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.
12
,
No.
1
,
M
a
r 202
1
, p
p.
5
58
~
56
6
IS
S
N:
20
88
-
8694
,
DOI: 10
.11
591/
ij
peds
.
v12.i
1
.
pp
5
58
-
56
6
558
Journ
al h
om
e
page
:
http:
//
ij
pe
ds
.i
aescore.c
om
Investig
ation of
re
li
ab
ilit
y assesse
ment
in
power el
ec
t
ro
nics
circ
uits using m
achine le
arning
So
umy
a Rani
Mestha
,
Pin
to Pius
A.J
.
Depa
rtment
o
f
E
le
c
tri
c
al a
nd
Ele
ct
roni
cs
Engi
n
eering,
NM
AM
Instit
ut
e
of
T
ec
hnol
ogy,
Karn
at
ak
a
,
India
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
28
, 20
20
Re
vised
Jan
12
, 2021
Accepte
d
Fe
b
2
, 2
0
21
Rec
en
t
adv
ances
in
power
elec
tr
onic
s
(PE)
a
nd
ma
ch
ine
le
arn
in
g
(ML)
hav
e
prompt
ed
the
technologi
sts
to
a
dapt
th
ese
n
ew
te
chno
logi
es
to
im
prove
the
rel
i
abi
l
it
y
of
PE
sys
te
ms.
Durin
g
the
p
roc
ess,
a
lot
of
inv
esti
ga
ti
ons
on
th
e
per
forma
n
ce
and
rel
i
abi
l
it
y
of
PE
sys
te
ms
is
c
arr
i
ed
out
.
Th
e
inten
ti
on
of
thi
s
pape
r
is
to
pr
e
sent
a
com
pre
h
ensive
study
of
adva
nc
es
in
t
he
field
o
f
rel
i
abi
l
it
y
of
PE
sys
te
m
s
using
ma
chi
n
e
learni
ng
.
Rec
en
t
publicati
ons
in
thi
s
reg
ard
ar
e
an
al
y
sed
and
f
indi
ngs
ar
e
t
abul
a
te
d
.
I
n
add
it
ion
to
this
,
l
it
e
rat
ur
e
s
publi
shed
in
th
e
pr
edi
c
ti
on
of
re
ma
in
ing
use
ful
l
ife
(RUL)
of
power
el
e
ct
roni
c com
p
onent
s
is d
iscussed
with
e
mpha
si
s on
it
s
li
m
it
a
ti
o
ns.
Ke
yw
or
d
s
:
Condit
ion
m
onit
or
in
g
Life cycle
M
ac
hin
e
l
ea
rn
i
ng
Power
e
le
ct
ron
ic
s
Pr
og
nosis
Re
mainin
g
u
se
fu
l
l
ife
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
:
Soum
ya
Ra
ni
M
est
ha
Dep
a
rteme
nt of Elect
rical
and
Ele
ct
ronics
E
nginee
rin
g
N
MAM
Insti
tute of Tech
nolo
gy
Nitt
e
574110,
Udu
pi D
ist
rict
,
K
a
rn
at
a
ka,
India
Emai
l
:
vs
.s
ou
my
a
@n
it
te
.edu
.in
1.
INTROD
U
CTION
Inve
ntion
of
th
yr
ist
or
in
the
ye
ar
1957
ha
s
c
reated
a
ne
w
e
ra
i
n
t
he
fiel
d
of
power
el
ect
r
on
ic
s
.
Since
then
,
pow
er
el
ect
ronics
has
f
ound
it
s
wa
y
i
n
the
wide
ra
nge
of
a
pp
li
cat
ion
s
rig
ht
f
r
om
powe
r
ge
ne
rati
on
to
end
-
us
er
c
ons
umpti
on
of
el
ect
rici
ty.
Thor
ough
rese
arch
a
nd
im
pro
veme
nts
in
the
semic
onduct
or
te
chnolo
gies,
conve
rter
ci
rc
uit
te
chnolo
gy
especial
ly
in
co
nt
ro
ll
e
d
recti
fiers
[
1
]
,
[
2]
ha
s
impro
ved
performa
nce
of
t
he
powe
r
e
le
ct
ro
nics
s
ys
t
ems
with
res
pe
ct
to
ef
fici
en
cy
an
d
switc
hi
ng
s
peed
s
.
P
ow
e
r
el
ect
ro
nics
c
omp
on
e
nts
ar
e
mainly
us
e
d
in
powe
r
co
nver
sion
s
ys
te
ms
due
to
t
heir
s
witc
hin
g
capa
bili
ty
an
d
eff
ic
ie
nc
y.
H
oweve
r,
t
hese
com
pone
nts
te
nd
t
o
get
ex
pose
d
to
c
urre
nt
surge
s,
high
te
m
per
at
ure
s
an
d
con
ti
nu
ous
s
witc
hin
g
ope
rati
ons
le
a
ding
to
the
possibil
it
y
of
po
wer
el
ect
r
on
ic
s
co
mpo
ne
nts
fail
in
g
t
o
operate
in the e
xpect
ed
manne
r.
Ow
i
ng
to
the
s
afety
re
qu
ir
em
ents,
t
he
a
utom
otive
(E
V)
an
d
aer
ospace
in
dustrie
s
ha
ve
broug
ht
in
the
string
e
nt
nor
m
s
in
the
fiel
d
of
reli
abili
ty
of
powe
r
el
ect
r
onic
s
sy
ste
m
s.
Y
antao
et
al
[3]
mentio
ns
t
hat
powe
r
semic
onduct
or
as
well
as
el
ect
ro
lyti
c
capaci
t
or
s
a
re
m
os
t
s
usc
eptible
to
fail
ur
e
s.
Fail
ur
e
of
an
y
of
the
se,
may
be on
e
or m
or
e
co
m
pone
nts c
ou
l
d be a catas
tro
ph
e
pro
vid
e
d
a
pprop
riat
e f
ault ha
nd
li
ng
mecha
nisms a
r
e not i
n
place.
As
per
the
st
udy
c
onduct
ed
[
4]
on
P
V
m
odules
,
power
i
nverters
accounte
d
f
or
37%
of
unsch
edu
le
d
mainte
na
nce i
nc
idents
by c
ompone
nt and c
on
tribu
te
d
f
or
59% of
un
s
che
du
le
d
mainte
na
nc
e ex
penditu
res.
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
In
vest
ig
atio
n o
f rel
iab
il
it
y a
ss
esseme
nt in p
ow
er ele
ct
ro
nics
ci
rcuits u
sin
g
...
(
Soumya
Rani Mest
ha
)
559
2.
NOTIO
N OF
RELIABIL
IT
Y
I
N
P
OWER
EL
ECTRO
N
ICS
In
po
wer
el
ect
ronics
ci
r
cuits
(P
EC)
,
fau
lt
s
c
an
be
ei
ther
in
trinsic
(
chi
p
re
la
te
d
-
m
os
tl
y
oc
cur
due
to
high
c
urre
nt
or
vo
lt
ag
e
)
or
ext
rinsic
(
pa
ckag
e
relat
ed
-
mo
stl
y
occur
du
e
to
t
her
m
o
–
mec
han
ic
al
s
tress).
Re
li
abili
ty
in
PEC
was
i
ntr
oduce
d
as
ea
rly
as
1950s
[5].
As
me
ntio
ned
in
[
6]
,
reli
abili
ty
is
the
prob
a
bili
ty
of
any p
art o
r
the
en
ti
re
s
ys
te
m
that
c
on
ti
nues
to w
or
k
with
ou
t
an
y
i
nterru
ption ov
e
r
a
pe
rio
d
of
ti
me.
Re
li
abili
ty
may be
de
fine
d
as
(
1)
R(t)
=
−
/
(1)
wh
e
re
M
TBF
= M
ea
n
ti
me
be
tween
fail
ures
.
The
reli
abili
ty
functi
on
R(t
)
ve
rsu
s
ti
me
[0,
t]
is
pl
otted
i
n
Fig
ur
e
1
the
s
hap
e
of
w
hich
resem
bles
bathtub
wh
ic
h
is
the
li
fe
-
c
ycl
e
of
a
co
mpo
ne
nt.
The
gr
a
ph
has
three
disti
nct
phases,
na
mely,
burn
-
in
,
us
ef
ul
li
fe
and
t
he
we
ar
-
ou
t
per
i
od
s
[7,
8].
E
very
c
ompone
nt
w
hich
co
mes
ou
t
of
ass
sem
bly
li
ne
is
r
olled
ou
t
afte
r
the
exec
utio
n
of
exte
ns
ive
te
sti
ng
processe
s
to
handle
t
he
i
nf
a
nt
-
m
ort
al
it
y
rate.
H
ow
e
ver,
de
fects
do
cr
eep
in
durin
g
t
he design as
well
as
pro
duct
ion p
has
es lea
ding t
o
in
crease in
the
f
a
il
ur
e rate
dur
i
ng the
f
ir
st p
has
e.
Figure
1. Fail
ure rate c
urve
as
a fu
nction o
f
t
ime
On
ce
t
he
c
ompone
nt
s
uccess
fu
ll
y
c
omplet
es
the
first
pha
se,
t
he
rate
of
fail
ure
re
mains
flat
f
or
a
portio
n
of
ti
me
w
hic
h
i
nd
ic
at
e
s
the
sta
b
il
iz
at
i
on
i
n
t
he
healt
h
of
t
he
co
mpo
nen
t.
Po
st
us
ef
ul
li
fe
ph
ase
,
f
ai
lure
rate
inc
reases
ex
pone
ntial
ly.
Howe
ver
,
by
t
his
ti
me
t
he
co
mpo
nen
t
migh
t
ha
ve
c
omplet
ed
it
s
inten
ded
pur
po
se
.
A
lot
of
rese
ar
ch
has
bee
n
al
r
eady
car
ried
out
by
resea
rch
e
rs
to
ma
ke
th
e
powe
r
el
ect
ron
ic
s
sy
ste
ms
reli
able,
e
nsur
e
high
a
vaila
bi
li
ty
with
lo
ng
l
ifet
ime
an
d
re
qu
i
rin
g
very
le
ss
mainte
nan
c
e
cost.
Va
rio
us
fa
ult
-
tolerant
desi
gn
an
d
c
on
t
ro
l
s
trat
egies,
patte
rn
rec
ogniti
on
al
gorithms
ha
ve
been
pr
opose
d
for
maki
ng
P
E
sy
ste
ms
reli
ab
l
e
[
9
]
-
[
13].
I
nd
us
trie
s
a
re
fo
c
us
sin
g
on
Desi
gn
f
or
Re
li
abili
ty
[
14]
rathe
r
than
dep
e
ndin
g
on
us
ua
l
wa
y
of
te
sti
ng
for
reli
abili
ty.
Along
with
these
,
re
cent
a
dv
a
nces
in
M
ac
hin
e
Learn
i
ngs
(ML)
ha
ve
sh
ow
n
gr
e
at
pote
ntial
in
ma
king
powe
r
el
ect
ronics
s
ys
te
ms
m
ore
reli
able
[
15
]
-
[
18].
Condit
ion
m
onit
or
i
ng
(CM)
[
19]
is
a
process
of
obs
erv
i
ng
op
e
rati
ng
c
ha
racteri
sti
cs
of
an
el
ect
ric
al
sy
ste
m
t
o
de
te
ct
an
y
a
noma
ly
in
it
s
char
act
erist
ic
s.
For
C
M
,
it
is
imperati
ve
to
ha
ve
decisi
on
maki
ng
al
gorith
ms,
t
hat
de
ci
de
base
d
on
these
current
meas
urements a
nd
histor
ic
al
d
at
a.
In
Fi
gure
2,
th
e
diff
e
ren
ce
be
tween
diag
no
si
s
a
nd
pro
gnos
i
s
is
de
picte
d.
Assessi
ng
the
pr
ese
nt
healt
h
of
a
co
mpo
nent
an
d
predict
in
g
t
he
f
uture
he
al
th
is
te
rme
d
as
P
rog
nosis
[
20]
w
her
ea
s
D
ia
gnos
is
is
the
pr
ocess
of
i
den
ti
f
ying
t
he
natur
e
of
fa
il
ur
e
by
e
xter
na
l
examinati
on
.
For
a
s
uccess
fu
l
C
M
s
ys
te
m
,
accu
rate
pro
gnos
is
plays
imp
or
ta
nt
r
ole.
The
as
sessme
nt
can be ca
rr
ie
d o
ut usi
ng se
nsor
d
at
a
obta
ine
d by m
on
it
ori
ng
a.
com
pone
nt’s
usa
ge
rate
an
d
per
i
od,
a
mb
ie
nt
t
empe
ratu
re
and
humidit
y,
vibrat
ion
a
nd
s
ho
c
k
colle
ct
iv
el
y
te
rmed as c
ompone
nt’s
li
fe
c
ycle en
vir
onm
ent
b.
div
e
rg
e
nce
of
op
e
rati
ng
par
a
mete
rs fr
om
t
he
ir u
s
ual
value
s ch
a
racteri
zed
as
performa
nc
e d
e
gr
a
datio
n
c.
mate
rial
d
isi
nt
egr
at
in
g, oxidi
zat
ion
, i
ncr
ea
s
e in ele
ct
rical
re
sist
ance or t
hresh
old
volt
age.
The
data
s
o
obta
ined
ca
n
th
en
be
a
nalyse
d
us
i
ng
pro
gnos
ti
c
al
gorith
ms,
pre
domina
ntly,
mac
hine
le
arn
in
g
base
d
on
w
hich
c
on
cl
us
io
ns
ca
n
be
draw
n,
the
de
ta
il
s
of
w
hich
will
be
discusse
d
i
n
t
he
s
ubs
equ
e
nt
sect
ion
s
of
th
is
pa
pe
r.
O
utco
me o
f
the
al
gor
it
hm
ca
n
t
he
n
be
us
e
d
f
or
ma
intenanc
e forec
ast
ing
,
fa
ult de
te
ct
ion
and ad
va
nced
warnin
g of fail
ur
es
.
R(t)
Burn
-
in
Useful
Life
t
Wear
out
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
1
,
Ma
rch
20
21
:
5
58
–
56
6
560
Figure
2. Dif
fe
ren
ce
b
et
ween
diag
nosis an
d pro
gnos
is
Figure
3
giv
es
a
glim
ps
e
of
numb
e
r
of
pu
bl
ic
at
ion
s
in
t
he
fiel
d
of
P
ow
e
r
S
ys
te
ms
reli
abili
ty
us
i
ng
M
L
a
ppr
oach
for
the
la
st
te
n
year
s
.
It
ca
n
be
obse
rv
e
d
th
at
M
L
ap
proac
h
in
r
el
ia
bili
ty
has
ga
r
ner
e
d
mu
c
h
more
interest
since
2017.
It
is
a
cl
ear
ind
ic
at
ion
that
s
ci
entifi
c
and
r
esearch
c
omm
un
it
y
has
fou
nd
t
he
pros
pect
a
nd
po
te
ntial
in
ML’s
abili
ty
in
the
fiel
d
of
po
wer
s
ys
te
ms
r
el
ia
bili
ty.
T
here
ha
ve
been
s
ever
al
su
r
ve
ys
publis
hed
for
reli
abili
ty
of
el
ect
rical
sy
ste
m
s
[21
]
-
[
27]
.
H
ow
e
ver,
this
pa
per
give
s
a
broa
d
ove
rv
ie
w
of
progn
os
ti
c
or
proac
ti
ve
m
et
hods
li
mit
ing
the
sc
op
e
to
t
he
us
e
of
M
L
for
reli
abili
ty
i
n
powe
r
el
ect
ro
nics
sy
ste
ms
.
Figure
3. Cha
rt r
e
pr
ese
nts
numb
e
r o
f p
ub
li
c
at
ion
s i
n
the
f
i
el
d
of
p
ow
e
r
s
ys
te
ms
reli
abili
ty usi
ng
M
L
appr
oach. S
our
ce:
ieee
xp
l
or
e.i
eee.o
rg
3.
PRO
GNOSIS
BY
MACHI
N
E L
EAR
NI
N
G
PE
s
ys
te
m
’s
mainte
na
nce
pl
ays
ke
y
ro
le
i
n
t
he
safet
y
of
pe
rs
onnel
an
d
eq
uipment
.
If
the
s
ys
te
m
sh
oul
d
pro
vide
bu
si
ness
c
onti
nu
it
y
of
ser
vice
with
high
eff
ic
ie
nc
y,
th
e
total
cost
of
ownershi
p
na
turall
y
increases
.
M
ai
ntena
nce
ac
ti
vi
ti
es
can
be
broa
dly
cl
assi
fied
i
nto
t
hr
e
e
typ
e
s
Re
act
ive,
P
re
ven
ti
ve
an
d
Pr
e
dicti
ve
wh
i
ch
a
re
summa
rized
in
Table
1
[
28
]
,
[
29
].
Fr
om
t
he
ta
ble
,
it
ca
n
be
in
f
err
e
d
t
hat,
pre
dicti
ve
mainte
na
nce
ha
s
cl
ear
ad
van
t
age
ove
r
ot
her
typ
es
of
maint
enan
ce
a
ppr
oa
ch
es.
Feldma
n
et
al
.
[30]
st
udy
on
a
disp
la
y
s
ys
te
m
of
Boei
ng
73
7
plane
re
vealed
that,
a
R
OI
of
3.5:1
ac
heie
ve
d
wh
e
n
t
he
pr
e
dicti
ve
mai
ntenan
ce
was
em
ploye
d
instea
d
of
re
ac
ti
ve
mainte
na
nc
e.
Also,
in
power
c
onveter
s
sh
ort
ci
rc
uit
an
d
de
gradati
on
fau
lt
s
do not
tri
gg
e
r
a
ny f
a
ult p
ro
te
ct
ion
mecha
nism
which is
an i
de
al
scenar
i
o for
pr
e
dicti
ve
ma
intenanc
e.
Total
Pro
du
ct
i
ve
M
ai
nte
nan
c
e
(TPM
)
[
31]
endo
rsed
as
Ja
pan
e
se
a
ppro
a
ch
to
ef
fecti
ve
mainte
na
nce
mana
geme
nt
de
velo
ped
by
Demin
g
to
e
nhance
ov
e
rall
equ
i
pm
e
nt
e
ffec
t
iveness
(
O
EE)
w
hich
te
nd
t
o
us
e
pr
e
dicti
ve
mai
ntena
nce a
ppr
oa
ches.
Th
e
OE
E can
b
e
d
e
fin
ed
as
OEE
=
Avail
abili
ty
⨯
Pe
rformance Rat
e
⨯
Qu
al
it
y
Ra
te
(2)
w
he
re
A
=
(
RA
−
D
)
RA
⨯
100
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
In
vest
ig
atio
n o
f rel
iab
il
it
y a
ss
esseme
nt in p
ow
er ele
ct
ro
nics
ci
rcuits u
sin
g
...
(
Soumya
Rani Mest
ha
)
561
wh
e
re
A:
Av
ai
la
bili
ty,
RA: R
equ
i
red A
vaila
bili
ty,
D:
D
owntime
PR
=
DC
T
⨯
OT
⨯
100
wh
e
re
PR:
Perf
ormance
Rat
e,
DCT: De
sig
n C
ycle Time
,
OT:
Operati
ng T
ime
QR
=
PI
−
QD
PI
⨯
100
wh
e
re
QR: Q
ua
li
ty Rat
e, PI
:
Pr
od
uctio
n
I
np
ut, QD:
Q
ualit
y Defect,
PI
:
P
rod
uction I
nput
Table
1.
Re
act
i
ve,
p
rev
e
ntive
and
p
re
dicti
ve mai
nte
na
nce t
ypes
Mainten
an
ce T
y
p
e
s
Descripti
o
n
Ap
p
licatio
n
s
Reactiv
e M
ain
ten
a
n
ce
(RM)
Co
rr
ectiv
e bas
ed
,
u
su
ally
r
efer
r
ed
to
as repair
that resto
res the
requ
ired fun
ctio
n
of a
faulty
ite
m
;
Ad
v
an
tag
es
,
Low co
s
t
,
Disad
v
an
tag
es:
a.
Co
st ass
o
ciated
with
replacin
g
the
failed
p
art
co
u
ld
be m
o
re
o
win
g
to th
e m
ain
t
en
an
ce of spare pa
rts inv
en
to
ry
b
.
Po
ss
ib
le secon
d
ary equ
ip
m
en
t da
m
ag
e du
e to th
e c
a
scad
in
g
eff
ect
a.
Sm
all pa
rts and
eq
u
ip
m
en
t
b.
No
n
-
critical
eq
u
ip
m
en
t
c.
Equ
ip
m
en
t un
lik
ely
to fail
d.
Red
u
n
d
an
t
system
s
Preven
tiv
e
Mainten
an
ce (
P
M)
Diag
n
o
stics
bas
ed
,
av
o
id
s an
y
po
ss
i
b
l
e f
ailu
re
by
r
eg
u
lar
in
sp
ectio
n
con
d
u
ct
ed
du
ring
a
sch
ed
u
led
sh
u
td
o
wn
/
still
wo
rkin
g
to
m
in
im
iz
e its i
m
p
act on
bu
sin
ess
o
p
eration
s
Ad
v
an
tag
es:
a.
Bath
tu
b
curv
e c
an
be u
sed
to
p
redict fa
ilu
re
r
at
e of th
e
eq
u
ip
m
en
t b. Flexi
b
ility
allows
f
o
r
th
e adju
stment o
f
m
a
in
ten
an
ce
p
eriod
icity
Disad
v
an
tag
es:
a.
Sin
ce
wear
-
o
u
t
p
eriod
is bas
ed
on
theo
ry rather
th
an
actu
al
d
ata,
PM
beco
m
es
an
exp
en
siv
e strategy
b
.
Labo
r
in
ten
siv
e
a.
Mos
t
fr
eq
u
en
tly
us
ed
eq
u
ip
m
en
ts
b.
Co
n
su
m
ab
les
c.
Kin
d
of equ
ip
m
en
ts h
av
in
g
a his
to
ry o
f
failur
e
s
d.
Manu
facturer
recom
m
en
d
atio
n
s
Predictiv
e M
ain
ten
an
ce
(PdM)
It
is ab
o
u
t equ
ip
m
en
t con
d
itio
n
m
o
n
i
to
ring
us
in
g
adv
a
n
ced s
en
so
r
an
d
ins
trumentatio
n
tech
n
o
lo
g
ies, an
d
its r
ep
etitiv
e anal
y
sis
us
in
g
p
redictiv
e algo
rithms
Ad
v
an
tag
es:
a.
Tho
u
g
h
PM
r
eq
u
ires hig
h
inv
estment, it
is
wo
rth th
e
m
o
n
ey
sin
ce it
p
rov
id
es ex
ten
d
ed
lif
e to th
e
eq
u
ip
m
en
t.
b
.
Prov
id
es a pree
m
p
tiv
e app
roach
f
o
r
safegu
ardin
g
the
eq
u
ip
m
e
n
t.
c.
Red
u
ces th
e do
wn
tim
e of
the eq
u
ip
m
en
t.
Disad
v
an
tag
es:
a.
Incr
eased
inv
estment in
diag
n
o
stic eq
u
ip
m
en
t
a.
Equ
ip
m
en
t with ra
n
d
o
m
failure patte
rns
b.
Critical equ
ip
m
en
t
c.
Kin
d
of equ
ip
m
en
ts th
at
are
less
likely
to
w
ear
an
d
tear
In
p
a
rtic
ular,
P
M
h
as g
ive
n
ri
se
to
a
c
ollec
ti
on
of
met
hodolog
ie
s
,
na
mely,
p
r
obabili
sti
c
appr
oach
a
nd
a
fu
ll
y
data
dri
ven
a
ppr
oac
h
that
reli
es
upon
M
L
[
32]
.
I
n
a
nu
ts
hell,
ML
com
pr
ise
s
of
a
var
ie
ty
of
sta
ti
sti
cal
,
pro
bab
il
ist
ic
a
nd
opti
miza
ti
on
te
ch
niques
that
le
ar
ns
fro
m
t
he
set
of
da
ta
an
d
be
co
m
es
intel
li
gen
t
enou
gh
t
o
make
j
udgeme
nts
with
out
huma
n
inte
rv
e
nt
ion
.
M
L
al
go
rithms
with
e
mphasis
on
non
-
li
near
m
odel
s
li
ke
su
pp
or
t
vect
or
mac
hin
es
(
S
VM),
decisi
on
trees,
lo
gisti
c
re
gr
e
ssio
n
and
arti
fici
al
neural
ne
tw
orks
a
s
pr
e
dicti
ve
mod
el
li
ng
t
oo
ls
ha
ve
great
er
pred
ic
ti
ve
pe
rform
ance
an
d
are
quit
e
po
pu
la
r
a
mong
researc
he
rs
[
33
]
,
[
34]
.
Ta
ble
2
a
ptly
s
ummari
z
es
the
rece
nt
publica
ti
ons
on
M
L
al
gorith
ms
us
e
d
in
pro
gn
os
is
of
PE
ci
rc
uits.
It
is
obse
rv
e
d
th
at
a
c
ombinati
on
of
M
L
al
gorith
ms
is
us
e
d
t
o
boost
the
ef
fici
enc
y
of
the
a
ppro
ac
h.
F
or
exam
ple,
S
V
M
is
co
mputat
io
nally
hea
vy,
he
nce
requires
more
trai
ning
t
ime.
By
intr
od
ucin
g
le
ast
s
qu
are
t
o
the cost
fun
ct
i
on, th
e
comp
ut
at
ion
al
c
omplexit
y
is
reduce
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
1
,
Ma
rch
20
21
:
5
58
–
56
6
562
Table
2.
Lit
era
ture revie
w on
M
L a
ppr
oac
h
f
or r
el
ia
bili
ty in
PE
s
ys
te
m
’s
Sl.
No
Refere
n
ces
Machin
e L
e
arnin
g
Techn
iq
u
es
Co
m
p
o
n
en
t/Sy
ste
m
Pros
and
Co
n
s (As claimed
by
the
resp
ectiv
e auth
o
rs)
1
Aravid
Sai
S
arathi
Vasan
,
et
al [
35
]
-
[
37
]
Least Squ
are
SV
M
(L
S
-
SVM)
Ban
d
p
ass
and
L
o
w
Pass
f
ilte
rs
•
Used
to ev
alu
ate R
UP
•
Ear
ly
f
au
lt detectio
n
and
iso
latio
n
•
Decr
eases
comple
x
ity
2
Xi
-
Sh
an
Z
h
an
g
et.al,
[
38
]
Su
p
p
o
rt
Vector M
achi
n
es
Co
m
p
lex
electr
o
n
i
c
sy
stem
Biq
u
ad
f
ilter
•
Lon
g
ter
m
op
er
ati
o
n
of the
sy
stem
•
Reliab
ility
analy
sis
us
in
g
Kap
lan
-
Meie
r
(
K
M)
and
K
ernel
d
en
sity
E
stimation
(
KD
E)
3
LanHai and
LiuHo
n
g
-
d
a,
et.
al
,
[
39
]
SVM
an
d
P
rincip
al
Co
m
p
o
n
en
t Analysis (PCA)
Three
-
Ph
ase r
ectif
ier
circuits
•
Im
p
rov
es g
en
erali
zatio
n
ability
•
Cap
ab
le of locatin
g
faults
p
recisely
4
Jian
ch
en
Wan
g
,
et
al [
40
]
Ch
ao
s th
eo
ry an
d
Particle
Swar
m
Opti
m
izati
o
n
(CPSO)
-
SV
M
Elliptical fi
lter
circ
u
it
•
Im
p
rov
es ef
ficienc
y
•
Execu
tio
n
tim
e is l
ess
5
Sh
ao
wei Ch
en
,
et
al,
[
41
]
Gen
etic Algo
rithm
(
GA
)
-
SVM
Qu
ad
hig
h
pas
s filter
circuit
•
Preven
ts d
ep
en
d
en
ce of la
rge
trainin
g
sam
p
les
•
Better su
ccess
rate
of
d
iag
n
o
si
b
ilty
6
Qin
g
feng
M
a,
e
t al
[
42
]
Decisio
n
T
ree
(
D
T
)
an
d
BSVM
Sallen
-
k
ey
ban
d
p
ass
filter
Activ
e ban
d
-
sto
p
filter
circuit
•
Execu
tio
n
tim
e is l
ess
•
Testin
g
acc
u
racy i
s h
ig
h
8
Tang
Jin
g
y
u
an
,
et
al [
43
]
SVM
an
d
Adab
o
o
st
Two
-
stag
e f
o
u
r
op
-
am
p
biq
u
ad
low
-
p
ass
filter
•
Clas
sificatio
n
acc
u
racy is hi
g
h
10
W
EI
H
E,
et al
,
[44
]
Mehrd
ad
Big
larbeg
ian
,
et al
[
45
]
Naïv
e Bay
es Clas
s
ifier
Op
am
p
biq
u
ad
f
ilt
er
circuit.
Galliu
m
Nitride
(GaN
)
trans
isto
rs.
•
Ef
fe
ctiv
e f
au
lt
diag
n
o
sing
•
Hig
h
latency
•
Enh
an
ces sy
stem
r
eliab
ility
11
Pio
tr
Bils
k
i [
4
6]
Ran
d
o
m
Forest (R
F)
5
th
ord
er
lo
wp
ass
filter
.
•
Used
to d
etect par
am
et
ric
f
au
lts
•
Hig
h
acc
u
racy
13
Seo
n
g
m
in
Heo,
[
47
]
Mehrd
ad
Big
larbeg
ian
[
4
8
]
ANN
Recu
rr
en
t Neur
al
Netwo
rk
(RNN)
Neu
ral
n
etwo
rk
class
ifier
s
-
Tenn
ess
ee E
ast
m
a
n
(T
E
)
Galliu
m
Nitride
(GaN
)
p
o
wer
co
n
v
erter
s
•
Increased
f
au
lt det
ectio
n
accuracy
•
Better f
au
lt detecti
o
n
and
class
ification
15
Q.
Su
n
,
et
al,
[
49
]
Cro
w Sea
rch Alg
o
rithm
-
LSSV
M
Cap
acito
r
-
o
p
en
loo
p
Bo
o
st co
n
v
erter
•
Hig
h
compu
tatio
n
al efficien
cy
•
Go
o
d
estim
atio
n
a
ccuracy
16
W
.
Ch
en
,
et al
[
50
]
PCA (Uns
u
p
ervis
e
d
alg
o
rithm)
SiC
-
M
OSFE
T
•
Used
f
o
r
o
f
fline as
well
as
o
n
lin
e f
au
lt detecti
o
n
17
B. Gou
,et
al
[
51
]
IGBT
3
-
p
h
ase
PW
M
in
v
erter
Ran
d
o
m
Vector
Fu
n
ctio
n
al L
ink
(RVFL
)
n
etwo
rk
•
Fau
lt pred
ictio
n
ac
cu
racy of
9
8
.83
%
•
Ap
p
lied
to n
o
n
-
lin
ear
sy
stems
4.
REM
AI
NING
U
SEF
UL LIF
E (RUL)
Fo
r
a
n
ef
fici
en
t
pro
gnos
is
,
e
sti
mati
on
of
RU
L
plays
a
crit
ic
al
r
ole.
RUL
c
an
be
def
i
ned
a
s
numb
e
r
o
f
pro
du
ct
ive
hours
le
ft
in
a
co
mpon
e
nt
at
a
point
of
ti
me
w
hile
it
is
op
e
ra
ti
ng
.
It
can
be
al
so
te
rme
d
as
us
ef
ul
ti
me
le
ft
ti
ll
ne
xt
mai
ntena
nce
.
Ba
se
d
on
ho
w
t
he
avail
able
in
formati
on
is
use
d,
t
he
pro
gnos
ti
c
meth
od
ologies
are
cl
assifi
e
d
i
nto
model
or
phys
ic
s
-
dr
ive
n,
data
or
mac
hine
le
arn
i
ng
-
dr
iv
en
a
nd
hybr
i
d
appr
oach
es
[
5
2
]
-
[
54
]
.
In
data
dr
ive
n
method
ology,
degra
dation
c
ha
racteri
sti
cs
ar
e
co
mputed
ba
sed
on
t
he
c
hr
onologica
l
se
nsor
data
to
trai
n
t
he
s
yst
em
model
tha
t
may
be
us
ed
to
co
mpute
R
UL
of
t
he
co
m
pone
n
t.
Widel
y
ap
plied
al
gorithms
include
Ga
us
si
an
proce
ss
[
55
]
,
[
56
],
S
V
M
,
Least
S
quare
SVM
(L
SS
V
M
)
[
57
]
,
ne
ur
a
l
net
work
s
[
58
]
,
[
59
],
gamma
proces
ses
[
60
]
an
d
Hidden
M
a
rkov
Mod
el
s
(
HMMs)
[
61
].
Physics
base
d
ap
proac
h
de
man
ds
su
bst
antia
l
pr
i
or
underst
an
din
g
a
bou
t
ph
ys
ic
al
sy
ste
ms
w
hich
is
rar
e
t
o
fi
nd
in
pr
act
i
ce.
T
he
mat
he
mati
cal
models
are
bui
lt
on
first
pr
i
nc
iple
or
c
ompr
ehensi
on
of
c
omp
on
e
nt’s
fail
ur
e
mec
han
is
m
.
E
yr
in
g
model
[
62
],
Weib
ull
distri
bu
ti
on
[
6
3
],
pa
rtic
le
filt
er
[
6
4],
Ba
yesia
n
infe
re
nce
-
bas
ed
meth
o
ds
[
65
]
a
re
some
of
t
he
commo
nly
us
e
d
al
go
rithms
in
phys
ic
al
m
odel
ing
a
ppro
ac
h.
H
ybrid
m
odel
s
are
t
he
c
ombinati
on
of
both
t
he
Data
dri
ve
n
ap
proac
h
an
d
P
hysic
al
modeli
ng
ba
sed
a
ppr
oa
ch.
In
case
of
non
-
li
near
s
yst
ems,
hy
br
id
models
can
scal
e
from
com
pone
nt
l
e
vel
to
s
ys
te
m
le
vel
[
66
].
Ba
s
ed
on
t
he
f
ai
lure
modes
,
a
com
pone
nt
ca
n
hav
e
var
i
ou
s
d
et
e
rio
rati
on curve
s
wh
ic
h
mi
gh
t
re
su
lt
in varie
d
R
UL
[
67
].
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
In
vest
ig
atio
n o
f rel
iab
il
it
y a
ss
esseme
nt in p
ow
er ele
ct
ro
nics
ci
rcuits u
sin
g
...
(
Soumya
Rani Mest
ha
)
563
The
fo
ll
owin
g
sect
ion
discu
sses
the
wor
k
carrie
d
out
on
P
dM
f
or
rel
ia
bili
ty
assess
ment
of
PE
sy
ste
ms
.
Vasa
n,
et
al
[
35
]
use
d
LS
SVM
al
gorith
m
to
a
ddr
ess
the
c
on
ce
r
ns
of
t
he
ci
rc
uit
fail
ur
e
by
pre
dicti
ng
and
is
olati
ng
f
aults.
T
hey
al
so
est
imat
ed
t
he
RU
P
(
(Re
mainin
g
Usefu
l
Performance
)
by
us
in
g
Ba
yesian
M
onte
Ca
rlo
a
ppr
oach
f
or
th
e
filt
er
ci
rcu
it
s
.
T
hus,
ai
ding
the
pr
e
ve
ntion
of
s
ys
te
m
fail
ur
es
[
36
]
,
[
3
7
]
.
Xi
-
Sh
a
n
Z
hang,
e
t
al
[
38
]
pro
pose
d
fa
ult
pro
gnos
ti
c
te
ch
ni
qu
e
to
reali
ze
healt
h
ma
na
ge
ment
of
t
he
c
omplex
el
ect
ro
nic
e
qu
ipment
us
in
g
SVM
al
gorith
m.
Re
li
abili
ty
analysis
is
done
us
i
ng
Ka
plan
-
M
ei
e
r
(
K
M
)
a
nd
Kernel
de
ns
it
y
Esti
mati
on
(
K
DE)
te
c
hn
i
que.
Com
bin
in
g
S
VM
al
gorith
m
and
Pr
i
ncipal
Com
pone
nt
A
nalysis
(P
CA
)
it
is
po
s
sible
to
locat
e
the
po
sit
io
n
of
the
po
wer
s
yst
em
fa
ults.
It
will
al
so
hel
p
i
n
i
den
ti
f
y
in
g
the
t
yp
e
of
t
he
fa
ult
an
d
re
duce
the
i
nt
err
upti
on
of
t
he
s
ys
te
m
[
39
]
.
Jia
nc
hen
Wa
ng,
et
al
[
40
]
pro
posed
C
ha
os
theo
ry
and
Partic
le
Sw
ar
m
Op
ti
m
iz
at
ion
(CPS
O
)
-
S
V
M
to
e
nhance
t
he
s
yst
em
performa
nce
by
reduci
ng
t
he
execu
ti
on
ti
me
.
Re
li
abili
ty
an
al
ys
is
can
al
s
o
be
done
us
in
g
genet
ic
al
gorithm
(GA).
GA
can
al
so
be
use
d
to
increase
t
he
s
uccess
rate
in
fau
lt
dia
gnosi
s
[
41
]
.
Usi
ng
decisi
on
tree
al
gorithm,
e
xe
cution
ti
me
can
be
minimi
zed
,
the
reby
im
pro
vin
i
g
t
he
e
ff
ic
ie
nc
y
of
the
s
ys
te
m
[
42
]
.
C
ombinin
g
SVM
an
d
A
da
boos
t
al
gorith
m
yields
bette
r
r
el
ia
bili
ty
and
high
cl
assifi
ca
ti
on
acc
uracy
[
43
]
.
A
pro
ba
bili
sti
c
cl
assifi
er,
Naïve
Ba
ye
s
(
NB)
al
gorithm
pro
vi
des
acc
ur
at
e
r
esults
a
nd
co
nsumes
le
ss
trai
ni
ng
ti
me
[
44
]
,
[
45
].
T
he
a
ppr
oa
ch
is
us
e
d
to
detect
par
a
metri
c
fa
ul
ts
i
n
the
fifth
orde
r
lo
w
pas
s
filt
er,
RF
is
the
f
av
or
a
ble
cl
assifi
cat
ion
ap
proac
h
with
high
eff
ic
ie
nc
y
eve
n
on
t
he
quit
e
small
data
se
ts
[
4
6].
Ne
ura
l
netw
ork
cl
as
sifie
r
a
nd
Gall
ium
Nitride
(
GaN
)
conve
rters ar
e
us
e
d
in r
el
ia
bil
it
y
analysis fo
r
b
et
te
r
fa
ult cl
as
sific
at
ion
a
nd
detect
ion. Ga
N
-
base
d dev
ic
es
hav
e
incre
dib
le
pe
rformance
a
nd
e
xh
i
bit
bette
r
m
at
erial
prop
e
rtie
s
when
c
omp
ared
t
o
th
os
e
dev
ic
es
made
up
of
sil
ic
on
.
Using
GaN
de
vice
w
ou
l
d
be
highly
us
ef
ul
f
or
po
wer
e
ngineer
s
in
en
han
ci
ng
t
he
reli
abili
ty
of
th
e
sy
ste
m
[
47
]
,
[
48
].
C
row
Se
arch
Al
gorith
m
-
LS
SVM
is
novel
a
ppr
oac
h
wh
ic
h
yield
s
high
c
omp
utati
on
al
eff
ic
ie
nc
y
f
or
boos
t
co
nverte
rs
[
49
].
A
n
un
su
pe
r
vised
al
gorith
m
is
use
d
f
or
fau
lt
pro
gnos
is
wh
e
re
onli
ne
a
s
well
as
offli
ne
fa
ults
can
be
detect
ed
[
50
].
Fast
F
ourier
T
ran
s
f
or
m
s
(
FF
T)
is
us
e
d
by
IG
BT
3
-
phase
PWM
inv
e
rter
to
e
xtr
act
the f
a
ult f
re
qu
e
nc
y
s
pectr
um o
f
t
hr
ee
-
pha
se cu
rr
e
nts.
5.
RESU
LT
S
AND DI
SCUS
S
ION
M
ore
tha
n
150
pap
e
rs
wer
e
r
eviewe
d
an
d
67
of
the
m
are
mentio
ned
i
n
the
re
fer
e
nce
to
exp
la
in
t
he
sign
ific
a
nce
of
machine
le
a
r
ning
in
the
reli
abili
ty
do
mai
n.
M
L’
s
us
e
i
n
PEC
reli
abil
it
y
comes
with
both
chall
enges
an
d
opport
un
it
ie
s.
Pr
og
nosis
re
quires
li
ve
co
nd
it
ion
m
onit
or
i
ng
wh
ic
h
ma
y
be
a
chall
eng
e
i
n
it
sel
f
du
e
t
o
acce
ssi
bili
ty
an
d
en
vi
ronme
nt
co
ndit
ion
s.
M
L
al
gorith
ms
are
no
t
scable
in
a
way
t
hat
a
pa
r
ti
cular
al
gorithm
is
tr
ai
ned
an
d
te
ste
d
for
l
ow
e
r
rated
dev
ic
e
ma
y
no
t
be
su
it
able
for
higher
rate
d
dev
ic
e
.
M
aj
ori
ty
of
the
rev
ie
we
d
pap
e
rs
hav
e
publis
hed
th
ei
r
resu
lt
s
ba
sed
on
t
he
la
borat
ory
co
ndit
ion
s
or
us
in
g
sim
ulati
on
so
ft
war
e
. Ho
w
ever
real
-
w
or
ld
scena
rios ma
y va
ry.
6.
CONCL
US
I
O
N
W
it
h
t
he
c
omplexit
ie
s
in
vo
l
ved
in
PE
S
yst
ems,
t
heir
sa
fety,
mainte
na
nce
a
n
d
reli
a
bi
li
ty
are
the
major
c
oncer
n
s
.
This
pap
e
r
f
ocuses
on
provi
din
g
a
re
view
of
reli
abili
ty
assessme
nt
for
P
E
sy
ste
ms
us
in
g
M
L
te
chn
iq
ues
.
Th
e
adv
a
ntage
s,
disad
va
ntages
and
a
ppli
cat
ion
s
of
var
io
us
typ
es
of
maint
enan
ce
sche
m
es
are
discusse
d
i
n
de
ta
il
.
A
p
ara
di
gm
s
hifts
to
w
ard
s
us
e
of
ML
has
been
ob
serv
e
d
in
t
he
appr
oach
of
ha
nd
li
ng
reli
abili
ty
co
nc
ern
s
in
PEC
.
S
ever
al
M
L
al
gorith
m
s
ha
ve
pro
ve
n
thei
r
e
ff
i
c
a
cy
i
n
t
he
a
re
a
of
reli
abili
ty
and
in
the
bette
r
fa
ul
t
predict
io
n
m
od
el
s
.
Pr
e
dicti
on
of
fau
lt
s
ta
ke
cauti
onar
y
measu
res
to
a
vo
i
d
sig
nifica
nt
a
nd
insubstanti
al
lo
sses
in
the
s
ys
t
em
.
C
ombinin
g
M
L
al
gorith
m
yield
s
bette
r
res
ults
in
ac
hi
evin
g
highly
re
li
able
PE
s
ys
te
ms
.
Fi
nd
i
ng
R
UL
it
sel
f
is
a
c
halle
ngin
g
ta
s
k
.
H
oweve
r
,
it
pro
vid
es
an
in
sig
ht
into
t
he
healt
h
of
th
e
sy
ste
m
.
T
his
li
te
ratur
e
re
view
has
bee
n
de
ve
lop
e
d
to
in
vest
igate
va
rio
us
methods
in
as
s
essing
the
reli
abili
ty
of PE system
s
us
in
g
ML
ap
proach f
or
the
be
nef
it
of
powe
r e
ng
i
neer
s
and
r
esearche
rs
.
REFERE
NCE
S
[1]
J.
D.
van
Wyk
a
nd
F.
C.
Lee,
“On
a
future
for
po
wer
el
e
ct
roni
cs
,
”
IEE
E
J.
Eme
rg.
Sel
.
Top.
Powe
r E
lectron
,
vol
.
1
,
no.
2
,
pp
.
59
–
72
,
2013
.
[2]
J.
C
.
B
al
d
a
an
d
A.
Mantoo
th
,
“Power
-
semic
onduct
or
dev
ic
e
s
and
co
mponents
for
new
p
ower
conve
rt
er
deve
lop
me
nts:
a
key
en
abler
fo
r
ult
rah
igh
eff
ic
i
e
ncy
power
el
e
ct
r
onic
s
,
”
IEEE
Po
wer
El
e
ct
ron.
M
ag
,
vol
.
3
,
no.
2
,
pp.
53
–
56
,
2016
.
[3]
Y
.
Song
and
B
.
W
ang,
“Surve
y
on
Rel
i
abi
l
it
y
of
Pow
er
El
e
c
troni
c
Sys
tems
,
”
IEEE
Tr
ansacti
ons
On
Power
El
e
ct
ronics
,
vol
.
28,
no
.
1,
Jan
.
2
013
.
[4]
H.
W
ang
et
al
.
,
“
Tra
nsi
ti
oning
t
o
Phys
ic
s
-
of
-
Fai
lure
as
a
R
el
i
abili
ty
Drive
r
in
Pow
er
E
le
c
troni
cs
,
”
in
IE
EE
Journ
al
of
Eme
rging
and
Selec
t
ed
Topi
cs
in
Pow
er
Elec
tronic
s
,
vo
l. 2, no. 1, pp. 97
–
114,
Mar
.
2014
.
[5]
L.
M.
Moore
an
d
H.
N.
Pos
t,
“F
ive
yea
rs
of
ope
ra
ti
ng
exp
erienc
e
at
a
l
arg
e
,
util
it
y
-
sca
le
photov
olt
aic
generating
pla
nt
,
”
Prog.
Ph
otov
olt.: Res. Ap
pl
,
vo
l.
16
,
no.
3
,
pp.
249
–
259
,
20
08
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
1
,
Ma
rch
20
21
:
5
58
–
56
6
564
[6]
IEE
E
Standa
r
d
Dict
ion
ary
of
Elec
tr
ic
a
l and Elec
troni
cs
Te
rms
,
I
EE
E
,
Std
100
-
19
96.
[7]
R.
K.
Mobley
,
“
An i
ntroduction to
pr
ed
ic
t
ive m
a
i
nte
nan
ce
,”
El
sev
ie
r
.
2002.
[8]
M
.
S.
Alvar
e
z
-
Alvar
ado
et
al
.
,
“Bathtub
cur
v
e
as
a
ma
rkov
ian
proc
ess
to
de
scribe
the
re
liab
il
it
y
of
rep
ai
rab
le
com
ponen
ts,”
IE
T Gene
ration, Tr
ansm
ission Distr
ibut
ion
,
vol
.
12
,
no.
21
,
pp
.
5683
–
5689,
2018
.
[9]
R.
Peuget
e
t
al
.
,
“Faul
t
detec
ti
on
and
isol
at
ion
on
a
PWM
inv
ert
e
r
by
knowledge
-
base
d
model,
”
I
EE
E
Tr
ans.
Ind
.
Appl
,
vol
.
34
,
no
.
6
,
pp
.
1318
–
13
26,
1998
.
[10]
D.
Diallo
e
t
al
.
,
“Faul
t
d
etec
t
ion
and
dia
g
nosis
in
an
induction
machine
d
rive:
A
p
at
t
ern
re
cogni
t
io
n
appr
oa
ch
base
d
on
conc
ord
ia sta
tor
m
ea
n
cur
r
ent ve
c
tor,”
I
EE
E
Tr
ans.
Ene
rgy
Co
nv
,
vo
l. 20, no. 3
,
pp
.
512
–
519
,
S
ep.
2005
.
[11]
S.
Khomfoi
e
t
a
l
.
,
“Fault
di
agno
sis
and
rec
onfig
ura
ti
on
for
mult
il
ev
el
inv
erter
d
rive
using
AI
-
ba
sed
te
chn
ique
s,
”
IEE
E
Tr
ans.
Ind
.
E
lectron
,
vol. 5
4,
no
.
6
,
pp
.
295
4
–
2968,
De
c. 20
07.
[12]
Q.
T
.
An
e
t
a
l
.
,
“Switc
hing
func
ti
on
mode
lb
ase
d
fast
-
d
ia
gnosti
c
me
thod
of
open
-
sw
it
ch
f
aul
ts
in
i
nver
te
rs
without
sensor
s,”
IE
EE
Tr
ans.
Powe
r E
l
ec
tron
,
vol
.
26
,
no.
1
,
pp
.
119
–
1
26,
Jan
.
2010
.
[13]
O.
Wallmark
e
t
al
.
,
“Cont
ro
l
a
lg
orit
hms
for
a
fa
ult
-
tolera
n
t
PM
SM
drive
,
”
IE
EE
Tr
ans.
Ind.
Elec
tron
,
vol
.
54,
no
.
4,
pp
.
1973
–
198
0,
Aug.
2007.
[14]
P.
O'Connor
e
t
a
l
.
,
“Pra
ct
i
cal
r
el
i
abi
li
ty engi
ne
ering,
”
the 5 editio
n,
John
Wi
l
ey
&
Sons
,
2012.
[15]
Y.
Chen
e
t
al
.
,
“Mode
li
ng
and
opti
mization
of
com
pl
ex
buil
d
in
g
ene
rgy
sys
tem
s
with
dee
p
n
eu
ral
ne
tworks,”
i
n
Signal
s,
S
yste
ms
,
and
Computers,
2017
51st
Asi
lo
mar
Confe
renc
e
on.
IE
EE
,
2017
,
pp.
1368
–
13
73
.
[16]
X.
Y.
Chen
et
a
l
.
,
“An
unsuperv
ised
dee
p
learni
ng
appr
oac
h
for
sce
nar
io
fo
recast
s,”
Powe
r
S
ystem
s
Computati
on
Confe
renc
e
(
PS
CC)
,
2018.
[17]
T.
Hong
et
al
.
,
“Proba
b
il
ist
ic
ene
rgy
fore
ca
sti
ng:
Glob
al
en
er
gy
fore
ca
sting
com
pe
ti
ti
on
201
4
and
beyon
d,
”
Inte
rnational
Jo
urnal
of Forec
as
ti
ng
,
vol. 32
,
no
.
3,
pp.
896
–
913,
2016.
[18]
Y.
W
ang
et
a
l
.
,
“De
ep
learni
ng
-
base
d
socio
-
d
emographi
c
infor
m
at
ion
ide
n
ti
fica
tion
from
sma
rt
me
t
er
d
at
a
,
”
IE
E
E
Tr
ansacti
ons on Sm
art Grid
,
201
8.
[19]
Y.
Han
e
t
al
.
,
“
Condit
ion
monit
oring
te
chn
ique
s
for
e
le
c
trica
l
eq
uipm
ent
-
a
liter
ature
survey
,
”
I
E
EE
Tr
ans.
Pow
er
Del
,
vo
l. 18, no.
1,
pp
.
4
–
13
,
Jan
.
2003.
[20]
T.
Bro
the
r
ton
e
t al
.
,
“Prognosis o
f
fau
l
ts i
n
gas
tur
bine
engi
nes
,
”
in
Proc. I
E
EE A
er
osp.
Conf
.
,
2000
,
pp
.
163
–
171
.
[21]
Y
.
R
an
et
a
l
.
,
“
A
Survey
of
Predictive
Ma
int
e
nanc
e
:
Sys
te
ms,
Purpos
es
and
Approac
hes
,
”
ar
Xiv:
1912
.
0738
3
,
2019
.
[22]
S.
Yang
et
a
l
.
,
“
An
industry
-
bas
ed
survey
o
f
re
libili
ty
in
power
e
le
c
troni
c
conv
erters,”
T
h
e
ECC
E
Conf.
R
ec
.
,
San
Jos
e,
CA,
Sep
.
2
009.
[23]
S.
Khan
et
al
.
,
“
A
rev
i
ew
on
the
applic
at
ion
of
d
ee
p
learni
ng
in
sys
te
m
h
ealth
manage
m
ent
,
”
Me
c
hanic
al
Syst
em
s
and
Signal P
roc
essing
,
vol
.
107
,
pp.
241
–
265
,
20
18.
[24]
S.
Zha
ng
et
al
.
,
“Ma
ch
ine
le
a
rn
ing
and
d
ee
p
learni
ng
al
gori
th
ms
for
be
ari
ng
fau
lt
d
ia
gnosti
cs
-
a
co
mpre
hensiv
e
rev
ie
w,
” arXiv
p
rep
rint a
rXiv:
19
01.
08247,
2019.
[25]
S.
Kat
ipa
mu
la
et
al
.,
“Me
thods
f
or
fau
lt
detec
t
io
n,
di
agnostics,
a
nd
pr
ognostic
s
f
or
buil
d
ing
sys
tems
:
a
r
evi
ew,
p
a
rt
i,
”
Hvac
&R
Re
s
earc
h
,
vo
l. 11, n
o.
1
,
pp
.
3
–
25
,
2
005.
[26]
Y.
L
ei
et
al
.
,
“
Mac
hine
ry
he
al
t
h
prognostic
s:
A
sys
te
ma
t
ic
re
vie
w
from
da
ta
ac
qu
isit
ion
to
rul
pre
d
ic
t
ion,
”
Me
chanical
S
yst
ems and
Signal
Proce
ss
ing
,
vo
l.
104,
pp
.
799
–
83
4,
2018
.
[27]
Y.
Peng
e
t
al
.
,
“Curr
ent
st
atus
of
machine
prognostic
s
in
condi
t
ion
-
base
d
ma
in
te
n
anc
e
:
a
r
evi
ew
,
”
T
he
Inte
rnational
Jo
urnal
of Adv
an
c
ed
Manufa
ct
urin
g
Technol
og
y
,
v
ol.
50
,
no
.
1
–
4
,
p
p.
297
–
313
,
201
0.
[28]
Bara
j
as
et
al
.,
“Re
a
l
-
Time
Dia
gnostic
s,
Progn
ostic
s
and
Hea
l
th
Mana
g
em
en
t
for
L
arg
e
-
Sc
ale
Manufa
ct
ur
in
g
Maint
en
anc
e
S
ystem
s
,
”
Proc
e
edi
ngs
of
th
e
2008
In
te
rnat
ional
Manu
fac
t
uring
Scienc
e
and
Engi
n
ee
rin
g
Confe
renc
e
,
Evans
ton
IL
,
Oct
.
20
08
.
[29]
X.
Jin
et
al
.
,
“The
Present
St
at
u
s
and
Future
Gr
owth
of
Mainte
nanc
e
in
US
Ma
nufa
ct
ur
i
ng:
R
esult
s
from
a
Pilo
t
Survey
,
”
Manuf
act
uring
Revi
ew
,
vol
.
3
,
pp.
1
–
10
,
2016
in
X.
Jin
,
B
.
A.
We
iss,
D
.
Siege
l
,
J
.
Lee,
a
nd
J
.
Ni
,
Pr
ese
nt
Stat
us
and
Fut
ur
e
Gr
owth
o
f
Advance
d
Maintena
nce
Technol
og
y
Strate
gy
in
US
Manufac
turing
,
vol.
7
,
no.
12
,
p
p.
1
–
18
,
2016
.
[30]
Feldm
an
et
al
.
,
“T
he
Ana
lysis
of
Ret
urn
on
Inv
e
stme
nt
for
PH
M
Applie
d
to
Elec
troni
c
Sys
tems
,
”
Proce
ed
ings
of
the
Inte
rnat
ion
al
Conf
ere
nc
e
on
Pronosti
cs
and
Health
Manage
men
t
,
Denv
er,
C
O.
Oct
.
2008
.
htt
p://ieeexp
lore.
ie
e
e.
org/do
cume
nt/
4711415/
[31]
R.
K.
Mobley
,
“
An i
ntrodu
c
ti
on
to
pre
d
ic
t
ive m
a
i
nte
nan
ce
,”
El
sev
ie
r
.
2002.
[32]
M.
Kraus
et
al
.
,
“
Forec
ast
ing
re
ma
ini
ng
use
ful
l
ife
:
in
te
rpr
etable
dee
p
l
ea
rning
a
pproa
ch
vi
a
var
i
at
ion
al
Bay
esia
n
infe
ren
ce
s,
”
De
c
is.
Support.
Syst
,
vol.
125
,
2019.
[33]
Das
K
.
et
al
.,
“
A
Survey
on
Mac
hin
e
Learni
ng
:
Conce
p
t,
Algo
rit
hms
and
Appl
ic
a
ti
ons,
”
Inte
rn
ati
onal
Journal
of
Innov
ative Re
se
arch
in
Comput
e
r and
Comm
unication Engi
ne
erin
g
,
v
ol
.
5
,
no
.
2
,
2
017.
[34]
D.
Dele
n
et
al
.
,
“
A
com
par
ativ
e
analysis
of
m
ac
hin
e
learni
ng
sys
te
ms
for
measuring
the
im
p
ac
t
of
knowl
edg
e
ma
nag
em
en
t
pr
a
ct
i
ce
s
,
”
De
ci
sion
Support
Syst
ems
,
vol.
54
,
no.
2,
pp.
1150
–
1160
,
2012
.
[35]
A
.
S
.
S
.
Vasan
et
al
.
,
“Di
agnost
ic
and
Prognos
tic
Methods
for
Analog
El
e
ct
roni
c
s
Circ
uit
s
,
”
I
EEE
transacti
on
on
Industrial
E
le
c
tronic
s
,
vo
l.
60,
no
.
11
,
Nov.
2
013
.
[36]
B
.
Long
et
al
.
,
“
Resea
rch
on
Fea
ture
s
for
Diagno
stic
s
of
Filt
ere
d
Analog
Circ
u
it
s
Based
on
LS
-
SVM
,
”
2011
IE
EE
AUTO
TEST
C
O
N
,
Ba
lt
i
more,
20
11,
pp
.
360
–
366
.
[37]
A
.
S
.
S
.
Vasan
et
al
.
,
“
Expe
r
imental
V
al
id
at
ion
of
LS
-
SV
M
Ba
sed
Faul
t
Ide
nt
if
ic
a
ti
on
in
Ana
lo
g
Cir
cui
ts
Us
ing
Freque
ncy
Fe
at
u
res
,
”
Engi
n
ee
rin
g
Asset
Manag
e
ment
,
pp
.
629
–
6
41,
2011
.
[38]
X
.
S
.
Zha
ng
et
al
.
,
“
Fau
lt
Prognos
ti
c
Te
chnogy
of
Comp
le
x
Elec
tron
ic
Equi
p
m
ent
for
PH
M
,
”
2
013
Inte
rnat
ion
al
Confe
renc
e
on
Qualit
y
,
R
eliability,
R
isk, Mai
n
tenance
,
and
Sa
fe
t
y
Eng
ine
ering
,
C
hengdu,
2013
,
p
p.
1790
–
1792
.
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
In
vest
ig
atio
n o
f rel
iab
il
it
y a
ss
esseme
nt in p
ow
er ele
ct
ro
nics
ci
rcuits u
sin
g
...
(
Soumya
Rani Mest
ha
)
565
[39]
L
.
Hong
-
da
et
al
.
,
“In
t
el
l
ige
nt
Fault
Diagnos
is
Method
in
Control
le
d
Rect
ifi
ers
Ba
sed
on
Support
Vec
to
r
Mac
hine
s,
”
201
0
Sec
ond
WRI
G
lobal
Congress
on
Intelli
g
ent Sys
te
ms
,
Wuh
an, 2
010,
pp
.
235
–
23
8
.
[40]
J
.
W
ang
et
al
.
,
“
Analog
c
irc
u
it
s
Fault
s
Diagnosis
Method
Based
on
CP
SO
-
SVM
,
”
2011
Inte
rnat
i
onal
Conf
ere
nc
e
on
Quality,
Reli
abil
ity,
Ri
sk
,
Ma
int
enan
ce
,
and
S
afe
t
y
Eng
ineerin
g
,
Xi’a
n
,
20
11,
p
p.
465
–
468
.
[41]
S.
Chen
e
t
a
l
.
,
“A
new
an
al
og
ci
rcu
it
fau
l
t
diagnos
is
appr
oa
ch
base
d
on
GA
-
SVM
,
”
2013
I
E
EE
Inte
rnat
iona
l
Confe
renc
e
of
I
EE
E
Re
g
ion
10
(
TENCON
2013)
,
Xi’a
n,
2013
,
pp
.
1
–
4
.
[42]
Q
.
Ma
et
al
.
,
“
A
new
de
ci
sion
tr
ee
appr
o
ac
h
of
s
upport
v
ector
m
ac
hin
e
for
analo
g
c
irc
u
it
fau
lt
di
agnosis
,
”
Ana
lo
g
Inte
gr Circ
Sig
Proce
ss
,
vol
.
88
,
pp.
455
–
463,
20
16
.
[43]
T
.
Jingyuan
et
al
.
,
“
Analog
C
irc
u
it
Fault
Di
agnosi
s
Us
ing
AdaBoost
and
SV
M
,
”
20
08
Inte
rnationa
l
Confe
renc
e
on
Comm
unic
ati
ons,
Circu
it
s and
Sy
stems
,
Fujia
n
,
20
08,
pp
.
1184
–
11
87
.
[44]
W
.
H
e
et
al
.
,
“
A
Naive
-
B
aye
s
-
Based
Faul
t
Di
agnosis
Approa
ch
for
Analog
Circ
uit
by
Us
ing
Ima
g
e
-
Orien
ted
Feat
ure
Ext
r
acti
on
and
Se
lecti
on
Techni
que
,
”
IE
EE
Ac
c
ess
,
vo
l.
8,
pp
.
5065
–
507
9,
2020
.
[45]
M
.
Bigl
a
rbe
gi
a
n
et
al
.
,
“
On
c
ondit
ion
moni
to
ring
of
high
fr
eque
ncy
power
GaN
conve
r
ters
with
ada
p
ti
v
e
prognostic
s,
”
20
18
IEEE
Appl
i
e
d
Powe
r
Elec
tro
nic
s
Conf
ere
nc
e
and
E
xposition
(AP
EC)
,
San
Antonio,
TX,
2018,
pp.
1272
–
1279
.
[46]
P
.
Bil
ski,
“
Appli
ca
t
ion
o
f
Rando
m
Forest
t
o
t
h
e
Fault
Det
ec
t
ion
i
n
Analog
Cir
cu
it
s
,
”
XI
IME
KO
World
Congress
“Measur
eme
nt
i
n
Re
search
and
I
ndustry”
,
2015
.
[47]
S
.
Heo, “
Faul
t
D
et
e
ct
io
n
and
Clas
sific
at
ion
Us
ing
AN
N
,
”
IF
AC P
apersO
nLine
,
vo
l.
51
,
no
.
18
,
pp
.
470
–
47
5
,
2018
.
[48]
M
.
Big
la
rb
egi
an
e
t
al
.
,
“
Sc
al
ab
l
e
R
el
i
abi
l
it
y
M
onit
oring
of
Ga
N
Pow
er
Conv
ert
er
Through
Rec
urre
n
t
Neur
al
Ne
tworks,
”
201
8
IEEE Energy
Conve
rs
ion
Con
gress
and
Ex
pos
it
ion (ECCE)
,
P
ortl
and
,
OR,
201
8,
pp
.
7271
–
7
27
7
.
[49]
Q.
Sun
et
al
.
,
“
Condit
ion
Monitori
ng
and
Prognos
is
of
Powe
r
Converters
Based
on
CS
A
-
LSS
VM
,
”
2017
Inte
rnational
C
onfe
renc
e
on
S
e
nsing,
Diagnost
i
cs,
Prognostic
s,
and
Control
(S
DPC)
,
Shangh
ai,
2017
,
pp.
524
–
529.
[50]
W.
Ch
en
et
a
l
,
“
Data
-
Driv
en
A
pproa
ch
for
Fau
lt
Prognos
is
of
SiC
MO
SF
ET
s,
”
in
IEEE
Tr
ansacti
ons
on
Pow
er
El
e
ct
ronics
,
vol
.
35,
no.
4,
pp.
40
48
–
4062,
Apr
.
2
020.
[51]
B.
Gou
et
a
l
.
,
“
An
Online
Dat
a
-
drive
n
Method
f
or
Simul
ta
n
eous
Diagnosis
of
IGBT
and
Curr
ent
Sensor
Fault
of
3
-
Phase
PWM
Inv
e
rte
r
in
Induc
t
io
n
Motor
Drive
s
,
”
in
IE
EE
Tr
ansacti
ons
on
Power
Elec
troni
cs
,
vo
l.
35,
no
.
12,
pp
.
13281
–
13294,
2
020
.
[52]
M.
Pech
t
et
al
.
,
“
Phys
ic
s
-
of
-
fa
ilur
e
-
base
d
progn
ostic
s
for
e
lectr
onic
produc
ts,
”
Tr
ansacti
ons
of
the
Inst
it
ut
e
o
f
Me
asur
eme
nt
an
d
Control
,
vol
.
31
,
no
.
3
-
4
,
pp
.
3
09
–
322
,
2009
.
[53]
F.
O.
He
im
es
,
“
Rec
urre
n
t
neur
al
ne
tworks
for
r
e
ma
ini
ng
use
ful
l
ife
est
im
a
ti
on,
”
in
Proceedi
ngs
of
In
te
rnationa
l
Confe
renc
e
on
P
rognos
ti
cs
and
Healt
h
Manag
e
ment
,
2008
,
pp
.
1
–
6.
[54]
A.
Heng
et
al
.
,
“
Rotating
ma
c
hine
ry
prognosti
cs:
State
of
the
art,
ch
al
l
enge
s
and
opportun
ities,
”
Me
chan
ic
a
l
Syste
ms
and
Sig
nal
Proc
essing
,
vol.
23
,
no
.
3
,
pp
.
724
–
739
,
2009
.
[55]
M.
Seeg
er,
“
Ga
uss
ia
n
proc
esses
for
ma
ch
ine
l
e
arn
ing,
”
In
te
rna
ti
onal
Journal
o
f
Neural
Syst
ems
,
vol
.
14
,
no
.
2
,
pp.
69
–
106
,
200
4
.
[56]
S.
Mohanty
et
al
.
,
“
Mix
ed
Ga
uss
ia
n
proc
ess
and
sta
te
-
spa
ce
appr
oa
ch
fo
r
f
at
igu
e
c
rac
k
gr
owth
pre
di
ct
ion
,
”
Inte
rnational
W
orkshop
on
Stru
ct
ural
Hea
th
Mo
nit
oring
,
vol.
2
,
pp.
1108
–
1115
,
2007
.
[57]
V.
T
.
Tr
an
et
al
.
,
“Da
t
a
-
driven
appr
o
ac
h
to
ma
ch
ine
cond
it
i
on
prognosis
using
le
ast
squar
e
reg
r
ession
tree
,
”
Journal
of
Me
ch
anic
al
Scienc
e
a
nd
Technol
og
y
,
vol.
23
,
pp
.
1468
–
1
475
,
2009
.
[58]
K.
Chakr
abor
ty
et
al
.
,
“Fore
cas
ti
ng
th
e
b
eha
v
ior
of
multiva
r
i
at
e
time
seri
es
using
neur
a
l
n
e
tworks,
”
N
eural
Net
works
,
vo
l.
5
,
pp.
961
–
970
,
19
92
.
[59]
X.
Yao,
“
Evol
vi
ng
artificia
l
neu
r
al
n
et
works
,
”
Pr
oce
ed
ings o
f
the
IEE
E
,
vo
l.
87
,
n
o.
9
,
pp
.
1423
–
1
447
,
1999
.
[60]
D.
C.
M.
Dickso
n
et
a
l
.
,
“Ga
mma
proc
esses
and
fini
te
time
survi
val
proba
b
il
i
ties
,
”
Asti
n
Bulletin
,
vol.
23
,
no.
2
,
pp
.
259
–
272
,
1993
.
[61]
L.
R.
Rabi
n
er,
“
A
tut
orial
on
hid
den
Markov
mo
del
s
and
sel
ecte
d
appl
i
ca
t
ions
in
spee
ch
r
ec
ogni
t
ion,
”
Proceedi
n
gs
of
th
e
I
E
EE
,
vo
l.
77
,
no.
2
,
pp.
25
7
–
286
,
1989
.
[62]
M.
Jouin
et
al
.
,
“De
gra
da
ti
ons
ana
lysis
and
aging
mode
li
ng
for
hea
l
th
asses
sment
and
prognos
ti
cs
of
PEMF
C
,
”
Re
li
ab
il
i
ty E
ngin
ee
ring
&
S
yste
m
Safety
,
vol
.
148
,
pp.
78
–
95
,
2016
.
[63]
J.
B.
Ali
et
al
.
,
“Ac
cur
a
te
be
ar
ing
r
em
a
ini
ng
u
seful
li
fe
pr
edi
c
ti
on
b
ase
d
on
We
ibu
ll
distri
bu
ti
on
and
artifi
cial
neur
al ne
twor
k
,
”
Me
chan
ic
al
Sys
te
ms
and
Signa
l Proce
ss
ing
,
vo
l.
5657
,
pp
.
150
–
1
72
,
2015
.
[64]
M.
Jouin
et
al
.
,
“Par
ticle
fi
lt
e
r
-
b
ase
d
prognost
ic
s
:
Rev
ie
w,
discus
sion
and
per
sp
e
c
ti
ves,
”
M
ec
han
i
cal
S
yste
ms
and
Signal
Proce
ss
in
g
,
vol.
7273
,
pp.
2
–
31
,
2016
.
[65]
M.
E
.
T
ippi
ng
,
“
Sparse
Baye
si
an
l
ea
rn
ing
and
the
re
le
van
ce
vec
tor
m
ac
hin
e,
”
Journal
of
M
achi
ne
Learning
Re
search
,
vol
.
1
,
pp.
211
–
244
,
20
01
.
[66]
H.
Khorasgan
i
e
t
al
.
,
“
Methodologies
for
sys
te
m
-
le
ve
l
r
emaini
ng
useful
l
ife
pre
d
i
ct
ion
,”
R
el
iab
il
i
t
y
Engi
ne
ering
&
Syste
m Saf
et
y
,
v
ol.
154
,
pp
.
8
–
18
,
2016
.
[67]
Y
.
Mahe
r
et
al
.
,
“Surve
y
on
de
ep
learni
ng
app
lied
to
pre
d
ic
t
ive
m
a
int
en
anc
e
,
”
Int
ernati
onal
Journal
of
E
le
c
tric
al
&
Comput
er
Engi
n
ee
ring (
I
JE
C
E
)
,
v
ol.
10
,
n
o
.
6
,
pp
.
5592
–
5598
,
De
c
.
2020
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
1
,
Ma
rch
20
21
:
5
58
–
56
6
566
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Soumya
Ran
i
M
estha
recei
v
ed
h
er
B
ac
he
lor
of
E
ngine
er
ing
d
egr
e
e
in
E
lectr
i
cal
a
nd
Elec
troni
cs
Engi
ne
eri
ng
(20
09)
and
M
aste
r’
s
in
Pow
er
El
e
ct
roni
cs
(2011)
from
VTU
B
el
a
gaum
.
She
is
cur
ren
t
ly
per
sui
ng
doct
ora
l
prog
ram
a
t
NM
AM
I
T
Nitte.
Presen
tly
serving
as
As
sistant
Profess
or
i
n
Elec
tr
ic
a
l
an
d
El
e
ct
roni
cs
E
ngine
er
ing
Dep
art
m
ent
,
NM
AM
IT,
Nitte,
Ind
ia
.
Her
area
of
int
er
est
ar
e
pr
ed
omi
nantly
in
Ma
chi
ne
Learni
ng
,
Pow
er
El
e
ct
ron
i
cs,
Re
la
ys
and
H
igh
Volt
age.
Dr.
Pinto
Pius
A.
J.
r
ecei
ved
his
B.
E
degr
ee
from
Mys
ore
Univer
sity
(19
7
6)
in
Elec
tr
ical
Engi
ne
eri
ng,
Master
’s
from
Ma
ngal
ore
unive
rsi
ty
(1999)
and
P
hD
in
Pow
er
E
l
ec
tron
ic
s
from
Nati
ona
l
Instit
u
t
e
of
technogy
Karna
t
aka,
Surath
kal
(2008).
H
e
h
as
twent
y
-
six
ye
ars
of
industry
expe
ri
ence
and
13
ye
ars
expe
ri
e
nce
in
a
ca
m
edic
s.
Curr
en
tl
y
he
i
s
serving
as
a
Profess
or
in
E&E
Engi
ne
eri
ng
d
ep
art
m
ent
,
NM
AMIT
Nitte.
His
area
of
intere
st
inclu
des
power
e
lectr
onic
s,
el
e
ct
r
ic
vehi
c
le
s,
mot
ors,
and
dr
ive
s.
Evaluation Warning : The document was created with Spire.PDF for Python.