Вы находитесь на странице: 1из 7

Supervision and Safety for Technical Processes

10th
10th IFAC
IFAC
Warsaw, Symposium
Symposium on
on Fault
Fault Detection,
Detection,
10th IFACPoland,
Supervision and
August
Symposium
Safetyon29-31,
Fault
for
2018
Detection,
Technical Processesonline at www.sciencedirect.com
Available
Supervision
Supervision and
and Safety
10th IFAC Symposium
Safety for
on Technical
Fault
for Processes
Detection,
Technical Processes
Warsaw,
Warsaw, Poland,
Poland,
Supervision August
August 29-31,
29-31, 2018
2018
Warsaw, Poland, August 29-31, 2018 Processes
and Safety for Technical
Warsaw, Poland, August 29-31, 2018
ScienceDirect
A comparative featureIFAC PapersOnLine
analysis for51-24 (2018)pitting
gear 346–352 level classification by using
A
A comparative
A comparative
comparative acoustic
featureemission,
feature
feature analysis
analysis for
analysis vibration
for
for gear
gear and current
gear pitting
pitting
pitting level
level signals
level classification
classification
classification by
by using
by using
using
A comparative feature
acoustic
acoustic analysis
emission,
emission, for gear
vibration
vibration pitting
and level
current
and current
currentclassification
signals
signals by using
acoustic
René-Vinicio Sánchez*, Pabloemission,
Lucero*, vibration
Rafael E. and
Vásquez**, Marielasignals
Cerrada*, Diego Cabrera*. 
acoustic emission, vibration and current signals
René-Vinicio
René-Vinicio Sánchez*,
Sánchez*, Pablo
Pablo Lucero*,
Lucero*, Rafael E.
Rafael- E.
E. Vásquez**,
Vásquez**, Mariela
Mariela Cerrada*,
Cerrada*, Diego Cabrera*.
Cabrera*.
Diego Cabrera*.
René-Vinicio Sánchez*,
* GIDTEC, Pablo
Universidad Lucero*,
Politécnica Rafael
Salesiana Vásquez**,
UPS, Mariela
Cuenca, Ecuador Cerrada*,
(e-mail: Diego
rsanchezl@ups.edu.ec).
René-Vinicio Sánchez*,
**Department Pablo Lucero*,
of Mechanical Engineering,Rafael E. Vásquez**,
Universidad Mariela Cerrada*,
Pontificia Bolivariana, Diego Cabrera*.
Medellín, Colombia (e-mail: 
* GIDTEC,
* Universidad Politécnica
GIDTEC, Universidad Politécnica Salesiana -- UPS, UPS, Cuenca, Ecuador
Ecuador (e-mail: rsanchezl@ups.edu.ec).
rsanchezl@ups.edu.ec).
* GIDTEC, Universidad Politécnica Salesiana Salesiana - UPS, Cuenca,
rafael.vasquez@upb.edu.co).Cuenca, Ecuador (e-mail: (e-mail: rsanchezl@ups.edu.ec).
**Department
* GIDTEC,
**Department of Mechanical
of Mechanical
Universidad Engineering,
Politécnica
Engineering, Salesiana Universidad
-
UniversidadUPS, Pontificia
Cuenca,
Pontificia Bolivariana,
Ecuador
Bolivariana,(e-mail: Medellín, Colombia (e-mail:
rsanchezl@ups.edu.ec).
Medellín, Colombia (e-mail:
**Department of Mechanical Engineering, Universidad Pontificia Bolivariana, Medellín, Colombia (e-mail:
**Department of Mechanical Engineering, rafael.vasquez@upb.edu.co).
Universidad
rafael.vasquez@upb.edu.co).Pontificia Bolivariana, Medellín, Colombia (e-mail:
rafael.vasquez@upb.edu.co).
rafael.vasquez@upb.edu.co).
Abstract: this paper addresses the comparison of features, extracted in the time domain, from vibration,
acoustic
Abstract: emission,
this paper and current the
addresses signals,
comparisonfor the ofidentification
features, of eightin levels
extracted the time of domain,
severity from of pittingvibration, in a
Abstract:
Abstract:
gearbox. this
The paper
paper addresses
thisvibration, addresses
acoustic the comparison
theemission,
comparison and of features,
ofcurrent
features, extracted
extracted
signals were in the
infirst time
time domain,
the acquired domain,using from
from
a vibration,
vibration,
gearbox laba
acoustic emission,
thistest
paper and current
addresses signals,
the comparisonfor the ofidentification
features, of
extracted eight in levels
the time of severity of pitting in
Abstract:
acoustic
acoustic
experimental emission,
emission, and
and
bed. current
current
Then, twentysignals,
signals, for
for the
features the
wereidentification
identification
extracted inof
ofthe eight
eight
time levels
levels
domain of domain,
of severity
severity
from each from
of
ofsignal; vibration,
pitting
pitting in
in aa
these
gearbox.
acoustic The
emission,vibration,
and acoustic
current emission,
signals, for and current signalsofwere
the identification eightfirstlevels acquired
ofallows using
severity aa gearbox
pitting inlab
gearbox.
gearbox.
features are
experimental
The vibration,
Theranked
vibration,
test by Then,
bed.
acoustic
acoustic
Chi squared
twenty
emission,
emission,
and entered
features
and
and
were
current
into the signals
current signals were
were
KNNin classifier,
extracted the time
first
first acquired
acquired
which
domain from
using
using theof
each
gearbox
aevaluation
gearbox
signal; these ofa
lab
lab
gearbox.
experimental
experimental The
the classification vibration,
test bed.
test bed. acoustic
Then,
Then,for
accuracy twenty
twenty emission,
eachand features
features
acquired andwere
werecurrent
signal signals
extracted
extracted in were
the time first
in the timeandomain
and performing acquired
domain
analysis from
from usingeach
of each
the a gearbox
signal;
signal; these
features. theselab
The
features
experimental
features are
are ranked
test
ranked by
bed. Chi
by Then,
Chi squared
twenty
squared and entered
features
entered wereinto
into the
the KNN
extracted
KNN classifier,
inpitting
the time
classifier, which
domain
which allows
fromthe
allows the
each
the evaluation
signal;
evaluation these of
of
features
results are
indicate ranked
that by
the Chi squared
vibration and and
AE entered
signals into the
identified KNN
the classifier,level which
better allows
than the evaluation
current signal. of
the classification
the classification
features are ranked accuracy
by
accuracyChi for each
squared
for eachand acquired
entered
acquired signal
into
signal andKNN
the
and performing
classifier,
performing anwhich
an analysis
analysis allows of the
of the
the features.
evaluation
features. The
Theof
the classification accuracy for each acquired signal and performing an analysis of the features. The
©
the 2018,
results
results
Keywords: IFAC
indicate
classification
indicate fault(International
that
that the
diagnosis, Federation
the vibration
accuracyvibration
for
pitting and
each
and
fault,AE
AEof Automatic
signalssignal
acquired
signals
acoustic Control)
identified
and
identified
emission, Hosting
the pittingby
performing
the pitting
vibration,
results indicate that the vibration and AE signals identified the pitting level better than the current signal. Elsevier
level
an
level
current, better
analysis
better
spur Ltd.
than
thanAll
of
gearbox, rights
the
the
the currentreserved.
features.
current
fault signal.
signal.
level The
results indicate that the vibration and AE signals identified the pitting level better than the current signal.
Keywords: fault
Keywords: diagnosis, pitting fault, acoustic emission, vibration, current, spur gearbox, fault level
Keywords:1.fault fault diagnosis,
diagnosis, pitting
INTRODUCTION pitting fault,
fault, acoustic
acoustic emission, vibration,
vibration
emission, vibration, current,
current, spur
conditions gearbox,
gearbox,orfault
indicators
spur fault level
features
level (RMS and
Keywords: fault diagnosis, pitting fault, acoustic emission, vibration,
kurtosis) current, spur
as measures gearbox, monitoring
for effective fault level of pitting in
vibration
vibration conditions indicators or features (RMS and
Gearboxes are used1. 1. INTRODUCTION
INTRODUCTION spur gears.asconditions
vibration conditions
Limeasures
et al. (2016) indicators
indicators or
or features
used measurements featuresof(RMS (RMS
anpitting and
and
acoustic
1.within different industries due to their kurtosis)
INTRODUCTION vibration
kurtosis) asconditions
measures for
for effective
indicators
effective or monitoring
features
monitoring of
(RMS
of pitting in
and
in
characteristics to 1.
transmit INTRODUCTION
high torques in small spaces. Gear kurtosis)
emission as
(AE) measures
sensor for
and effective
an accelerometer monitoring for of
the pitting
diagnosis in
Gearboxes are used within different industries due to their spur
spur gears.
kurtosis)
gears. as Li
Li et
et al.
measures
al. (2016)
for
(2016) used
effective
used measurements
monitoring
measurements of
of of an
an acoustic
pitting
acousticin
Gearboxes
Gearboxes
failures represent are
are used
used within
within
nearly high different
different
60 torques
% of in industries
industries
thesmall due
due
failures to
to their
their
inGear spur
of
the emission gears.
eleven Li et
conditions al. (2016)
in the used
gearbox; measurements
the statistical of an acoustic
parameters
characteristics
Gearboxes are to
usedtransmit
within different industries spaces.
due to their spur gears.
emission (AE)
(AE)Li etsensor
al.
sensor and
(2016)
and an
used
an accelerometer
measurements
accelerometer for
for ofthe
the andiagnosis
acoustic
diagnosis
characteristics
characteristics
gearboxes’ to transmit
to transmitFaults
components. high torques
high arise
torques in small
in small
in gear teeth, spaces.
spaces.
and in Gear
Gear
can be of emission
of eleven (AE) sensor
the wavelet packetand an accelerometer
transform (WPT) areforfirst the diagnosis
produced
failures represent nearly 60 % of the failures the emission
of eleven conditions
(AE)
conditionssensorin in the
and
the gearbox;
an accelerometer
gearbox; the
the statistical
for the parameters
diagnosis
characteristics
failures
failures
described represent
represent
as to transmit
pitting, nearly
nearly
cracking,high
60 torques
%
% of
60 wear, ofandin
the small
thespalling spaces.
failures
failures inGear
in
(Chaari the
the
et of
fromeleven
the conditions
AE signal inandthethe gearbox;
vibratory the statistical
statistical
signal, parameters
parameters
respectively,
gearboxes’
failures components. Faults arise in gear teeth, and can be of
of the
eleven
thetheywavelet
conditions
wavelet packet
packet in thetransform
gearbox;
transform (WPT)
(WPT) are
the statistical
are first produced
firstparameters
produced
al., 2008;represent
gearboxes’
gearboxes’
described
components.
components.
Korka
as
nearly
et al.,
pitting,
Faults
Faults
2017;
cracking,
60Liang% of
arise
arise
wear,
in the2016).
gear
in al.,
etand
gear failures
teeth,
teeth,
spalling
and in
and
The
(Chaari
can
can the
be
be from
pittinget
of
then
of
the
the the
wavelet
AE
wavelet
packet
are evaluated
signal
packet and
transform
in
the
transform
(WPT)
a k-nearest
vibratory
(WPT)
are
neighbor
signal,
are
first produced
(KNN)
respectively,
first produced
gearboxes’
described ascomponents.
pitting, Faults
cracking, arise
wear, inandgear teeth,
spalling and
(Chaarican be
et from
from the
the
classifier AE
AE
where signal
signal
the and
and
vibratory the
the vibratory
vibratory
sensor exhibitssignal,
signal, a respectively,
respectively,
little bit better
described
fault can as
be pitting,
attributed cracking,
to wear, and
excessive contactspallingstress (Chaari
due et then
to from they
the AE are evaluated
signal and in
the aavibratory
k-nearest neighbor
signal, (KNN)
respectively,
al.,
al., 2008;
described
2008; Korka
as
Korka et
pitting,
et al.,
al., 2017;
cracking,
2017; Liang
wear,
Liang et
et al.,
and
al., 2016).
spalling
2016). The
(Chaari
The pitting
pittinget then
then they
they
classification are
are rateevaluated
evaluated
than the AEin
in a
sensork-nearest
k-nearest
in the neighbor
neighbor
experiments. (KNN)
(KNN)
al.,
heavy2008; loadKorka et al., 2017;
transmission and Liang et al., 2016).
misalignment (Ozturk The pitting
et al., classifier
then theywhere the
are evaluated vibratory in sensor
a k-nearest exhibits aa little bit(KNN)
better
fault
al.,
fault
fault
2010).
can
2008;
can
can
The
be
Korka
be attributed
attributed
becondition
attributed
to
et al.,monitoring
2017;
to excessive
Liang
to excessive
excessiveof et
the
contact
al.,
contact
contact 2016).
gearbox
stress
stress
stress
is The due
pitting
due
due
important
to
to classifier
classifier
to classification
where
where the
the vibratory
vibratory sensor
sensor exhibitsneighbor
exhibits a little
little bit
bit better
better
heavy
fault load
can be transmission
attributed to and misalignment
excessive contact (Ozturk
stress et
due al.,
to classifier
classification
Aouabdi where
et rate
rate
al. thethan
than
(2017) the
vibratory
the AE
AE
presentedsensor
sensor
sensor thein
in the
exhibits
the experiments.
a little
experiments.
performance bitofbetter
the
heavy
heavy
to detect load transmission
loadpitting
transmission
faults and
atand theirmisalignment
misalignment
early stage; (Ozturk
(Ozturk
this et
canet helpal., classification rate than the AE sensor
al., classification rate than the AE sensor in the experiments. in the experiments.
2010).
heavy
2010). The
load condition
transmission
Theunscheduled monitoring
condition monitoring and
monitoring of the
misalignment
of the gearbox
the gearbox
gearbox is
(Ozturk important
is importantet
important al., technical analysis of current electric signals, and their
2010).
reducing The condition stops, of
ensuring is
the continuity of Aouabdi
Aouabdi et al. (2017) presented the performance of the
to detect
2010). Thepitting
condition faults at
monitoring their early
the stage;
gearboxthis can help potential et
Aouabdi et
for al.
al. (2017)
(2017) presented
monitoring presented
and diagnosis the
the performance
performance
of a gearbox; of
of the
the
to
to detect
detect
operations pitting
pitting
and faults
faults
avoiding at their of
their
atcatastrophic early
early stage;
stage;
failures is
this
this
in important
can
can
the help
help
gear technical
Aouabdi
technical analysis
et al. (2017)
analysis of
of current
presented electric
theentropy signals,
performance and their
oftheir
the
reducing
to detect
reducing unscheduled
pitting
unscheduled faults stops,
at
stops,their ensuring
early
ensuring the
stage;
the continuity
this
continuitycan of
help
of technical
proposed analysis
method, of current
using current
multi-scale electric
electric signals,
signals, and and
and their
principal
reducing
system unscheduled
(Cerrada et al., stops,
2016; ensuring
Ozturk et al., the continuity of potential
2008). technical
potential for monitoring
analysis
for monitoring of and
current
and diagnosis
electric
diagnosis of
signals,
of aa gearbox;
and
gearbox; the
their
the
operations
reducing and avoidingstops, catastrophic
ensuringfailures in the gear potential
componentfor monitoringisanduseful
analysis, diagnosis for ofmonitoring
a gearbox; and the
operations unscheduled
operations and
and avoiding
avoiding catastrophic
catastrophic the continuity
failures
failures in
in the the gear of proposed
gear potential
proposed method,
for
method,monitoring using
using multi-scale
and diagnosis
multi-scale entropy
of
entropy a and principal
gearbox;
and the
principal
system
operations
system
Different (Cerrada
(Cerradaand
gear et al.,
avoiding
et
fault al., 2016;
2016; Ozturk
catastrophic
diagnosis Ozturk andet
et al., 2008).
failures
al., 2008).
condition in the
monitoring gear proposed
classification method,of five usingfault multi-scale
patterns entropy
via the and
motor principal
current
system (Cerrada et al., 2016; Ozturk et al., 2008). component
proposed
component analysis,
method,
analysis, using is
is useful
multi-scale
useful for
entropy
for monitoring
and
monitoring and
principal
and
system
techniques (Cerrada etbeen
have fault al., 2016;
developed Ozturk foretclassification
al., 2008). levels of classification component
signature analysis analysis, (MCSA). is useful He et via for monitoring
al. (2014) combinedand
Different gear diagnosis and condition monitoring component
classification of
of five
analysis,
five fault
is patterns
fault useful
patterns for the
via the motor
monitoring
motor current
and
current
Different
Different
gear failures. gear fault
gearLiufault diagnosis
et al.diagnosis
(2011) proposed and condition
and condition monitoring
a methodmonitoring based on classification
empirical mode of five fault
decomposition, patterns
fast via the
independent motor current
component
techniques
Different have
gear been developed for classification levels of signature
classification
signature analysis
of
analysis five (MCSA).
fault
(MCSA). He
patterns
He et
et al.
via
al. (2014)
the
(2014) motor combined
current
combined
techniques
techniques
the linear have fault
have been diagnosis
been
discriminant developed
developed
analysis and
for condition
classification
for support
and classificationvector monitoring
levels
levels
machine of signature
analysis, and
of empirical
signature
analysis
mode
analysis
(MCSA).
an decomposition,
entropy
(MCSA).
measure Hetoetpropose
He fast
al. (2014)
etindependent
combined
a hybrid-feature
(2014) component
al. analyze combined
gear
gear failures.
techniques
failures. haveLiu
Liu et
been
et al.
al. (2011)
developed
(2011) proposed
for
proposed aa method
classification
method based
levels
based on
of
on empirical
empirical
extraction mode
mode decomposition,
decomposition,
methodology; it is fast
fast
used independent
independent
to component
component
the stator
gear
(SVM) failures. Liu etfour
to discriminant
identify al. pitting
(2011) faultsproposedlevels a method
(baseline, based on analysis,
slight, empirical and
mode an entropy
decomposition, measure to
fast propose
independent aa hybrid-feature
component
the
gear
the linear
failures.
linear Liu
discriminant et al. analysis
(2011)
analysis and
and support
proposed
support a vector
method
vector machine
based
machine on analysis,
analysis,
current and
and
signals an
an forentropy
entropy
gear measure
measure
fault to
to
detection propose
propose
normal a hybrid-feature
hybrid-feature
condition and
the linear
moderate, discriminant
and severe) analysis
of planetaryand support
gearboxes vector machine
using the extraction
analysis, and methodology;
anatentropy it
measureis used to to
propose analyzeaelectric the
hybrid-featurestator
(SVM)
the linearto identify
discriminant four pitting faults levels (baseline, slight, extraction methodology; it
it isis used to
to analyze the stator
(SVM)
(SVM)
vibration to identify
to signal.
identifyZhao four analysis
four pitting
pitting
et al. and support
faults
faults
(2013) levels
levels
employed vector
(baseline,
(baseline, an machine
slight, extraction
pitting signals
slight, current
ordinal extraction
methodology;
failure for
methodology;
two different
gear fault
used
detection
it Urevc
is
speeds
used (2008)
analyze
of
tonormal
analyze
the motor
condition
stator
and
the stator
moderate,
(SVM)
moderate, and
to identify
and severe)
severe) of
four pitting
of planetary
faults the
planetary gearboxes
levels
gearboxes(baseline, using
using the
slight,
the current
rotation.signals
current signals for
for gear
Boshkoski gearand fault
fault detection
detection normal
normal condition
condition
evaluated and
and
the
moderate,
ranking-based and approach
severe) of preserve
to planetary gearboxes
ordinal using
information the pitting
current failure
signals at
for two
gear different
fault detectionspeeds of
normal electric
condition motor
and
vibration
moderate,
vibration signal.
and
signal. Zhao
severe)
Zhao et
of
et al.
al. (2013)
planetary
(2013) employed
gearboxes
employed an
using
an ordinal
the
ordinal pitting
pitting failure
failure
monitoringBoshkoski at
at
of progressivetwo
two different
different
pitting speeds
speeds
gear of
of
faults electric
electric
using motor
motor
electrical
vibration
and to signal. four
identify Zhao pitting
et al. faults
(2013) levels
employed of a anplanetary
ordinal rotation.
pitting and Urevc (2008) evaluated the
ranking-based
vibration
ranking-based
ranking-based
gearbox signal.
from
approach
Zhao
approach
approach
four
to
et
to
vibration
preserve
to al. (2013)the
preserve
preserve
signals.the ordinal
theemployed
ordinal information
an
ordinaletinformation
Korka ordinal
information
al. current failure
rotation.
rotation.
(2017) monitoring and Boshkoski
Boshkoskiat twoemissions,
sound different
and
and Urevc
Urevcand speeds (2008)
(2008) of electric
determined evaluated
evaluated thatmotor
the
the
and to
ranking-basedidentify four pitting faults levels of a planetary rotation.
monitoring of
of progressive
Boshkoski
progressive and pitting
Urevc
pitting gear
gear faults
(2008)
faults using
evaluated
using electrical
the
electrical
and
and
assessed
gearbox
to three approach
to identify
identify
from
four
four pitting
different
four
topitting
pitting
vibration
preserve
faults
faults the
levels
signals.
ordinal
levels
levels
of
Korka aof et
aainformation
ofsingle
al.
planetary
planetary
helical
(2017)
monitoring
electrical current
current
monitoring and
of progressive
of sound
has less pitting
progressive emissions,
sensitivity;
pitting
gear nonetheless,
and
gear
faults using electrical
determined
faults using
with that
that the
electrical
and to identify four pitting faults current and sound emissions, and determined that the
gearbox
gearbox
assessed
from
from
at
threethreefour
four vibration
vibration
different
different speeds
pitting by levels
signals.
signals.
levels
Korka
using
of
ofet
Korkavibrationeta al.
aa single
al.planetary
(2017)
(2017) current
sensitivity,
signal electrical
helical current
andit can
andcurrent
sound
sound has
emissions,as and
be considered
less sensitivity;
emissions,
a reliabledetermined
andtononetheless,
signal for
determined
that
with
the
pitting
that
that fault
the
gearbox
assessed from
three four
differentvibrationpitting signals.
levels Korka
of et
single al. (2017)
helical electrical
electrical
detection, current
current
and sound has
has less
less
signals sensitivity;
sensitivity;
are able nonetheless,
nonetheless,
detect the with
with
initial that
that
assessed
analysis; threeit wasdifferentdetermined pittingthatlevels
with theof aincreasesingle helical of the sensitivity,
electrical it can
current be
has considered
less as
sensitivity; aa reliable signal
nonetheless, for
withpitting
that
gearbox
assessed
gearbox at
three
at three different
different speeds
pitting by
levels using
of avibration
single signal
helical sensitivity, it can be considered as reliable signal for pitting
gearbox
pitting grade,at three
three different
vibrationspeeds
different
thedetermined speeds by
by using
amplitudes usingat the vibration
vibration
engagement signal sensitivity,
sensitivity,
it can be considered as a reliable signal for pitting
generation earliest.
signal detection, and sound
it can signals
be considered are as able to
a reliable detectsignal the initial fault
for pitting
analysis;
gearbox atit was that with the increase of the detection,
detection, and sound signals
signals are are able to
to detect
detect the the initial fault
analysis;
analysis;
frequencies it three
it was
was their
and different
determined
determined
harmonics speeds
that by
withusing
with
thatincrease the
thetoo. vibration
increase
increase
Amarnath signal
of
of the
the generation
and detection,
and sound
earliest.
and sound signals are
able
able to detect the
initial
initial
fault
fault
pitting
analysis;
pitting grade,
it
grade, wasthe
the vibration
determined
vibration amplitudes
that with
amplitudes at
the
at the engagement
increase
the engagement of the generation
Tan et al.earliest.
generation earliest.(2007) evaluated three condition monitoring
pitting
Lee grade,
(2015) the vibration
evaluated pitting amplitudes
in spur gears atunder
the engagement
fatigue test generation earliest.
frequencies
pitting
frequencies grade, and
and thetheir
their harmonics
vibration
harmonics increase
amplitudes
increase too.
at
too. theAmarnath
engagement
Amarnath and
and techniques on a spur gearbox in which natural pitting was
frequencies
conditions; andstudies
the their harmonics
considered increase
the gears too. Amarnath
lubricant film thicknessand Tan Tan et
et toal. (2007)
al.occur;
(2007) evaluated
evaluated three
three condition
condition monitoring
monitoring
Lee (2015)
frequencies
Lee (2015) evaluated
and their
evaluated pitting
harmonics
pitting in
in spur
increase
spur gears under
too.
under fatigue
Amarnath
fatigue test
and
test Tan
allowed et al. (2007) evaluated
it was observed three
that, condition
based on the monitoring
analysis
Lee (2015)
analysis, wear evaluated
mechanism pitting in spur
studies on gears
gear tooth undersurfaces,
fatigue test oil techniques
Tan al. on
et levels aa spur gearbox in which natural pitting was
conditions;
Lee (2015) the studies
evaluated considered the lubricant film thickness techniques
techniques
of RMS on(2007)
on a spur
only,spurtheevaluated
gearbox
gearbox
AE technique inthree
in which
which condition
was natural
natural
more sensitive monitoring
pitting
pitting was
was
for
conditions;
conditions;
degradation the
the studies pitting
studies
analysis usingin Fourier
considered
considered spur
the gears
the lubricant
lubricant under
transform film
film fatigue
thickness
thicknesstest allowed
infrared techniques
allowed to
to occur;
on
occur; a it
spur
it was
was observed
gearbox
observed in that,
which
that, based
natural
based on
on the analysis
pitting
the was
analysis
analysis,
conditions; wear mechanism
the mechanism
studies studies on gear tooth surfaces, oil allowed
detecting to occur;
and it was
monitoring observed that,
pitting based
than on the
vibrationanalysisor
analysis,
radiation wear
analysis,
degradation
wear
method analysisalongconsidered
mechanism with
using
studies
studies the
on lubricant
on
vibration
Fourier
gear
gearsignaltoothfilm
tooth
transform analysis.thickness
surfaces,
surfaces, oil
oil of
The
infrared
RMS
allowed
of RMS levels
to occur;
levels only,
only, the
itAnalysis
was
the AE technique
observed
AE technique that,was based
was more
more on sensitive
the analysis
sensitive for
for
analysis, wearanalysis
mechanism studies on geartransform of RMS levels
spectrometric only,
oil the AE technique
techniques. was On more
the sensitive
other for
hand,
degradation
degradation
results provided analysis goodusing
aalong using Fourier
Fourier
understanding oftooth
transform surfaces,
tribological infrared
infraredoil detecting and monitoring pitting than vibration or
and of RMS levels
radiation
radiation method
degradation method analysisalong usingwith
with vibration
Fourier signal
vibration transform
signal analysis.
analysis. The
infrared
detecting
detecting
The spectrometric andonly,
and the AE technique
monitoring
monitoring pitting was
pitting than
thanmorevibrationsensitive for
vibration or
or
radiation method along with vibration signal analysis. The detecting
spectrometric and oil
oil Analysis
monitoring
Analysis techniques.
pitting
techniques. On
than
On the
the other
vibration
other hand,
or
hand,
results
radiation
results provided
method
provided aaalong
good
good understanding
with vibration
understanding of
signal
of tribological
analysis.
tribological and
The
and spectrometric oil Analysis techniques. On the other hand,
Copyright © 2018 IFAC
results provided a good understanding of tribological and 346 spectrometric oil Analysis techniques. On the other hand,
results provided
2405-8963 © 2018, IFAC a good understanding
(International Federationof oftribological
Automatic Control) and Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
Copyright 2018 responsibility
IFAC 346Control.
of International Federation of Automatic
Copyright ©© 2018
2018 IFAC
IFAC 346
346
10.1016/j.ifacol.2018.09.600
Copyright © 2018 IFAC 346
IFAC SAFEPROCESS 2018
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352 347

Loutas et al. (2011) and Eftekharnejad and Mba (2011) Additionally, distance metrics, such as the Euclidean norm,
evaluated the use of vibration and AE signals with implicitly assign a greater weight to attributes with longer
application of advanced signal processing methods for ranges. Therefore, in the attributes normalization it is sought
monitoring wear and progressive pitting in gears. The AE to equalize the ranges of all attributes, in order to achieve the
monitoring did not present significant advantages for the same effect when calculating a distance metric (Aksoy and
diagnosis in the case of the monitoring of the normal wear Haralick, 2001). In this work a uniform normalization,
and pitting of the gears. between values of 0 to 1, called min-max normalization is
used as follows:
The works cited above present certain limitations: (i)
classification is done up to four pitting faults levels; (ii), the x  xmin ( x)
feature extraction process is usually done in the frequency or x'  (1)
time-frequency domains to obtain good results in the xmax ( x)  xmin ( x)
classification stage, which implies a higher computational
cost; (iii) the experiments only vary the speed or the load; and 2.3. Feature selection by using Chi-Square
(iv) the works do not evaluate simultaneously features in
time domain for the acoustic emission, vibration and current It is a method used to test the independence of two events.
signals, for the classification of fault severity levels. That is, it can be determined if the occurrence of a specific
attribute and a given class are dependent. Therefore,
Therefore, the present work evaluates the classification
attributes that are not dependent on the class will be
accuracy based on features only in time domain for vibration,
discarded (Plackett, 1983; Vinay et al., 2015); it is given by:
AE, and current signals. The most relevant features of each
signal are selected in order to classify nine levels of severity n
of the pitting failure in a spur gearbox by using KNN; the
experiments are carried out at three different loads and
(y
j 1
j  u j )2
(2)
x2  .
speeds. uj
The document is organized as follows. Section 2 presents the
concepts used in this work. Section 3 presents the Table 1. Time domain features
experimental procedure. Section 4 presents the procedure to # Name Formula
evaluate the classification accuracy of the three signals.
N i 1 ( xi  u ) 4
N
Section 5 shows the classification results using features in the
time domain. Section 5 presents some conclusions.
T1 Kurtosis
 i 1
N
( xi  u ) 2 
2

1

N
2. BACKGROUND T2 Mean xi
i 1
N
Shannon 2
 i 1 xi log( xi )
N 2
2.1 Feature Extraction
T3 Entropy
(SNE)
Faults or anomalies detection procedures in rotating
Modified 1

N
machinery have traditionally used methods for the extraction Wi | xi |
of characteristics from vibration signals such as those Mean N i 1

presented in (Zhao et al., 2011). On the other hand, there are T4 Absolute Wi  1; if 0.25N  n  0.75N
fields, such as the biomedical engineering, in which different Value
(MMAV) Wi  0.5; otherwise
techniques have been used to extract attributes from an
electromyogram (Phinyomark et al., 2009a).
T5
Root mean
square (RMS)

1
N
N
i 1
( xi ) 2 
These "new" techniques can be used together with those used
in vibration analysis to improve the performance of the
classification stage. To do so, we have two attributes groups, T6
Standard
deviation
1
N
 N
i 1
( xi  u ) 2 
the first has 16 (T1 - T16) attributes belonging to the (STD)

Log energy N 2
"classical" methods. The second group belongs to the T7 log( xi )
i 1
biomedical field were 4 (T17 - T20) of its methods have been entropy (LEE)
used in (Phinyomark et al., 2009b; Chowdhury et al., 2013). 1 N 
T8
Log detector
exp i1 log | xi | 
(LD) N 
Hence, in this work 20 feature-extraction methods are used
(see Table 1). These features are computed using raw signals Mean of 1 N
in the time domain. T9 absolute  | xi |
N i 1
(Mean-Abs)
2.2. Feature normalization n # { I such that |x i | p}
Sure entropy
T10   min( xi , p 2 )
2
(SE)
Since each feature of a data set is calculated by different i
methods, each feature will have different ranges.

347
IFAC SAFEPROCESS 2018
348
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352

Norm entropy
 3. METHODOLOGY
N
T11 i 1
| xi | p
(NE)
The methodology used in this work is presented in Fig. 1.
Threshold 1 ; if | x i | p , Each stage is described:
T12
entropy (TE) 0 ; elsewhere
 Stage 1: acquire signals from three different sensors:
max | X i |
T13 CPT1 vibration, acoustic emission, and motor current.
MMAV

i1 log(| x(i) | 1) Stage 2: W attributes are extracted from each signal, for
N

T14 CPT2 this work W=20, these attributes are proposed in Table 1.
N log( STD  1)
 Stage 3: a vector of length twenty is formed with each

N
exp( xi )
T15 CPT3
i 1 one of the attributes for each signal.
N * exp( STD )
 Stage 4: the attributes are ranked using the Chi-Square

N
| Xi | method for each signal.
T16 CPT4 i 1

N * exp( STD )  Stage 5: a selection of the Z best attributes by signal is


made. For this work Z = 10 was chosen because for the
Wilson

N
i 1
step(| xi  xi 1 | T ) good results, but the variation of the classification
T17 amplitude
(WAMP) accuracy will be presented respecting the 20 features.
Wave Length

N
T18 i 1
| xi 1  xi |  Stage 6: the KNN was used for classification, because it
(WL) allows a better analysis with respect to the separation
Slope sign

N
T19 i 1
step([(xi  xi 1 )  ( xi  xi 1 )] between classes. Three neighbors in total were used, and
change (SSC) the selected distance metric that showed the best result
Zero crossing

N
T20 i 1
step(sin g ( xi  xi 1 )) was cosine. This evaluation is done for each sensor. For
(ZC) training and test a k-folds approach was used with k =5.
 Stage 7: accuracy general results and results for each
2.4 K-nearest neighbor classifier class are presented.
It is a non-parametric method for pattern classification that
uses a set-based distance metrics, commonly a Euclidean
distance (Cover and Hart, 1967). It is called a lazy learner
because does not learn a discriminative function from
training data, this memorizes the data. KNN find k-samples
(neighbors), from training data, that are nearest to a new
sample that we want to classify and assigns the majority class
of its neighbors (Raschka, 2016).
Given a set of n samples
   
Pn  {( x1, y1 ), ( x2 , y2 ),..., ( xn , yi )} , where xi  R d
(i  1,2,..., n) is a feature vector with dimension d , and yi
is the corresponding class label.

To predict the class of a new unknown sample, x j , it can

calculate the distance with respect to each xi thus:

d ( x j , xi )   ( x j xi ) , (3)
Fig. 1. Methodological process
where  represents any metric used for the calculation of
distance. Usually it is the Euclidean distance but also has the 4. EXPERIMENTAL SETUP AND TEST PROCEDURE
measure of similarity such as Cosine, Chi-Square,
Correlation, Minkowsky1 among others measures (Hu et al., The diagram of the test bed is displayed in Fig. 2. It
2016). comprises a motor (Siemens 1LA7 096-6YA60 2Hp) that
 generates rotating movement; an inverter (Danfoss VLT 1.5
Then the class y j belonging to the new sample x j is equal kW) to control the motor speed. The motor is coupled to the
to the most frequent class of its k -neighbors with the least input shaft to the spur gearbox, the motion is transmitted to
 the output shaft to a pulley. The pulley transmits through a
distance, min d ( x j , xi ) , about xi .
belt the motion to the magnetic brake to simulate different

348
IFAC SAFEPROCESS 2018
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352 349

load, the load is controlled by DC power supplies (TDK- Table 2. Patterns of pitting faults levels
Lambda, GEN 100-15-IS510).
Pitting
The electrical current signal that powers the motor is sensed Conditions holes Diameter percentage
by the current transducer (C1) model LEM HXS 20-NP /
L0 healthy healthy healthy
SP30, the current signal is sampled at 25 kHz by module # 1
DAQ-9205. The vibration signal is sensed by the L1 2 1.5 4.16 %
accelerometer (A1), model IPC 603C01, is placed vertically
in the gearbox case, the signal is sampled at 50 kHz by L2 2 2 7.39 %
module # 2 NI cDAQ-9234. To acquire the acoustic emission L3 4 2 14.79 %
signal, the acoustic emission sensor (AE1) model Physical
acoustic WD was used, which was mounted laterally in the L4 4 2.5 23.11 %
gearbox, the signal of this sensor is amplified by the
preamplifier model MITRAS 0/2/4C. The signal from AE1 L5 6 2.5 34.66 %
was sampled at 1 MHz through module # 3 NI cDAQ-9223. L6 6 3 49.91 %
The three modules (1,2, 3) are coupled in the NI cDAQ-9188
chassis that send the data via Ethernet to the laptop (ASUS L7 8 3 66.55 %
ROG GL752VW-DH74), which has LabVIEW™ and
Matlab™ for the acquisition of signals. The configuration is L8 10 3 83.18 %
as shown in Fig. 2, two spur gears (number of teeth Z1 =32,
and Z2 = 48 with modulus 2.25 and impact angle 20°).

L0 L1 L2

L3 L4 L5

L6 L7 L8

Fig. 3. Detail of gear tooth pitting levels

5. RESULTS AND DISCUSSION

This section presents the results of the accuracy of the


diagnosis using the KNN classifier for the features extracted
Fig. 2. Test bed diagram in the time domain from the three signals. They were ranked,
presenting both general accuracy of the nine levels of severity
Nine conditions L0 to L8 are simulated in the test bed, and accuracy for each condition or class. As it was mentioned
healthy spur gearbox L0, and eight levels of the pitting fault in the methodology section, the classification accuracy was
in the gear Z1 of L1 to L8. The faults were created artificially be presented for the 20 ranked attributes (Fig 4), but only the
by means of electrical discharge machining (EDM), faults best 10 were be analyzed (Table 3) for their good results.
were created in six teeth for each level of pitting, and the
details of the faults are for the different levels of pitting are Fig. 4 shows the general classification accuracy for the
shown in Table 2 and in Fig. 3 the details of the pitting levels twenty features ranked by Chi Square for the three signals.
for each level are presented. The experiments were performed When entering the first four ranked features of each signal to
with three loads (0, 5, 10 VCC) and three constant speeds (6, the KNN classifier, the signal of AE obtains a better
12, 16 Hz), each experiment was repeated 15 times; performance in the classification accuracy than the signal of
therefore, for each condition L, 135 samples were obtained, vibration and current, from the fifth to the ninth feature,
as there are nine conditions, a total of 1215 samples were vibration signal has higher performance in the classification
obtained. Each dataset has vibration, current and acoustic than the signals of AE and current.
emission signals; all signals were acquired for 10 seconds.

349
IFAC SAFEPROCESS 2018
350
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352

From the tenth feature, the classification accuracy remains twelfth feature. The current signal reached its maximum
above 94 % for the vibration signal, while the classification accuracy of 69.5 % with eleven features, and after it loses
accuracy for the AE signal remains above 95 % from the accuracy with the addition of more features.

Fig. 4. General classification accuracy for the twenty features

Table 3 shows the results with respect to the accuracy 6. CONCLUSIONS


classification ratio the following is observed. For each
monitored signal, a feature is able to detect the normal In this work, eight levels of severity of pitting failure and
condition L0 with 100 % accuracy; WL for vibration, RMS healthy condition were evaluated by using the acquisition of
for AE, and SSC for current. The vibration signal identifies vibration, AE, and current signals for a spur gearbox. We
the healthy condition L0, better than the signals of AE and proceeded to extract twenty features in the time domain for
current. The classification ratio of the AE signal has better each of the acquired signals, then, the ranking was done using
performance for the fault patterns L1, L4, L7, L8. For the Chi-Square for the features of each signal, that were entered
fault patterns L2, L3, L5 and L6, the AE signal has better progressively to the KNN classifier in order to obtain the
performance in the classification for the first four features, general classification accuracy and for each condition.
from the fifth feature the vibration signal has better The features of the vibration (94 %) and AE signals (95 %)
performance in the classification ratio per fault. presented a better performance in the accuracy than the
To achieve a total accuracy equal to or greater than 95 % it is current features that did not exceed 70 % in the classification
necessary to use 6 attributes using the vibration signal and 10 accuracy. This can be attributed to the fact that the current
attributes using the AE signal. The current signal does not signal needs a change in torque to reflect a change in the
allow reaching this value. signal, a change that cannot be seen in a pitting failure as if it
could occur in a broken-tooth failure.
The ratio of total accuracy is improved significantly: kurtosis
(35.8 % to 70.2 %) and SNE (70.2 % to 85 %) for vibration, ACKNOWLEDGMENT
SE (11.1 % to 67.9 %) and Mean-Abs (67.9 % to 86.1 %) for
acoustic emission; and for current we have SE (40 % to 68.1 This work was funded by the Universidad Politécnica
%). Salesiana, under grant No. 003-002-2016-03-03. The

350
IFAC SAFEPROCESS 2018
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352 351

experimental work was developed at the GIDTEC research 77, 283–293.


group lab of the Universidad Politécnica Salesiana de https://doi.org/10.1016/j.ymssp.2016.02.007
Cuenca, Ecuador. Liang, X., Zhang, H., Liu, L., Zuo, M.J., 2016. The influence
of tooth pitting on the mesh stiffness of a pair of
REFERENCES external spur gears. Mech. Mach. Theory 106, 1–15.
https://doi.org/10.1016/j.mechmachtheory.2016.08.0
Aksoy, S., Haralick, R.M., 2001. Feature normalization and
05
likelihood-based similarity measures for image
Liu, Z., Zuo, M.J., Qu, J., Xu, H., 2011. Classification of gear
retrieval. Pattern Recognit. Lett. 22, 563–582.
damage levels in planetary gearboxes, in:
Amarnath, M., Lee, S.-K., 2015. Assessment of surface
Computational Intelligence for Measurement
contact fatigue failure in a spur geared system based
Systems and Applications (CIMSA), 2011 IEEE
on the tribological and vibration parameter analysis.
International Conference On. pp. 1–5.
Measurement 76, 32–44.
Loutas, T.H., Roulias, D., Pauly, E., Kostopoulos, V., 2011.
https://doi.org/10.1016/j.measurement.2015.08.020
The combined use of vibration, acoustic emission
Aouabdi, S., Taibi, M., Bouras, S., Boutasseta, N., 2017.
and oil debris on-line monitoring towards a more
Using multi-scale entropy and principal component
effective condition monitoring of rotating
analysis to monitor gears degradation via the motor
machinery. Mech. Syst. Signal Process. 25, 1339–
current signature analysis. Mech. Syst. Signal
1352. https://doi.org/10.1016/j.ymssp.2010.11.007
Process. 90, 298–316.
Ozturk, H., Sabuncu, M., Yesilyurt, I., 2008. Early Detection
https://doi.org/10.1016/j.ymssp.2016.12.027
of Pitting Damage in Gears using Mean Frequency
Boshkoski, P., Urevc, A., 2008. Monitoring pitting gear
of Scalogram. J. Vib. Control 14, 469–484.
faults using electrical current and sound emissions,
https://doi.org/10.1177/1077546307080026
in: 9th International PhD Workshop on Systems and
Ozturk, H., Yesilyurt, I., Sabuncu, M., 2010. Detection and
Control. Izola, Slovenia.
Advancement Monitoring of Distributed Pitting
Cerrada, M., Zurita, G., Cabrera, D., Sánchez, R.-V., Artés,
Failure in Gears. J. Nondestruct. Eval. 29, 63–73.
M., Li, C., 2016. Fault diagnosis in spur gears based
https://doi.org/10.1007/s10921-010-0066-4
on genetic algorithm and random forest. Mech. Syst.
Phinyomark, A., Limsakul, C., Phukpattaranont, P., 2009a. A
Signal Process. 70, 87–103.
novel feature extraction for robust EMG pattern
Chaari, F., Baccar, W., Abbes, M.S., Haddar, M., 2008.
recognition. ArXiv Prepr. ArXiv09123973.
Effect of spalling or tooth breakage on gearmesh
Phinyomark, A., Limsakul, C., Phukpattaranont, P., 2009b. A
stiffness and dynamic response of a one-stage spur
novel feature extraction for robust EMG pattern
gear transmission. Eur. J. Mech. - ASolids 27, 691–
recognition. ArXiv Prepr. ArXiv09123973.
705.
Plackett, R.L., 1983. Karl Pearson and the chi-squared test.
https://doi.org/10.1016/j.euromechsol.2007.11.005
Int. Stat. Rev. Int. Stat. 59–72.
Chowdhury, R., Reaz, M., Ali, M., Bakar, A., Chellappan,
Raschka, S., 2016. Python machine learning: unlock deeper
K., Chang, T., 2013. Surface Electromyography
insights into machine learning with this vital guide
Signal Processing and Classification Techniques.
to cutting-edge predictive analytics, Community
Sensors 13, 12431–12466.
experience distilled. Packt Publishing open source,
https://doi.org/10.3390/s130912431
Birmingham Mumbai.
Cover, T., Hart, P., 1967. Nearest neighbor pattern
Tan, C.K., Irving, P., Mba, D., 2007. A comparative
classification. IEEE Trans. Inf. Theory 13, 21–27.
experimental study on the diagnostic and prognostic
Eftekharnejad, B., Mba, D., 2011. Monitoring Natural Pitting
capabilities of acoustics emission, vibration and
Progress on Helical Gear Mesh Using Acoustic
spectrometric oil analysis for spur gears. Mech.
Emission and Vibration. Strain 47, 299–310.
Syst. Signal Process. 21, 208–233.
https://doi.org/10.1111/j.1475-1305.2010.00769.x
Vinay, V., Kumar, G.V., Kumar, K.P., 2015. Application of
He, Q., Ren, X., Jiang, G., Xie, P., 2014. A hybrid feature
chi square feature ranking technique and random
extraction methodology for gear pitting fault
forest classifier for fault classification of bearing
detection using motor stator current signal. Insight-
faults, in: Proceedings of the 22th International
Non-Destr. Test. Cond. Monit. 56, 326–333.
Congress on Sound and Vibration, Florence, Italy.
Hu, L.-Y., Huang, M.-W., Ke, S.-W., Tsai, C.-F., 2016. The
pp. 12–16.
distance function effect on k-nearest neighbor
Zhao, X., Zuo, M.J., Liu, Z., 2011. Diagnosis of pitting
classification for medical datasets. SpringerPlus 5,
damage levels of planet gears based on ordinal
1304. https://doi.org/10.1186/s40064-016-2941-7
ranking, in: Prognostics and Health Management
Korka, Z., Bara, A., Clavac, B., Filip, L., 2017. Gear Pitting
(PHM), 2011 IEEE Conference On. IEEE, pp. 1–8.
Assessment Using Vibration Signal Analysis.
Zhao, X., Zuo, M.J., Liu, Z., Hoseini, M.R., 2013. Diagnosis
Romanian J. Acoust. Vib. XIV, 44–49.
of artificially created surface damage levels of
Li, C., Sanchez, R.-V., Zurita, G., Cerrada, M., Cabrera, D.,
planet gear teeth using ordinal ranking.
Vásquez, R.E., 2016. Gearbox fault diagnosis based
Measurement 46, 132–144.
on deep random forest fusion of acoustic and
https://doi.org/10.1016/j.measurement.2012.05.031
vibratory signals. Mech. Syst. Signal Process. 76–

351
IFAC SAFEPROCESS 2018
352
Warsaw, Poland, August 29-31, 2018 René-Vinicio Sánchez et al. / IFAC PapersOnLine 51-24 (2018) 346–352

Signal Rank Name of Accuracy ratio for patter fault Ratio of


features total
accuracy

L0 L1 L2 L3 L4 L5 L6 L7 L8 Total

1 WL 100 0 0 0 0 0 0 0 0 11.1
2 Mean 89.6 48.1 21.5 24.4 32.2 28.1 37 30.4 10.5 35.8
3 Kurtosis 100 74.8 57.8 63 63.4 57.8 74.1 64.4 76.8 70.2
4 SNE 100 86.7 82.2 76.3 78.2 71.1 89.6 91.1 89.5 85
Vibration

5 WAMP 100 90.4 88.9 80.7 85.8 81.5 97 97.8 91.1 90.3
6 SE 100 93.3 99.3 86.7 91.1 94.8 100 98.5 94.1 95.3
7 LEE 100 92.6 99.3 88.9 91.8 94.1 100 97 92.5 95.1
8 LD 100 92.6 99.3 87.4 91.1 94.1 100 97 92.5 94.9
9 TE 100 92.6 99.3 86.7 91.1 93.3 100 97 93.3 94.8
10 NE 100 93.3 99.3 87.4 91.1 94.1 100 97 92.5 95
1 RMS 100 0 0 0 0 0 0 0 0 11.1
2 SE 54.8 74.1 63 60.7 66.7 60.7 78.5 88.9 63.7 67.9
3 Mean-Abs
65.2 94.8 83 77.8 97 73.3 91.1 94.8 97.8 86.1
4 SE
Emission

659 96.3 815 80 97 74.1 91.1 95.6 97.8 86.6


Acoustic

5 LEE 62.2 94.1 80 73.3 97 75.6 88.9 97.8 97 85.1


6 WAMP 71.9 94.8 83 77.8 97 77.8 90.4 95.6 99.3 87.5
7 NE 71.9 94.8 83.7 78.5 97 78.5 92.6 97 99.3 88.1
8 CPT4 75.6 94.8 83.7 79.3 97 78.5 92.6 97 100 88.7
9 Mean 85.2 96.3 85.9 86.7 97.8 83.7 94.1 96.3 100 91.8
10 CPT2 98.5 99.3 96.3 96.3 97.8 93.3 100 95.6 100 97.4
1 SSC 100 0 0 0 0 0 0 0 0 11.1
2 Mean 43.7 31.1 35.6 24.4 29.6 17 14.1 18.5 22.2 26.3
3 Mean-Abs
43.7 31.1 35.6 24.4 29.6 17 14.1 18.5 22.2 26.3
4 NE 43.7 30.4 41.5 31.1 34.1 19.3 23 20 26.7 30
Current
Motor

5 CPT3 43.7 31.9 42.2 31.9 32.6 20.7 23 20 25.9 30.2


6 RMS 45.9 35.6 43.7 34.8 35.6 22.2 21.5 24.4 32.6 32.9
7 LEE 48.1 35.6 44.4 36.3 36.3 21.5 25.2 25.2 34.8 34.2
8 WL 48.9 37.8 44.4 36.3 36.3 22.2 26.7 24.4 37 34.9
9 CPT1 42.2 47.4 54.8 38.5 41.5 28.9 18.5 31.1 57 40
10 SE 71.1 75.6 76.3 64.4 69.6 58.5 51.1 69.6 77 68.1
Table 3. Accuracy results by sensor, (%)

352

Вам также может понравиться