Int
ern
at
i
onal
Journ
al of
P
ower E
le
ctr
on
i
cs a
n
d
Drive
S
ystem
s
(
IJ
PEDS
)
Vo
l.
1
2
,
No.
2
,
Jun
202
1
,
pp.
120
5
~
121
5
IS
S
N:
20
88
-
8694
,
DOI: 10
.11
591/
ij
peds
.
v
1
2
.i
2
.
pp
120
5
-
121
5
1205
Journ
al h
om
e
page
:
http:
//
ij
pe
ds
.i
aescore.c
om
Machin
e le
arning b
ased multi
class
fault di
agnosis
to
ol for
vo
lt
age s
ource in
verter d
rive
n in
ductio
n motor
Jyothi R
, Te
jas H
olla, Um
a Rao
K, J
ayapal R
Depa
rtment
o
f
E
le
c
tri
c
al a
nd
Ele
ct
roni
cs
Engi
n
eering,
RV
Coll
eg
e
of Engin
ee
ring
,
Beng
al
uru
,
india
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
1
9
, 20
20
Re
vised
M
a
r
9
,
20
21
Accepte
d
Apr
23
, 20
21
AC
drive
s
ar
e
em
p
loye
d
in
proc
ess
industries
for
va
rying
applications
result
ing
in
a
wi
de
ran
g
e
of
r
ati
ngs.
The
entire
proc
ess
industry
has
see
n
a
par
adi
g
m
shif
t
from
m
anual
to
au
tomate
d
sys
te
ms.
The
m
aj
or
fa
ct
or
cont
ributing
to
t
his
is
the
adv
an
ce
d
power
el
e
ctronics
technolog
y
ena
bl
ing
po
wer
e
lectr
oni
c
dr
ive
s
for
sm
ooth
cont
rol
of
e
lectr
i
c
mot
ors
.
Indu
ction
mot
ors
are
most
co
mm
only
used
in
industr
ie
s.
Fa
ult
s
in
the
powe
r
e
lectr
oni
c
ci
rcu
it
s
m
ay o
ccur
per
iod
ic
a
ll
y
.
The
se
f
aul
ts
of
ten
go
unnoticed
a
s
the
y
r
are
ly
ca
use
a
com
pl
ete
shutdown
and
the
f
aul
t
l
eve
ls
ma
y
no
t
b
e
l
arg
e
enough
to
le
ad
to
a b
rea
kd
own
of th
e driv
e.
An
e
arl
y
d
etec
t
i
on
of th
ese
f
ault
s
is
req
uir
ed
to
pre
ven
t
th
ei
r
esc
alati
on
int
o
ma
jor
f
aul
ts
.
Th
e
dia
gnost
ic
too
l
for
detec
t
ion
of
fau
lt
s
r
equi
r
es
re
al
time
mon
it
o
ring
of
th
e
en
tire
drive
.
In
thi
s
wor
k,
d
et
a
iled
inve
stigation
of
diffe
ren
t
f
aul
ts
t
hat
ca
n
oc
cur
in
the
pow
er
el
e
ct
r
onic
ci
r
cui
t
of
an
industr
ia
l
driv
e
is
c
arr
i
ed
out
.
Ana
lysis
a
nd
im
p
act
of
f
a
ult
s
on
th
e
per
forma
n
ce
of
the
induc
t
ion
m
otor
is
pre
sent
e
d.
A
r
eal
ti
m
e
mon
it
oring
pla
tfor
m
is
prop
osed
to
detec
t
an
d
cl
assify
th
e
fa
ult
a
cc
ura
tely
using
machine
le
arn
ing.
A
d
iagnos
ti
c
tool
al
so
is
dev
el
oped
to
d
ispla
y
th
e
s
eve
ri
ty
an
d
loc
a
ti
on
of
th
e
f
a
ult
to
th
e
op
era
to
r
to ta
k
e
co
rre
c
tive
m
ea
sures.
Ke
yw
or
d
s
:
Diag
nosti
c too
l
F
ault cl
assifi
ca
ti
on
Ind
uction m
otor
M
ac
hin
e lea
rn
i
ng
M
ulti
cl
ass S
V
M
Re
li
abili
ty cen
te
red
Vo
lt
age
s
ource
inv
e
rter
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
BY
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Jy
ot
hi R
Dep
a
rtme
nt of
Ele
ct
rical
an
d
Ele
ct
ro
nics
E
nginee
rin
g
RV Colle
ge
of
En
gin
eeri
ng
RV V
i
dyani
kethan
P
os
t,
Mys
uru
R
oad, Be
ngal
uru
-
56
0059,
Ind
ia
Emai
l:
jyo
thi
r
@rvce.e
du.in
1.
INTROD
U
CTION
Ind
uction
mo
t
or
s
a
re
r
ugge
d
in
co
ns
tr
uctio
n
an
d
a
re
us
e
d
in
dr
i
ving
la
the
mac
hin
es
,
cru
s
he
rs
,
oil
extracti
ng
mil
ls,
te
xtil
e
indu
strie
s,
et
c.
F
or
the
sm
ooth
c
on
t
ro
l
of
t
he
i
nductio
n
m
otors,
powe
r
el
ec
tro
nic
conve
rters
a
re
employe
d.
I
n
proces
s
in
dustri
es
an
d
man
ufa
ct
ur
in
g
s
ect
ors,
the
monit
ori
ng
of
po
wer
el
ect
ronic
sy
ste
ms
is
ess
entia
l
as
it
aff
e
ct
s
the
s
ys
te
m'
s
pe
rformance
and
ef
fici
enc
y.
The
fa
ults
in
t
he
s
ys
te
m
a
re
to
be
identifie
d
on
oc
currence
t
o
a
vo
i
d
c
omplet
e
br
ea
kdow
n
of
the
pro
du
ct
io
n
pro
cess.
M
os
t
of
t
he
orga
niz
at
ion
s
now
a
da
ys
fo
l
low
pr
e
ven
ti
ve
mainte
na
nce
by
sc
he
du
li
ng
per
i
od
ic
al
mai
nte
na
nce
be
f
ore
fail
ur
e
occur
s
and
causes
da
mage
to
the
entire
s
ys
te
m
un
der
c
on
si
der
at
io
n.
I
f
t
he
pr
e
ve
ntiv
e
mainte
na
nce
is
sche
dule
d
e
arly,
it
migh
t
le
ad
t
o
wastage
of
us
e
fu
l
li
fe
of
mac
hin
e.
A
n
al
te
rnat
ive
appr
oach
is
reli
abili
ty
centere
d
mainte
nan
ce
.
r
el
ia
bili
ty
c
ent
ered
m
ai
nte
na
nce
is
c
ho
se
n
t
o
op
ti
mize
the
mainte
na
nce
pro
gr
am
by
m
on
it
ori
ng
th
e
e
ssentia
l
par
a
mete
rs a
nd
rec
ognizin
g t
he
fa
ults
that
af
f
ect
s
the
f
unct
io
n
of
an
AC
dri
ve s
ys
te
m.
D
ue
to
heavy
dow
nt
ime
costs
of
dev
ic
e
s
in
m
od
e
r
n
po
wer
i
ndus
trie
s,
there
is
a
need
for
real
ti
me
m
on
it
ori
ng
s
ys
te
m
to
detect
incipie
nt
fau
lt
s
befor
e
th
ey
escal
at
e a
nd casue
a
br
ea
kdown.
M
uc
h
of
t
he
re
search
pr
ese
nt
ed
i
n
the
li
te
ratur
e
is
fo
c
us
e
d
on
the
det
ect
io
n
of
fa
ults
in
t
he
i
nductio
n
mo
to
r
s
uc
h
as
r
otor
ba
r
fau
lt
,
e
ccentric
it
y
fa
ul
t.
Conve
ntio
na
l
appro
ac
hes
li
ke
m
otor
c
urre
nt
sig
natu
re
a
na
lysis
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
Dr
i
S
ys
t
,
V
ol
.
12
, N
o.
2
,
J
une
202
1
:
120
5
–
121
5
1206
(M
CS
A
)
te
ch
ni
qu
es
wer
e
e
m
ployed
i
n
earli
er
da
ys
f
or
fa
ul
t
diagnosis
in
t
hr
ee
phase
i
nductio
n
m
otors
.
Fault
fe
at
ur
e
was
e
xtracted
th
rou
gh
fast
fou
rier
tra
nsfo
rm
(F
FT
)
f
r
om
sam
ple
d
sta
tor
cu
rrents
to
t
rain
the
m
odel
us
in
g
mu
lt
il
ayer
perce
ptr
on,
s
uppo
rt
vecto
r
mac
hi
ne
(SV
M
)
[
1
]
,
[
7]
,
[
8
].
Rul
e
base
d
ap
pro
ach
li
ke
fu
zz
y
log
ic
te
chn
iq
ue
was
employe
d
f
or
f
ault
detect
io
n
in
the
I
nductio
n
mo
t
or
s
[
2
]
-
[6
]
.
Discrete
wav
e
le
t
transform
(
DWT
)
[8],
[
9]
an
d
pri
ncipal
c
omp
onent
a
nalysis
(PC
A)
we
re
util
iz
ed
t
o
ide
ntif
y
t
he
disco
ntin
uity
i
n
the
feat
ur
e
cause
d
by
the
fa
ults.
R
el
evan
ce
vect
or
mac
hin
e
(
R
V
M
)
wit
h
op
ti
mi
zat
ion
te
c
hn
i
ques
li
ke
e
voluti
onar
y
pa
rtic
le
s
w
ar
m
op
ti
miza
ti
on,
c
ucko
o
sea
rch
opti
miza
ti
on
wa
s
em
ployed
to
impro
ve
the
fa
ult
detect
io
n
as
the
DWT
in
volves
huge
data
an
d
com
pu
ta
ti
onal
ti
me
[
7],
[
13
].
Cl
ark
e
tra
ns
f
ormed
tw
o
dime
ns
io
nal
featu
re
s
al
ong
with
s
urface
error
wer
e
us
e
d
to
detect
the
f
ault.
A
ver
a
ge
c
urren
t
of
t
he
th
ree
sta
tor
c
urre
nts
al
so
was
use
d
to
detect
the
fau
lt
s
in the system
[11
]
-
[
15].
Detai
le
d
diag
nosti
c
to
ol
for
e
xh
a
us
ti
ve
fa
ults
in
the
powe
r
el
ect
ro
nics
dri
ve
of
a
n
in
duct
ion
m
oto
r
is
no
t
f
ound
in
t
he
li
te
ratu
re.
I
n
this
w
ork
,
th
e
va
rio
us
fa
ults
that
a
re
li
kel
y
to
occur
i
n
t
he
powe
r
el
ect
ronics
ci
rcu
it
s
hav
e
be
en
in
vestigat
e
d
to
detect
an
d
cl
assify
the
fa
ults
accu
ratel
y
and
ide
ntify
th
e
co
rr
ect
l
ocati
on
o
f
the
fa
ults.
The
numb
e
r
of
featur
es
has
bee
n
opti
miz
ed
to
im
pro
ve
the
e
ff
ic
ie
ncy
of
the
diagnostic
to
ol
an
d
al
s
o
enab
le
it
s
a
da
ptati
on
f
or
a
real
ti
me
mon
it
or
in
g.
Tw
o
machine
le
ar
ni
ng
al
gorith
ms
na
mely
m
ulti
la
yer
per
ce
ptr
on
ne
ural
netw
ork
a
nd
s
upport
vect
or
machi
ne
ha
ve
bee
n
fou
nd
to
giv
e
a
ccu
rate
resu
lt
s.
100
per
ce
nt
accurac
y
has
be
en
ac
hieve
d
f
or
detect
ion,
cl
assifi
cat
ion
a
nd
locat
io
n
of
t
he
fau
lt
.
Simi
la
r
app
li
ct
ai
ons
di
scusse
d
in
li
te
ratur
e
use
la
bv
ie
w
a
nd
intric
at
e
hardware
set
up.
T
he
novelty
of
this
pr
ese
nt
w
ork
is
the
us
e
of
I
oT
an
d
minimal
sen
sor
s
f
or
re
mo
te
m
on
it
ori
ng
of
t
he
dev
ic
es
.
Als
o,
a
us
e
r
-
fr
ie
nd
l
y
G
U
I
has
bee
n
de
velo
pe
d
to
disp
la
y
the
res
ults
as
a
n
ope
rati
ng
to
ol.
Re
li
abili
ty
centere
d
mai
ntenan
ce
w
ould
go
l
ong
wa
y
w
it
h
su
c
h
a
diag
no
sti
c
too
l
reducin
g
l
os
s
of pr
oduction t
ime.
The
pap
e
r
is
div
ide
d
int
o
8
Se
ct
ion
s.
Sect
ion
1
f
oc
us
es
on
the
di
ff
e
ren
t
ma
intenanc
e
strat
egy
an
d
the
conve
ntion
al
methods
a
vaila
ble
for
fa
ult
diagnosis.
In
Sec
ti
on
2,
t
he
st
ruct
ur
e
of
the
dri
ve
a
nd
va
rio
us
fa
ults
are
pr
e
sente
d.
Fault
detect
ion
an
d
cl
assifi
cat
ion
meth
odol
ogy
us
ed
is
pres
ented
in
sect
io
n
3.
Sect
io
n
4
pr
ese
nts
the
sim
ulati
on
model
a
nd
t
he
featur
e
ac
quisi
ti
on
.
Ha
r
dw
a
re
set
up
a
nd
it
s
sp
eci
ficat
io
ns
are
mentio
ne
d
in
th
e
Sect
ion
5.
Re
s
ults
a
nd
discussi
on
ar
e
pr
e
sen
te
d
i
n
Sect
ion
6.
Dia
gnos
ti
c
t
oo
l
wit
h
case
s
tu
dies
is
pr
e
se
nted
in
Sect
ion
7
a
nd c
on
cl
us
io
n
is
pr
esented
in Sec
ti
on
8.
2.
FAU
LT
S
I
N AN A
C D
RIVE SYSTE
M
Figure
1
re
pres
ents
the
st
ru
ct
ure
of
vo
lt
a
ge
s
ource
in
ve
rter
dr
i
ven
i
nductio
n
m
otor.
It
co
nsi
sts
of
t
he
three
p
hase
A
C
supp
l
y
,
co
nt
ro
ll
ed
or
unco
ntr
olle
d
recti
fier,
DC
li
nk
a
nd
a
n
i
nv
e
rter
c
apab
le
of
gen
e
rati
ng
var
ia
ble
volt
age
an
d
var
ia
bl
e
fr
eq
ue
ncy
with
P
WM
pulse
s.
diff
e
re
nt
typ
es
of
P
W
M
te
ch
niques
and
t
he
modu
la
ti
on
i
ndex
will
ha
ve
impact
on
in
ve
rter
li
ne
ou
t
pu
t
volt
ages
an
d
sta
tor
c
urre
nts
[18],
[19
].
The
three
ph
a
se
s
uppl
y
i
s
co
nnect
ed
to
a
diode
bri
dg
e
recti
fier
to
pro
duce
DC
volt
age
at
t
he
fro
nt
e
nd
.
T
he
re
ct
ifie
d
vo
lt
age
is
fe
d
t
hro
ugh
DC
li
nk
to
t
he
po
wer
inv
e
rter
ci
rc
uit
consi
sti
ng
o
f
s
ensiti
ve
po
wer
elec
tron
ic
s
witc
hing
dev
ic
es
su
c
h
a
s
I
GBTs
or
MOSFE
Ts
wh
ic
h
a
re
dr
iv
en
by
gate
dri
ve
ci
rcu
it
s
to
pr
ov
i
de
var
ia
ble
vol
ta
ge
an
d
fr
e
qu
e
nc
y
to
t
he
dy
namic
lo
ads
f
or
sp
ee
d
c
on
t
ro
l.
T
he
power
el
ect
r
on
ic
ci
rcu
it
ry
em
plo
ye
d
in
AC
dr
i
ves
are
li
able
to
numer
ou
s
t
yp
e
s
of
el
ect
rical
fau
l
ts
su
ch
as
fau
lt
s
in
the
recti
fier,
in
ver
te
r
,
fa
ults
at
the
mo
to
r
te
r
minal
s
and fa
ults in t
he
mech
a
nical
l
oad as
detai
le
d.
Figure
1
.
Str
uc
ture of
volt
age
so
urce i
nv
e
rter
drive
n
in
duct
ion m
otor
2.1.
Fau
lt
s in
rec
tifier
Bridg
e
recti
fier
s
are
us
ed
in
m
os
t
of
the
in
dus
tria
l
dr
i
ves.
Fa
ults
in
a
n
unco
ntr
olled
br
i
dge
r
ect
ifie
r
ca
n
be
cat
eg
or
iz
e
d
as
op
e
n
diode
fau
lt
an
d
di
ode
sh
ort
ci
rcu
it
f
ault.
I
n
case
o
f
op
e
n
diode
fa
ult
the
mag
nitud
e
of
the
DC
outp
ut
vo
lt
age
of
the
recti
fie
r
re
duce
s
thu
s
a
ff
ect
in
g
pe
rforma
nce
of
the
i
nv
e
rter
ci
rcu
it
.
I
f
di
odes
of
the
sa
me
le
g
a
re
s
horte
d,
t
he
current
th
rou
gh
t
he
ci
rcu
it
s
hoots
up
s
wiftly
an
d
pote
ntial
diff
e
re
nce
ac
ross
the
diodes
reduces
to
zer
o.
D
ue
t
o
s
hort circ
uit of
DC Li
nk ca
pacit
or
s
at
the
recti
fier fr
on
t e
nd, f
a
ults ca
n o
ccur
i
n
the drive
syst
em [2
0]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N:
20
88
-
8
694
Mac
hin
e le
ar
ni
ng base
d m
ulti
class fault
d
i
agno
sis
to
ol for
v
oltag
e
s
ou
rce
inverte
r
…
(
Jyo
thi R
)
1207
2.2.
Fau
lt
s
in in
ve
rter
I
n
ge
ne
r
a
l
,
v
a
r
i
a
bl
e
vo
l
t
a
ge
va
r
i
a
bl
e
f
r
e
qu
e
nc
y
i
nv
e
r
t
e
r
s
a
r
e
pr
e
f
e
r
r
e
d
ov
e
r
c
ur
r
e
nt
s
ou
r
c
e
i
nv
e
r
t
e
r
s
,
f
o
r
dr
i
ve
s
.
I
n
ve
r
t
e
r
s
a
r
e
a
l
s
o
s
us
c
e
pt
i
bl
e
t
o
op
e
n
c
i
r
c
ui
t
a
nd
s
h
or
t
c
i
r
c
ui
t
I
G
B
T
f
a
ul
t
s
.
P
r
ot
e
c
t
i
on
s
c
he
m
e
s
w
i
l
l
be
e
m
pl
oy
e
d
f
or
s
ho
r
t
c
i
r
c
ui
t
f
a
u
l
t
s
.
I
n
c
a
s
e
of
o
pe
n
I
G
B
T
f
a
ul
t
,
t
he
m
ot
o
r
w
i
l
l
s
t
i
ll
be
i
n
r
u
nn
i
ng
c
o
nd
i
t
i
on
b
ut
a
t
r
e
du
c
e
d
s
pe
e
d
[20
]
-
[
22]
.
T
h
e
ha
r
m
o
ni
c
s
c
om
p
on
e
nt
i
n
th
e
s
t
a
t
or
c
ur
r
e
nt
of
c
or
r
e
s
po
nd
i
n
g
ph
a
s
e
i
n
c
r
e
a
s
e
s
r
e
s
ul
t
i
ng
i
n
a
l
ow
e
r
e
f
f
i
c
i
e
nc
y
du
e
t
o
P
Q
di
s
t
ur
ba
nc
e
.
S
h
o
r
t
c
i
r
c
ui
t
f
a
ul
t
of
I
G
B
T
r
e
s
ul
t
s
i
n
l
a
r
ge
a
m
o
un
t
o
f
c
ur
r
e
nt
f
l
o
w
t
h
r
ou
gh
t
he
i
nv
e
r
t
e
r
c
i
r
c
ui
t
.
2.3.
Fau
lt
s
at
mo
t
or termi
na
l
The
th
ree
-
phas
e
var
ia
ble
fr
e
quenc
y,
var
ia
bl
e
ou
t
pu
t
volt
a
ge
f
rom
the
V
SI
is
fe
d
to
th
e
three
-
phase
mo
to
r
in
put
te
rmin
al
.
Faults
li
ke
SLG
fau
lt
,
L
-
L
fau
lt
,
li
ne
s
op
e
n
fau
lt
at
the
AC
po
wer
li
ne
c
ou
l
d
cause
dis
tur
ba
nces
at
the
m
otor
te
r
minal
.
I
n
case
of
s
hort
ci
rc
uit
of
t
he
li
nes,
th
e
inducti
on
m
ot
or
will
stop
runn
i
ng
and
ve
r
y
high
current
flo
w
s
t
hro
ugh t
he s
hort
-
ci
rc
uited
te
r
minals
a
nd
the
rev
e
rse c
urre
nt
causes
da
ma
ge
to
t
he
diodes a
nd IG
BTs.
Faults
ca
n
al
s
o
occ
ur
in
the
in
du
ct
io
n
m
oto
r
du
e
to
da
mage
i
n
sta
to
r
winding
s
,
bea
rin
gs
,
s
qu
i
rr
el
ca
ge
ro
t
or
brok
e
n
ba
rs,
ins
ulati
on
fail
ur
e
[
21
]
-
[
23]
.
The
inci
pie
nt
fau
lt
s
do
no
t
ca
us
e
the
m
ot
or
to
st
op
unli
ke
s
hort
ci
rcu
it
fa
ults.
The
in
duct
io
n
mo
to
rs
operate
eve
n
on
t
he
oc
currence
of
in
ci
pient
fa
ults
.
T
he
reli
abili
ty
of
t
he
entire
syst
em
te
nd
t
o
re
duce
and
dama
ge
t
he
machin
es
ov
er
the
lo
ng
r
un
.
He
nce
,
the
y
a
re
da
nger
ous
a
s
they
may be
undete
ct
ed
f
or a lo
ng
ti
me, sile
ntly c
ausin
g dama
ge
to
the
m
otor
.
Fault
diag
nosis
plays
a
cr
uc
ia
l
ro
le
in
in
du
st
rial
proces
ses
wh
ic
h
in
volves
fa
ult
de
te
ct
ion
a
nd
cl
assifi
cat
ion
(FDC)
b
ase
d
on
locat
io
n
an
d
it
s
se
ver
it
y. The
o
pe
rato
r
al
so
ne
eds
to
be
in
f
orme
d
a
bout
the
fa
ult
thr
ough
an
inte
rf
ace
to
ta
ke
pr
ecauti
onar
y
m
easur
e
s
or
to
ta
ke
imme
diate
act
ion
to
a
vo
i
d
the
co
ns
e
qu
e
nc
es
of
fail
ur
e a
nd
da
mage
to t
he
m
achine. T
he
e
nt
ire p
r
ocess
is t
erme
d
as
reli
ab
il
it
y
centered mai
ntena
nce
.
Amongst
numer
ous
ap
pr
oach
e
s
f
ollo
we
d
f
or
t
he
reli
a
bili
ty
centere
d
mainte
na
nce
,
machine
le
arn
i
ng
a
ppears
to
be
an
encou
rag
i
ng
to
ol.
In
t
he
propose
d
wor
k,
tw
o
mac
hin
e
le
ar
ning
al
gorithms
na
me
ly
m
ulti
-
la
yer
pe
rcep
tr
on
-
ne
ural
netw
ork
(MLP
-
N
N)
a
nd
s
upport
vector
machi
ne
(
S
VM)
a
re
em
pl
oy
e
d
[22
]
.
Data
acq
uisit
ion
i
s
pe
rformed
t
hro
ugh
m
ulti
ple
sens
ors
to
m
onit
or
t
he
co
ndit
ion
of
the
p
ower
el
ect
ronic
sy
ste
m.
O
ft
en
t
he
da
ta
colle
ct
ed
thr
ough
phys
ic
al
se
nso
rs
may
no
t
yiel
d
good
featu
re
extracti
on.
I
n
s
uch
cases
,
a
ddit
ion
al
featur
e
s
are
ge
ner
at
ed
us
in
g
the meas
ur
e
d d
at
a to obtai
n t
he
h
id
de
n feat
ur
es. T
his
data is t
erme
d
as
synt
hesized
d
at
a.
In
t
he
propose
d
w
ork
,
AC
dri
ve
is
sim
ulate
d
us
in
g
MATL
AB
/
Simuli
nk.
Var
i
ou
s
ty
pes
of
open
ci
r
cui
t
fau
lt
s
s
uc
h
as
diode
op
e
n
fa
ul
t
in
the
recti
fier,
open
I
GBT
fa
ult
in
the
in
ver
te
r
are
creat
ed
a
nd
co
rr
e
spondin
g
el
ect
rical
par
a
mete
rs
li
ke
sta
tor
c
urre
nts,
i
nverter
li
ne
outpu
t
volt
ages
a
nd
their
TH
Ds
an
d
the
mec
ha
nical
par
a
mete
rs
li
ke
s
peed
a
nd
t
orq
ue
of
the
m
otor
are
obta
ined
t
o
de
te
rmine
t
he
eff
ect
of
the
el
e
ct
rical
fa
ults.
F
ault's
detect
ion,
cl
ass
ific
at
ion
a
n
d
point
of
occ
urre
nce
of
fa
ult
are
identifie
d
usi
ng
mac
hin
e
le
ar
ning
a
nd
dia
gnos
ti
c
too
l
is
de
velo
pe
d
us
in
g
pyth
on
to
dis
play
th
e
inf
ormat
io
n
about
t
he
se
ve
rity
of
fa
ult
to
the
operato
r
t
o
ta
ke
correct
ive
mea
su
res
.
3.
FAU
LT
DET
ECTION
AN
D CLAS
SIFI
CA
TI
ON S
YST
EM
The
blo
c
k
diag
ram
of
fa
ult
d
et
ect
ion
a
nd
cl
as
sific
at
ion
(
FDC
)
sy
ste
m
is
as
sh
ow
n
in
Fig
ure
2.
M
ac
hine
le
arn
in
g
base
d
FD
C
s
ys
te
m
in
vo
l
ves
the
ste
ps
.
The
data
c
ollec
ti
on
f
rom
the
sy
ste
m,
f
eat
ur
e
eng
i
neer
i
ng
wh
ic
h
involves
t
he i
de
ntific
at
ion
of
par
a
mete
rs
that
are
ef
fici
ent
in
the
detec
ti
on
of
fa
ults,
data
tr
ai
nin
g
an
d c
hoos
in
g
a
su
it
able
m
odel
,
trai
ning
an
d
evaluati
ng
the
model
an
d
pa
ra
mete
r
tu
ning
f
or
eff
ic
ie
nt
pr
e
di
ct
ion
an
d
dia
gnos
is
of
fa
ults
in
the
sy
ste
m
.
T
he
r
aw
data
colle
c
te
d
f
rom
the
s
ys
te
m
is
no
t
oft
en
s
uitable
to
trai
n
the
m
o
de
l
and
evaluate
t
he
pa
rameters
f
or
cl
assifi
cat
ion
of
fau
lt
s.
He
nce
,
featur
e
e
ng
i
ne
erin
g
plays
a
s
ign
ific
a
nt
r
ole
in
the
f
ault
detect
io
n
and
cl
assifi
cat
ion.
F
or
a
real
ti
me,
m
onit
or
i
ng
too
l,
t
he
feat
ures
ha
ve
to
be
opti
mize
d
to
im
pro
ve
com
pu
ta
ti
onal
powe
r,
mem
ory
us
a
ge
an
d
re
su
lt
in
s
horter
trai
ning
ti
me
with
minimal
chan
ce
of
ov
e
r
fitt
ing
.
Cl
assifi
cat
ion
of f
a
ults f
rom t
he
sel
ect
ed
f
ea
tures
is
kn
own as su
pe
rv
ise
d
l
earn
i
ng appr
oa
ch.
The
di
ff
e
ren
t
machine
le
a
rn
i
ng
al
go
rithms
app
li
cable
f
or
cl
assifi
cat
ion
of
t
he
fa
ults
in
t
he
sy
ste
m
a
re
li
near
classi
fier
s li
ke
l
ogist
ic
r
egr
es
sio
n,
deci
sion
t
ree,
s
upport
v
ect
or
m
ac
hin
e a
nd
neura
l
n
et
w
ork.
Mul
ti
cl
ass
s
upport
v
ect
or
m
achine
al
gorithm
has
been
extensi
ve
ly
use
d
f
or
cl
assifi
c
at
ion
a
nd
outl
ie
r
detect
ion
in
ma
ny
app
li
cat
io
ns
.
M
ulti
la
yer
perce
ptr
on
ne
ural
netw
ork
(
M
LP
-
N
N)
al
gorith
m
co
ns
ist
s
of
s
et
of
softwa
re
neur
on
s
arr
a
ng
e
d
i
n
la
ye
rs
with
an
or
ga
nized
c
onnect
ion
for
c
ommu
nicat
ion
.
M
LP
-
NN
us
es
gra
dient
desce
nt
al
go
rithm
for
le
a
rn
i
ng
t
hro
ugh
loss
f
un
ct
io
n.
In
th
e
c
urre
nt
rese
arch
work,
MLP
-
NN
a
nd
S
VM
a
re
prefe
rr
e
d
f
or
qu
a
ntit
at
ive m
ulti
class el
ect
rical
f
ault detec
ti
on
a
nd classi
f
ic
at
ion
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
Dr
i
S
ys
t
,
V
ol
.
12
, N
o.
2
,
J
une
202
1
:
120
5
–
121
5
1208
Figure
2
.
Fa
ult
detect
io
n
a
nd
cl
assifi
cat
ion
s
ys
te
m
4.
SIMULATI
O
N MO
DEL
Fau
lt
detect
io
n
and
cl
assi
ficat
ion
sy
ste
m
de
ve
lop
e
d
f
or
volt
age
s
ource
in
vert
er
dr
i
ven
in
duct
ion
m
otor
(V
S
I
-
I
M
)
are
di
scusse
d
in
this
secti
on. T
he
si
mu
la
ti
on is
car
ried
out i
n M
A
TLAB
/
S
im
ulin
k plat
f
or
m
, t
he
d
at
a
extracte
d
is
tra
ined
a
nd
te
ste
d
in
google
c
o
la
b
e
nv
i
ronme
nt
for
bo
t
h
th
e
m
achine
le
a
rn
i
ng
al
go
rithms
na
mel
y
M
LP
-
N
N
a
nd
SVM.
T
w
o
dif
fer
e
nt
ind
ic
at
ors
are
disc
us
se
d
for
fa
ult
detect
ion
a
nd
cl
assifi
cat
ion
in
the
V
SI
-
IM
sy
ste
m.
one
us
i
ng
insta
ntane
ous
valu
es
of
th
e
pa
rameter
s
a
nd
a
no
t
her
us
ing
R
M
S
Val
ue
s.
T
he
S
i
m
ulin
k
model
us
e
d
f
or e
xtrac
ti
ng
the
d
at
a
fr
om
volt
age s
ou
rce
dr
ive
n
i
nduc
ti
on
m
otor is
as sho
wn in Fi
gure
3.
Figure
3
.
Sim
ul
ink
m
odel
of
VS
I
drive
n I
nduction m
otor
The
se
ver
it
y
of
ope
n
ci
rc
uit
in
ci
pient
fau
lt
s
is
a
majo
r
co
nc
ern,
Var
i
ous
open
ci
rc
uit
fau
l
ts
are
diod
e
op
e
n
ci
rcu
it
,
I
GBT
open
ci
rc
uit,
cr
os
s
di
od
e
an
d
c
r
os
s
IGB
T
ope
n
ci
rc
ui
t
fau
lt
s.
These
are
di
vid
e
d
in
to
24
cl
asses
.
Feat
ur
es
are
us
e
d
to
di
sti
ng
uis
h
bet
w
een
the
var
io
us
fau
lt
s.
Feat
ur
e
s
su
c
h
as
,
i)
rec
ti
fier
o
ut
pu
t
volt
age
(Vdc)
,
ii
)
i
nver
te
r
li
ne
outp
ut
vo
lt
age
s
(
Vab,
Vb
c
,
Vca
)
,
ii
i
)
i
nduction
mo
t
or
sta
to
r
cu
rre
nts
(I
a
,
I
b
,
I
c
)
,
iv)
th
e
inv
e
rter
li
ne
outp
ut
volt
ages
an
d
sta
tor
cu
rr
e
nts
T
HD
s
,
and
v)
s
peed
are
use
d
f
or
f
ault
detect
ion
an
d
cl
assifi
cat
ion
usi
ng
machi
ne l
earn
i
ng
al
gorithms
[
15
]
,
[
16]
.
The
data
is
obta
ined f
r
om
t
he
simulat
io
n m
odel
as
sh
ow
n
i
n
Fig
ure
3
buil
t
us
in
g
MATL
AB
2019a
Sim
ulink
platf
orm
by
c
reati
ng
va
rio
us
ty
pes
of
f
ault
s
at
a
samplin
g
f
requ
ency
of
1e
-
5
s
econds
.
T
he
na
ture
of
t
he
data
trie
d
f
or
acc
ur
at
e
detect
io
n
a
nd
cl
assifi
cat
io
n
a
re
of tw
o
ty
pes
.
T
he data
is
pr
e
processe
d by no
r
mali
sat
ion
tech
nique.
a.
In
sta
ntane
ous
v
al
ues
The
i
ns
ta
nta
ne
ou
s
values
of
inv
e
rter
outp
ut
li
ne
vo
lt
a
ges
and
in
duct
ion
mo
to
r
sta
tor
c
urren
ts
an
d
th
e
recti
fied
outp
ut
volt
age
a
nd
T
HD
s
are
ta
ke
n
for
a
wi
ndow
of
1
sec
ond
wit
h
the
fa
ults
pe
rsi
ste
nt
th
rou
ghout
the
wind
ow.
T
he
li
mit
at
ion
of
insta
n
ta
ne
ous
values
for
feat
ur
e
e
xtracti
on
i
s
the
vo
l
um
i
nous
siz
e
of
t
he
da
ta
.
This
w
ou
l
d
be
a
majo
r
disa
dvantage
f
or
real
ti
me
ap
plica
ti
on
.
He
nce
in
vest
igati
on
wa
s
ca
r
ried
out
t
o
redu
ce
the
numb
e
r
of
featur
e
s
with
out
sac
rifici
ng
t
he
cl
assi
ficat
ion
acc
uracy
.
It
was
obser
ve
d
t
hat
th
ree
featu
re
s
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N:
20
88
-
8
694
Mac
hin
e le
ar
ni
ng base
d m
ulti
class fault
d
i
agno
sis
to
ol for
v
oltag
e
s
ou
rce
inverte
r
…
(
Jyo
thi R
)
1209
namely
t
he
t
hree
sta
to
r
c
urre
nts
a
re
suffici
e
nt
for
detect
io
n
a
nd
cl
assifi
cat
ion
.
Ot
her
fe
at
ur
es
are
fou
nd
t
o
be red
unda
nt.
b.
RMS
v
al
ue
s
In
t
his
ap
proac
h,
R
M
S
values
of
t
he
sta
to
r
cu
rr
e
nt
phas
or
s
w
ere
us
e
d
i
ns
te
a
d
of
in
sta
nt
ane
ou
s
val
ues.
He
nce
the total
numbe
r
of
featu
res were
six, as ea
ch ph
a
sor is
re
pr
ese
nted
by m
agn
it
ude a
nd it
s angle.
5.
HARD
WA
RE
The
ha
rdwar
e
set
up
of
VS
I
dr
i
ven
in
du
ct
i
on
m
otor
is
as
sho
wn
in
Fig
ur
e
4.
T
w
o
s
w
it
ches
we
re
employe
d
to
cr
eat
e
cro
ss
di
ode
open
ci
rcu
it
f
aults
an
d
the
ga
ti
ng
pu
lse
s
w
ere
re
move
d
to
the
I
GBTs
t
o
create
op
e
n
ci
rc
uit
I
GBT
fa
ults.
T
he
s
pecifica
ti
ons
of
t
he
hard
war
e
set
up
f
or
ext
racti
on
of
curre
nt
featu
r
es
is
as
mentio
ned
in
t
he
ta
ble
1
.
Th
e
sta
tor
c
urre
nt
RMS
data
w
as
extra
ct
e
d
f
r
om
t
he
V
SI
dr
iven
s
ys
te
m
th
rou
gh
ub
i
do
ts
IoT
pla
tform
. It
was o
bs
er
ve
d
to
b
e
in vali
datio
n wit
h
the
simulat
ion data.
Table
1
.
Sp
eci
f
ic
at
ion
s
of
t
he
VS
I
drive
n
i
nduction m
otor
Co
m
p
o
n
en
ts
Details
Dio
d
es
6
Nos
.
1
2
0
0
V,
3
0
A (
for
three
ph
ase
Dio
d
e Brid
g
e r
ecti
fier
)
-
RHR
G3
0
1
2
0
DC Link
Cap
acito
r
100uF
IGBT
6
Nos
.
1
2
0
0
V,
5
0
A
(f
o
r
Voltag
e So
u
rce
Inv
erte
r)
-
KG
T25
N1
2
0
KDA
Micr
o
co
n
troller
Ardu
in
o
Uno
sen
so
rs
3
Nos
.
Hall
effect
Cu
rr
en
t sen
so
rs
-
A
CS7
1
6
w,
Tem
p
e
ra
tu
re
sen
so
r
-
DT
H1
1
Wi
-
Fi
m
o
d
u
le
ESP82
6
6
No
d
em
cu
IoT Plat
form
Ub
id
o
ts
Figure
4
.
V
oltage S
ource
in
ve
rter
dr
i
ven in
duct
ion m
otor
ha
rdwar
e
set
up
6.
RESU
LT
S
AND DI
SCUS
S
ION
The vari
ous
dimens
i
on
s
of t
he
algorit
hms f
or the
fa
ult dete
ct
ion
a
nd c
la
ss
ific
at
ion
a
re
:
a.
Ty
pe of
machi
ne
le
ar
ning
Tw
o
ty
pes
are
us
e
d
namely
m
ulti
la
yer
pe
rce
ptr
on
–
neural
ne
twork
(
M
LP
-
NN)
a
nd
s
uppo
rt
vect
or
mac
hin
e
(S
V
M).
b.
Ty
pe of
data
Tw
o
ty
pes o
f d
at
a n
amel
y
inst
antane
ous a
nd
RMS val
ue
s
w
ere em
ploye
d
t
o
trai
n
the
m
odel
.
c.
Numbe
r of
feat
ur
es
14
feat
ur
es
na
mely
recti
fied
vo
lt
age
,
i
nv
e
rt
er
li
ne
outp
ut
vo
lt
age
s,
sta
tor
cu
rr
e
nts
a
nd
t
heir
c
orres
pondin
g
THDs,
s
peed
i
nst
antane
ou
s
dat
a
wer
e
us
e
d
init
ia
ll
y.
It
was
ob
serv
e
d
th
rou
gh
inv
est
igati
on
th
at
sta
tor
curre
nt
ph
a
sors
a
re
s
uffici
ent
to
detec
t
and
cl
assi
fy
the
fa
ults
.
Henc
e
six
par
a
mete
r
s
namel
y i
nd
uc
ti
on
m
otor
sta
t
or
currents
phas
ors w
e
re c
hosen
to
trai
n
the
m
odel
f
or R
M
S
dat
a.
The
data
was
s
plit
in
the
70
-
30
rati
o.
M
L
P
-
N
N
wa
s
trai
ne
d
with
pa
ramete
r
.
M
L
P
a
cc
ur
ac
y
is
ta
bula
te
d
as
me
ntion
e
d
in
T
a
ble
2
for
an
opti
mum
hy
pe
r
par
a
mete
r
tu
ning
of
hi
dd
e
n
la
yer
s
,
l
earn
i
ng
rat
e,
s
olv
e
r,
act
ivati
on
func
ti
on
for
14
a
nd
3
featu
res
res
pecti
vely
f
or
instanta
neous
da
ta
.
SVM
accu
racy
is
al
so
ta
bula
t
e
d
in
T
a
ble
2
for
an
opti
mal
hype
r
pa
rameter
tu
ning
of
ke
rn
el
,
C
a
nd
ga
mma
with
14
an
d
3
f
eat
ur
es
re
sp
ect
ively
for
i
n
sta
nta
neous
d
at
a.
F
or R
M
S
data
with
s
ta
tor
c
urre
nt
phas
or
s
wer
e
sel
ect
ed
as
6 feat
ur
es
wh
e
re
eac
h st
at
or
current
co
ns
ist
s
of
mag
nitu
de
an
d
phase
a
ng
le
[
13
]
,
[
14]
.
T
he
acc
uracy
was
f
ound
to
be
100%
with
hype
r
par
a
mete
r
t
un
i
ng a
s
mentio
ne
d
in
Ta
ble 2.
To
eval
uate
t
he
r
obus
t
ness
of
the
m
odel
a
nd
to
c
onve
y
t
he
balance
betwe
en
t
he
preci
sio
n
a
nd
recall
,
F1
sc
or
e
was
c
al
culat
ed.
T
he
F1
sc
or
e
of
0.9
58
a
nd
0.9
44
w
as
obta
ined
f
or
25
cl
asse
s,
14
f
eat
ur
es
a
nd
3
f
eat
ur
es
resp
ect
ivel
y re
pr
ese
ntin
g
the
accurate cl
assif
ic
at
ion
of
diff
e
ren
t t
ypes
of f
a
ults f
or
a
n
i
ns
ta
ntane
ous d
at
a
u
sin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
Dr
i
S
ys
t
,
V
ol
.
12
, N
o.
2
,
J
une
202
1
:
120
5
–
121
5
1210
M
LP
.
T
he
F
1
scor
e
of
0.96
and
0.9
56
was
obta
ined
f
or
25
cl
asses,
14
featur
e
s
a
nd
3
featu
res
res
pe
ct
ively
represe
nting t
he
accu
rate cl
as
sific
a
ti
on
of d
i
ff
e
ren
t t
yp
es
of
f
a
ults f
or
a
n
i
nst
antane
ou
s
d
at
a u
si
ng S
VM.
Table
2
.
O
ptimum h
yp
e
r para
mete
rs for in
sta
ntane
ous a
nd
RMS val
ue
s
Ins
tan
tan
eo
u
s Valu
es ML
P
-
NN:
Featu
res
Op
tim
u
m
Hy
p
er
p
arameters
Accuracy
F1
-
sco
re
Hid
d
en
layers
Lear
n
in
g
r
ate
So
lv
er
Alp
h
a
Activ
atio
n
Fun
ctio
n
14
(50
,10
0
,
5
0
)
Co
n
stan
t
Ad
am
0
.05
tan
h
9
5
.6 %
0
.95
8
3
(50
,10
0
,
5
0
)
Co
n
stan
t
Ad
am
0
.00
0
1
relu
9
5
.4%
0
.94
4
Ins
tan
tan
eo
u
s v
alu
es SVM
Featu
res
Op
tim
u
m
Hy
p
er
p
arameters
Accuracy
F1
-
sco
re
Kernel
C
Gam
m
a
14
rbf
1000
0
.00
1
9
5
.4%
0
.96
3
R
bf
1000
1
9
5
.08
%
0
.95
6
RMS valu
es M
LP
-
NN
Featu
res
Op
tim
u
m
Hy
p
er
p
arameters
Accuracy
F1
-
sco
re
Hid
d
en
layers
Lear
n
in
g
r
ate
So
lv
er
Alp
h
a
Activ
atio
n
Fun
ctio
n
6
(50
,50
,50
)
Co
n
stan
t
Ad
am
0
.00
0
1
tan
h
100%
1
RMS valu
es SV
M
Featu
res
Kernel
C
Gam
m
a
Accuracy
F1
-
sco
re
6
rbf
1
1
100%
1
The
F
1
Sc
ore of
1
w
as
obta
i
ned
for
al
l
25
c
la
sses
with
acc
ur
at
e
cl
assifi
ca
ti
on
of d
iffe
re
nt
fau
lt
typ
e
s
,
locat
ion
a
nd
se
ver
it
y
f
or
R
MS
sta
tor
c
urre
nt
ph
as
ors
us
i
ng
M
LP
a
nd
S
V
M
in
c
om
pa
rison
with
i
ns
ta
nt
aneous
data
the
li
mit
ation
of
the
insta
ntane
ous
value
meth
od
is
the
vo
l
um
in
ous
da
ta
and
the
co
m
pu
ta
ti
on
ti
me.
Th
ough
the
accu
racy
w
as
f
ound
t
o
be
great
er
t
ha
n
95%
,
it
was
a
patte
rn
-
base
d
ap
proac
h.
H
ence
,
a
nother
appr
oach
wh
ic
h
was
us
e
d
for
trai
ning
a
nd
te
sti
ng
the
model
wa
s
by
us
in
g
R
M
S
val
ues
of
the
ma
gnit
ud
e
of
s
ta
tor
cu
rr
e
nts
on
l
y.
B
ut
the
accurac
y
was
l
imi
te
d
to
onl
y
51%
in
both
M
LP
a
nd
S
V
M
al
go
rithms.
Hence
,
sta
to
r
current
ph
a
sors
wer
e
c
on
si
de
re
d
for
tr
ai
nin
g
an
d
te
sti
ng
the
model.
The
m
odel
wa
s
simulat
ed
f
or
a
ll
25
cl
asses
of
fau
lt
s
with
vo
lt
age
ra
ng
i
ng
f
rom
390V
to
426
V
f
o
r
0.5
HP
,
squi
rr
el
ca
ge
3
ph
a
se
inducti
on
m
otor
rate
d
f
or
415V,
50Hz.
Fr
om
the
num
be
r
of
case
stu
di
es
carrie
d
out,
i
t
was
obser
ve
d
that
minor
fa
ult
s
are
detect
ed
a
s
distor
ti
on
in
the
c
on
c
ordi
a
patte
rns
of
st
at
or
c
urren
ts
.
The
performa
nc
e
of
t
he
AC
dri
ve
s
ys
te
m
is
sign
ific
a
ntly
c
aptu
red
by
the
phaso
r
s
ta
tor
c
urre
nts
flow
i
ng
in
t
he
ci
rcu
it
as
obser
ve
d
by
the
two
-
dimensi
onal
st
at
or
cu
rr
e
nts
obta
ined
by
Cl
ar
ke’
s
tra
ns
f
ormat
io
n.
A
pl
ot
of
cu
rr
e
nt
con
c
ordia
patte
rn
s
f
or
dif
fer
e
nt
typ
e
s
of
el
ect
r
ic
al
fau
lt
s
i
n
vo
lt
age
so
urce
i
nv
e
rter
are
as
sho
wn
in
Fi
gur
e
5
(a
)
to
Fi
gure
5
(
e
).
It
can
be
ob
serv
e
d
t
hat
co
ncor
dia
patte
r
ns
are
disti
nc
t
f
or
eac
h
ty
pe
of
fa
ult.
He
nce,
t
he
phasor
sta
tor
cu
r
ren
ts
ca
rry
t
he
signa
ture
of
t
he
fau
lt
s
a
nd
c
an
be
eff
ect
ivel
y use
d for
detect
ion
and cla
ssific
at
ion o
f faults
[4
]
,
[
25]
[26]
.
Figure
5
.
Cu
rr
e
nt
co
nc
ordia
pa
tt
ern
s
for diff
eren
t t
yp
es
of e
le
ct
rical
f
aults
in
volt
age
sour
ce inv
e
rter
The
c
omp
utati
on
al
t
rainin
g
ti
me
f
or
the
i
ns
t
antane
ous
valu
es
was
m
or
e
t
han
one
hour
us
in
g
M
LP
al
gorithm
t
o
obta
in
acc
ur
at
e
cl
assifi
cat
ion
of
fa
ults
as
mor
e
num
ber
of
hyper
pa
rameters
ar
e
to
be
t
uned
du
e
to
vo
l
um
in
ous
da
ta
.
The
co
mput
at
ion
al
trai
ni
ng
ti
m
e
f
or
t
he
R
M
S
valu
es
was
12.
38
min
utes
for
M
L
P
a
nd
18.52
seco
nd
s
for
S
VM.
T
he
sta
to
r
cu
rr
e
nt
phas
or
a
ppr
oach
w
as
fou
nd
t
o
be
mo
r
e
be
nef
ic
i
al
even
for
pr
act
ic
al
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N:
20
88
-
8
694
Mac
hin
e le
ar
ni
ng base
d m
ulti
class fault
d
i
agno
sis
to
ol for
v
oltag
e
s
ou
rce
inverte
r
…
(
Jyo
thi R
)
1211
impleme
ntati
on
as
it
in
vo
l
ve
s
minimal
num
ber
of
sen
sor
da
ta
.
Both
M
L
P
and
m
ulti
cl
ass
SVM
gav
e
ac
cur
at
e
resu
lt
s
of
100%
with
e
xact
de
te
ct
ion
,
cl
assi
ficat
ion
an
d
l
oc
at
ion
of
the
f
aults.
Si
nce
t
he
in
pu
t
is
ta
ke
n
over
a
wide ran
ge
a
nd
num
ber
of cas
e stud
ie
s
h
a
ve been
con
du
ct
e
d wit
h diff
e
rent
p
ara
mete
r
set
s for
SVM a
nd
M
L
P,
the
100%
accu
racy
ca
nnot
be
at
tribu
te
d
t
o
over
fitt
ing
.
Ra
t
her
it
in
dicat
es
the
eff
ic
ac
y
of
the
met
hod
a
nd
t
he
natu
re
of
data
colle
ct
ion
.
Mul
ti
cl
ass
SVM
i
s
m
or
e
prefe
ra
bl
e
tha
n
M
LP
for
f
ault
detect
ion
a
nd
cl
assifi
cat
ion
as the c
omp
utati
on
al
ti
me is l
ess.
7.
DEVELOP
M
ENT OF
D
I
A
GNOST
IC
TOOL
Af
te
r
an
accu
r
at
e
detect
ion
a
nd
c
la
ssific
at
io
n
of
fa
ults,
it
is
essenti
al
to
t
ake
c
orrect
ive
measu
re
to
avo
i
d
cat
ast
r
ophic
e
ff
ect
s
t
o
the
syst
em.
H
ence
a
diag
nos
ti
c
too
l
was
de
velo
ped
us
i
ng
T
Kinter
pac
kag
e
i
n
py
t
hon
to
vis
ua
ll
y
dis
play
th
e
in
formati
on
to
the
operato
r
.
I
n
Diag
nosti
c
T
oo
l
,
the
R
M
S
va
lues
of
sta
tor
currents
,
i
nvert
er
li
ne
outp
ut volt
ages alo
ng
with the
wav
e
f
orms, Conc
ord
ia
cu
rr
e
nt p
at
te
rn
a
nd
t
he
s
pee
d
pl
ot
are
disp
la
ye
d.
7.1.
Ca
se
studie
s
7.1.1.
Healt
hy co
ndi
tion
The
diag
nosti
c
too
l
f
or
the
he
al
thy
c
onditi
on
of
the
VSI
dri
ve
n
in
duct
io
n
m
otor
is
as
s
how
n
in
the
Fig
ure
6
(a)
a
nd
Fig
ur
e
6
(
b).
The
RM
S
val
ue
s
of
the
c
urr
e
nt
s
and
vo
lt
ages
are
balan
ced
.
T
he
cu
rr
e
nt
co
nc
ordia
patte
rn
f
or
hea
lt
hy
sy
ste
m
with
P
W
M
-
VSI
dr
i
ven
in
duct
io
n
mo
t
or
i
s
f
ou
nd
to
be
re
gu
la
r
s
ymmet
rical
patte
rn
al
ong
iB
et
a
a
xi
s
an
d
the
s
pee
d
is
al
so
fou
nd
to
be
c
onsta
nt
.
P
re
dicti
on
fa
ult
ty
pe
an
d
lo
cat
ion
are
dis
pl
ayed
nu
ll
.
(a)
(b)
Figure
6.
Healt
hy con
diti
on
of
VSI
dr
i
ven in
du
ct
io
n mot
or
7.1.2.
Single I
GBT
open faul
t
The
dia
gnos
ti
c
to
ol
for
t
he
sing
le
I
GBT
ope
n
fa
ult
co
ndit
ion
of
the
VS
I
dr
i
ven
in
duct
io
n
mo
t
or
is
a
s
sh
ow
n
in
t
he
Fi
gure
7(a)
a
nd
Figure
7(b)
.
T
he
RM
S
val
ue
s
of
t
he
c
urrent
s
and
volt
a
ges
are
un
balance
d.
T
he
current
Co
ncord
ia
patte
r
n
f
or
I
GBT
open
fa
ult
with
P
W
M
-
VS
I
dr
iv
en
i
nductio
n
mo
to
r
is
f
ound
t
o
be
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
Dr
i
S
ys
t
,
V
ol
.
12
, N
o.
2
,
J
une
202
1
:
120
5
–
121
5
1212
a
sy
m
metri
cal
and
t
he
s
peed
is
al
so
f
ound
to
be
fluctuati
ng
a
f
fecti
ng
t
he
loa
d.
P
re
dicti
on
fa
ult
typ
e
a
nd
loc
at
ion
are
dis
playe
d
e
xactl
y
.
IG
BT
5
is
the
thir
d
le
g
uppe
r
switc
h
wh
ic
h
co
nduc
ts
durin
g
posit
ive
hal
f
cyc
le
.
I
n
case
of
I
GBT5
ope
n
fau
lt
,
it
ca
n
be
cl
early
visu
al
iz
ed
that
t
he
po
si
ti
ve
half
cycle
of
phase
c
sta
to
r
cu
rr
e
nt
is
missi
ng
in the wa
veforms.
(
a
)
(b)
Figure
7
.
U
nhe
al
t
hy
c
onditi
on
fa
ult pr
e
dicti
on
for IGBT
5 o
pen fa
ult
7.1.3.
Cro
s
s IGBTs
op
en
fault
The
diag
nosti
c
too
l
f
or
t
he c
r
os
s
IG
BT
ope
n
ci
rcu
it
f
ault
c
onditi
on
of
the
VS
I
dri
ve
n i
nduction
m
otor
is
as
sho
wn
i
n
t
he
Fi
gure
8
(a
)
and
Fig
ur
e
8
(
b)
.
T
he
R
M
S
val
ues
of
t
he
c
ur
re
nts
are
un
balan
ced
a
nd
the
i
nverter
li
ne
outp
ut
vol
ta
ges
are
al
so
unbalance
d
w
hi
ch
caus
es
inc
r
ea
sed
vibrat
io
n
in
t
he
in
duct
ion
mo
t
or
le
a
di
ng
t
o
sever
e
da
ma
ge
if
e
xisted
f
or
a
long
r
un
an
d
a
lso
inc
rease
d
T
HD
s
l
eadi
ng
to
powe
r
qual
it
y
issues.
The
c
urren
t
c
on
c
ordia
patte
rn
f
or
I
GBT
op
en
fau
lt
with
P
W
M
VS
I
dr
ive
n
in
duct
io
n
m
ot
or
is
al
so
f
ound
to
be
as
ymm
et
rical
and
t
he
s
peed
is
al
so
f
ound
to
be
fl
uctuati
ng
aff
ect
in
g
the
l
oa
d.
P
re
dicti
on
f
ault
typ
e
a
nd
locat
ion
are
dis
playe
d
in
the
to
ol
.
IGB
T3 i
s
the
sec
ond
le
g
uppe
r
s
witc
h
w
hich c
onduct
s
duri
ng
po
sit
ive
hal
f c
yc
le
.
In
case
of
I
GBT2
and
I
GBT
3
op
en
fau
lt
,
i
t
can
be
cl
earl
y
visua
li
zed
that
t
he
po
sit
ive
hal
f
c
yc
le
of
ph
ase
b
a
nd
ne
gative
hal
f
c
ycle
of phase
c
in
ve
rter
ou
t
pu
t c
urr
ents are
is miss
ing
i
n
the
w
a
ve
forms
.
(a)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N:
20
88
-
8
694
Mac
hin
e le
ar
ni
ng base
d m
ulti
class fault
d
i
agno
sis
to
ol for
v
oltag
e
s
ou
rce
inverte
r
…
(
Jyo
thi R
)
1213
(b)
Figure
8
.
U
nhe
al
thy
c
onditi
on
fa
ult pr
e
dicti
o
n
for IGBT
3
a
nd IGBT
2 op
e
n fault
8.
CONCL
US
I
O
N
Robust
an
d
ac
cur
at
e
fa
ult
di
agnostic
to
ols
hav
e
bee
n
de
ve
lop
e
d
f
or
fa
ults
in
the
in
du
c
ti
on
m
otor
.
Howe
ver,
in
m
od
e
r
n
powe
r
el
ect
ronic
dr
i
ves
,
there
is
a
nee
d
to
de
velo
p
a
diag
nosti
c
too
l
for
detect
io
n
of
fau
lt
s
in the Po
wer el
ect
ronics circ
uit. Th
ese
f
a
ults
are not easi
ly
predict
able,
as t
he
in
duct
ion m
otor often
cont
inu
es
to
r
un.
T
he
y
m
ay
le
ad
to
cat
a
strophic
e
ff
ect
s.
He
nce,
an
e
arly
detect
ion
is
necessa
r
y.
In
this
w
ork
,
det
ai
ls
of
Detect
ion
an
d
cl
assifi
cat
ion
of
fa
ults
in
i
nverter
dr
ive
n
i
nductio
n
mo
t
or
are
pr
ese
nted
us
i
ng
MLP
-
N
N
a
nd
mu
lt
ic
la
ss
S
V
M
.
D
i
ff
e
ren
t
se
ts o
f feat
ures
w
ere i
nv
est
igate
d usin
g b
oth i
nst
antane
ou
s
values
a
nd R
M
S
va
lues
of
i
nv
e
rter
outpu
t
li
ne
volt
ag
es
an
d i
nd
uctio
n m
otor
sta
to
r
currents
. W
he
n
instanta
ne
ous
values
we
re
us
ed,
t
he
three
sta
tor
c
ur
ren
ts
car
r
y
t
he
sign
at
ur
e
of
va
rio
us
fa
ults
a
nd
are
s
uffici
ent
to
co
rr
ect
ly
cl
as
sify
the
m.
H
ow
eve
r,
the
data
is
vol
um
in
ous
a
nd
t
he
co
mputat
io
n
ti
me
is
al
so
hi
gh
.
A
n
al
te
rn
a
ti
ve
to
us
e
of
i
ns
ta
nta
neous
va
lues
is
us
e
of
R
M
S
values
of
sta
to
r
c
urre
nt
phas
or
s
.
The
acc
ur
ac
y
i
s
al
so
f
ound
to
be
10
0
pe
rce
nt
with
mini
mal
f
eat
ur
es
and
a
diag
no
st
ic
too
l
is
deve
l
op
e
d
as
a
n
i
ndic
at
or
to
t
he
op
e
rato
r.
The
la
boratory
prot
otype
dev
el
oped
f
or
monit
or
i
ng
t
he
powe
r
el
ect
r
onic
s
ci
rc
uitry
with
mi
nimal
s
ens
or
s
has
al
s
o
bee
n
prese
nte
d.
T
he
w
ork
w
ou
l
d
be
eff
ect
ive
in
re
al
isi
ng
t
he
co
nc
ept
of
reli
abili
ty
ce
ntred
ma
intenanc
e
.
F
ur
t
her
e
xtensi
on
of
the
wor
k
is
bei
ng
carried
out
us
i
ng PCA
for hi
gh
dimensi
onal
it
y
re
duct
ion.
REFERE
NCE
S
[1]
Shu
-
Ying
Li
,
an
d
Le
i
Xue
,
“
M
otor's
ea
rly
fau
l
t
dia
gnosis
base
d
on
support
ve
ctor
ma
ch
ine,
”
A
MIMA
2018
IO
P
Publ
ishing
,
IO
P
Conf
.
Seri
es:
Mate
rials
S
c
ie
nc
e
and
Eng
ine
ering
382
(
2018)
032047
,
pp
1
-
4,
201
8
DO
I:
10.
1088/17
57
-
899X/382/3/032047
[2]
Khire
ddine,
M.S
.
Sli
ma
ne
,
N
.
Ab
desseme
d
and
Y
assine
Makhlouf
i,
M.
T,
“
Fau
lt
d
e
te
c
ti
on
and
d
ia
g
nosis
in
indu
ct
io
n
mot
or
using
ar
ti
ficia
l
int
e
ll
ig
e
nce
techniq
u
e,
”
MATEC
We
b
of
Confe
r
enc
e
s
,
2014,
vol
1
6,
pp
.
1
-
5
,
DO
I:
10.
1051/matecc
onf/20141610004.
[3]
Furqan
As
ghar
,
Muhamm
ad
Tal
ha,
and
Sung
Ho
Kim,
“
Compa
r
ative
study
of
thr
e
e
fau
lt
d
ia
gnosti
c
me
thods
for
thr
e
e
phase
inve
r
te
r
w
it
h
indu
ct
ion
mo
tor
,
”
Inte
rnat
ion
al
Journal
of
Fu
zz
y
Logic
and
In
te
lligent
Syste
ms
,
vol
.
17,
no.
4
,
pp.
245
-
256
,
De
ce
mb
er
2017
,
D
OI:
10.
5391/IJFI
S.2017.
17.
4
.
245.
[4]
Faeka
Kha
te
r
,
S
eba
h
M.
I.,
and
Os
am
a
M,
“
Faul
t
dia
gnost
ic
s
in
an
inv
ert
er
fe
eding
an
induc
t
ion
mot
or
using
fu
z
zy
l
ogic,
”
Journal
of
El
e
ct
ric
al
Sy
stems
and
Infor
mation
Te
chnology
,
vo
l.
4
,
no.
1,
pp
.
10
-
17,
May
2017
,
DO
I:
10.
1016/j.je
si
t.
2
016.
10.
005
.
[5]
Shorouk
Os
sam
aIb
rah
i
m
,
kha
le
d
Nagdy
Faris
,
an
d
Esa
m
Abo
Elz
aha
b
,
“
Implementat
ion
of
fu
zz
y
mode
li
ng
sys
tem
for
fau
lt
s
d
et
e
ct
i
on
and
dia
gnosi
s
in
thre
e
phase
induc
ti
on
mot
or
drive
sys
te
m
,
”
J
ournal
of
Elec
tri
cal
Syst
ems
an
d
Information
Tec
hnology
,
vo
l. 2,
no.
1
,
pp
.
27
-
46
,
2015
,
DO
I:
10.
1
016/j
.
j
esit.2015.03.004
.
[6]
V.
Gomat
hy
and
Selva
pe
rumal
S
,
“
Fau
lt
d
et
e
ct
io
n
and
c
la
ss
ifica
t
i
on
with
opt
im
i
zation
t
ec
hniqu
es
f
or
a
thr
ee
-
ph
ase
single
-
inv
ert
er
ci
rcu
it
,
”
Journal
of
Powe
r
Elec
troni
cs
,
v
ol
.
16
,
pp
.
1097
-
1109,
May
2016
,
D
OI:
10.
6113/JP
E.
20
16.
16.
3
.
1097
.
[7]
H.
Yang,
J
.
Zh
ao
and
F.
Wu,
“
Cur
ren
t
sim
ilarit
y
ba
sed
fau
l
t
di
agnosis
f
or
indu
ct
ion
mot
or
driv
es
wit
h
discrete
wav
ele
t
tra
nsform,
”
201
6
Prognosti
cs
an
d
Syst
em
Hea
lt
h
Manage
ment
Co
nfe
renc
e
(PHM
-
Chengdu)
,
Chen
gdu,
2016
,
pp.
1
-
6,
DO
I:
10
.
1109
/PHM
.
2016.
7819841.
[8]
Furqan
As
ghar
,
Muhamm
ad
Ta
l
ha,
Sung
Ho
Ki
m,
”
Neura
l
ne
twork
base
d
fau
lt
det
e
ct
ion
and
di
agnosis
sys
te
m
f
or
thre
e
-
ph
ase
inv
ert
er
in
v
ariab
le
spe
ed
driv
e
with
induc
t
io
n
mot
or
”
,
Jour
nal
of
Control
Sc
ie
nc
e
and
Engi
ne
ering
,
vol
.
2016
,
pp
.
1
-
12
,
2016.
DO
I:
10.
1
155/2016/
12863
18.
[9]
Dong
-
Eok
Kim
and
Dong
-
Choo
n
Lee,
“
Faul
t
di
a
gnosis
of
thr
ee
-
phase
PWM
in
ver
te
rs
using
wa
vel
e
t
and
SV
M,
”
IEE
E
Inte
rnat
io
nal
Symposium
on
Industrial Elect
ronics
,
2008,
p
p.
329
-
334
,
DO
I:
10.
1109
/ISIE.
2
008.
4676998.
[10]
We
i
,
Li,
Xu,
a
nd
Huang,
“
A
rev
ie
w
of
e
arl
y
fau
lt
di
agnosis
appr
oac
h
es
an
d
the
ir
appl
i
ca
t
i
ons
in
ro
ta
t
ing
ma
ch
ine
ry,
”
Entr
opy
,
vol
.
21
,
no
.
4
,
p
.
409
,
Apr.
2019
,
DO
I:
10
.
3
390/e
21040409
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
Dr
i
S
ys
t
,
V
ol
.
12
, N
o.
2
,
J
une
202
1
:
120
5
–
121
5
1214
[11]
Ubale
,
M.R
.
,
Dh
uma
l
e,
R.
B
.
,
and
Lokh
ande,
S,
“
Open
sw
it
ch
f
ault
d
ia
gnosis
in
thr
ee
phase
inv
erter
using
di
agnosti
c
var
ia
b
le
m
et
hod
,
”
In
te
rnationa
l
J
ournal
of
Re
sear
ch
in
Engi
n
ee
rin
g
and
Techno
log
y
,
vol
.
02,
no
.
1
2,
pp
.
636
-
640,
2013.
[12]
H.
Zha
o
,
W.
D
e
ng,
R.
Yao,
M.
Sun,
Y.
Luo
,
an
d
C.
Dong,
“
Stu
dy
on
a
nov
el
f
aul
t
d
ia
gnosis
m
et
hod
b
ase
d
on
int
egr
at
ing
EM
D,
fu
zz
y
en
tropy
,
im
prov
ed
PS
O
and
SV
M,
”
Jour
nal
o
f
Vi
broeng
i
nee
ring
,
vo
l.
19,
no.
4
,
pp.
2562
-
2577,
Jun.
2017
,
DO
I:
10.
21595
/
jve
.
2017
.
18052
.
[13]
Jlassi,
J.
O.
Estima,
S.
K.
El
Khi
l,
N
.
M.
Be
ll
a
aj a
nd
A.
J.
M.
C
ar
doso,
“
A
robust
observe
r
-
base
d
me
thod
for
IGB
Ts
and
cur
ren
t
sensors
fau
l
t
di
agnosis
in
vo
lt
ag
e
-
so
urc
e
inv
ert
ers
of
PM
S
M
dr
ive
s,
”
in
I
EE
E
Tr
ansac
ti
ons
on
Industr
y
Appl
ic
a
ti
ons
,
vo
l.
53
,
no
.
3
,
pp
.
2
894
-
2905,
May
-
June
2017,
DO
I:
10.
1109/
TIA.
20
16.
2616398.
[14]
Jian
-
jian,
Z
.
,
Yo
ng,
C.
,
Zha
ng
-
Y
ong,
C.
,
and
An
ji
an
,
Z.
,
“
Op
en
-
sw
it
ch
fau
l
t
diagnos
is
me
thod
i
n
volt
ag
e
-
source
inve
rt
ers
base
d
on
phase
cur
ren
ts
,
”
I
EE
E
Acce
ss
,
vol.
7
,
p
p.
63619
-
636
25,
2019
,
DO
I:
10.
1109/ACCESS
.
2019.
2913164
.
[15]
R.
Jyothi,
R.
Jay
apa
l
and K. U
.
R
ao,
“
Sever
i
ty an
d
im
pa
ct
of
fau
l
t
s
on
cur
ren
t
har
moni
cs
in
inve
r
t
er
-
fed
AC
driv
es,
”
2016
IEEE
Inno
vat
i
ve
Smar
t
Gr
id
Techno
logi
es
-
Asia
(ISGT
-
A
sia)
,
Melbourn
e,
VIC,
2016
,
pp
.
401
-
405,
DO
I:
10.
1109/ISGT
-
As
ia
.
2016.
779641
9.
[16]
R.
Jyothi
,
Shaik
Mohamm
ed
Z
ah
id
,
Moh
am
m
ed
Sufyan
Z
ai
n
,
Re
shma
Begu
m;
Syed
Afz
al
and
K.
Um
a
R
ao
,
e
t
al
.
,
“
Automatic
fau
l
t
di
a
gnosis
sys
te
m
for
voltage
source
inv
erter
drive
n
induc
t
ion
mot
or
,
”
2019
3
rd
Inte
rnationa
l
Confe
renc
e
on
R
ec
en
t
De
ve
lopm
e
nts
in
Con
trol,
A
utomati
on
&
Po
wer
Engi
n
ee
ring
(RDCAPE)
,
NO
IDA
,
India,
2019
,
pp.
477
-
481
,
DO
I:
10.
1109
/RDCAP
E47089.
2019.
8979041
.
[17]
Shali
ni
Vasisht
a
,
and
KR
Rekha,
“
A
Survey
:
spa
ce
v
ec
tor
PWM
(SV
PWM)
in
3φ
voltage
source
inve
rt
er
(VS
I)
,
”
Inte
rnational
Jo
urnal
of
Elec
tri
c
al
and
Computer
Engi
ne
ering
(IJ
ECE
)
,
vo
l.
8
,
no.
1,
pp.
11
-
18
,
Feb
rua
ry
2018
,
DO
I
:
10.
11591/ijece.
v
8i1.
pp11
-
18
.
[18]
Akhile
sh
Sharm
a,
Anandh
N
.
,
an
d
Sarsing
Gao
,
“
Modulat
ion
index
eff
e
ct
on
inve
r
te
r
base
d
inducti
on
mot
or
driv
e
,
”
Inte
rnational
Jo
urnal
of
Powe
r
El
e
ct
ronics
and
Dr
iv
e
Syst
em
(IJ
PE
DS),
vo
l.
11,
no.
4,
pp.
1785
-
1798,
De
c
2020
,
DO
I:
10.
11591/ijpeds.
v11.
i4.
pp1
785
-
1798
.
[19]
Amine
Moham
e
d
Kheli
f,
Aze
ddi
ne
Bendiabdella
h,
and
B
il
a
l
Dja
ma
l
Eddi
ne
Cher
if,
“
Short
-
ci
r
cui
t
fau
lt
d
ia
gnosis
of
the
DC
-
Li
nk
c
ap
ac
i
tor
and
it
s
i
mp
ac
t
on
an
el
e
ct
r
ical
drive
sys
tem
,
”
Int
ernati
onal
Jo
urnal
of
E
lectrical
and
Computer
Engi
ne
ering
(I
J
ECE
),
vol
.
10
,
n
o.
3
,
pp
.
2807
-
2
814,
June
2020
,
DO
I:
10.
11591/ijece.
v10i3
.
pp28
07
-
2814
.
[20]
Rekha
SN
,
P
Ar
una
Jeya
n
thy,
an
d
D
Deva
ra
j,
“
R
el
ev
anc
e
ve
ct
or
ma
ch
ine
b
ase
d
f
aul
t
class
ifi
catio
n
in
wind
ene
rg
y
conve
rsion
sys
tem
,
”
In
te
rnationa
l
Journal
of
E
lec
tric
al
and
Comp
ute
r
Engi
n
ee
ring
(I
JE
CE)
,
vol
.
9,
no.
3
,
pp.
1506
-
1513,
June
2019
,
DO
I:
10
.
11591
/i
jece
.
v9i3
.
pp15
06
-
1513
.
[21]
Khadidj
a
El
Me
rra
oui,
Abde
llaz
iz
Ferdjouni,
an
d
M’hame
d
Bo
unekhl
a
,
“
R
ea
l
ti
me
obse
rve
r
b
ase
d
stat
or
f
aul
t
dia
gnosis
for
IM
,
”
Int
ernati
onal
Journal
o
f
Elec
t
rical
and
Comp
ute
r
Engi
ne
erin
g
(IJ
ECE
)
,
vol.
10,
no
.
1,
pp
.
21
0
-
222,
Feb
2020
,
DO
I:
10.
11591/ijece.
v10i1
.
pp21
0
-
222
.
[22]
La
ch
ta
r
Sa
la
h
,
G
hoggal
Ade
l,
Ko
uss
a
Khale
d,
Bo
ura
iou
Ahm
ed,
a
nd
Attoui
Iss
a
m
,
“
Broken
ro
tor
ba
r
fau
l
t
di
agnosti
c
for
DTC
Fed
in
duct
ion
mo
tor
u
sing
stat
o
r
instantane
ous
com
p
l
ex
appa
r
ent
po
wer
env
el
op
e
si
gnat
ure
an
al
ysis
,
”
Inte
rnational
Jo
urnal
of
Powe
r
El
e
ct
ronics
and
Dr
iv
e
Syst
em
(IJ
PE
DS),
vo
l.
10,
no.
1,
pp.
1187
-
1196,
Sep
2019
,
10.
11591/ijpeds.
v10.
i3.
pp1187
-
1
196
.
[23]
N.
F.
Fadz
ai
l
,
S.
Mat
Z
al
i
,
“
Fau
l
t
d
et
e
ct
ion
and
cl
assifi
ca
t
ion
in
wind
turb
ine
by
using
artifi
ci
a
l
neur
al
ne
twork
,
”
Inte
rnational
Jo
urnal
of
Powe
r
El
e
ct
ronics
and
Dr
iv
e
Syst
em
(IJ
PE
DS),
vo
l.
10,
no
.
3,
pp.
1687
-
1693,
Sep
2019
,
DO
I:
10.
11591/ijpeds.
v10.
i3.
pp1
687
-
1693
.
[24]
A.
O.
Di
To
mm
a
so,
F.
Genduso,
R.
Mi
ce
l
i
and
G.
R.
Ga
ll
uz
zo,
“
C
urre
nt
fau
lt
signa
ture
s
of
Vol
ta
ge
Source
Inv
erters
in
diff
ere
nt
ref
er
enc
e
fra
me
s,
”
20
16
Inte
rnationa
l
Symposium
on
Powe
r
Elec
troni
cs,
E
le
c
tric
a
l
Dr
iv
es,
Aut
omat
ion
and
Moti
on
(SP
EE
DAM)
,
Anac
a
pri,
2016
,
pp
.
59
9
-
604,
DO
I:
10
.
1109/SP
EE
DA
M.2016.
752597
7
.
[25]
Yong
C,
Zha
ng
J
J,
Chen
ZY
,
“
Cu
rre
nt
observe
r
-
b
a
sed
onli
n
e
open
-
sw
it
ch
f
aul
t
di
ag
nosis
for
vo
lt
ag
e
-
source
inve
rt
er
”
,
ISA
Tr
ans
act
ion
s
,
pp.
445
-
453
,
Apr
2020
,
DO
I:
10.
1016/j.isa
tr
a.2019.09.
019
.
[26]
Chouhan
Abhisa
r,
Gangsa
r
Puru
shotta
m
,
Porw
al
Raj
ku
ma
r
,
Mec
hef
ske
Chr
istop
her
K
,
“
Art
ificial
n
eur
al
ne
twork
base
d
fau
l
t
di
ag
nostic
s
for
thr
ee
phase
indu
ct
io
n
mot
ors
und
er
simi
l
ar
oper
at
i
ng
condi
t
ions”,
Vi
broengin
ee
rin
g
PR
OCEDIA
202
0,
Vol.
30,
p
p.
55
-
60
,
Apr
2020,
D
OI:
10.
21595/vp
.
2020.
21334
.
BIOGR
AP
HI
ES OF
A
UTH
ORS
J
y
o
t
h
i
R
r
e
c
e
i
v
e
d
B
.
E
i
n
E
l
e
c
t
r
i
c
a
l
a
n
d
E
l
e
c
t
r
o
n
i
c
s
E
n
g
i
n
e
e
r
i
n
g
a
n
d
M
.
T
e
c
h
i
n
P
o
w
e
r
E
l
e
c
t
r
o
n
i
c
s
F
r
o
m
V
i
s
v
e
s
v
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
i
n
t
h
e
y
e
a
r
2
0
0
9
a
n
d
2
0
1
3
r
e
s
p
e
c
t
i
v
e
l
y
.
S
h
e
i
s
c
u
r
r
e
n
t
l
y
p
u
r
su
i
n
g
P
h
.
D
i
n
V
i
sv
e
s
v
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
.
S
h
e
h
a
s
8
y
e
a
r
s
o
f
t
e
a
c
h
i
n
g
e
x
p
e
r
i
e
n
c
e
a
n
d
c
u
r
r
e
n
t
l
y
w
o
r
k
i
n
g
a
s
A
s
s
i
s
t
a
n
t
P
r
o
f
e
ss
o
r
i
n
D
e
p
a
r
t
m
e
n
t
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
E
l
e
c
t
r
o
n
i
c
s
E
n
g
i
n
e
e
r
i
n
g
,
R
V
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
a
n
g
a
l
o
r
e
.
H
e
r
r
e
s
e
a
r
c
h
i
n
t
e
r
e
s
t
s
i
n
c
l
u
d
e
P
o
w
e
r
E
l
e
c
t
r
o
n
i
c
s
,
D
r
i
v
e
s
a
n
d
F
a
u
l
t
d
i
a
g
n
o
s
i
s
u
s
i
n
g
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
.
S
h
e
h
a
s
a
u
t
h
o
r
e
d
5
a
r
t
i
c
l
e
s
i
n
t
h
e
a
r
e
a
o
f
P
o
w
e
r
E
l
e
c
t
r
o
n
i
c
s
a
n
d
D
r
i
v
e
s
.
E
-
m
a
i
l
:
j
y
o
t
h
i
r
@
r
v
c
e
.
e
d
u
.
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.