
and RF Low convergence
and excellent for Roller bearings Rotating machines [135]
+ rate
multi-objective functions
PSO
Mean square
correlation Predict complex and Demands labeled
coefficient (MSCC) nonlinear combinations data for high Ball bearings Rotating machines [136]
+ of inputs accuracy
PSO
Naive Bayes Feature selection Nonlinear data
classifier methodology used to decreases the
+ select features by mutual performance of the
LDA information model
Bi-spectrum
Able to deal with Overfitting and Metallic, ceramic,
analysis
complex and nonlinear training time are and hybrid Rotating machines [139]
+
problems high bearings
SVM
Adaptive stacked A noise-assisted
encoder methodology Computational
Rolling bearings Induction motors [140]
+ implemented for roller cost is very high
Softmax layer bearing fault diagnosis
Energies 2023, 16, 296 46 of 54
Table 3. Cont.

Technique Type
Convolutional
neural network Provides high accuracy Demands very
+ rate for detecting large datasets for Rolling bearing Servo motors [141]
Continuous bearings features training
wavelet transform
Multilevel
Able to detect important Slower due to
1D-CNN
attributes without max-pooling of Rolling bearings Electric motors [142]
+
supervision data
Vibration signals
Support vector
High accuracy in the Highly dependent
machines Three-phase
detection of faults due to on the size of the Ball bearings [143]
+ induction motors
margin separation dataset
Fiber Bragg grating
Maximal overlap
discrete wavelet Powerful algorithm to
Demands high
transform represent faulty bearing Three-phase
computational Ball bearings [144]
+ signals in redundant induction motors
power
Image edge forms
detection
Convolutional
neural network Demands fewer number
Computationally
+ of parameters than other Rolling bearings Induction motors [145]
expensive
Fusion wireless DL algorithms
sensor
Has the ability to Does not provide
MUSIC algorithm Three-phase
provide extra exact information
+ Ball bearings induction [146]
information about faulty for the localization
Pre-fault frequency machines
signals of faulty elements
5. Recommendations and Suggestions

The successful implementation of different approaches to implementing conventional,
statistical, and artificial intelligence-based approaches demands physically powerful cor-
relations between different variables and makes a strong feature extraction capability for
excellent results. For researchers and engineers, to implement conventional, statistical,
and artificial intelligence-based approaches for bearing fault identification, we have the
following suggestions.
• Make a proper environmental setup to examine the working conditions of all possible
working combinations; for example, select the fixed operating point for varying loads
at different speeds.
• For high-speed applications, in which chances of faults are complicated by external
sources such as in the case of electrical motors fed by external sources, such as variable
frequency devices (VFD), deep learning algorithms are proposed for accurate results.
• Sensors that are very helpful for recording the abnormal conditions produced in
symmetry due to certain conditions are mounted close to the bearings for accurate
results, and for machine learning/deep learning-based approaches, multiple sensors
are required for processing of high-level data.
• The dataset which is used to implement machine learning/deep learning models must
have used labeled data for high-level accuracy because unlabeled data in the dataset
provides unsatisfactory results.
Energies 2023, 16, 296 47 of 54
5.1. Gaps and Thoughts for Future Investigation

Although major developments are made in the field of fault detection and diagnosis
of multiple issues found in rotating electrical machines, various types of issues are still
unsolved. The following issues discussed below need the concentration of researchers and
engineers to tackle them:
5.2. Current Challenges

• In the case of distributed faults in bearings, roughness and waviness forms of bearing
faults have a huge untold impact on the frequency spectrum. For that reason, fault
characteristics and frequency components are not addressed properly.
• For generating datasets, the majority of the work is performed in laboratories to train
machine learning and deep learning models, so there is a need to make arrangements
to record the real data for making datasets.
• The collection of data from different sources makes it easier to generate datasets, but
accurate and excellent results demand labeled data, and that is why machine learning
and deep learning models are not able to give high-accuracy results.
• The data imbalance is mostly due to data collected when bearings are in healthy
condition, and limited samples are collected during faulty conditions, producing an
imbalance of data in datasets, which also affects the performance of models.
• Usually, vibration signals are used to accumulate information through imposing
accelerometers in laboratories, but real series of data nearly have lots of noise in their
measurements. This is an open query for researchers: is it feasible to generate datasets
from practical work without noisy measurements?
5.3. Future Work Directions

• To implement transfer learning as a promising approach instead of vibration signals,
which have an excellent ability to transfer knowledge, and learn from past experiences
to identify unexpected faulty conditions in real-world applications.
• Distinctive transfer learning-to-know strategies include domain randomization and
area adaptation, which can successfully increase the range of the source domain
(present datasets) and facilitate quicker studying and higher overall performance in
the goal area (real-world applications).
• The datasets, which have limited labeled data points and are imbalanced, are proposed
to implement semi-supervised learning to make possible the full utilization of data
points for excellent results.
• The data augmentation approach is employed to solve the unbalancing in the dataset
and is also used to identify the fake data points in the dataset, which badly affect the
performance of artificial intelligence techniques.
• To introduce a few short learning techniques in combination with transfer learning,
which addresses the data unbalance and scarcity in datasets and is very helpful for
generating datasets from practical, real-world applications?
• The data points in datasets that have noise and are measured by vibration signals are
suggested to use other types of sensors such as acoustic emission sensors and imple-
ment sensor fusion techniques to improve the strength of data points in the dataset.
6. Conclusions
This paper comprehensively reviewed state-of-the-art existing prose by employing
conventional, statistical, and artificial intelligence (machine learning and deep learning)-
based approaches for the diagnosis of all major types of bearing faults in electrical ma-
chines. From the previous literature, different fault detection and diagnosis (FDD)-based
approaches (conventional, statistical, and artificial intelligence) were discussed, namely,
vibration measurements, acoustic emission, temperature analysis, thermal imaging, spec-
tral analysis, wavelet techniques, signal processing, and other mentioned techniques used
for searching for bearing faults in rotating machines. Bearing fault diagnosis approaches
Energies 2023, 16, 296 48 of 54
were discussed by defining their merits, demerits, applications, and specific type of ab-
normal conditions that occurred on the bearings. Conventional techniques highlighted
the previous literature-based methodologies to trace the abnormal conditions in bearings
and also reviewed the advancements in research with hybrid mechanisms. Current and
vibration analysis were most commonly employed with other techniques to characterize
faults due to their robust and noninvasive nature. Conventional techniques also pointed
out the time-varying conditions on variable speed drives and large locomotives’ linear
bearing faults using high-frequency analysis. Mostly statistical approaches were used
in the extraction of useful information and featured reduction analysis such as principal
component analysis (PCA) and linear discriminant analysis (LDA) and were the important
statistical parameters employed. Statistical approaches (skewness, kurtosis, root mean
square values, stochastic resonance analysis, empirical mode decomposition, curvilinear
component analysis) implemented data analysis and descriptive analysis-based approaches
to diagnose all types of bearing faults in different applications. Statistical approaches
mainly employed feature reduction techniques to optimize the parameters for excellent
results in diagnosing the bearing faults. All the statistical techniques were discussed in the
form of a table with their specific applications, merits and demerits, and bearing fault types
taking place on bearings. Artificial intelligence-based approaches such as machine learning
and deep learning demand large numbers of data points for training, and these algorithms
have the capability to extract attributes for excellent results. Artificial intelligence-based
approaches (machine learning and deep learning) were mostly discussed with all types
of algorithms (support vector machines, K-nearest neighbors, random forest, fuzzy logic,
neural networks, deep belief network, artificial bee colony) with different optimization
and feature reduction techniques for the diagnosis of bearing faults in electrical machines.
Machine learning and deep learning algorithms employed the Case Western Reserve Uni-
versity (CWRU) dataset to validate their performance for the diagnosis of bearing faults.
Feature reduction techniques such as principal component analysis (PCA) and linear dis-
criminant analysis (LDA) were employed in the datasets, and trained algorithms provided
more efficient results for bearing fault diagnosis. Similarly, artificial intelligence-based
approaches (machine learning and deep learning) were summarily discussed in the form
of a table to discuss their merits and demerits, bearing fault type, applications, and their
relevant references to search out the related articles. In the end, some recommendations
and suggestions about current challenges and future directions were addressed briefly.
Author Contributions: Conceptualization, M.A.K. and B.A.; methodology, M.A.K.; software, M.A.K.;
validation, B.A. and T.V.; formal analysis, K.K.; investigation, A.K.; resources, A.K.; data curation,
M.A.K.; writing—original draft preparation, M.A.K. and B.A.; writing—review and editing, M.A.K.;
visualization, T.V.; supervision, B.A. All authors have read and agreed to the published version of
the manuscript.
Funding: The “Industrial Internet methods for electrical energy conversion systems monitoring and
diagnostics” benefits from a 993,000 € grant from Iceland, Liechtenstein and Norway through the EEA
Grants. The aim of the project is to provide the research in field of energy conversion systems and to
develop artificial intelligence and virtual emulator-based prognostic and diagnostic methodologies
for these systems. Project contract with the Research Council of Lithuania (LMTLT) No is S-BMT-21-5
(LT08-2-LMT-K-01-040).
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Nandi, S.; Toliyat, H.A.; Li, X. Condition Monitoring and Fault Diagnosis of Electrical Motors—A Review. IEEE Trans. Energy
Convers. 2005, 20, 719–729. [CrossRef]
2. Liu, Z.; Zhang, P.; He, S.; Huang, J. A Review of Modeling and Diagnostic Techniques for Eccentricity Fault in Electric Machines.
Energies 2021, 14, 4296. [CrossRef]
Energies 2023, 16, 296 49 of 54
3. Tiboni, M.; Remino, C.; Bussola, R.; Amici, C. A Review on Vibration-Based Condition Monitoring of Rotating Machinery. Appl.
Sci. 2022, 12, 972. [CrossRef]
4. Gritli, Y.; Bellini, A.; Rossi, C.; Casadei, D.; Filippetti, F.; Capolino, G.-A. Condition monitoring of mechanical faults in induction
machines from electrical signatures: Review of different techniques. In Proceedings of the 2017 IEEE 11th International Symposium
on Diagnostics for Electrical Machines, Power Electronics and Drives, SDEMPED 2017, Tinos, Greece, 29 August–1 September
2017; pp. 77–84. [CrossRef]
5. Frosini, L. Monitoring and Diagnostics of Electrical Machines and Drives: A State of the Art. In Proceedings of the IEEE Workshop
on Electrical Machines Design, Control and Diagnosis (WEMDCD), Athens, Greece, 8–9 April 2019; pp. 169–176. [CrossRef]
6. Capolino, G.-A.; Antonino-Daviu, J.A.; Riera-Guasp, M. Modern Diagnostics Techniques for Electrical Machines, Power Electron-
ics, and Drives. IEEE Trans. Ind. Electron. 2015, 62, 1738–1745. [CrossRef]
7. Basak, D.; Tiwari, A.; Das, S.P. Fault diagnosis and condition monitoring of electrical machines—A Review. In Proceedings of the
IEEE International Conference on Industrial Technology, Mumbai, India, 15–17 December 2006; pp. 3061–3066. [CrossRef]
8. He, Y.; Zhang, X.; Friswell, M.I. Defect Diagnosis for Rolling Element Bearings Using Acoustic Emission. J. Vib. Acoust. 2009, 131,
061012. [CrossRef]
9. Trajin, B.; Regnier, J.; Faucher, J. Bearing fault indicator in induction machine using stator current spectral analysis. In Proceedings
of the 2008 4th IET Conference on Power Electronics, Machines and Drives, York, UK, 2–4 April 2008. [CrossRef]
10. Moundekar, N.D.; Deshmukh, B.D. Study of failure modes of rolling bearings: A review. Int. J. Mod. Eng. Res. (IJMER) 2014,
4, 139.
11. Hamadache, M.; Jung, J.H.; Park, J.; Youn, B.D. A comprehensive review of artificial intelligence-based approaches for rolling
element bearing PHM: Shallow and deep learning. JMST Adv. 2019, 1, 125–151. [CrossRef]
12. Frosini, L. Novel Diagnostic Techniques for Rotating Electrical Machines—A Review. Energies 2020, 13, 5066. [CrossRef]
13. Eren, L.; Devaney, M.J. Bearing Damage Detection via Wavelet Packet Decomposition of the Stator Current. IEEE Trans. Instrum.
Meas. 2004, 53, 431–436. [CrossRef]
14. Cung, L.E.; Hien, B.M.; Son, N.T. Gear and bearing fault detection using wavelet packet and Hilbert method via acoustic signals.
ASEAN Eng. J. 2016, 6, 5–13.
15. Huang, L.; Hu, N.; Yang, Y.; Chen, L.; Wen, J.; Shen, G. Study on Electromagnetic–Dynamic Coupled Modeling Method—Detection
by Stator Current of the Induction Motors with Bearing Faults. Machines 2022, 10, 682. [CrossRef]
16. Rosero, J.A.; Cusido, J.; Garcia, A.; Ortega, J.A.; Romeral, L. Broken bearings and eccentricity fault detection for a permanent
magnet synchronous motor. In Proceedings of the IECON 2006—32nd Annual Conference on IEEE Industrial Electronics, Paris,
France, 6–10 November 2006; pp. 964–969.
17. Rosero, J.; Cusido, J.; Ortega, J.A.; Romeral, L.; Garcia, A. PMSM Bearing Fault Detection by means of Fourier and Wavelet
transform. In Proceedings of the IECON 2007—33rd Annual Conference of the IEEE Industrial Electronics Society, Taipei, Taiwan,
5–8 November 2007; pp. 1163–1168.
18. Rosero, J.; Romeral, L.; Rosero, E.; Urresty, J. Fault Detection in dynamic conditions by means of Discrete Wavelet Decomposition
for PMSM running under Bearing Damage. In Proceedings of the IEEE Applied Power Electronics Conference and Exposition,
Washington, DC, USA, 15–19 February 2009; pp. 951–956.
19. Vilhekar, T.G.; Ballal, M.S.; Suryawanshi, H.M. Application of Multiple Parks Vector Approach for Detection of Multiple Faults in
Induction Motors. J. Power Electron. 2017, 17, 972–982. [CrossRef]
20. Silva, J.L.; Cardoso, A.M. Bearing failures diagnosis in three-phase induction motors by extended Park’s vector approach. In
Proceedings of the 31st Annual Conference of IEEE Industrial Electronics Society, Raleigh, NC, USA, 6–10 November 2005.
21. Salem, S.B.; Touti, W.; Bacha, K.; Chaari, A. Induction motor mechanical fault identification using Park’s vector approach. In
International Conference on Electrical Engineering and Software Applications; IEEE: Piscataway Township, NJ, USA, 2013; pp. 1–6.
22. Corne, B.; Vervisch, B.; Derammelaere, S.; Knockaert, J.; Desmet, J. The reflection of evolving bearing faults in the stator current’s
extended park vector approach for induction machines. Mech. Syst. Signal Process. 2018, 107, 168–182. [CrossRef]
23. Bessous, N.; Zouzou, S.E.; Bentrah, W.; Sbaa, S.; Sahraoui, M. Diagnosis of bearing defects in induction motors using discrete
wavelet transform. Int. J. Syst. Assur. Eng. Manag. 2018, 9, 335–343. [CrossRef]
24. Harlişca, C.; Szabó, L.; Frosini, L.; Albini, A. Bearing faults detection in induction machines based on statistical processing of the
stray fluxes measurements. In Proceedings of the IEEE International Symposium on Diagnostics for Electric Machines, Power
Electronics and Drives (SDEMPED), Valencia, Spain, 27–30 August 2013; pp. 371–376.
25. Frosini, L.; Magnaghi, M.; Albini, A.; Magrotti, G. A new diagnostic instrument to detect generalized roughness in rolling
bearings for induction motors. In Proceedings of the IEEE 10th International Symposium on Diagnostics for Electrical Machines,
Power Electronics and Drives (SDEMPED), Guarda, Portugal, 1–4 September 2015; pp. 239–245.
26. Zamudio-Ramírez, I.; Osornio-Ríos, R.A.; Antonino-Daviu, J.A.; Quijano-Lopez, A. Smart-Sensor for the Automatic Detection of
Electromechanical Faults in Induction Motors Based on the Transient Stray Flux Analysis. Sensors 2020, 20, 1477. [CrossRef]
27. Dalvand, F.; Kalantar, A.; Safizadeh, M.S. A Novel Bearing Condition Monitoring Method in Induction Motors Based on
Instantaneous Frequency of Motor Voltage. IEEE Trans. Ind. Electron. 2015, 63, 364–376. [CrossRef]
28. Zanini, R.A.A.; Freitas, T.; Silva, L.N.; Gutierres, A.; Santos, G. Evaluation of the Method of Failure Analysis in Angular Contact
Bearings through the Envelope Technique. In SAE International 21st SAE Brasil International Congress and Exhibition; SAE Technical
Paper Series; SAE International: Warrendale, PA, USA, 2012. [CrossRef]
Energies 2023, 16, 296 50 of 54
29. Immovilli, F.; Cocconcelli, M.; Bellini, A.; Rubini, R. Detection of Generalized-Roughness Bearing Fault by Spectral-Kurtosis
Energy of Vibration or Current Signals. IEEE Trans. Ind. Electron. 2009, 56, 4710–4717. [CrossRef]
30. Gong, X.; Qiao, W. Current-based online bearing fault diagnosis for direct-drive wind turbines via spectrum analysis and impulse
detection. In Proceedings of the IEEE Power Electronics and Machines in Wind Applications, Denver, CO, USA, 16–18 July 2012;
pp. 1–6.
31. Attaianese, C.; De Falco, P.; Del Pizzo, A.; Di Noia, L.P. Bearing Failure Prognostic Method Based on High Frequency Inductance
Variation in Electric Railway Traction Motors. In Proceedings of the IEEE Texas Power and Energy Conference (TPEC), virtually,
2–5 February 2021; pp. 1–5. [CrossRef]
32. Tischmacher, H. Bearing Wear Condition Identification on Converter-fed Motors. In Proceedings of the IEEE International
Symposium on Power Electronics, Electrical Drives, Automation and Motion (SPEEDAM), Amalfi, Italy, 20–22 June 2018;
pp. 19–25. [CrossRef]
33. Kudelina, K.; Baraškova, T.; Shirokova, V.; Vaimann, T.; Rassõlkin, A. Fault detecting accuracy of mechanical damages in rolling
bearings. Machines 2022, 10, 86. [CrossRef]
34. Nivesrangsan, P.; Jantarajirojkul, D. Bearing fault monitoring by comparison with main bearing frequency components using
vibration signal. In Proceedings of the IEEE International Conference on Business and Industrial Research (ICBIR), Bangkok,
Thailand, 17–18 May 2018; pp. 292–296.
35. Ibrahim, A.; El Badaoui, M.; Guillet, F.; Bonnardot, F. A new bearing fault detection method in induction machines based on
instantaneous power factor. IEEE Trans. Ind. Electron. 2008, 55, 4252–4259. [CrossRef]
36. Zarei, J.; Tajeddini, M.A.; Karimi, H.R. Vibration analysis for bearing fault detection and classification using an intelligent filter.
Mechatronics 2014, 24, 151–157. [CrossRef]
37. Song, L.; Wang, H.; Chen, P. Vibration-Based Intelligent Fault Diagnosis for Roller Bearings in Low-Speed Rotating Machinery.
IEEE Trans. Instrum. Meas. 2018, 67, 1887–1899. [CrossRef]
38. Zhang, L.; Zhang, F.; Qin, Z.; Han, Q.; Wang, T.; Chu, F. Piezoelectric energy harvester for rolling bearings with capability of
self-powered condition monitoring. Energy 2022, 238, 121770. [CrossRef]
39. Lu, S.; Wang, X. A New Methodology to Estimate the Rotating Phase of a BLDC Motor with Its Application in Variable-Speed
Bearing Fault Diagnosis. IEEE Trans. Power Electron. 2017, 33, 3399–3410. [CrossRef]
40. Esfahani, E.T.; Wang, S.; Sundararajan, V. Multisensor Wireless System for Eccentricity and Bearing Fault Detection in Induction
Motors. IEEE/ASME Trans. Mechatronics 2013, 19, 818–826. [CrossRef]
41. Cheng, Y.; Wang, Z.; Zhang, W. A Novel Condition-Monitoring Method for Axle-Box Bearings of High-Speed Trains Using
Temperature Sensor Signals. IEEE Sensors J. 2018, 19, 205–213. [CrossRef]
42. Bendjama, H.; Bouhouche, S.; Moussaoui, A.K. Wavelet transform for bearing faults diagnosis. Adv. Control. Eng. 2013, 85–88.
43. Jiang, F.; Zhu, Z.; Li, W. An Improved VMD with Empirical Mode Decomposition and Its Application in Incipient Fault Detection
of Rolling Bearing. IEEE Access 2018, 6, 44483–44493. [CrossRef]
44. Djeddi, M.; Granjon, P.; Leprettre, B. Bearing fault diagnosis in induction machine based on current analysis using high-resolution
technique. In Proceedings of the IEEE International Symposium on Diagnostics for Electric Machines, Power Electronics and
Drives, Cracow, Poland, 6–8 September 2007; pp. 23–28.
45. Wilson, K.W. Probabilistic inter-disturbance interval estimation for bearing fault diagnosis. In Proceedings of the IEEE In-
ternational Symposium on Diagnostics for Electric Machines, Power Electronics and Drives IEEE International Symposium
on Diagnostics for Electric Machines, Power Electronics and Drives, Cargese, France, 31 August–3 September 2009; pp. 1–6.
[CrossRef]
46. Soares, M.N.; Gyselinck, J.; Mollet, Y.; Peeters, C.; Gioia, N.; Helsen, J. Vibration-based rotor-side-converter open-switch-fault
detection in DFIGs for wind turbines. In Proceedings of the IEEE International Conference on Prognostics and Health Management
(ICPHM), Seattle, WA, USA, 11–13 June 2018; pp. 1–6.
47. Niu, J.; Lu, S.; Liu, Y.; Zhao, J.; Wang, Q. Intelligent Bearing Fault Diagnosis Based on Tacholess Order Tracking for a Variable-
Speed AC Electric Machine. IEEE Sensors J. 2018, 19, 1850–1861. [CrossRef]
48. Arun, P.; Lincon, S.A.; Prabhakaran, N. Non-intrusive detection and characterization of bearing faults from the temporal features
of vibration. Aust. J. Mech. Eng. 2020, 18 (Suppl. 1), S132–S139. [CrossRef]
49. Zhang, Y.; Wang, A.; Zuo, H. Roller bearing performance degradation assessment based on fusion of multiple features of
electrostatic sensors. Sensors 2019, 19, 824. [CrossRef]
50. Harlişca, C.; Szabó, L.; Frosini, L.; Albini, A. Diagnosis of rolling bearings faults in electric machines through stray magnetic flux
monitoring. In Proceedings of the IEEE International Symposium on Advanced Topics in Electrical Engineering (Atee), Bucharest,
Romania, 23–25 May 2013; pp. 1–6.
51. Chen, X.; Xu, W.; Liu, Y.; Islam, R. Bearing Corrosion Failure Diagnosis of Doubly Fed Induction Generator in Wind Turbines
Based on Stator Current Analysis. IEEE Trans. Ind. Electron. 2019, 67, 3419–3430. [CrossRef]
52. Jung, J.; Park, Y.; Lee, S.B.; Cho, C.; Kim, K.; Wiedenbrug, E.; Teska, M. Monitoring of journal bearing faults based on motor
current signature analysis for induction motors. In Proceedings of the IEEE Energy Conversion Congress and Exposition (ECCE),
Montreal, QC, Canada, 20–24 September 2015; pp. 300–307.
53. Immovilli, F.; Bianchini, C.; Cocconcelli, M.; Bellini, A.; Rubini, R. Bearing Fault Model for Induction Motor with Externally
Induced Vibration. IEEE Trans. Ind. Electron. 2012, 60, 3408–3418. [CrossRef]
Energies 2023, 16, 296 51 of 54
54. Lu, S.; He, Q.; Yuan, T.; Kong, F. Online Fault Diagnosis of Motor Bearing via Stochastic-Resonance-Based Adaptive Filter in an
Embedded System. IEEE Trans. Syst. Man, Cybern. Syst. 2016, 47, 1111–1122. [CrossRef]
55. Lu, S.; He, Q.; Zhao, J. Bearing fault diagnosis of a permanent magnet synchronous motor via a fast and online order analysis
method in an embedded system. Mech. Syst. Signal Process. 2018, 113, 36–49. [CrossRef]
56. Skrimpas, G.A.; Dragiev, I.G.; Hilmisson, R.; Sweeney, C.W.; Jensen, B.B.; Mijatovic, N.; Holbøll, J. Detection of generator bearing
inner race creep by means of vibration and temperature analysis. In Proceedings of the IEEE 10th International Symposium
on Diagnostics for Electrical Machines, Power Electronics and Drives (SDEMPED) Guarda, Portugal, 1–4 September 2015;
pp. 303–309.
57. Perez-Ramirez, C.A.; Valtierra-Rodriguez, M.; Dominguez-Gonzalez, A.; Amezquita-Sanchez, J.P.; Camarena-Martinez, D.;
Romero-Troncoso, R.J. Homogeneity-based approach for bearing fault detection in induction motors by means of vibrations. In
Proceedings of the IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC), Ixtapa, Mexico, 8–10
November 2017; pp. 1–5. [CrossRef]
58. Ciszewski, T. Induction motor bearings diagnostic indicators based on MCSA and normalized triple covariance. In Proceedings
of the IEEE 11th International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives (SDEMPED),
Tinos, Greece, 29 August–1 September 2017; pp. 498–502. [CrossRef]
59. Zhang, B.; Miao, Y.; Lin, J.; Yi, Y. Adaptive maximum second-order cyclostationarity blind deconvolution and its application for
locomotive bearing fault diagnosis. Mech. Syst. Signal Process. 2021, 158, 107736. [CrossRef]
60. Frosini, L.; Harlişca, C.; Szabó, L. Induction machine bearing fault detection by means of statistical processing of the stray flux
measurement. IEEE Trans. Ind. Electron. 2014, 62, 1846–1854. [CrossRef]
61. Amarnath, M.; Krishna, I.P. Empirical mode decomposition of acoustic signals for diagnosis of faults in gears and rolling element
bearings. IET Sci. Meas. Technol. 2012, 6, 279–287. [CrossRef]
62. Liu, Y.; Dai, Z.; Lu, S.; Liu, F.; Zhao, J.; Shen, J. Enhanced Bearing Fault Detection Using Step-Varying Vibrational Resonance
Based on Duffing Oscillator Nonlinear System. Shock Vib. 2017, 2017, 5716296. [CrossRef]
63. Ciabattoni, L.; Ferracuti, F.; Freddi, A.; Monteriu, A. Statistical spectral analysis for fault diagnosis of rotating machines. IEEE
Trans. Ind. Electron. 2017, 65, 4301–4310. [CrossRef]
64. Hamadache, M.; Lee, D.; Veluvolu, K.C. Rotor speed-based bearing fault diagnosis (RSB-BFD) under variable speed and constant
load. IEEE Trans. Ind. Electron. 2015, 62, 6486–6495. [CrossRef]
65. Deutsch, J.; He, M.; He, D. Remaining useful life prediction of hybrid ceramic bearings using an integrated deep learning and
particle filter approach. Appl. Sci. 2017, 7, 649. [CrossRef]
66. Li, L.; Cui, Y.; Chen, R.; Liu, X. Optimal SES Selection Based on SVD and Its Application to Incipient Bearing Fault Diagnosis.
Shock. Vib. 2018, 2018, 8067416. [CrossRef]
67. Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Bearings Ball Fault Detection Using KullbackLeibler Divergence in the EMD
Framework. In Proceedings of the IEEE 2018 Prognostics and System Health Management Conference (PHM-Chongqing),
Chongqing, China, 26–28 October 2018; pp. 729–734. [CrossRef]
68. Liao, Z.; Song, X.; Jia, B.; Chen, P. Bearing fault feature enhancement and diagnosis based on statistical filtering and 1.5-dimensional
symmetric difference analytic energy spectrum. IEEE Sens. J. 2021, 21, 9959–9968. [CrossRef]
69. Ambrożkiewicz, B.; Syta, A.; Meier, N.; Litak, G.; Georgiadis, A. Radial internal clearance analysis in ball bearings. Eksploat.
Niezawodn. 2021, 23. [CrossRef]
70. Wang, X.; Lu, S.; Zhang, S. Rotating Angle Estimation for Hybrid Stepper Motors with Application to Bearing Fault Diagnosis.
71. Leite, V.C.M.N.; Da Silva, J.G.B.; Veloso, G.F.C.; Da Silva, L.E.B.; Lambert-Torres, G.; Bonaldi, E.L.; De Lacerda De Oliveira, L.E.
Detection of Localized Bearing Faults in Induction Machines by Spectral Kurtosis and Envelope Analysis of Stator Current. IEEE
72. Li, H.; Liu, T.; Wu, X.; Chen, Q. A Bearing Fault Diagnosis Method Based on Enhanced Singular Value Decomposition. IEEE
Trans. Ind. Informatics 2020, 17, 3220–3230. [CrossRef]
73. Lu, S.; Guo, J.; He, Q.; Liu, F.; Liu, Y.; Zhao, J. A Novel Contactless Angular Resampling Method for Motor Bearing Fault Diagnosis
Under Variable Speed. IEEE Trans. Instrum. Meas. 2016, 65, 2538–2550. [CrossRef]
74. Singleton, R.K.; Strangas, E.G.; Aviyente, S. Time-frequency complexity based remaining useful life (RUL) estimation for bearing
faults. In Proceedings of the 2013 9th IEEE International Symposium on Diagnostics for Electric Machines, Power Electronics and
Drives (SDEMPED), Valencia, Spain, 27–30 August 2013; pp. 600–606. [CrossRef]
75. Shan, Z.; Wang, Z.; Yang, J.; Zhou, D.; Liu, H. Extracting non-stationary signal under strong noise background: Time-varying
system analysis. J. Vib. Control 2022. [CrossRef]
76. Mollasalehi, E.; Wood, D.; Sun, Q. Indicative fault diagnosis of wind turbine generator bearings using tower sound and vibration.
Energies 2017, 10, 1853. [CrossRef]
77. Jin, X.; Fan, J.; Chow, T.W.S. Fault Detection for Rolling-Element Bearings Using Multivariate Statistical Process Control Methods.
78. Sun, R.-B.; Yang, Z.-B.; Zhai, Z.; Chen, X.-F. Sparse representation based on parametric impulsive dictionary design for bearing
fault diagnosis. Mech. Syst. Signal Process. 2019, 122, 737–753. [CrossRef]
Energies 2023, 16, 296 52 of 54
79. Harmouche, J.; Delpha, C.; Diallo, D. Improved Fault Diagnosis of Ball Bearings Based on the Global Spectrum of Vibration
Signals. IEEE Trans. Energy Convers. 2014, 30, 376–383. [CrossRef]
80. Picot, A.; Obeid, Z.; Régnier, J.; Poignant, S.; Darnis, O.; Maussion, P. Statistic-based spectral indicator for bearing fault detection
in permanent-magnet synchronous machines using the stator current. Mech. Syst. Signal Process. 2014, 46, 424–441. [CrossRef]
81. Kundu, P.; Chopra, S.; Lad, B.K. Multiple failure behaviors identification and remaining useful life prediction of ball bearings. J.
Intell. Manuf. 2019, 30, 1795–1807. [CrossRef]
82. Singleton, R.K.; Strangas, E.G.; Aviyente, S. Extended Kalman filtering for remaining-useful-life estimation of bearings. IEEE
83. Li, Y.-F.; Zuo, M.; Feng, K.; Chen, Y.-J. Detection of Bearing Faults Using a Novel Adaptive Morphological Update Lifting Wavelet.
Chin. J. Mech. Eng. 2017, 30, 1305–1313. [CrossRef]
84. Zhou, W.; Habetler, T.G.; Harley, R.G. Bearing fault detection via stator current noise cancellation and statistical control. IEEE
85. Laha, S.K. Enhancement of fault diagnosis of rolling element bearing using maximum kurtosis fast nonlocal means denoising.
Measurement 2017, 100, 157–163. [CrossRef]
86. Aburakhia, S.A.; Myers, R.; Shami, A. A Hybrid Method for Condition Monitoring and Fault Diagnosis of Rolling Bearings with
Low System Delay. IEEE Trans. Instrum. Meas. 2022, 71, 1–13. [CrossRef]
87. Bazan, G.H.; Goedtel, A.; Castoldi, M.F.; Godoy, W.F.; Duque-Perez, O.; Morinigo-Sotelo, D. Mutual information and meta-
heuristic classifiers applied to bearing fault diagnosis in three-phase induction motors. Appl. Sci. 2020, 11, 314. [CrossRef]
88. Pham, M.T.; Kim, J.-M.; Kim, C.H. Intelligent Fault Diagnosis Method Using Acoustic Emission Signals for Bearings under
Complex Working Conditions. Appl. Sci. 2020, 10, 7068. [CrossRef]
89. Liang, M.; Cao, P.; Tang, J. Rolling bearing fault diagnosis based on feature fusion with parallel convolutional neural network. Int.
J. Adv. Manuf. Technol. 2021, 112, 819–831. [CrossRef]
90. Duong, B.P.; Kim, J.Y.; Jeong, I.; Im, K.; Kim, C.H.; Kim, J.M. A Deep-Learning-Based Bearing Fault Diagnosis Using Defect
Signature Wavelet Image Visualization. Appl. Sci. 2020, 10, 8800. [CrossRef]
91. Minervini, M.; Hausman, S.; Frosini, L. Transfer learning technique for automatic bearing fault diagnosis in induction motors.
In 2021 IEEE 13th International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives (SDEMPED); IEEE:
Piscataway Township, NJ, USA, 2021; Volume 1, pp. 186–192.
92. Eren, L. Bearing Fault Detection by One-Dimensional Convolutional Neural Networks. Math. Probl. Eng. 2017, 2017, 8617315.
[CrossRef]
93. Piltan, F.; Kim, J.-M. Bearing Fault Identification Using Machine Learning and Adaptive Cascade Fault Observer. Appl. Sci. 2020,
10, 5827. [CrossRef]
94. Van, M.; Hoang, D.T.; Kang, H.J. Bearing Fault Diagnosis Using a Particle Swarm Optimization-Least Squares Wavelet Support
Vector Machine Classifier. Sensors 2020, 20, 3422. [CrossRef]
95. Shenfield, A.; Howarth, M. A Novel Deep Learning Model for the Detection and Identification of Rolling Element-Bearing Faults.
Sensors 2020, 20, 5112. [CrossRef]
96. Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Bearing Fault Severity Classification Based on EMD-KLD: A Comparative Study for
Incipient Ball Faults. In Proceedings of the Prognostics and Health Management Conference (PHM-Besançon), Besancon, France,
4–7 May 2020; pp. 257–262.
97. Toma, R.N.; Kim, J.-M. Bearing Fault Classification of Induction Motors Using Discrete Wavelet Transform and Ensemble Machine
Learning Algorithms. Appl. Sci. 2020, 10, 5251. [CrossRef]
98. Hoang, D.T.; Kang, H.J. Deep belief network and dempster-shafer evidence theory for bearing fault diagnos. In Proceedings of
the IEEE international symposium on industrial electronics (ISIE), Cairns, Australia, 12–15 June 2018; pp. 841–846.
99. Zamudio-Ramirez, I.; Osornio-Rios, R.; Antonino-Daviu, J.; Cureño-Osornio, J.; Saucedo-Dorantes, J.-J. Gradual Wear Diagnosis
of Outer-Race Rolling Bearing Faults through Artificial Intelligence Methods and Stray Flux Signals. Electronics 2021, 10, 1486.
[CrossRef]
100. Liang, M.; Su, D.; Hu, D.; Ge, M. A Novel Faults Diagnosis Method for Rolling Element Bearings Based on ELCD and Extreme
Learning Machine. Shock Vib. 2018, 2018, 1891453. [CrossRef]
101. Gharesi, N.; Arefi, M.M.; Razavi-Far, R.; Zarei, J.; Yin, S. A neuro-wavelet based approach for diagnosing bearing defects. Adv.
Eng. Informatics 2020, 46, 101172. [CrossRef]
102. Zhao, Z.; Xu, Q.; Jia, M. Improved shuffled frog leaping algorithm-based BP neural network and its application in bearing early
fault diagnosis. Neural Comput. Appl. 2016, 27, 375–385. [CrossRef]
103. Tao, S.; Zhang, T.; Yang, J.; Wang, X.; Lu, W. Bearing fault diagnosis method based on stacked autoencoder and softmax regression.
In 2015 34th Chinese Control Conference (CCC); IEEE: Piscataway Township, NJ, USA, 2015; pp. 6331–6335.
104. Yin, H.; Li, Z.; Zuo, J.; Liu, H.; Yang, K.; Li, F. Wasserstein Generative Adversarial Network and Convolutional Neural Network
(WG-CNN) for Bearing Fault Diagnosis. Math. Probl. Eng. 2020, 2020, 2604191. [CrossRef]
105. Gunerkar, R.S.; Jalan, A.K.; Belgamwar, S.U. Fault diagnosis of rolling element bearing based on artificial neural network. J. Mech.
Sci. Technol. 2019, 33, 505–511. [CrossRef]
106. Al-Raheem, K.F.; Abdul-Karem, W. Rolling bearing fault diagnostics using artificial neural networks based on Laplace wavelet
analysis. Int. J. Eng. Sci. Technol. 2010, 2. [CrossRef]
Energies 2023, 16, 296 53 of 54
107. Islam, R.; Khan, S.A.; Kim, J.-M. Discriminant Feature Distribution Analysis-Based Hybrid Feature Selection for Online Bearing
Fault Diagnosis in Induction Motors. J. Sensors 2016, 2016, 7145715. [CrossRef]
108. Ahmed, H.; Nandi, A.K. Compressive Sampling and Feature Ranking Framework for Bearing Fault Classification with Vibration
Signals. IEEE Access 2018, 6, 44731–44746. [CrossRef]
109. Ettefagh, M.M.; Ghaemi, M.; Asr, M.Y. Bearing fault diagnosis using hybrid genetic algorithm K-means clustering. In Proceedings
of the IEEE International Symposium on Innovations in Intelligent Systems and Applications (INISTA) Proceedings, Alberobello,
Italy, 23–25 June 2014; pp. 84–89.
110. Kang, M.; Kim, J.; Kim, J.M. An FPGA-based multicore system for real-time bearing fault diagnosis using ultrasampling rate AE
signals. IEEE Trans. Ind. Electron. 2014, 62, 2319–2329. [CrossRef]
111. Appana, D.K.; Prosvirin, A.; Kim, J.-M. Reliable fault diagnosis of bearings with varying rotational speeds using envelope
spectrum and convolution neural networks. Soft Comput. 2018, 22, 6719–6729. [CrossRef]
112. Hasan, M.J.; Kim, J.M. Bearing fault diagnosis under variable rotational speeds using stock-well transform-based vibration
imaging and transfer learning. Appl. Sci. 2018, 8, 2357. [CrossRef]
113. Zhang, H.; Li, C.; Wei, Q.; Zhang, Y. Fault detection and diagnosis of the air handling unit via combining the feature sparse
representation based dynamic SFA and the LSTM network. Energy Build. 2022, 269, 112241. [CrossRef]
114. Piltan, F.; Prosvirin, A.E.; Jeong, I.; Im, K.; Kim, J.-M. Rolling-Element Bearing Fault Diagnosis Using Advanced Machine
Learning-Based Observer. Appl. Sci. 2019, 9, 5404. [CrossRef]
115. Deng, W.; Zhang, S.; Zhao, H.; Yang, X. A Novel Fault Diagnosis Method Based on Integrating Empirical Wavelet Transform and
Fuzzy Entropy for Motor Bearing. IEEE Access 2018, 6, 35042–35056. [CrossRef]
116. Song, Q.; Zhao, S.; Wang, M. On the Accuracy of Fault Diagnosis for Rolling Element Bearings Using Improved DFA and
Multi-Sensor Data Fusion Method. Sensors 2020, 20, 6465. [CrossRef]
117. Toma, R.N.; Prosvirin, A.E.; Kim, J.-M. Bearing Fault Diagnosis of Induction Motors Using a Genetic Algorithm and Machine
Learning Classifiers. Sensors 2020, 20, 1884. [CrossRef]
118. Huo, Z.; Zhang, Y.; Francq, P.; Shu, L.; Huang, J. Incipient Fault Diagnosis of Roller Bearing Using Optimized Wavelet Transform
Based Multi-Speed Vibration Signatures. IEEE Access 2017, 5, 19442–19456. [CrossRef]
119. Choudhary, A.; Mian, T.; Fatima, S. Convolutional neural network based bearing fault diagnosis of rotating machine using
thermal images. Measurement 2021, 176, 109196. [CrossRef]
120. Li, G.; Deng, C.; Wu, J.; Xu, X.; Shao, X.; Wang, Y. Sensor Data-Driven Bearing Fault Diagnosis Based on Deep Convolutional
Neural Networks and S-Transform. Sensors 2019, 19, 2750. [CrossRef]
121. Sadoughi, M.; Hu, C. Physics-Based Convolutional Neural Network for Fault Diagnosis of Rolling Element Bearings. IEEE
Sensors J. 2019, 19, 4181–4192. [CrossRef]
122. Zhou, Y.; Long, X.; Sun, M.; Chen, Z. Bearing fault diagnosis based on Gramian angular field and DenseNet. Math. Biosci. Eng.
2022, 19, 14086–14101. [CrossRef]
123. Magar, R.; Ghule, L.; Li, J.; Zhao, Y.; Farimani, A.B. Fault-Net: A deep convolutional neural network for bearing fault classification.
IEEE Access 2021, 9, 25189–25199. [CrossRef]
124. Sohaib, M.; Kim, J.-M. Hierarchical Radial Basis Function Based Multiclass Support Vector Machines and a Hybrid Feature Pool
for Bearings Fault Diagnosis. In Proceedings of the IEEE International Conference on Electrical Engineering (ICEE), Lahore,
Pakistan, 15–16 February 2018; pp. 1–6. [CrossRef]
125. Gao, Y.; Liu, X.; Xiang, J. FEM Simulation-Based Generative Adversarial Networks to Detect Bearing Faults. IEEE Trans. Ind.
Informatics 2020, 16, 4961–4971. [CrossRef]
126. Wang, Z.; Yao, L.; Chen, G.; Ding, J. Modified multiscale weighted permutation entropy and optimized support vector machine
method for rolling bearing fault diagnosis with complex signals. ISA Trans. 2021, 114, 470–484. [CrossRef]
127. Zheng, J.; Dong, Z.; Pan, H.; Ni, Q.; Liu, T.; Zhang, J. Composite multi-scale weighted permutation entropy and extreme learning
machine based intelligent fault diagnosis for rolling bearing. Measurement 2019, 143, 69–80. [CrossRef]
128. William, P.; Hoffman, M. Identification of bearing faults using time domain zero-crossings. Mech. Syst. Signal Process. 2011, 25,
3078–3088. [CrossRef]
129. Pan, Z.; Meng, Z.; Chen, Z.; Gao, W.; Shi, Y. A two-stage method based on extreme learning machine for predicting the remaining
useful life of rolling-element bearings. Mech. Syst. Signal Process. 2020, 144, 106899. [CrossRef]
130. Lu, Y.; Xie, R.; Liang, S.Y. CEEMD-assisted kernel support vector machines for bearing diagnosis. Int. J. Adv. Manuf. Technol. 2020,
106, 3063–3070. [CrossRef]
131. Ng, S.S.; Tse, P.W.; Tsui, K.L. A one-versus-all class binarization strategy for bearing diagnostics of concurrent defects. Sensors
2014, 14, 1295–1321. [CrossRef]
132. Loutas, T.H.; Roulias, D.; Georgoulas, G. Remaining useful life estimation in rolling bearings utilizing data-driven probabilistic
E-support vectors regression. IEEE Trans. Reliab. 2013, 62, 821–832. [CrossRef]
133. Chen, Z.; Deng, S.; Chen, X.; Li, C.; Sánchez, R.V.; Qin, H. Deep neural networks-based rolling bearing fault diagnosis.
Microelectron. Reliab. 2017, 75, 327–333. [CrossRef]
134. Soualhi, A.; Medjaher, K.; Zerhouni, N. Bearing health monitoring based on Hilbert–Huang transform, support vector machine,
and regression. IEEE Trans. Instrum. Meas. 2014, 64, 52–62. [CrossRef]
Energies 2023, 16, 296 54 of 54
135. Wang, B.; Zhang, X.; Xing, S.; Sun, C.; Chen, X. Sparse representation theory for support vector machine kernel function selection
and its application in high-speed bearing fault diagnosis. ISA Trans. 2021, 118, 207–218. [CrossRef] [PubMed]
136. Mbo’o, C.P.; Hameyer, K. Fault diagnosis of bearing damage by means of the linear discriminant analysis of stator current features
from the frequency selection. IEEE Trans. Ind. Appl. 2016, 52, 3861–3868. [CrossRef]
137. Zhu, H.; Cheng, J.; Zhang, C.; Wu, J.; Shao, X. Stacked pruning sparse denoising auto-encoder based intelligent fault diagnosis of
rolling bearings. Appl. Soft Comput. 2020, 88, 106060. [CrossRef]
138. Saidi, L.; Ali, J.B.; Fnaiech, F. Application of higher order spectral features and support vector machines for bearing faults
classification. ISA Trans. 2015, 54, 193–206. [CrossRef]
139. Saucedo-Dorantes, J.J.; Arellano-Espitia, F.; Delgado-Prieto, M.; Osornio-Rios, R.A. Diagnosis methodology based on deep feature
learning for fault identification in metallic, hybrid and ceramic bearings. Sensors 2021, 21, 5832. [CrossRef]
140. Hu, L.; Wang, L.; Chen, Y.; Hu, N.; Jiang, Y. Bearing Fault Diagnosis Using Piecewise Aggregate Approximation and Complete
Ensemble Empirical Mode Decomposition with Adaptive Noise. Sensors 2022, 22, 6599. [CrossRef]
141. Kaya, Y.; Kuncan, F.; Ertunç, H.M. A new automatic bearing fault size diagnosis using time-frequency images of CWT and deep
transfer learning methods. Turk. J. Electr. Eng. Comput. Sci. 2022, 30, 1851–1867. [CrossRef]
142. Ozcan, I.H.; Devecioglu, O.C.; Ince, T.; Eren, L.; Askar, M. Enhanced bearing fault detection using multichannel, multilevel 1D
CNN classifier. Electr. Eng. 2022, 104, 435–447. [CrossRef]
143. Brusamarello, B.; Da Silva, J.C.C.; De Morais Sousa, K.; Guarneri, G.A. Bearing fault detection in three-phase induction motors
using support vector machine and fiber Bragg grating. IEEE Sensors J. 2022. [CrossRef]
144. Avina-Corral, V.; de Jesus Rangel-Magdaleno, J.; Peregrina-Barreto, H.; Ramirez-Cortes, J.M. Bearing Fault Detection in ASD-
Powered Induction Machine Using MODWT and Image Edge Detection. IEEE Access 2022, 10, 24181–24193. [CrossRef]
145. Hu, J.; Deng, S. Rolling bearing fault diagnosis based on wireless sensor network data fusion. Comput. Commun. 2022, 181,
404–411. [CrossRef]
146. Deekshit, K.K.C.; Madhav, G.V. Progressive Bearing Fault Detection in a Three-Phase Induction Motor Using S-Transform via
Pre-Fault Frequency Cancellation. In Advanced Practical Approaches to Web Mining Techniques and Application; IGI Global: Hershey,
PA, USA, 2022; pp. 209–228.
147. AlShorman, O.; Alkahatni, F.; Masadeh, M.; Irfan, M.; Glowacz, A.; Althobiani, F.; Kozik, J.; Glowacz, W. Sounds and acoustic
emission-based early fault diagnosis of induction motor: A review study. Adv. Mech. Eng. 2021, 13. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.

The Bearing Faults Detection Methods For Electrica

Uploaded by

Copyright:

Available Formats

The Bearing Faults Detection Methods For Electrica

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Bearing Faults Detection Methods For Electrica

Uploaded by

Copyright:

Available Formats

energies

1 Department of Electrical Power Engineering, The Islamia University of Bahawalpur,

Energies 2023, 16, 296. https://doi.org/10.3390/en16010296 https://www.mdpi.com/journal/energies

Energies 2023, 16, x FOR PEER REVIEW 2 of 54

Figure 2. The major classifications of rolling bearings.

Energies 2023, 16,Energies 2023, 16, x FOR PEER REVIEW

In [51], the authors proposed a scheme to diagnose corrosion failures in bearings

Figure 17. The

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

In [61], the authors implemented empirical mode decomposition (EMD) of acoustic

Energies 2023, 16, x FOR PEER REVIEW 18 of 54

Energies 2023, 16, x FOR PEER REVIEW

Energies 2023, 16, x FOR PEER REVIEW 21 of 54

Energies 2023, 16, x FOR PEER REVIEW 22 of 54

Energies 2023, 16, x FOR PEER REVIEW 24 of 54

Table 2. A summary of different statistical techniques for bearing fault diagnosis.

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

4. Artificial Intelligence (AI) Techniques for Diagnosis of Bearing Faults

Energies 2023, 16, x FOR PEER REVIEW 28 of 54

Energies 2023, 16, x FOR PEER REVIEW 29 of 54

In [101], the authors suggested an intelligent neuro-wavelet neural network in combi-

simulated vibration signals.

Energies 2023, 16, x FOR PEER REVIEW 31 of 54

Figure 42. (a)

Energies 2023, 16, x FOR PEER REVIEW 34 of 54

In [120], the authors proposed a sensor-driven technique to diagnose the bearing

In [129], the authors suggested the incorporation of a multivariable feedback extreme

Energies 2023, 16, x FOR PEER REVIEW 41 of 54

In [144], the authors investigated abnormal conditions of ball bearings by employing

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

Technique/Assisting Bearing Fault

5. Recommendations and Suggestions

5.1. Gaps and Thoughts for Future Investigation

5.2. Current Challenges

5.3. Future Work Directions

You might also like