I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
3334
~
334
2
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
33
34
-
334
2
3334
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
Op
t
imiz
e
d
f
a
u
lt
d
e
t
e
c
t
io
n
i
n
b
e
ar
in
gs of
r
ot
a
t
in
g m
a
c
h
i
n
e
s
vi
a
b
at
c
h
n
o
r
m
al
iz
at
io
n
-
i
n
t
e
g
r
at
e
d
b
id
ir
e
c
t
io
n
al
g
at
e
d
r
e
c
u
r
r
e
n
t
u
n
it
n
e
t
w
or
k
s
S
u
j
it
Ku
m
ar
1
,
M
an
is
h
K
u
m
ar
2
,
C
h
e
t
an
B
ar
d
e
3
,
P
r
ak
as
h
Ranj
an
3
1
D
e
pa
r
tm
e
nt
of
E
le
c
tr
ic
a
l
E
ngi
ne
e
r
in
g, G
ove
r
nme
nt
E
ngi
ne
e
r
in
g C
ol
le
ge
B
a
nka
,
S
c
i
e
nc
e
,
T
e
c
hnol
ogy a
nd
T
e
c
hni
c
a
l
E
duc
a
ti
on
D
e
pa
r
tm
e
nt
, B
ih
a
r
E
ngi
ne
e
r
in
g U
ni
ve
r
s
it
y, B
a
nka
, I
ndi
a
2
D
e
pa
r
tm
e
nt
of
E
le
c
tr
ic
a
l
E
ngi
ne
e
r
in
g, I
ndi
a
n I
ns
ti
tu
te
of
T
e
c
hnol
ogy, P
a
tn
a
, I
nd
ia
3
D
e
pa
r
tm
e
nt
of
E
le
c
tr
oni
c
s
a
nd C
omm
uni
c
a
ti
on E
ngi
ne
e
r
in
g, I
ndi
a
n I
ns
ti
tu
te
of
I
nf
or
ma
ti
on T
e
c
hnol
ogy, B
ha
ga
lp
ur
, I
ndi
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
S
e
p
8,
2024
R
e
vis
e
d
J
un
12,
2025
Ac
c
e
pted
J
ul
10,
2025
Mo
t
o
r
i
s
co
mm
o
n
l
y
u
s
e
d
i
n
i
n
d
u
s
t
ri
a
l
ap
p
l
i
ca
t
i
o
n
s
.
A
l
t
h
o
u
g
h
m
o
t
o
rs
ar
e
freq
u
e
n
t
l
y
fo
u
n
d
t
o
h
av
e
b
ear
i
n
g
p
ro
b
l
ems
,
t
h
i
s
cau
s
es
a
s
eri
o
u
s
s
afet
y
ri
s
k
t
o
i
n
d
u
s
t
r
i
al
p
ro
d
u
c
t
i
o
n
.
T
rad
i
ti
o
n
al
l
y
,
fa
u
l
t
d
i
ag
n
o
s
t
i
cs
me
t
h
o
d
s
o
f
t
e
n
req
u
i
red
o
n
l
y
s
i
g
n
a
l
p
r
o
ces
s
i
n
g
t
ec
h
n
i
q
u
es
a
n
d
are
i
n
eff
ect
i
v
e.
T
o
o
v
erco
m
e
t
h
i
s
p
r
o
b
l
em,
d
eep
l
ear
n
i
n
g
(D
L
)
h
a
s
b
een
recen
t
l
y
d
e
v
el
o
p
e
d
rap
i
d
l
y
an
d
ach
i
e
v
ed
remar
k
ab
l
e
re
s
u
l
t
s
i
n
fa
u
l
t
d
i
ag
n
o
s
i
s
.
T
h
e
i
n
t
el
l
i
g
e
n
t
fa
u
l
t
d
i
ag
n
o
s
i
s
an
d
c
l
as
s
i
f
i
cat
i
o
n
o
f
r
o
l
l
i
n
g
b
e
ari
n
g
fa
u
l
t
s
b
a
s
ed
o
n
en
s
emb
l
e
emp
i
r
i
ca
l
mo
d
e
d
e
c
o
mp
o
s
i
t
i
o
n
(E
E
M
D
)
an
d
b
at
c
h
n
o
rmal
i
zat
i
o
n
(B
N
)
,
p
ri
n
ci
p
a
l
co
mp
o
n
e
n
t
a
n
al
y
s
i
s
(PCA
)
b
as
e
d
s
t
ac
k
ed
b
i
d
i
rec
t
i
o
n
a
l
-
g
at
e
d
recu
r
re
n
t
u
n
i
t
(Bi
-
G
RU
)
n
eu
r
al
n
e
t
w
o
rk
,
i
s
p
ro
p
o
s
ed
i
n
t
h
i
s
p
a
p
er.
BN
i
s
i
n
t
ro
d
u
ce
d
t
o
i
mp
r
o
v
e
t
h
e
fas
t
co
n
v
er
g
en
ce
o
f
g
a
t
ed
rec
u
rren
t
u
n
i
t
(G
RU
)
.
E
E
M
D
i
s
ap
p
l
i
e
d
t
o
el
i
mi
n
at
e
t
h
e
n
o
i
s
e
i
n
t
erfere
n
ce
fr
o
m
t
h
e
v
i
b
rat
i
o
n
al
s
i
g
n
a
l
,
an
d
t
h
e
n
i
mp
o
rt
a
n
t
fea
t
u
re
s
are
s
el
ec
t
ed
u
s
i
n
g
t
h
e
c
o
rrel
a
t
i
o
n
co
eff
i
ci
e
n
t
v
al
u
e.
N
ex
t
,
PCA
i
s
u
t
i
l
i
ze
d
fo
r
d
i
me
n
s
i
o
n
al
i
t
y
red
u
c
t
i
o
n
t
o
ret
ai
n
o
n
l
y
t
h
e
es
s
e
n
t
i
al
.
Fi
n
a
l
l
y
,
t
h
e
BN
b
a
s
ed
s
t
ack
e
d
Bi
-
G
RU
mo
d
el
i
s
d
ev
e
l
o
p
ed
t
o
cl
as
s
i
f
y
fau
l
t
s
b
a
s
ed
o
n
ex
t
ract
e
d
feat
u
re
s
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
e
l
co
rrec
t
l
y
cl
as
s
i
f
i
es
t
h
e
d
i
ff
eren
t
t
y
p
e
s
o
f
fau
l
t
s
i
n
rea
l
o
p
erat
i
n
g
co
n
d
i
t
i
o
n
s
an
d
a
l
s
o
co
mp
are
d
w
i
t
h
e
x
i
s
t
i
n
g
t
ech
n
i
q
u
e
s
.
K
e
y
w
o
r
d
s
:
B
a
tch
nor
maliza
ti
on
B
i
d
i
r
e
c
t
i
on
a
l
ga
t
e
d
r
e
c
u
r
r
e
n
t
u
n
it
De
e
p
lea
r
ning
E
ns
e
mbl
e
e
mpi
r
ica
l
mode
de
c
ompos
it
ion
F
a
ult
diagnos
is
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
S
uji
t
Kuma
r
De
pa
r
tm
e
nt
of
E
lec
tr
ica
l
E
nginee
r
ing,
Gove
r
nmen
t
E
nginee
r
ing
C
oll
e
ge
B
a
nka
,
S
c
ienc
e
T
e
c
hnology
a
nd
T
e
c
hnica
l
E
duc
a
ti
on
De
pa
r
tm
e
nt
,
B
ihar
E
nginee
r
ing
Unive
r
s
it
y
B
a
nka
,
813102,
B
ihar
,
I
ndia
E
mail:
s
uji
tar
ya
n0412@gmail
.
c
om
1.
I
NT
RODU
C
T
I
ON
M
a
c
hine
f
a
ult
diagnos
is
is
e
s
s
e
nti
a
l
f
or
de
tec
ti
n
g
a
nd
c
las
s
if
ying
f
a
il
ur
e
s
in
r
otating
e
quipm
e
nt,
whic
h
a
r
e
e
s
pe
c
ially
pr
one
to
de
f
e
c
ts
li
ke
be
a
r
ing,
ge
a
r
,
a
nd
s
tator
f
a
ult
s
[
1
]
,
[
2]
.
T
he
s
e
f
a
ult
s
o
f
ten
ge
ne
r
a
te
unique
vibr
a
ti
on
pa
tt
e
r
ns
that
c
a
n
indi
c
a
te
the
mac
hine’
s
he
a
lt
h
s
tatus
.
C
ondit
ion
-
ba
s
e
d
moni
tor
ing
(
C
B
M
)
ha
s
be
c
ome
a
p
r
e
f
e
r
r
e
d
maintena
nc
e
s
tr
a
tegy
due
to
it
s
a
bil
i
ty
to
de
tec
t
p
r
oblems
e
a
r
ly
,
mi
n
im
ize
d
owntim
e
,
a
nd
r
e
duc
e
maintena
nc
e
c
os
ts
[
3
]
.
R
e
s
e
a
r
c
he
r
s
ha
ve
incr
e
a
s
ingl
y
tu
r
ne
d
to
a
r
ti
f
icia
l
int
e
ll
igenc
e
(
AI
)
a
nd
e
xpe
r
t
s
ys
tems
to
e
nha
nc
e
the
r
e
li
a
bil
it
y
a
nd
a
c
c
ur
a
c
y
of
s
uc
h
moni
tor
ing
tec
hniques
.
How
e
ve
r
,
s
ign
a
l
nois
e
r
e
mains
a
major
obs
tac
le,
c
ompl
ica
ti
ng
f
a
ult
de
tec
t
ion
e
f
f
or
ts
[
4]
,
[
5]
.
T
r
a
dit
ional
tec
hniques
li
ke
f
a
s
t
F
our
ier
t
r
a
ns
f
or
m
(
F
F
T
)
,
e
nve
lope
a
na
lys
is
,
a
nd
high
-
f
r
e
que
nc
y
r
e
s
on
a
nc
e
methods
ha
ve
b
e
e
n
wide
ly
us
e
d
[
6]
–
[
8]
,
though
thei
r
e
f
f
e
c
ti
ve
ne
s
s
is
of
ten
li
mi
ted
in
c
ompl
e
x,
non
-
li
ne
a
r
e
nvir
onments
[
9]
.
R
e
c
e
nt
methods
,
including
wa
ve
let
tr
a
ns
f
or
ms
a
nd
e
mpi
r
ica
l
mode
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
e
d
faul
t
de
tec
ti
on
in
be
ar
ings
of
r
otat
ing
m
ac
hines
v
ia
batch
nor
maliz
ati
on
…
(
Suji
t
K
umar
)
3335
de
c
ompos
it
ion
(
E
M
D)
,
of
f
e
r
im
p
r
ove
ments
but
s
t
il
l
f
a
c
e
c
ha
ll
e
nge
s
r
e
late
d
to
ba
s
is
f
unc
ti
on
s
e
lec
ti
on
[
10]
.
E
ns
e
mbl
e
e
mpi
r
ica
l
mode
de
c
ompos
it
ion
(
E
E
M
D)
a
ddr
e
s
s
e
s
thes
e
s
hor
tcomings
by
r
e
duc
ing
mode
a
li
a
s
ing,
ther
e
by
e
nha
nc
ing
diagnos
ti
c
a
c
c
ur
a
c
y
in
nois
y
c
ondit
ions
[
10
]
,
[
11
]
.
W
he
n
c
ombi
ne
d
wi
th
de
e
p
lea
r
nin
g
(
DL
)
a
ppr
oa
c
he
s
s
uc
h
a
s
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
R
NN
)
,
long
s
hor
t
-
ter
m
memor
y
(
L
S
T
M
)
,
a
n
d
ga
ted
r
e
c
ur
r
e
nt
unit
(
GR
U)
,
whic
h
a
r
e
we
ll
-
s
uit
e
d
f
or
ti
me
-
s
e
r
ies
a
na
ly
s
is
,
the
ove
r
a
ll
diagnos
ti
c
pe
r
f
or
manc
e
im
pr
ove
s
s
igni
f
ica
ntl
y
[
12
]
,
[
13]
.
GR
U,
in
p
a
r
ti
c
ular
,
o
f
f
e
r
s
c
omput
a
ti
ona
l
a
dva
ntage
s
,
a
nd
ba
tch
nor
maliza
ti
on
(
B
N)
he
lps
s
pe
e
d
up
ne
twor
k
tr
a
ini
ng
[
14]
.
T
his
innovation
ha
s
s
igni
f
ica
ntl
y
im
pr
ove
d
the
pe
r
f
or
manc
e
o
f
ne
ur
a
l
ne
twor
ks
in
va
r
ious
dom
a
ins
,
including
f
a
ult
c
las
s
if
ica
ti
on
in
r
otating
m
a
c
hiner
y.
T
hir
ukova
ll
ur
u
e
t
al.
[
15]
e
mpl
oye
d
a
n
a
utoenc
od
e
r
f
or
f
a
ult
pr
e
diction
,
a
c
hieving
good
a
c
c
ur
a
c
y.
C
he
n
a
nd
L
i
[
16]
a
ppli
e
d
s
tatis
ti
c
a
l
be
a
r
ing
s
ignals
to
a
s
pa
r
s
e
a
utoenc
ode
r
,
c
ombi
ning
it
with
a
de
e
p
be
li
e
f
ne
t
wor
k
f
or
f
a
ult
c
las
s
if
ica
ti
on.
Ne
ur
a
l
n
e
twor
ks
ha
ve
a
ls
o
p
r
ove
n
e
f
f
e
c
ti
ve
in
a
ddr
e
s
s
ing
c
ompl
e
x
s
e
que
nti
a
l
d
a
ta,
with
L
S
T
M
ne
twor
ks
be
ing
us
e
d
to
c
a
lcula
te
the
r
e
maining
us
e
f
ul
li
f
e
(
R
UL
)
of
mac
hines
a
nd
identif
y
f
a
ult
pr
oba
bil
it
ies
[
17]
,
[
18]
.
Yu
e
t
al.
[
19]
f
ur
ther
de
mons
tr
a
ted
that
L
S
T
M
models
c
ould
a
c
hieve
f
a
ult
diagnos
is
a
c
c
ur
a
c
y
up
to
99%
by
a
utom
a
ti
c
a
ll
y
e
xtr
a
c
t
ing
dyna
mi
c
inf
o
r
mation
f
r
om
r
a
w
da
ta.
I
n
a
ddit
ion,
C
he
n
e
t
al.
[
20]
a
ppli
e
d
c
onvolut
ional
ne
ur
a
l
n
e
twor
ks
(
C
NN
)
to
e
xtr
a
c
t
f
a
ult
f
e
a
tur
e
s
f
r
om
r
a
w
da
ta,
f
oll
owe
d
by
L
S
T
M
f
o
r
f
a
ult
identif
ica
ti
on
.
Hua
ng
e
t
al.
[
21]
uti
li
z
e
d
E
M
D
f
o
r
nois
e
r
e
duc
ti
o
n
a
nd
a
c
onvolut
ional
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
C
R
NN
)
f
or
c
las
s
if
ying
r
oll
ing
be
a
r
ing
f
a
ult
s
.
T
he
r
e
s
e
a
r
c
h
in
[
22]
,
[
23]
e
mpl
oye
d
E
E
M
D
to
e
xtr
a
c
t
e
ne
r
gy
e
ntr
opy
a
s
input
f
e
a
tur
e
s
,
late
r
us
ing
s
uppor
t
ve
c
tor
mac
hine
s
(
S
VM
)
f
or
f
a
ult
c
las
s
if
ica
ti
on.
Hinc
hi
a
nd
T
kiouat
[
2
4]
de
ve
loped
a
c
onvolut
ional
long
s
hor
t
-
ter
m
memor
y
(
C
L
S
T
M
)
ne
u
r
a
l
ne
twor
k,
us
ing
C
NN
f
or
f
e
a
tur
e
e
xtr
a
c
ti
on
a
nd
L
S
T
M
f
or
pr
e
dicting
R
UL
.
P
e
ng
e
t
al.
[
25]
pr
opos
e
d
a
f
a
ult
diagnos
is
method
ba
s
e
d
on
a
bidi
r
e
c
ti
ona
l
-
ga
ted
r
e
c
ur
r
e
nt
unit
(
B
i
-
GR
U)
,
whic
h
e
f
f
icie
ntl
y
c
a
ptur
e
s
dyna
mi
c
inf
or
mation
f
r
om
ti
me
-
s
e
r
ies
vibr
a
ti
on
da
ta.
S
im
il
a
r
ly
,
Z
hiwe
i
[
26]
de
s
i
gne
d
a
one
-
dim
e
ns
ional
c
onvolut
ional
(
1DC
NN
)
-
GR
U
model
to
ha
ndle
s
e
que
nti
a
l
da
ta
f
or
f
a
ult
diagnos
is
.
W
a
ng
e
t
al
.
[
27]
p
r
opos
e
d
a
B
i
-
GR
U
model
that
e
l
im
inate
s
the
ne
e
d
f
or
p
r
e
-
pr
oc
e
s
s
ing
a
nd
a
c
hieve
s
s
upe
r
ior
r
e
s
ult
s
in
f
a
ult
c
las
s
if
ica
ti
on.
I
n
thi
s
wor
k
,
a
B
i
-
GR
U
ne
ur
a
l
ne
twor
ks
is
pr
opo
s
e
d
to
diagnos
e
the
f
a
ult
s
.
T
he
model
is
pr
opos
e
d
to
c
las
s
if
y
dif
f
e
r
e
nt
types
of
f
a
ult
s
in
r
otating
ma
c
hiner
y
unde
r
va
r
ying
ope
r
a
ti
ona
l
c
ondi
ti
ons
.
T
h
e
a
im
of
thi
s
wor
k
a
r
e
a
s
f
oll
ows
.
F
ir
s
t,
the
vib
r
a
ti
on
s
ignal
is
tr
a
ns
f
or
med
int
o
bo
th
the
ti
me
a
nd
f
r
e
que
nc
y
d
omains
,
a
nd
E
E
M
D
is
a
ppli
e
d
to
obtain
int
r
ins
ic
mode
f
u
nc
ti
ons
(
I
M
F
s
)
.
C
or
r
e
lation
c
oe
f
f
icie
nts
a
r
e
us
e
d
to
s
e
lec
t
im
por
tant
f
e
a
tur
e
s
ba
s
e
d
on
thei
r
s
igni
f
ica
nc
e
a
nd
pr
incipa
l
c
omponent
a
na
lys
is
(
P
C
A)
is
us
e
d
f
or
f
e
a
tur
e
s
e
xtr
a
c
ti
on.
S
e
c
ond,
a
B
i
-
GR
U
ne
twor
k
is
uti
li
z
e
d
to
lea
r
n
thes
e
f
e
a
tur
e
s
,
with
BN
e
mpl
oye
d
to
e
nha
nc
e
the
model's
tr
a
ini
ng
s
pe
e
d
a
nd
a
c
c
ur
a
c
y.
F
inally,
the
de
ve
loped
model
is
c
ompar
e
d
with
other
mac
hine
lea
r
ning
tec
hniques
,
de
mons
tr
a
ti
ng
it
s
s
upe
r
ior
pe
r
f
or
man
c
e
in
f
a
ult
c
las
s
if
ica
ti
on.
T
his
r
e
s
e
a
r
c
h
pr
opos
e
d
a
highl
y
e
f
f
icie
nt
f
a
ult
diagnos
is
f
r
a
mew
or
k
that
a
ddr
e
s
s
e
s
ke
y
li
mi
tations
in
c
onve
nti
ona
l
methods
.
B
y
in
tegr
a
ti
ng
E
E
M
D,
c
or
r
e
lation
c
oe
f
f
icie
nt
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on,
a
nd
B
i
-
GR
U
with
BN
,
the
de
ve
loped
model
a
c
hieve
s
im
pr
ove
d
f
a
ult
c
las
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
a
nd
f
a
s
ter
tr
a
ini
ng
ti
mes
,
making
it
a
va
luable
tool
f
or
i
ndus
tr
ial
a
ppli
c
a
ti
ons
.
T
he
innovative
a
s
pe
c
ts
of
thi
s
wor
k
li
e
in
it
s
a
bil
it
y
to
non
-
s
tationar
y
s
ignals
,
pr
ovidi
ng
a
r
obus
t
s
olut
ion
f
or
r
e
a
l
-
wor
ld
f
a
ult
diagnos
is
.
2.
P
ROP
OS
E
D
M
E
T
HO
D
OL
OG
Y
I
n
thi
s
r
e
s
e
a
r
c
h,
a
n
opti
m
ize
d
f
a
ult
de
tec
ti
on
met
hod
f
or
r
oll
ing
be
a
r
ings
in
r
otating
mac
hines
wa
s
de
ve
loped
us
ing
a
B
N
-
int
e
gr
a
ted
s
tac
ke
d
B
i
-
GR
U
ne
ur
a
l
ne
twor
k
model.
I
ni
ti
a
ll
y,
vibr
a
ti
on
s
ign
a
ls
we
r
e
obtaine
d
f
r
om
be
a
r
ings
unde
r
nor
mal
a
nd
f
a
ult
y
c
ondit
ions
a
t
va
r
ious
ope
r
a
ti
ng
s
pe
e
ds
.
T
he
s
e
s
ign
a
ls
we
r
e
f
ir
s
t
c
onve
r
ted
int
o
ti
me
a
nd
f
r
e
que
nc
y
domains
f
or
a
na
lys
is
.
T
o
r
e
move
nois
e
a
nd
de
c
ompos
e
the
s
ignals
,
E
E
M
D
wa
s
a
ppli
e
d,
r
e
s
ult
ing
in
mul
ti
ple
I
M
F
s
.
T
o
e
ns
ur
e
that
only
the
mos
t
r
e
leva
nt
a
nd
nois
e
-
f
r
e
e
f
e
a
tur
e
s
we
r
e
s
e
lec
ted
f
or
c
las
s
if
ica
ti
on,
the
c
or
r
e
lation
c
o
e
f
f
icie
nts
be
twe
e
n
the
I
M
F
s
a
nd
the
r
a
w
vibr
a
ti
o
n
s
ignals
we
r
e
c
a
lcula
ted.
T
his
a
ll
owe
d
f
or
the
s
e
lec
ti
on
of
the
be
s
t
I
M
F
s
f
or
f
a
ult
diagnos
is
.
Ne
xt,
P
C
A
is
a
p
pli
e
d
f
or
dim
e
ns
ionalit
y
r
e
du
c
ti
on,
p
r
e
s
e
r
ving
only
the
mo
s
t
s
igni
f
ica
nt
f
e
a
tur
e
s
f
r
om
the
I
M
F
da
ta
c
or
r
e
s
po
nding
to
f
ive
dis
ti
nc
t
f
a
ult
c
ondit
ions
.
T
he
e
xtr
a
c
ted
f
e
a
tu
r
e
s
we
r
e
then
input
int
o
a
s
tac
ke
d
B
i
-
GR
U
mode
l,
whic
h
wa
s
e
nha
nc
e
d
by
the
incor
por
a
ti
on
of
B
N
to
a
c
c
e
ler
a
te
c
onve
r
ge
nc
e
a
nd
im
pr
ove
the
lea
r
ning
p
r
oc
e
s
s
.
T
he
a
r
c
hit
e
c
tur
e
wa
s
t
r
a
ined
us
ing
s
e
ve
r
a
l
hype
r
pa
r
a
m
e
ter
s
,
including
the
Ada
m
opti
mi
z
e
r
,
mea
n
s
qua
r
e
d
e
r
r
o
r
a
s
the
los
s
f
unc
ti
on,
a
ba
tch
s
ize
of
50,
a
dr
opou
t
r
a
t
e
of
0
.
2,
50
e
poc
hs
,
a
nd
a
lea
r
ning
r
a
te
o
f
0
.
01.
T
he
model
e
f
f
e
c
ti
ve
ly
ha
ndled
s
e
que
nti
a
l
da
ta
a
nd
e
xploi
ted
bidi
r
e
c
ti
ona
l
de
pe
nde
nc
ies
f
or
mor
e
a
c
c
ur
a
te
f
a
ult
c
las
s
if
ica
ti
on.
T
o
e
va
luate
it
s
pe
r
f
or
manc
e
,
the
model
wa
s
tr
a
ined,
tes
ted,
a
nd
va
li
da
ted
with
a
be
a
r
ing
da
tas
e
t.
R
e
s
ult
s
we
r
e
a
s
s
e
s
s
e
d
us
ing
a
c
onf
u
s
ion
matr
ix,
r
e
ve
a
li
ng
high
a
c
c
ur
a
c
y
in
c
las
s
if
ying
va
r
ious
be
a
r
ing
c
ondit
ions
.
Additi
ona
ll
y
,
r
e
c
e
iver
ope
r
a
ti
ng
c
ha
r
a
c
ter
is
ti
c
(
R
OC
)
c
ur
ve
s
we
r
e
us
e
d
to
e
va
luate
the
model's
pe
r
f
or
manc
e
a
c
r
os
s
dif
f
e
r
e
nt
thr
e
s
holds
,
c
onf
ir
mi
ng
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
f
a
ult
de
tec
ti
on.
T
he
vibr
a
ti
on
s
ignals
f
r
om
the
f
a
n
-
e
nd
(
F
E
)
a
nd
dr
ive
-
e
nd
(
DE
)
be
a
r
ings
,
c
oll
e
c
ted
f
r
om
a
da
ta
r
e
pos
it
or
y
a
t
htt
p:/
/engine
e
r
ing.
c
a
s
e
.
e
du/bea
r
ingdata
c
e
nter
,
r
e
pr
e
s
e
nt
nor
mal
a
nd
f
a
ult
y
c
ondit
ions
a
t
va
r
ying
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
333
4
-
334
2
3336
s
pe
e
ds
of
1730,
1750
,
1772
,
a
nd
1797
r
pm
a
s
s
hown
in
T
a
ble
1.
T
he
s
e
s
ignals
a
r
e
obs
e
r
ve
d
to
c
ont
a
in
high
leve
ls
of
s
tationar
it
y
a
nd
nois
e
,
pos
ing
s
igni
f
ica
nt
c
ha
ll
e
nge
s
in
f
a
ult
identif
ica
ti
on
us
ing
c
onve
nti
on
a
l
f
e
a
tur
e
e
xtr
a
c
ti
on
tec
hniques
.
As
s
hown
in
the
method
ology
a
t
F
igur
e
1
,
E
E
M
D
wa
s
e
mpl
oye
d
f
or
b
oth
nois
e
r
e
moval
a
nd
the
e
xtr
a
c
ti
on
of
I
M
F
s
without
mo
de
mi
xing.
T
he
I
M
F
s
with
low
non
-
s
tationar
it
y
a
nd
high
c
or
r
e
lation
with
the
r
a
w
s
ignals
we
r
e
s
e
lec
ted
a
s
f
e
a
tur
e
s
.
T
he
s
e
f
il
ter
e
d
f
e
a
tu
r
e
s
we
r
e
then
f
e
d
s
e
que
nti
a
ll
y
int
o
a
s
tac
ke
d
B
i
-
GR
U
mod
e
l
f
or
c
las
s
if
ying
be
a
r
ing
c
ondit
ions
.
T
he
r
a
w
vibr
a
ti
on
da
ta
f
r
om
F
E
a
nd
DE
be
a
r
ings
unde
r
dif
f
e
r
e
nt
c
ondit
ions
a
nd
s
pe
e
ds
we
r
e
a
na
lyze
d
in
both
t
im
e
a
nd
f
r
e
que
nc
y
domains
.
F
r
e
que
nc
y
s
pe
c
tr
um
a
na
lys
is
is
a
c
omm
on
tec
hn
ique
to
identif
y
be
a
r
ing
de
f
e
c
t
f
r
e
qu
e
nc
y
c
omponents
by
a
pplyi
ng
the
F
F
T
.
I
n
thi
s
wor
k
,
the
o
r
igi
na
l
vib
r
a
ti
on
s
ignals
we
r
e
c
onve
r
ted
int
o
the
f
r
e
que
nc
y
-
a
mpl
it
ude
domain,
a
nd
E
E
M
D
wa
s
a
ppli
e
d
to
de
c
ompos
e
t
he
s
ignals
int
o
s
e
v
e
r
a
l
I
M
F
s
(
I
M
F
1
to
14)
a
nd
r
e
s
iduals
.
E
a
c
h
I
M
F
s
howe
d
dif
f
e
r
e
nt
f
r
e
que
nc
y
c
omponents
,
with
high
-
f
r
e
que
nc
y
c
ontent
s
hif
ti
ng
to
low
-
f
r
e
que
nc
y
c
ontent
dur
ing
de
c
ompos
it
ion.
Nois
e
r
e
moval
i
mpr
ove
d
a
t
higher
de
c
ompos
it
ion
leve
ls
,
a
nd
by
I
M
F
14
,
f
r
e
que
nc
y
c
omponents
we
r
e
we
ll
is
olate
d.
T
a
ble
1.
R
oll
ing
be
a
r
ing
s
tate
B
e
a
r
in
g
s
ta
te
(
A
ppr
ox
mot
or
s
pe
e
d (
r
pm)
=
1730,
1750,
1772,
1797
N
o.
F
a
ul
t
di
a
me
te
r
(
in
c
he
s
)
F
a
ul
t
lo
c
a
ti
on
1
-
N
or
ma
l
c
ondi
ti
on
(
N
C
)
(
C
la
s
s
0)
2
0.007
I
nne
r
r
a
c
e
f
a
ul
t
(
I
R
F
007)
(
C
la
s
s
1)
3
0.021
I
nne
r
r
a
c
e
f
a
ul
t
(
I
R
F
021)
(
C
la
s
s
2)
4
0.007
O
ut
e
r
r
a
c
e
f
a
ul
t
(
O
R
F
007)
(
C
la
s
s
3)
5
0.007
O
ut
e
r
r
a
c
e
f
a
ul
t
@
(
6:
00)
a
(
O
R
F
007@
6)
(
C
la
s
s
4)
6
0.014
O
ut
e
r
r
a
c
e
f
a
ul
t
@
(
12:
00)
a
(
O
R
F
014@
12)
(
C
la
s
s
5)
F
igur
e
1.
C
ombi
na
ti
ona
l
f
r
a
mew
or
k
f
or
c
las
s
if
ica
ti
on
of
be
a
r
ing
f
a
ult
s
2.
1.
F
e
a
t
u
r
e
s
e
lec
t
ion
Any
c
las
s
if
ica
ti
on
model
pe
r
f
or
ms
be
s
t
whe
n
tr
a
ined
on
s
igni
f
ica
nt
f
e
a
tur
e
s
while
a
voidi
ng
nois
e
.
T
hough
E
E
M
D
e
f
f
e
c
ti
ve
ly
de
c
ompos
e
s
s
ignals
,
i
t
incr
e
a
s
e
s
input
da
ta.
T
o
a
dd
r
e
s
s
thi
s
,
c
o
r
r
e
lation
c
oe
f
f
icie
nts
be
twe
e
n
the
de
c
ompos
e
d
I
M
F
s
a
nd
r
a
w
s
ignals
a
r
e
c
a
lcul
a
ted
to
s
e
lec
t
the
be
s
t
de
noi
s
e
d
a
nd
highl
y
c
or
r
e
late
d
I
M
F
s
.
T
he
a
ppli
c
a
ti
on
of
E
E
M
D
a
nd
f
e
a
tur
e
s
e
lec
ti
on
us
ing
c
or
r
e
lation
c
oe
f
f
icie
nt
f
inally
ha
s
given
a
s
e
t
of
8
I
M
F
f
e
a
tur
e
s
e
a
c
h
of
s
a
mpl
e
length
15,
000
f
or
s
ix
be
a
r
ing
c
ondit
ions
[
8
×
6
×
1500
0]
.
2.
2.
F
e
a
t
u
r
e
e
xt
r
ac
t
ion
an
d
d
im
e
n
s
ion
ali
t
y
r
e
d
u
c
t
ion
P
C
A
wa
s
pe
r
f
or
med
on
the
ini
ti
a
l
f
e
a
tur
e
s
pa
c
e
of
[
8
×
6
×
15000
]
in
or
de
r
to
r
e
duc
e
the
dim
e
ns
ion
a
nd
a
ls
o
to
f
u
r
ther
r
e
move
the
da
ta
r
e
dunda
nc
y.
Al
l
the
s
e
lec
ted
I
M
F
s
ha
ve
be
e
n
r
e
duc
e
d
a
long
two
p
r
incipa
l
c
omponents
s
ince
they
c
a
ptur
e
d
mos
t
of
the
va
r
ia
nc
e
in
the
da
ta
a
nd
the
r
e
s
ult
e
d
da
ta
s
ize
is
of
[
2
×
6
×
15000
]
f
or
e
a
c
h
of
the
be
a
r
ing
c
ondit
ion
.
T
he
r
e
duc
e
d
f
e
a
tur
e
ve
c
tor
s
f
or
a
ll
s
ix
c
ondit
ions
a
r
e
f
e
d
a
s
input
f
or
tr
a
ini
ng
the
ne
u
r
a
l
ne
twor
k
.
2.
3.
F
a
u
lt
d
iagn
os
is
b
as
e
d
on
b
at
c
h
n
or
m
ali
z
at
i
on
s
t
ac
k
e
d
b
i
d
ire
c
t
ion
al
-
ga
t
e
d
r
e
c
u
r
r
e
n
t
u
n
it
T
he
f
a
ult
d
iagnos
is
a
lgor
it
hm
is
d
ivi
de
d
int
o
t
wo
s
e
c
ti
ons
.
T
he
f
ir
s
t
is
to
c
a
ptur
e
the
dyna
mi
c
inf
or
mation
f
r
o
m
r
a
w
da
ta
a
nd
the
s
e
c
ond
is
to
de
ve
lop
a
DL
c
las
s
if
ier
model
f
or
c
las
s
if
ying
the
va
r
ious
types
of
be
a
r
ing
f
a
ult
s
unde
r
di
f
f
e
r
e
nt
c
ondit
ions
.
T
he
f
r
a
mew
or
k
of
the
pr
opos
e
d
a
lgor
it
hm
is
s
hown
in
F
igur
e
2.
T
he
f
oll
owing
s
teps
a
r
e
given:
i)
C
oll
e
c
t
the
s
e
ns
or
s
da
ta.
T
he
n,
the
da
ta
is
pr
e
pr
oc
e
s
s
e
d
a
nd
s
c
a
led
(
r
a
nge
s
f
r
om
0
to
1)
.
ii)
Applica
ti
on
of
E
E
M
D
on
vibr
a
ti
ona
l
s
ignal
whic
h
a
na
lys
e
s
in
ti
me
-
f
r
e
que
nc
y
domain.
iii)
S
e
lec
ti
on
of
f
e
a
tur
e
s
is
done
by
c
or
r
e
lation
c
oe
f
f
ici
e
nt.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
e
d
faul
t
de
tec
ti
on
in
be
ar
ings
of
r
otat
ing
m
ac
hines
v
ia
batch
nor
maliz
ati
on
…
(
Suji
t
K
umar
)
3337
iv)
R
e
duc
ti
on
of
high
d
im
e
ns
ion
f
e
a
tur
e
s
pa
c
e
int
o
lo
w
dim
e
ns
ion
us
ing
P
C
A.
v)
S
pli
t
the
pr
e
pa
r
e
d
da
tas
e
t
int
o
tr
a
in,
va
li
da
ti
on
a
nd
tes
t
da
ta
vi)
BN
is
us
e
d
to
s
pe
e
d
up
tr
a
ini
ng,
s
tabili
z
e
t
he
lea
r
ning
pr
oc
e
s
s
,
a
nd
potentially
im
p
r
ove
the
ge
ne
r
a
li
z
a
ti
on
of
the
ne
ur
a
l
ne
twor
k
vii
)
T
r
a
in
a
nd
de
ve
lop
the
B
N
ba
s
e
d
s
tac
ke
d
B
i
-
GR
U.
vii
i)
T
he
pe
r
f
or
manc
e
of
the
pr
o
pos
e
d
a
lgor
it
hm
is
c
o
nf
ir
med
by
a
c
c
ur
a
c
y,
model
los
s
,
c
onf
us
ion
matr
ix,
a
nd
R
OC
a
r
e
a
unde
r
the
c
u
r
ve
(
AUC
)
c
u
r
ve
.
F
igur
e
2.
F
r
a
mew
or
k
of
pr
opos
e
d
a
lgor
it
h
m
2.
4.
S
t
ac
k
e
d
b
i
d
ire
c
t
ion
al
-
ga
t
e
d
r
e
c
u
r
r
e
n
t
u
n
it
m
od
e
l
f
or
c
las
s
if
icat
ion
of
b
e
ar
in
g
c
on
d
it
io
n
s
T
he
s
tac
ke
d
B
i
-
GR
U
model
[
27
]
,
c
ompos
e
d
o
f
tw
o
GR
U
laye
r
s
in
s
e
que
nc
e
,
leve
r
a
ge
s
inf
or
mation
f
r
om
both
ti
me
d
ir
e
c
ti
ons
to
c
las
s
if
y
be
a
r
ing
c
ondit
ions
.
I
nc
r
e
a
s
ing
the
number
o
f
B
i
-
GR
U
laye
r
s
theor
e
ti
c
a
ll
y
e
nha
nc
e
s
f
e
a
tur
e
e
xtr
a
c
ti
on
a
nd
i
mpr
ove
s
f
a
ult
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y.
How
e
ve
r
,
a
dding
mul
t
ipl
e
GR
U
laye
r
s
a
ls
o
incr
e
a
s
e
s
tr
a
ini
ng
ti
me
a
nd
r
is
ks
ove
r
f
it
ti
ng.
T
o
maintain
e
f
f
e
c
ti
ve
pr
oc
e
s
s
ing,
the
a
r
gument
‘
r
e
tur
n_s
e
que
nc
e
s
’
is
s
e
t
to
‘
T
r
ue
’
,
e
ns
ur
ing
the
output
of
e
a
c
h
GR
U
laye
r
is
r
e
s
ha
pe
d
int
o
a
3D
a
r
r
a
y
a
nd
pa
s
s
e
d
to
the
ne
xt
laye
r
.
I
n
thi
s
wor
k,
f
our
di
f
f
e
r
e
nt
types
o
f
mu
lt
i
-
laye
r
e
d,
B
N
ba
s
e
d
s
tac
ke
d
Bi
-
GR
U
model
s
we
r
e
tr
a
ined
to
c
las
s
if
y
the
c
ondi
ti
ons
of
r
oll
e
r
be
a
r
ings
,
wi
th
their
pe
r
f
o
r
manc
e
s
c
ompar
e
d
to
one
a
nother
.
T
he
be
a
r
ing
da
tas
e
t
wa
s
s
pli
tt
e
d
in
to
tr
a
ini
ng
,
tes
ti
ng,
a
nd
va
li
da
ti
o
n
s
e
ts
.
Dur
ing
e
a
c
h
e
poc
h,
the
model
wa
s
tr
a
ined
us
ing
the
tr
a
ini
ng
da
tas
e
t
a
nd
a
utom
a
ti
c
a
ll
y
va
li
da
ted
with
2%
of
the
tr
a
ined
da
ta
to
pr
e
ve
nt
ove
r
f
i
tt
ing
a
nd
im
p
r
ove
pa
r
a
mete
r
s
e
lec
ti
on.
T
he
hype
r
pa
r
a
mete
r
s
us
e
d
f
o
r
tr
a
ini
ng
a
r
e
a
s
f
oll
ows
:
Ada
m
opti
mi
z
e
r
,
mea
n
s
qua
r
e
d
e
r
r
o
r
los
s
f
unc
ti
o
n,
ba
tch
s
ize
of
50,
d
r
opout
r
a
te
of
0
.
2,
50
e
poc
hs
,
a
nd
a
lea
r
ning
r
a
te
of
0.
01
.
T
he
e
nti
r
e
f
r
a
mew
or
k
is
de
ve
loped
a
nd
tr
a
ined
us
ing
the
P
ython
pr
og
r
a
mm
ing
langua
ge
,
with
Ke
r
a
s
a
nd
T
e
ns
or
F
low
1
.
0
li
b
r
a
r
ies
f
or
im
pleme
ntati
on.
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
I
n
r
e
c
e
nt
ye
a
r
s
,
r
e
s
e
a
r
c
he
r
s
ha
ve
looked
a
t
di
f
f
e
r
e
nt
methods
f
or
diagnos
ing
f
a
ult
s
in
r
otating
mac
hines
.
T
he
y
of
ten
us
e
tec
hniques
li
ke
wa
ve
let
tr
a
ns
f
or
m
a
nd
E
M
D,
including
a
va
r
iation
c
a
ll
e
d
E
E
M
D.
How
e
ve
r
,
thes
e
methods
s
tr
ug
gle
with
nois
e
a
nd
mode
mi
xing,
whic
h
c
a
n
make
them
les
s
e
f
f
e
c
ti
ve
in
r
e
a
l
-
wor
ld
s
it
ua
ti
ons
.
T
r
a
dit
ional
mac
hine
lea
r
nin
g
methods
a
ls
o
r
e
ly
on
manua
ll
y
s
e
lec
ti
ng
f
e
a
tur
e
s
,
whic
h
c
a
n
lea
d
to
poor
pe
r
f
or
manc
e
if
the
f
e
a
tur
e
s
a
r
e
not
c
hos
e
n
c
or
r
e
c
tl
y.
T
his
s
tudy
int
r
oduc
e
s
a
ne
w
method
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
333
4
-
334
2
3338
that
us
e
s
E
E
M
D
to
r
e
move
nois
e
a
nd
a
s
tac
ke
d
B
i
-
GR
U
ne
ur
a
l
ne
twor
k
with
B
N
f
or
be
tt
e
r
f
e
a
tur
e
s
e
lec
ti
on
a
nd
c
las
s
if
ica
ti
on.
W
e
f
ound
that
E
E
M
D
gr
e
a
tl
y
r
e
duc
e
s
nois
e
in
vibr
a
ti
on
s
ignals
,
im
pr
oving
the
q
ua
li
ty
of
da
ta
f
or
c
las
s
i
f
ica
ti
on.
B
y
us
ing
c
or
r
e
lation
c
oe
f
f
i
c
ients
,
we
s
e
lec
ted
the
mos
t
im
por
tant
f
e
a
tur
e
s
f
r
om
thes
e
s
ignals
.
T
he
B
N
-
ba
s
e
d
B
i
-
G
R
U
model
a
c
hieve
d
high
a
c
c
ur
a
c
y
in
identif
ying
di
f
f
e
r
e
nt
types
of
be
a
r
in
g
f
a
ult
s
.
I
t
a
ls
o
tr
a
ined
f
a
s
ter
a
nd
pe
r
f
or
med
be
tt
e
r
than
tr
a
dit
ional
methods
li
ke
C
NN
a
nd
L
S
T
M
.
How
e
ve
r
,
ther
e
a
r
e
s
ome
li
mi
tations
,
s
uc
h
a
s
the
da
tas
e
t
be
ing
c
oll
e
c
ted
unde
r
c
ontr
oll
e
d
c
ondit
ions
,
whic
h
may
not
r
e
pr
e
s
e
nt
r
e
a
l
-
wor
ld
s
c
e
na
r
ios
.
F
utur
e
r
e
s
e
a
r
c
h
s
hould
f
oc
us
on
im
pr
oving
f
e
a
tur
e
s
e
lec
ti
on
to
a
ddr
e
s
s
t
he
s
e
is
s
ue
s
.
T
a
ble
2
c
ompar
e
s
the
tes
ti
ng
a
c
c
ur
a
c
y
of
va
r
ious
DL
models
,
s
howing
that
L
S
T
M
,
B
i
-
L
S
T
M
,
GR
U,
a
nd
B
i
-
GR
U
a
c
hieve
d
moder
a
te
a
c
c
ur
a
c
y
(
82.
92
t
o
86.
80%
)
,
while
the
pr
opos
e
d
B
N
-
ba
s
e
d
s
tac
ke
d
B
i
-
GR
U
ne
twor
k
outper
f
or
med
a
ll
other
s
with
a
pe
r
f
e
c
t
10
0
%
a
c
c
ur
a
c
y.
T
he
ke
y
f
indi
ngs
de
mons
tr
a
te
that
a
pplyi
ng
E
E
M
D
to
pr
e
pr
oc
e
s
s
vibr
a
ti
on
s
ignals
e
f
f
e
c
ti
ve
ly
r
e
duc
e
s
nois
e
a
nd
e
nha
nc
e
s
the
qua
li
ty
of
input
da
ta
f
or
c
las
s
if
ica
ti
on.
T
he
pr
opos
e
d
method
r
e
s
ult
e
d
in
a
s
igni
f
ica
ntl
y
higher
pr
opo
r
ti
on
of
im
po
r
tant
f
e
a
tur
e
s
be
ing
s
e
lec
ted
thr
ough
the
c
or
r
e
lation
c
oe
f
f
icie
nt
f
r
om
th
e
de
c
ompos
e
d
s
ignals
,
c
ompar
e
d
to
tr
a
dit
ional
a
ppr
oa
c
he
s
.
T
he
B
N
-
ba
s
e
d
B
i
-
G
R
U
model
a
ls
o
e
xhibi
ted
f
a
s
ter
c
onve
r
ge
nc
e
a
nd
s
upe
r
ior
f
a
ult
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
c
ompar
e
d
to
e
xis
ti
ng
methods
s
uc
h
a
s
C
NN
,
L
S
T
M
,
a
nd
S
VM
,
making
it
pa
r
ti
c
ula
r
ly
s
uit
a
ble
f
o
r
r
e
a
l
-
ti
me
indus
tr
ial
a
ppli
c
a
ti
ons
.
T
a
ble
2.
Ac
c
ur
a
c
y
of
c
las
s
if
ica
ti
on
models
M
ode
ls
T
e
s
ti
ng
a
c
c
ur
a
c
y (
%
)
L
S
T
M
84.90
Bi
-
L
S
T
M
86.80
G
R
U
82.92
Bi
-
G
R
U
ne
twor
k
83
BN
-
P
C
A
ba
s
e
d
s
ta
c
ke
d
B
i
-
G
R
U
ne
twor
k (
pr
opos
e
d
)
100
T
he
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
of
the
B
N
-
ba
s
e
d
s
t
a
c
ke
d
B
i
-
GR
U
model
wa
s
c
ompar
e
d
with
other
mac
hine
lea
r
ning
a
nd
DL
models
f
r
om
the
li
ter
a
tur
e
[
28]
–
[
31
]
.
T
a
ble
3
s
hows
that
the
pr
opos
e
d
model
outper
f
or
med
e
xis
ti
ng
methods
,
a
c
hieving
s
upe
r
io
r
r
e
s
ult
s
c
ompar
e
d
to
the
1D
-
C
NN
-
L
S
T
M
(
97.
69
%
)
,
S
VM
(
56.
2%
)
,
r
a
ndom
f
o
r
e
s
t
(
55.
5
%
)
,
R
NN
(
60
.
1%
)
,
XG
B
oos
t
(
94%
)
,
ne
u
r
a
l
ne
twor
k
(
55.
5%
)
,
Attenti
on
L
S
T
M
(
84.
73%
)
,
a
nd
L
S
T
M
(
91.
79
)
models
.
Additi
ona
l
ly,
the
R
OC
c
ur
ve
,
a
ke
y
e
va
luation
metr
ic,
wa
s
us
e
d
to
a
s
s
e
s
s
the
model’
s
f
a
ult
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
F
igur
e
3
indi
c
a
tes
a
s
tr
ong
t
r
ue
pos
it
ive
r
a
te
,
w
it
h
AU
C
va
lues
f
or
e
a
c
h
f
a
ult
c
las
s
r
a
nging
f
r
om
0.
82
to
0
.
93,
c
on
f
ir
mi
ng
the
mo
de
l's
r
e
li
a
bil
it
y
f
or
be
a
r
ing
c
ondit
ion
c
las
s
if
ica
ti
on
us
ing
r
a
w
vibr
a
ti
on
da
ta
.
T
h
is
s
tu
dy
e
x
a
m
in
e
d
a
c
om
pr
e
h
e
ns
iv
e
f
a
u
l
t
d
ia
gn
os
is
m
od
e
l
us
in
g
t
he
p
r
o
pos
e
d
s
tac
ke
d
B
i
-
GR
U
a
r
c
h
it
e
c
t
u
r
e
w
it
h
E
E
M
D
f
o
r
f
e
a
t
u
r
e
s
e
lec
ti
on
.
H
ow
e
ve
r
,
f
u
r
the
r
r
e
s
e
a
r
c
h
may
be
ne
e
d
e
d
t
o
va
l
ida
te
i
ts
e
f
f
e
c
t
iv
e
ne
s
s
,
pa
r
ti
c
u
la
r
ly
r
e
ga
r
d
in
g
v
a
r
y
in
g
r
e
a
l
-
w
o
r
l
d
i
nd
us
tr
ia
l
c
on
di
t
io
ns
a
n
d
t
he
pr
e
s
e
nc
e
o
f
a
d
d
it
io
na
l
n
o
is
e
s
o
u
r
c
e
s
.
W
h
il
e
t
he
E
E
M
D
a
n
d
c
or
r
e
la
t
io
n
c
oe
f
f
i
c
i
e
n
t
m
e
t
ho
ds
we
r
e
be
ne
f
ic
ial
f
o
r
s
e
lec
ti
ng
r
e
le
va
n
t
f
e
a
t
ur
e
s
,
th
e
inc
r
e
a
s
e
d
nu
mb
e
r
o
f
i
np
ut
s
i
gn
a
ls
m
a
y
l
e
a
d
to
h
ig
he
r
c
om
pu
ta
t
io
na
l
d
e
m
a
n
ds
,
w
h
ic
h
f
ut
u
r
e
r
e
s
e
a
r
c
h
s
h
ou
ld
a
d
d
r
e
s
s
by
op
t
im
iz
in
g
f
e
a
tu
r
e
s
e
lec
ti
o
n
f
u
r
the
r
.
Ou
r
s
tu
dy
de
m
ons
t
r
a
t
e
s
tha
t
th
e
B
N
-
P
C
A
ba
s
e
d
s
tac
ke
d
B
i
-
GR
U
mo
de
l
is
m
or
e
r
e
s
i
li
e
n
t
th
a
n
t
r
a
di
t
io
na
l
f
a
ul
t
de
t
e
c
t
io
n
me
th
ods
f
o
r
be
a
r
i
ng
d
iag
n
os
is
i
n
r
o
ta
ti
ng
m
a
c
hi
ne
s
.
F
ut
u
r
e
s
t
ud
ies
m
a
y
i
nv
e
s
t
ig
a
t
e
hy
b
r
i
d
m
ode
ls
t
ha
t
c
om
b
ine
DL
w
it
h
e
x
pe
r
t
k
no
wl
e
dg
e
-
ba
s
e
d
s
ys
t
e
ms
a
nd
e
xp
lo
r
e
f
e
a
s
i
bl
e
me
th
ods
f
o
r
pr
od
uc
in
g
m
o
r
e
c
o
m
pu
ta
ti
on
a
l
ly
e
f
f
ici
e
n
t
a
lg
or
i
th
ms
t
ha
t
m
a
in
ta
in
h
i
gh
c
las
s
if
ic
a
t
io
n
a
c
c
u
r
a
c
y
wh
il
e
mi
n
im
iz
in
g
t
r
a
ini
n
g
ti
me
,
p
a
r
t
icu
la
r
ly
in
r
e
a
l
-
t
im
e
a
pp
l
ica
ti
on
s
w
he
r
e
d
a
t
a
is
c
on
t
in
uo
us
l
y
s
t
r
e
a
me
d
.
F
ig
ur
e
4
s
ho
ws
th
e
c
on
f
us
io
n
ma
t
r
ix
o
f
th
e
p
r
o
pos
e
d
m
ode
l
,
wh
ic
h
c
o
r
r
e
c
t
ly
c
las
s
i
f
i
e
s
t
he
d
i
f
f
e
r
e
n
t
f
a
ul
t
c
o
nd
it
io
ns
of
r
o
ll
e
r
be
a
r
i
ngs
.
T
he
tes
t
in
g
r
e
s
ul
ts
,
dis
pl
a
ye
d
in
F
ig
u
r
e
5
,
s
h
ows
th
e
e
nh
a
nc
e
d
p
e
r
f
o
r
m
a
n
c
e
o
f
th
e
s
t
a
c
k
e
d
B
i
-
GR
U
m
od
e
l
in
c
l
a
s
s
if
y
in
g
b
e
a
r
in
g
c
o
nd
i
ti
ons
us
i
ng
p
r
o
c
e
s
s
d
a
t
a
.
T
a
ble
3.
C
ompar
is
on
of
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
M
e
th
ods
T
e
s
ti
ng
a
c
c
ur
a
c
y (
%
)
R
N
N
[
29]
60.1
S
V
M
[
30]
56.2
X
G
B
oos
t
[
31]
94
R
a
ndom
f
or
e
s
t
[
29
]
55.5
N
e
ur
a
l
n
e
twor
k
[
29]
85
A
tt
e
nt
io
n L
S
T
M
[
28]
84.73
1D
-
C
N
N
-
L
S
T
M
[
28]
97.69
L
S
T
M
[
30]
91.79
BN
-
P
C
A
ba
s
e
d
s
ta
c
ke
d
B
i
-
G
R
U
(
p
r
opos
e
d)
100
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
e
d
faul
t
de
tec
ti
on
in
be
ar
ings
of
r
otat
ing
m
ac
hines
v
ia
batch
nor
maliz
ati
on
…
(
Suji
t
K
umar
)
3339
F
igur
e
3.
R
OC
c
ur
ve
s
of
pr
opos
e
d
model
F
igur
e
4.
C
onf
us
ion
matr
ix
of
pr
opos
e
d
model
F
igur
e
5.
T
e
s
ti
ng
a
c
c
ur
a
c
y
o
f
p
r
opos
e
d
model
7
8
.
9
9
8
2
.
8
2
9
3
.
1
5
100
0
20
40
60
80
100
120
1
2
3
4
T
E
S
T
I
N
G
A
C
C
U
R
A
C
Y
(
)
%
BI
-
G
R
U
L
A
Y
E
R
S
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
333
4
-
334
2
3340
4.
CONC
L
USI
ON
R
oll
ing
be
a
r
ing
f
a
il
u
r
e
s
a
r
e
c
omm
on
f
a
ult
s
in
r
o
tating
mac
hines
.
I
n
thi
s
pa
pe
r
,
a
B
N
-
P
C
A
-
ba
s
e
d
s
tac
ke
d
B
i
-
G
R
U
model
is
de
ve
loped.
T
o
ha
ndle
non
-
s
tationar
y
s
ignals
,
E
E
M
D
is
e
mpl
oye
d
a
s
a
powe
r
f
ul
tool
to
de
c
ompos
e
vib
r
a
ti
ona
l
s
ignals
int
o
mul
t
ipl
e
I
M
F
s
.
T
he
c
or
r
e
lation
c
oe
f
f
icie
nt
tec
hnique
is
then
a
ppli
e
d
to
s
e
lec
t
f
e
a
tur
e
s
f
r
om
thes
e
I
M
F
s
.
B
N
is
us
e
d
to
a
c
c
e
ler
a
te
model
tr
a
ini
ng
a
nd
e
ns
ur
e
f
a
s
t
c
onve
r
ge
nc
e
,
a
nd
P
C
A
is
us
e
d
f
or
f
e
a
tur
e
e
xtr
a
c
ti
on.
T
he
pr
opos
e
d
model
a
c
c
ur
a
tely
c
las
s
if
ies
dif
f
e
r
e
nt
be
a
r
ing
f
a
ult
c
ondit
ions
unde
r
va
r
ious
mot
or
r
u
nning
s
pe
e
ds
a
nd
ha
s
a
ls
o
be
e
n
c
ompar
e
d
with
e
xis
ti
ng
methods
.
R
e
c
e
nt
obs
e
r
va
ti
ons
indi
c
a
te
that
the
a
ppli
c
a
ti
on
of
E
E
M
D
s
igni
f
ica
ntl
y
r
e
duc
e
s
nois
e
a
nd
e
nha
nc
e
s
f
e
a
tur
e
s
e
lec
ti
on
in
the
f
a
ult
diagnos
is
o
f
be
a
r
ings
.
Ou
r
f
indi
ngs
p
r
ovide
c
onc
lus
ive
e
videnc
e
that
thi
s
a
ppr
oa
c
h
is
a
s
s
oc
iate
d
with
f
a
s
ter
c
onve
r
ge
nc
e
a
nd
s
upe
r
ior
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y,
not
only
c
ompar
e
d
to
e
xis
ti
ng
tec
hniques
but
a
ls
o
in
the
c
ontext
of
r
e
a
l
-
ti
me
moni
tor
ing
a
nd
f
a
ult
diagnos
is
in
i
ndus
tr
ial
e
nvir
onments
.
T
his
wor
k
de
s
c
r
ibes
the
im
po
r
ta
nc
e
of
int
e
gr
a
ti
ng
a
dva
nc
e
d
s
ignal
pr
oc
e
s
s
ing
a
nd
DL
methods
f
or
e
f
f
e
c
ti
ve
f
a
ult
de
tec
ti
on.
F
utu
r
e
wo
r
k
c
a
n
e
xtend
th
is
a
ppr
oa
c
h
f
o
r
im
pleme
ntation
in
r
e
a
l
-
ti
me
f
a
ult
diagnos
is
.
AC
KNOWL
E
DGM
E
N
T
S
T
he
a
uthor
s
wo
uld
li
ke
to
thank
C
a
s
e
W
e
s
ter
n
R
e
s
e
r
ve
Unive
r
s
it
y
f
or
the
pr
ovided
mot
or
be
a
r
ing
da
tas
e
t
on
it
s
we
bs
it
e
.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
he
a
uthor
s
de
c
lar
e
that
no
f
inanc
ial
s
uppor
t
wa
s
r
e
c
e
ived
f
r
om
a
ny
a
ge
nc
y.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
S
uji
t
Kuma
r
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
M
a
nis
h
Kuma
r
✓
✓
✓
C
he
tan
B
a
r
de
✓
✓
✓
✓
P
r
a
ka
s
h
R
a
njan
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
Author
s
s
tate
no
c
onf
li
c
t
of
int
e
r
e
s
t.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
P
ubli
c
ly
a
va
il
a
ble
da
tas
e
t
ha
s
be
e
n
r
e
f
e
r
r
e
d
in
the
manus
c
r
ipt
.
RE
F
E
RE
NC
E
S
[
1]
S
.
M
a
ni
ka
nda
n
a
nd
K
.
D
ur
a
iv
e
lu
,
“
F
a
ul
t
di
a
gnos
is
of
va
r
io
us
r
ot
a
ti
ng
e
qui
pme
nt
us
in
g
ma
c
hi
ne
le
a
r
ni
ng
a
ppr
oa
c
he
s
–
a
r
e
vi
e
w
,”
P
r
o
c
e
e
di
ngs
of
th
e
I
ns
ti
tu
ti
on
of
M
e
c
hani
c
al
E
ngi
ne
e
r
s
,
P
ar
t
E
:
J
our
nal
of
P
r
oc
e
s
s
M
e
c
hani
c
al
E
ngi
ne
e
r
in
g
,
vol
.
235,
no.
2,
pp. 629
–
642, 2021, doi:
10.1177/095440892
0971976.
[
2]
R
.
M
. S
ou
z
a
,
E
. G
.
S
.
N
a
s
c
im
e
n
to
,
U
.
A
.
M
ir
a
n
da
,
W
.
J
.
D
.
S
il
v
a
,
a
nd
H
. A
. L
e
p
ik
s
on
,
“
D
e
e
p
le
a
r
ni
ng
f
or
di
a
gn
o
s
i
s
a
n
d
c
l
a
s
s
if
ic
a
t
io
n
o
f
f
a
ul
t
s
i
n
in
du
s
t
r
i
a
l
r
o
ta
ti
ng
m
a
c
hi
n
e
r
y
,
”
C
o
m
put
e
r
s
a
nd
I
n
du
s
t
r
i
al
E
n
gi
ne
e
r
in
g
,
v
ol
.
1
53
,
20
21
,
do
i:
1
0.
10
16
/j
.
c
ie
.2
02
0.
10
70
60
.
[
3]
S
.
G
a
w
de
,
S
.
P
a
ti
l,
S
.
K
uma
r
,
P
.
K
a
ma
t,
K
.
K
ot
e
c
ha
,
a
nd
A
.
A
br
a
ha
m,
“
M
ul
ti
-
f
a
ul
t
di
a
gnos
is
of
in
dus
tr
ia
l
r
ot
a
ti
ng
ma
c
hi
ne
s
us
in
g
da
ta
-
dr
iv
e
n
a
p
pr
oa
c
h :
a
r
e
vi
e
w
of
two
de
c
a
de
s
of
r
e
s
e
a
r
c
h,”
E
ngi
ne
e
r
in
g
A
ppl
ic
at
io
ns
of
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
,
vol
.
123,
2023, doi:
10.1016/j
.e
nga
ppa
i.
2023.106139.
[4
]
V
.
S
i
n
g
h
,
P
.
G
a
n
g
s
a
r
,
R
.
P
o
r
w
a
l
,
a
n
d
A
.
A
t
u
l
k
a
r
,
“
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
p
p
l
i
c
a
t
i
o
n
i
n
f
a
u
l
t
d
i
a
g
n
o
s
t
i
c
s
o
f
r
o
t
a
t
i
n
g
i
n
d
u
s
t
r
i
a
l
m
a
c
h
i
n
e
s
:
a
s
t
a
t
e
-
of
-
t
h
e
-
a
r
t
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
M
a
n
u
f
a
c
t
u
r
i
n
g
,
v
o
l
.
3
4
,
n
o
.
3
,
p
p
.
9
3
1
–
960, 2023, doi:
10.1007/
s
10845
-
021
-
01861
-
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
e
d
faul
t
de
tec
ti
on
in
be
ar
ings
of
r
otat
ing
m
ac
hines
v
ia
batch
nor
maliz
ati
on
…
(
Suji
t
K
umar
)
3341
[
5]
D
.
Y
.
P
im
e
nov,
A
.
B
us
ti
ll
o,
S
.
W
oj
c
ie
c
how
s
ki
,
V
.
S
.
S
ha
r
ma
, M
.
K
.
G
upt
a
,
a
nd
M
.
K
unt
oğl
u,
“
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
s
ys
te
ms
f
o
r
to
ol
c
ondi
ti
on
moni
to
r
in
g
in
ma
c
hi
ni
ng:
a
na
ly
s
is
a
nd
c
r
it
ic
a
l
r
e
vi
e
w
,”
J
our
nal
of
I
nt
e
ll
ig
e
nt
M
anuf
ac
tu
r
in
g
,
vol
.
34,
no
.
5,
pp. 2079
–
2121, 2023, doi:
10.1007/s
10845
-
022
-
01923
-
2.
[
6]
E
.
R
a
nya
l,
A
.
S
a
dhu,
a
nd
K
.
J
a
in
,
“
R
oa
d
c
ondi
ti
on
moni
to
r
in
g
us
in
g
s
ma
r
t
s
e
ns
in
g
a
nd
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
:
a
r
e
vi
e
w
,”
Se
ns
or
s
,
vol
. 22, no. 8, 2022, doi
:
10.3390/s
22083044.
[
7]
M
. R
. P
. E
le
nc
he
z
hi
a
n, V
. V
a
dl
a
mudi
, R
. R
a
ih
a
n, K
. R
e
if
s
ni
d
e
r
, a
nd E
. R
e
if
s
ni
de
r
, “
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
i
n r
e
a
l
-
ti
me
di
a
gno
s
ti
c
s
a
nd
pr
ognos
ti
c
s
of
c
ompos
it
e
ma
te
r
ia
ls
a
nd
it
s
unc
e
r
ta
in
ti
e
s
-
a
r
e
vi
e
w
,”
Sm
ar
t
M
at
e
r
ia
ls
and
St
r
uc
tu
r
e
s
,
vol
.
30,
no.
8,
2021,
doi
:
10.1
088/
1361
-
665X/a
c
099f
.
[
8]
D
.
L
i,
Y
.
W
a
ng,
J
.
W
a
ng,
C
.
W
a
ng,
a
nd
Y
.
D
ua
n,
“
R
e
c
e
nt
a
dv
a
nc
e
s
in
s
e
n
s
or
f
a
ul
t
di
a
gnos
is
:
a
r
e
vi
e
w
,
”
Se
ns
o
r
s
and
A
c
tu
at
or
s
,
A
:
P
hy
s
ic
al
, vol
. 309, 2020, doi:
10.1016/j
.s
na
.2020.111990.
[
9]
J
.
G
e
r
tl
e
r
,
“
F
a
ul
t
de
te
c
ti
on
a
nd
di
a
gnos
i
s
,”
in
E
nc
y
c
lo
pe
di
a
of
Sy
s
te
m
s
and
C
ont
r
ol
,
C
ha
m,
S
w
it
z
e
r
la
nd:
S
pr
in
ge
r
,
2021,
pp. 764
–
769
, doi
:
10.1007/978
-
3
-
030
-
44184
-
5_223.
[
10]
S
.
Z
ha
ng,
S
.
Z
ha
ng,
B
.
W
a
ng,
a
nd
T
.
G
.
H
a
be
tl
e
r
,
“
D
e
e
p
le
a
r
ni
ng
a
lg
or
it
hms
f
or
be
a
r
in
g
f
a
ul
t
di
a
gnos
ti
c
s
—
a
c
ompr
e
he
n
s
iv
e
r
e
vi
e
w
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 29857
–
29881, 2020, doi:
10.
1109/AC
C
E
S
S
.2020.2972859.
[
11]
M
.
H
.
M
.
G
ha
z
a
li
a
nd
W
.
R
a
hi
ma
n,
“
V
ib
r
a
ti
on
a
na
ly
s
is
f
or
ma
c
hi
ne
moni
to
r
in
g
a
nd
di
a
gnos
is
:
a
s
y
s
te
ma
ti
c
r
e
vi
e
w
,”
Shoc
k
and
V
ib
r
at
io
n
, vol
. 2021, no. 1, 2021, doi
:
10.1155/2021/
9469318.
[
12]
D
.
N
e
upa
ne
a
nd
J
.
S
e
ok,
“
B
e
a
r
in
g
f
a
ul
t
de
te
c
ti
on
a
nd
di
a
gnos
i
s
us
in
g
c
a
s
e
w
e
s
te
r
n
r
e
s
e
r
ve
uni
ve
r
s
it
y
da
ta
s
e
t
w
it
h
de
e
p
le
a
r
n
i
ng
a
ppr
oa
c
he
s
:
A
r
e
vi
e
w
,
”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 93155
–
93178,
2020, doi:
10.1109/AC
C
E
S
S
.2020.2990528.
[
13]
Y
.
Z
ha
ng,
T
.
Z
hou,
X
.
H
ua
ng,
L
.
C
a
o,
a
nd
Q
.
Z
hou,
“
F
a
ul
t
di
a
gnos
is
of
r
ot
a
ti
ng
ma
c
hi
ne
r
y
ba
s
e
d
on
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
ks
,”
M
e
as
ur
e
m
e
nt
, vol
. 171, 2021, doi:
10.1016/j
.me
a
s
ur
e
me
nt
.2020.108774.
[
14]
S
.
I
of
f
e
a
nd
C
.
S
z
e
ge
dy,
“
B
a
tc
h
nor
ma
li
z
a
ti
on:
a
c
c
e
le
r
a
ti
ng
d
e
e
p
ne
twor
k
tr
a
in
in
g
by
r
e
duc
in
g
in
te
r
na
l
c
ova
r
ia
te
s
hi
f
t,
”
in
3
2nd
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on M
ac
hi
ne
L
e
ar
ni
ng, I
C
M
L
2015
, 2015, pp. 448
–
456.
[
15]
R
.
T
hi
r
ukova
ll
ur
u,
S
.
D
ix
it
,
R
.
K
.
S
e
va
kul
a
,
N
.
K
.
V
e
r
ma
,
a
nd
A
.
S
a
lo
ur
,
“
G
e
ne
r
a
ti
ng
f
e
a
tu
r
e
s
e
ts
f
or
f
a
ul
t
di
a
gnos
is
u
s
in
g
de
noi
s
in
g
s
ta
c
ke
d
a
ut
o
-
e
nc
ode
r
,”
in
2016
I
E
E
E
I
n
te
r
nat
io
nal
C
onf
e
r
e
nc
e
on
P
r
ognos
ti
c
s
and
H
e
al
th
M
anage
m
e
nt
(
I
C
P
H
M
)
,
2016, pp. 1
–
7
, doi
:
10.1109/I
C
P
H
M
.2016.7542865.
[
16]
Z
.
C
he
n
a
nd
W
.
L
i,
“
M
ul
ti
s
e
n
s
or
f
e
a
tu
r
e
f
us
io
n
f
or
be
a
r
in
g
f
a
ul
t
di
a
gnos
is
u
s
in
g
s
pa
r
s
e
a
ut
oe
nc
od
e
r
a
nd
d
e
e
p
be
li
e
f
ne
twor
k,”
I
E
E
E
T
r
ans
ac
ti
ons
on I
ns
tr
um
e
nt
at
io
n and M
e
as
u
r
e
m
e
nt
, vol
.
66, no. 7, pp. 1693
–
1702, 2017, doi:
10.1109/T
I
M
.2017.2669947.
[
17]
K
.
G
r
e
f
f
, R
. K
.
S
r
iv
a
s
t
a
va
, J
. K
o
ut
ni
k,
B
. R
.
S
t
e
u
ne
br
i
nk
, a
nd
J
.
S
c
hmi
dh
ub
e
r
, “
L
S
T
M
:
A
s
e
a
r
c
h
s
p
a
c
e
o
dy
s
s
e
y
,
”
I
E
E
E
T
r
an
s
ac
ti
o
n
s
o
n
N
e
u
r
al
N
e
t
w
o
r
k
s
an
d
L
e
a
r
n
in
g
Sy
s
te
m
s
,
v
ol
.
28
,
no
.
10
,
pp
.
22
2
2
–
22
32
,
20
17
,
doi
:
10
.1
10
9/
T
N
N
L
S
.
20
16
.2
58
29
24
.
[
18]
M
.
Y
ua
n,
Y
.
W
u,
a
nd
L
.
L
in
,
“
F
a
ul
t
di
a
gnos
i
s
a
nd
r
e
ma
in
in
g
us
e
f
ul
li
f
e
e
s
ti
ma
ti
on
of
a
e
r
o
e
ngi
ne
u
s
in
g
L
S
T
M
n
e
ur
a
l
ne
two
r
k,”
in
2016 I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on A
ir
c
r
af
t
U
ti
li
ty
Sy
s
te
m
s
(
A
U
S)
, 2016, pp. 135
–
140
, doi
:
10.1109/AUS.2016.
77480
35.
[
19]
L
.
Y
u,
J
.
Q
u,
F
.
G
a
o,
a
nd
Y
.
T
ia
n,
“
A
nove
l
hi
e
r
a
r
c
hi
c
a
l
a
lg
o
r
it
hm
f
or
be
a
r
in
g
f
a
ul
t
di
a
gnos
is
ba
s
e
d
o
n
s
ta
c
k
e
d
L
S
T
M
,”
Sh
oc
k
and V
ib
r
at
io
n
, vol
. 2019, no. 1, 2019, doi
:
10.1155/2019/
27562
84.
[
20]
X
.
C
he
n,
B
.
Z
ha
ng,
a
nd
D
.
G
a
o,
“
B
e
a
r
in
g
f
a
ul
t
di
a
gnos
i
s
ba
s
e
on
mul
ti
-
s
c
a
le
C
N
N
a
nd
L
S
T
M
mod
e
l,
”
J
ou
r
nal
of
I
nt
e
ll
ig
e
nt
M
anuf
ac
tu
r
in
g
, vol
. 32, no. 4, pp. 971
–
987
, 2021, doi:
10.1007/s
10845
-
020
-
01600
-
2.
[
21]
M
.
H
ua
ng,
T
.
H
u
a
ng,
Y
.
Z
ha
o,
a
nd
W
.
D
a
i,
“
F
a
ul
t
di
a
gnos
i
s
of
r
ol
li
ng
be
a
r
in
g
ba
s
e
d
on
e
mpi
r
ic
a
l
mode
de
c
ompos
it
io
n
a
nd
c
onvolut
io
na
l
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k,”
I
O
P
C
onf
e
r
e
nc
e
S
e
r
ie
s
:
M
at
e
r
ia
ls
Sc
ie
nc
e
and
E
ngi
n
e
e
r
in
g
,
vol
.
1043,
no.
4,
20
21,
doi
:
10.1088/1757
-
899X/1043/4/
042015.
[
22]
X
.
Z
h
a
n
g
,
Y
.
L
i
a
n
g
,
J
.
Z
h
o
u
,
a
n
d
Y
.
Z
a
n
g
,
“
A
n
o
v
e
l
b
e
a
r
i
n
g
f
a
u
lt
d
i
a
g
n
o
s
i
s
m
o
d
e
l
i
n
t
e
g
r
a
t
e
d
p
e
r
m
u
t
a
t
i
o
n
e
n
t
r
o
p
y
,
e
n
s
e
m
b
l
e
e
m
p
i
r
i
c
a
l
m
o
d
e
d
e
c
o
m
p
o
s
i
t
i
o
n
a
n
d
o
p
t
i
m
i
z
e
d
S
V
M
,
”
M
e
a
s
u
r
e
m
e
n
t
,
v
o
l
.
6
9
,
p
p
.
1
6
4
–
179, 2015, doi:
10.1016/
j.m
e
a
s
ur
e
m
e
nt.201
5.03.017.
[
23]
J
.
Z
h
e
n
g
,
H
.
P
a
n
,
a
n
d
J
.
C
h
e
n
g
,
“
R
o
l
l
i
n
g
b
e
a
r
i
n
g
f
a
u
l
t
d
e
t
e
c
t
i
o
n
a
n
d
d
i
a
g
n
o
s
i
s
b
a
s
e
d
o
n
c
o
m
p
o
s
i
t
e
m
u
l
t
i
s
c
a
l
e
f
u
z
z
y
e
n
t
r
o
p
y
a
n
d
e
n
s
e
m
b
l
e
s
u
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
s
,
”
M
e
c
h
a
n
i
c
a
l
S
y
s
t
e
m
s
a
n
d
S
i
g
n
a
l
P
r
o
c
e
s
s
i
n
g
,
v
o
l
.
8
5
,
p
p
.
7
4
6
–
759, 2017, doi:
10.1016/
j.ym
s
s
p.20
16.09.010.
[
24]
A
.
Z
.
H
in
c
hi
a
nd
M
.
T
ki
oua
t,
“
R
ol
li
ng
e
le
me
nt
be
a
r
in
g
r
e
ma
in
in
g
us
e
f
ul
li
f
e
e
s
ti
ma
ti
on
ba
s
e
d
on a
c
onvolut
io
na
l
lo
ng
-
s
hor
t
-
te
r
m
me
mor
y ne
twor
k,”
P
r
oc
e
di
a C
om
put
e
r
Sc
i
e
nc
e
, vol
. 127, pp. 1
23
–
132, 2018, doi:
10.1016/j
.pr
oc
s
.2018.01.106.
[
25]
P
.
P
e
n
g,
W
.
Z
h
a
n
g,
Y
.
Z
ha
ng
,
Y
.
X
u,
H
.
W
a
n
g,
a
n
d
H
.
Z
h
a
ng,
“
C
o
s
t
s
e
n
s
i
ti
v
e
a
c
ti
v
e
l
e
a
r
n
in
g
u
s
i
ng
bi
di
r
e
c
ti
on
a
l
g
a
t
e
d
r
e
c
u
r
r
e
n
t
n
e
ur
a
l
n
e
t
w
or
k
s
f
or
i
mb
a
l
a
n
c
e
d
f
a
u
lt
d
ia
gn
o
s
i
s
,”
N
e
u
r
o
c
om
pu
ti
ng
,
v
ol
.
4
07
,
pp
.
23
2
–
24
5,
2
02
0,
d
oi
:
10
.1
01
6/
j
.n
e
uc
om
.2
02
0.
04
.07
5
.
[
26]
L
.
Z
hi
w
e
i,
“
B
e
a
r
in
g
f
a
ul
t
di
a
gnos
is
of
e
nd
-
to
-
e
nd
mode
l
de
s
ig
n
ba
s
e
d
on
1D
C
N
N
-
G
R
U
ne
twor
k,”
D
is
c
r
e
te
D
y
nam
ic
s
in
N
at
ur
e
and Soc
ie
ty
, vol
. 2022, no. 1, 2022, doi
:
10.1155/2022/
7167821.
[
27]
Y
.
W
a
ng,
D
.
Z
he
ng,
a
nd
R
.
J
ia
,
“
F
a
ul
t
di
a
gnos
is
me
th
od
f
or
M
M
C
-
H
V
D
C
ba
s
e
d
on
B
i
-
G
R
U
ne
ur
a
l
ne
twor
k,”
E
ne
r
gi
e
s
,
vol
.
15,
no. 3, 2022, doi:
10.3390/en1503099
4.
[
28]
Q
.
W
u,
T
.
Z
ong,
W
.
C
he
ng,
Y
.
L
i,
a
nd
P
.
L
i,
“
N
e
twor
k
c
o
ns
tr
uc
ti
on
f
or
be
a
r
in
g
f
a
ul
t
di
a
gnos
is
ba
s
e
d
on
double
a
tt
e
nt
io
n
me
c
ha
ni
s
m,”
C
o
m
put
at
io
nal
I
nt
e
ll
ig
e
nc
e
and N
e
ur
o
s
c
ie
nc
e
, vo
l.
2022, no. 1, 2022, doi
:
10.1155/2022/
3987480.
[
29]
Y
.
F
a
n,
C
.
Z
ha
ng,
Y
.
X
ue
,
J
.
W
a
ng,
a
nd
F
.
G
u,
“
A
be
a
r
in
g
f
a
u
lt
di
a
gnos
is
u
s
in
g
a
s
uppor
t
ve
c
to
r
ma
c
hi
ne
opt
im
is
e
d
by
th
e
s
e
lf
-
r
e
gul
a
ti
ng pa
r
ti
c
le
s
w
a
r
m,”
Shoc
k
and V
ib
r
at
io
n
, vol
. 2020, no.
1, 2020, doi:
10.1155/2020/
9096852.
[
30]
K
.
G
u,
Y
.
Z
ha
ng,
X
.
L
iu
,
H
.
L
i,
a
nd
M
.
R
e
n,
“
D
W
T
-
L
S
T
M
-
ba
s
e
d
f
a
ul
t
di
a
gnos
is
of
r
ol
li
ng
be
a
r
in
gs
w
it
h
mul
ti
-
s
e
ns
or
s
,”
E
le
c
tr
oni
c
s
, vol
. 10, no. 17, 2021, doi
:
10.3390/e
le
c
tr
oni
c
s
1017
2076.
[
31]
X
.
X
u,
D
.
C
a
o,
Y
.
Z
hou,
a
nd
J
.
G
a
o,
“
A
ppl
ic
a
ti
on
o
f
ne
ur
a
l
ne
twor
k
a
lg
or
it
hm
in
f
a
ul
t
di
a
gnos
is
of
me
c
ha
ni
c
a
l
in
te
ll
ig
e
n
c
e
,
”
M
e
c
hani
c
al
Sy
s
te
m
s
and Signal
P
r
oc
e
s
s
in
g
, vol
. 141, 2020, doi:
10.1016/j
.yms
s
p.2020.106625.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Suji
t
Kum
a
r
i
s
w
o
rk
i
n
g
as
t
h
e
A
s
s
i
s
t
an
t
Pro
f
es
s
o
r
i
n
t
h
e
D
ep
ar
t
men
t
o
f
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
a
t
G
o
v
ern
me
n
t
E
n
g
i
n
eeri
n
g
Co
l
l
eg
e,
Ban
k
a.
H
e
co
mp
l
et
e
d
h
i
s
Ph
.
D
.
d
eg
ree
i
n
J
a
n
u
ar
y
2
0
2
5
fro
m
t
h
e
N
a
t
i
o
n
a
l
In
s
t
i
t
u
t
e
o
f
T
ech
n
o
l
o
g
y
,
N
ag
a
l
an
d
,
7
9
7
1
0
3
,
In
d
i
a.
H
e
recei
v
ed
h
i
s
Ma
s
t
er'
s
d
e
g
ree
i
n
t
h
e
y
ear
2
0
1
8
fr
o
m
t
h
e
N
at
i
o
n
al
In
s
t
i
t
u
t
e
o
f
T
ech
n
o
l
o
g
y
,
U
t
t
arak
h
an
d
,
In
d
i
a.
H
i
s
res
earc
h
i
n
t
ere
s
t
s
i
n
c
l
u
d
e
t
h
e
ap
p
l
i
ca
t
i
o
n
o
f
ar
t
i
f
i
ci
al
i
n
t
el
l
i
g
en
ce
i
n
fau
l
t
d
i
ag
n
o
s
i
s
.
H
e
h
as
p
u
b
l
i
s
h
e
d
mo
re
t
h
a
n
1
0
p
ap
er
s
i
n
rep
u
t
e
d
j
o
u
rn
a
l
s
.
H
e
can
b
e
co
n
t
ac
t
e
d
at
emai
l
:
s
u
j
i
t
ar
y
an
0
4
1
2
@
g
ma
i
l
.
co
m.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
4,
Augus
t
2025
:
333
4
-
334
2
3342
M
a
ni
s
h
Kum
a
r
recei
v
ed
a
B.
T
ec
h
.
d
eg
ree
i
n
2
0
2
3
fr
o
m
t
h
e
Mu
zaffar
p
u
r
I
n
s
t
i
t
u
t
e
o
f
T
ech
n
o
l
o
g
y
,
Mu
zaffar
p
u
r,
Bi
h
ar,
In
d
i
a.
H
e
co
mp
l
et
e
d
h
i
s
M.
T
ech
.
fro
m
t
h
e
In
d
i
an
I
n
s
t
i
t
u
t
e
o
f
T
ec
h
n
o
l
o
g
y
Pat
n
a.
H
e
h
a
s
p
u
b
l
i
s
h
e
d
mo
re
t
h
a
n
4
p
ap
er
s
i
n
rep
u
t
e
d
j
o
u
rn
al
s
.
H
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
man
i
s
h
k
u
mar1
8
1
6
0
5
@
g
ma
i
l
.
co
m.
D
r.
C
heta
n
B
a
rde
rece
i
v
e
d
t
h
e
B.
E
.
d
eg
ree
i
n
E
l
e
ct
ro
n
i
c
s
an
d
C
o
mmu
n
i
ca
t
i
o
n
E
n
g
i
n
eeri
n
g
fr
o
m
t
h
e
RG
PV
U
n
i
v
er
s
i
t
y
,
B
h
o
p
al
,
I
n
d
i
a,
i
n
2
0
0
8
,
a
n
d
t
h
e
M.
T
ec
h
.
d
eg
rees
i
n
N
an
o
t
ec
h
n
o
l
o
g
y
fro
m
t
h
e
V
I
T
U
n
i
v
er
s
i
t
y
,
V
e
l
l
o
re,
In
d
i
a,
i
n
2
0
1
3
.
H
e
recei
v
ed
Ph
.
D
.
d
e
g
ree
fro
m
t
h
e
N
at
i
o
n
a
l
In
s
t
i
t
u
t
e
o
f
T
ec
h
n
o
l
o
g
y
,
J
ams
h
ed
p
u
r
,
In
d
i
a.
H
i
s
res
ear
ch
i
n
t
ere
s
t
i
n
cl
u
d
e
s
Z
O
R
A
n
t
e
n
n
a,
met
ama
t
eri
a
l
s
a
n
d
i
t
s
a
p
p
l
i
ca
t
i
o
n
s
,
el
ect
ro
mag
n
et
i
c
w
av
e
a
b
s
o
rb
er
s
.
Pres
en
t
l
y
,
h
e
i
s
w
o
rk
i
n
g
as
an
A
s
s
i
s
t
a
n
t
Pr
o
fes
s
o
r
i
n
t
h
e
D
e
p
art
m
en
t
o
f
E
l
ect
r
o
n
i
cs
a
n
d
Co
mm
u
n
i
cat
i
o
n
E
n
g
i
n
eeri
n
g
,
In
d
i
a
n
In
s
t
i
t
u
t
e
o
f
I
n
fo
rma
t
i
o
n
T
ech
n
o
l
o
g
y
Bh
a
g
al
p
u
r,
Bi
h
ar
I
n
d
i
a.
H
e
h
a
s
p
u
b
l
i
s
h
ed
mo
re
t
h
an
2
5
p
a
p
ers
i
n
rep
u
t
e
d
j
o
u
r
n
a
l
s
.
H
e
can
b
e
c
o
n
t
act
e
d
at
emai
l
:
c
b
ard
e.
ece@
i
i
i
t
b
h
.
ac.
i
n
.
D
r.
P
ra
ka
s
h
R
a
nja
n
recei
v
ed
t
h
e
B.
E
.
d
eg
ree
i
n
E
l
e
ct
ro
n
i
c
s
an
d
Co
mm
u
n
i
cat
i
o
n
E
n
g
i
n
eeri
n
g
fr
o
m
t
h
e
A
n
n
a
U
n
i
v
er
s
i
t
y
,
Ch
e
n
n
a
i
,
In
d
i
a,
i
n
2
0
0
9
,
an
d
t
h
e
M.
T
ec
h
.
d
eg
ree
s
i
n
E
l
ec
t
ro
n
i
c
s
an
d
Co
mmu
n
i
ca
t
i
o
n
E
n
g
i
n
eer
i
n
g
fro
m
t
h
e
Y
MCA
U
n
i
v
er
s
i
t
y
,
H
ary
a
n
a,
In
d
i
a,
i
n
2
0
1
2
.
In
2
0
1
9
,
h
e
co
m
p
l
e
t
ed
h
i
s
Ph
.
D
.
at
t
h
e
N
a
t
i
o
n
a
l
In
s
t
i
t
u
t
e
o
f
T
ec
h
n
o
l
o
g
y
,
J
ams
h
ed
p
u
r,
In
d
i
a
an
d
s
u
b
s
eq
u
en
t
l
y
,
j
o
i
n
ed
D
ep
art
me
n
t
o
f
E
l
ect
r
o
n
i
c
s
an
d
Co
mm
u
n
i
cat
i
o
n
E
n
g
i
n
eeri
n
g
,
In
d
i
an
In
s
t
i
t
u
t
e
o
f
I
n
fo
rma
t
i
o
n
T
ec
h
n
o
l
o
g
y
,
Bh
ag
a
l
p
u
r,
as
A
s
s
i
s
t
a
n
t
Pro
fe
s
s
o
r.
H
e
i
s
a
g
rad
u
at
e
s
t
u
d
en
t
memb
er
o
f
IE
E
E
.
H
i
s
res
ea
rch
i
n
t
ere
s
t
i
n
c
l
u
d
es
met
a
s
u
rface
s
,
met
amat
e
ri
a
l
s
an
d
i
t
s
ap
p
l
i
ca
t
i
o
n
s
,
el
ect
ro
ma
g
n
e
t
i
c
w
a
v
e
ab
s
o
rb
er
s
,
Z
O
R
an
t
en
n
a
,
an
d
s
o
ft
co
m
p
u
t
i
n
g
o
p
t
i
mi
zat
i
o
n
t
ech
n
i
q
u
e
s
.
H
e
can
b
e
c
o
n
t
act
e
d
at
emai
l
:
p
ran
j
a
n
.
ece@
i
i
i
t
b
h
.
ac.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.