T
E
L
KO
M
N
I
KA
T
e
lec
om
m
u
n
icat
ion
,
Com
p
u
t
i
n
g,
E
lec
t
r
on
ics
an
d
Cont
r
ol
Vol.
18
,
No.
1
,
F
e
br
ua
r
y
2020
,
pp.
251
~
257
I
S
S
N:
1693
-
6930,
a
c
c
r
e
dit
e
d
F
ir
s
t
G
r
a
de
by
Ke
me
nr
is
tekdikti
,
De
c
r
e
e
No:
21/E
/KP
T
/2018
DO
I
:
10.
12928/
T
E
L
KO
M
NI
KA
.
v18i1.
9627
251
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
A
n
al
og c
ir
c
u
it
f
au
lt
d
ia
gn
osi
s
vi
a FOA
-
L
S
S
V
M
Yu
We
n
xin
Co
l
l
eg
e
o
f
E
l
ec
t
ri
ca
l
&
In
fo
rma
t
i
o
n
E
n
g
i
n
eeri
n
g
,
H
u
n
a
n
U
n
i
v
ers
i
t
y
,
P.
R.
Ch
i
n
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
Apr
1
7
,
201
8
R
e
vis
e
d
Oc
t
15
,
20
19
Ac
c
e
pted
Nov
3
,
20
19
At
pr
e
s
e
nt,
the
r
e
s
e
a
r
c
h
on
f
a
ult
de
tec
ti
on
a
nd
diag
nos
is
tec
hnology
is
ve
r
y
s
igni
f
ica
nt
to
im
p
r
ove
the
r
e
li
a
bil
it
y
of
the
e
q
uipm
e
nt,
whic
h
c
a
n
gr
e
a
tl
y
im
pr
ove
the
s
a
f
e
ty
a
nd
e
f
f
icie
nc
y
of
the
e
q
uipm
e
nt.
T
his
pa
pe
r
pr
opos
e
s
a
ne
w
f
a
ult
de
tec
ti
on
a
nd
diagnos
is
mea
ns
b
a
s
e
d
on
the
F
OA
-
L
S
S
VM
a
lgor
it
hm.
E
xpe
r
im
e
ntal
r
e
s
ult
s
de
mons
tr
a
te
that
the
a
lgor
it
hm
is
e
f
f
e
c
ti
ve
f
or
the
de
tec
ti
on
a
nd
d
iagnos
is
of
a
na
log
c
ir
c
uit
f
a
ult
s
.
I
n
a
ddit
ion,
the
model
a
ls
o
de
mons
tr
a
te
good
ge
ne
r
a
li
z
a
ti
on
a
bil
it
y.
K
e
y
w
o
r
d
s
:
Ana
log
c
ir
c
uit
F
OA
L
S
S
VM
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
:
Yu
W
e
nxin
,
C
oll
e
ge
of
E
lec
tr
ica
l
&
I
nf
or
mation
E
nginee
r
ing,
Huna
n
Unive
r
s
it
y,
C
ha
ngs
ha
,
4100
82,
P
.
R
.
C
hina
.
E
mail:
13874894700@163
.
c
om
1.
I
NT
RODU
C
T
I
ON
Ac
c
or
ding
to
s
tatis
ti
c
s
,
a
t
pr
e
s
e
nt,
80
%
of
de
vice
s
in
e
lec
tr
onic
s
ys
tems
a
r
e
digi
tal,
but
80%
o
f
f
a
ult
s
oc
c
ur
on
a
n
a
log
de
vice
s
.
At
the
s
a
me
ti
me,
the
tes
t
c
os
t
of
the
a
na
log
c
ir
c
uit
pa
r
t
a
c
c
ount
s
f
or
80%
of
the
tot
a
l
tes
t
c
os
t,
thenc
e
,
it
is
ve
r
y
im
por
tant
to
c
a
r
r
y
out
dis
c
us
s
on
f
a
ult
diagnos
is
of
a
na
log
c
ir
c
uit
s
.
I
n
r
e
c
e
nt
ye
a
r
s
,
many
s
c
holar
s
ha
ve
c
ondu
c
ted
e
xtens
ive
r
e
s
e
a
r
c
h
in
the
f
ield
of
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
a
nd
ha
ve
a
c
hieve
d
ma
ny
e
xc
e
ll
e
nt
r
e
s
ult
s
[
1
-
9]
.
How
e
ve
r
,
t
he
a
na
log
c
ir
c
uit
it
s
e
lf
ha
s
the
c
ha
r
a
c
ter
is
ti
c
s
of
p
oor
f
a
ult
model,
c
omponent
to
ler
a
nc
e
,
f
a
ult
pa
r
a
mete
r
c
onti
nuit
y
a
nd
c
ir
c
uit
nonli
ne
a
r
it
y
.
S
uc
h
c
ha
r
a
c
ter
is
ti
c
s
make
the
de
ve
lopm
e
nt
of
a
na
log
c
i
r
c
uit
f
a
ult
diagnos
i
s
tec
hnology
s
low,
a
nd
ther
e
is
s
ti
ll
no
p
r
a
c
ti
c
a
l
method.
I
n
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
,
the
e
xtr
a
c
ti
on
o
f
f
a
ult
f
e
a
tur
e
s
is
a
ve
r
y
im
por
tant
li
nk
,
a
nd
the
qua
li
ty
of
the
e
xtr
a
c
ti
on
r
e
s
ult
s
will
dir
e
c
tl
y
a
f
f
e
c
t
the
f
inal
diagnos
is
a
c
c
ur
a
c
y
r
a
te
.
O
r
din
a
r
y
f
e
a
tur
e
e
xtr
a
c
ti
on
methods
mainly
include
P
C
A,
wa
ve
let
a
na
lys
is
,
ke
r
ne
l
a
na
lys
is
,
e
tc.
[
10
-
13]
.
T
he
s
e
methods
ha
ve
their
li
mi
tations
.
F
or
e
xa
mpl
e
,
the
P
C
A
method
is
only
s
u
it
a
ble
f
or
li
ne
a
r
f
e
a
tur
e
e
xtr
a
c
ti
on.
W
a
ve
let
a
na
lys
is
a
nd
nuc
lea
r
a
na
lys
is
invo
lve
the
s
e
lec
ti
on
a
nd
c
ons
ider
a
ti
on
of
many
f
a
c
tor
s
s
uc
h
a
s
wa
ve
let
ba
s
e
a
nd
nuc
lea
r
pa
r
a
mete
r
s
,
whic
h
a
r
e
gr
e
a
tl
y
in
f
luenc
e
d
by
e
xpe
r
i
e
nc
e
.
M
or
e
ove
r
,
in
e
s
s
e
nc
e
,
thes
e
a
na
lys
i
s
method
s
a
nd
da
ta
a
r
e
is
olate
d
f
r
om
e
a
c
h
other
,
a
nd
it
is
dif
f
icult
to
e
ns
ur
e
t
ha
t
the
e
xtr
a
c
ted
f
e
a
tur
e
s
a
r
e
the
e
s
s
e
nti
a
l
c
ha
r
a
c
ter
is
ti
c
s
of
the
da
ta.
F
a
ult
c
las
s
if
ica
ti
on
a
nd
identi
f
ica
ti
on
is
a
nother
k
e
y
point
of
f
a
ult
diagnos
is
f
or
a
na
log
c
ir
c
uit
s
.
I
n
r
e
c
e
nt
ye
a
r
s
,
the
c
onti
nuous
de
ve
lopm
e
nt
of
va
r
ious
a
r
ti
f
icia
l
int
e
ll
igenc
e
a
l
gor
it
hms
ha
s
g
a
ve
bir
th
a
ne
w
idea
s
f
or
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
.
T
he
y
a
r
e
n
e
ur
a
l
ne
twor
ks
(
NN
)
[
14
-
19
]
,
s
uppor
t
ve
c
tor
mac
hine
(
S
VM
)
[
20
-
24]
,
de
e
p
lea
r
n
ing
[
25
-
27]
a
nd
s
o
on.
T
he
main
idea
o
f
the
f
a
ult
diagnos
is
method
of
ne
ur
a
l
ne
twor
k
is
:
t
he
map
ping
be
twe
e
n
f
a
ult
s
ympt
oms
a
nd
f
a
ult
t
ype
s
is
e
s
tablis
he
d
thr
ough
lea
r
ning
be
twe
e
n
ne
twor
k
laye
r
s
.
T
he
node
s
of
the
input
laye
r
a
r
e
c
a
us
e
d
to
c
or
r
e
s
p
ond
to
f
a
ult
s
ympt
oms
,
a
nd
the
node
s
o
f
the
out
put
laye
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
251
-
257
252
c
or
r
e
s
pond
to
f
a
ult
types
.
T
hus
,
the
r
e
a
s
oning
p
r
oc
e
s
s
f
r
om
f
a
ult
s
ympt
om
to
f
a
ult
type
c
a
n
be
r
e
a
li
z
e
d.
T
he
ne
ur
a
l
ne
twor
k
c
a
n
s
e
t
the
ne
twor
k
s
tr
uc
tur
e
a
c
c
or
ding
to
r
e
quir
e
ments
a
nd
a
ppr
oxim
a
te
th
e
nonli
ne
a
r
f
unc
ti
on
with
a
r
bi
tr
a
r
y
pr
e
c
is
ion.
B
ut
the
lea
r
ning
of
the
ne
twor
k
r
e
qui
r
e
s
a
la
r
ge
numbe
r
o
f
c
ir
c
ui
t
f
a
il
u
r
e
s
a
mpl
e
s
.
T
he
r
e
f
or
e
,
f
or
s
ys
tems
that
c
a
nnot
obtain
a
lar
ge
a
mount
o
f
f
a
ult
da
ta,
the
us
e
of
ne
ur
a
l
ne
tw
or
ks
will
be
li
mi
ted.
At
the
s
a
me
ti
me,
how
to
e
ns
ur
e
the
int
e
gr
it
y
a
nd
typi
c
a
li
ty
of
the
f
a
ult
s
a
mpl
e
a
nd
the
c
onv
e
r
ge
nc
e
,
tr
a
ini
ng
s
pe
e
d
a
nd
r
e
a
l
-
ti
me
diagnos
is
of
the
meth
od
a
r
e
the
bott
lene
c
ks
r
e
s
tr
icting
the
de
ve
lopm
e
nt
of
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
tec
hnology
ba
s
e
d
on
ne
ur
a
l
n
e
twor
k.
As
a
pa
tt
e
r
n
r
e
c
ognit
ion
method
ba
s
e
d
on
s
tatis
ti
c
a
l
lea
r
ning
theor
y
,
S
VM
ha
s
many
unique
a
dva
ntag
e
s
,
f
or
e
xa
mpl
e
s
olvi
ng
s
mall
s
a
mpl
e
s
,
n
onli
ne
a
r
a
nd
high
-
dim
e
ns
ional
pa
tt
e
r
n
r
e
c
ognit
ion,
a
nd
c
a
n
be
a
ppli
e
d
to
othe
r
mac
hine
lea
r
ning
pr
oblems
,
s
uc
h
a
s
f
unc
ti
on
f
it
ti
ng
.
How
e
ve
r
,
whe
n
c
ons
tr
u
c
ti
ng
the
opti
mal
c
las
s
if
ica
ti
on
hype
r
plane
,
S
VM
only
pa
ys
a
tt
e
nti
on
to
the
s
e
pa
r
a
bil
it
y
be
twe
e
n
the
da
ta
c
la
s
s
e
s
a
nd
ignor
e
s
the
s
tr
uc
tur
a
l
inf
o
r
mation
of
the
da
ta
withi
n
the
c
las
s
.
T
his
r
e
s
ult
s
in
the
c
las
s
if
ica
ti
on
bounda
r
y
o
f
the
da
ta
be
ing
too
s
moot
h
whe
n
the
da
ta
ha
s
a
n
onli
ne
a
r
manif
old
s
tr
uc
tur
e
,
whic
h
s
e
r
ious
ly
a
f
f
e
c
ts
the
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
of
the
S
VM
.
I
n
p
r
a
c
ti
c
a
l
pr
oblems
,
mos
t
of
the
s
a
mpl
e
s
a
r
e
highl
y
c
or
r
e
late
d,
that
is
,
they
a
r
e
a
t
lea
s
t
pa
r
ti
a
ll
y
dis
tr
ibut
e
d
on
a
low
-
dim
e
ns
ional
manif
old.
I
n
pa
r
t
icula
r
,
the
r
e
is
of
ten
a
n
onli
ne
a
r
r
e
lations
hip
b
e
twe
e
n
the
output
of
the
ge
ne
r
a
l
c
ir
c
uit
a
nd
the
f
a
ult
mec
ha
nis
m
of
the
c
ir
c
uit
.
T
he
r
e
f
or
e
,
the
tr
a
dit
ional
S
VM
only
pa
ys
a
tt
e
nti
on
to
the
int
e
r
-
c
las
s
s
pa
c
ing
inf
or
mation,
whic
h
is
not
e
nough
f
or
t
he
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
c
las
s
if
ica
ti
on
pr
oblem.
A
t
pr
e
s
e
nt,
the
r
e
s
e
a
r
c
h
r
e
s
ult
s
ba
s
e
d
on
de
e
p
lea
r
ning
a
r
e
r
e
latively
f
e
w
in
a
na
log
c
ir
c
uit
f
a
ult
diagnos
is
.
T
he
di
f
f
iculty
in
the
f
ield
of
f
a
ult
diag
nos
is
li
e
s
in
the
a
djus
tm
e
nt
o
f
pa
r
a
mete
r
s
.
T
he
p
a
r
a
mete
r
s
e
lec
ti
on
a
f
f
e
c
ts
the
a
c
c
ur
a
c
y
of
f
a
ult
s
ign
e
xt
r
a
c
t
ion.
T
he
r
e
is
no
s
ys
tema
ti
c
theor
e
ti
c
a
l
s
ys
tem
to
guide
the
a
djus
tm
e
nt
of
de
e
p
lea
r
ning
pa
r
a
mete
r
s
.
T
he
a
dj
us
tm
e
nt
of
r
e
leva
nt
pa
r
a
mete
r
s
of
ten
ne
e
ds
to
be
s
e
lec
ted
a
c
c
or
ding
to
a
c
tual
e
xpe
r
ienc
e
.
De
e
p
lea
r
ning
t
r
a
ini
ng
is
ti
me
c
ons
umi
ng.
F
o
r
m
a
c
hine
lea
r
ning,
the
ve
r
if
ica
ti
on
p
r
oc
e
s
s
of
model
c
or
r
e
c
tnes
s
is
c
ompl
e
x
a
nd
th
e
f
e
a
tur
e
s
f
ound
a
r
e
not
int
uit
ive
e
nough.
F
a
ult
diagnos
is
r
e
quir
e
s
the
model
to
ident
if
y
the
type
of
f
a
ult
in
a
ti
mely
a
nd
r
a
pid
manne
r
.
T
h
is
is
a
dif
f
icult
point
to
ove
r
c
ome
in
th
e
a
ppli
c
a
ti
on
of
the
de
e
p
lea
r
ning
method.
I
n
thi
s
pa
pe
r
,
we
we
r
e
ins
pir
e
d
to
r
e
c
e
ive
the
a
bove
me
thod,
we
pr
e
s
e
nt
F
OA
-
L
S
S
VM
model
f
or
c
i
r
c
uit
f
a
ult
diag
nos
is
.
T
he
e
xa
mpl
e
o
f
S
a
ll
e
n
-
Ke
y
ba
nd
pa
s
s
f
il
te
r
c
ir
c
uit
dis
play
that
our
r
e
s
ult
ing
diagnos
ti
c
s
ys
tem
c
a
n
e
f
f
e
c
ti
ve
ly
c
las
s
if
y
the
f
a
ult
y
c
omponents
of
a
na
log
c
ir
c
uit
s
whe
n
it
is
tes
ted,
a
nd
it
ha
s
a
c
ompetit
ive
c
las
s
if
ica
ti
on
pe
r
f
o
r
manc
e
.
2.
F
L
Y
OP
T
I
M
I
Z
AT
I
ON
AL
GO
RI
T
HM
F
r
uit
F
ly
Optim
iza
ti
on
Algor
it
hm
(
F
OA
)
[
28
,
29
]
is
a
n
e
mer
ging
s
wa
r
m
int
e
ll
igent
opti
mi
z
a
ti
on
a
lgor
it
hm
ba
s
e
d
on
the
bioni
c
s
p
r
inciple
o
f
f
r
uit
f
ly
f
o
r
a
ging
be
ha
vio
r
.
I
t
is
ba
s
e
d
on
the
f
ood
s
e
a
r
c
hing
be
ha
vior
of
the
f
r
uit
f
ly
.
I
n
c
ompar
is
on
to
a
ny
ot
he
r
s
pe
c
ies
,
f
r
uit
f
ly
ha
s
e
xc
e
pti
ona
l
olf
a
c
tor
y
a
n
d
vis
ua
l
s
e
ns
e
s
.
T
he
or
ga
n
r
e
s
pons
ibl
e
f
or
the
s
e
ns
e
of
s
mel
l
with
-
in
f
r
uit
f
li
e
s
c
a
n
s
e
a
r
c
h
of
a
ll
kinds
of
s
mells
f
loating
in
the
a
ir
,
a
ls
o
it
is
a
ble
to
s
mell
the
f
ood
tas
te
that
i
s
ne
a
r
ly
40
km.
I
t
ha
s
a
buil
t
-
in
ol
f
a
c
tor
y
or
ga
n
th
a
t
a
ll
ows
them
to
pick
up
di
f
f
e
r
e
nt
odo
r
mol
e
c
ules
in
the
a
i
r
a
n
d
to
de
ter
mi
ne
the
s
our
c
e
of
their
f
ood.
T
he
r
e
a
f
t
e
r
,
it
ge
ts
c
los
e
r
to
the
s
our
c
e
s
,
a
nd
it
s
s
ha
r
p
e
ye
s
ight
wa
s
us
e
d
to
f
ind
f
ood,
a
ls
o
it
us
e
s
the
wa
y
ba
c
k
to
it
s
s
wa
r
m.
F
OA
's
ope
r
a
ti
on
is
s
im
ple,
e
a
s
y
to
im
pleme
nt
,
a
nd
ha
s
s
tr
o
ng
loca
l
s
e
a
r
c
h
c
a
pa
bil
it
ies
.
T
he
s
teps
f
or
a
n
it
e
r
a
t
iv
e
s
e
a
r
c
h
f
or
f
ood
by
the
Dr
os
ophil
a
population
a
r
e
a
s
f
oll
o
ws
:
-
S
tep
1.
De
f
ine
a
f
r
uit
f
ly
s
wa
r
m’
s
loca
ti
on
r
a
ndoml
y
.
_
;
_
(
1
)
-
S
tep
2.
Give
f
r
ui
t
f
ly
indi
viduals
r
a
ndom
dis
tanc
e
a
nd
dir
e
c
ti
on
to
s
e
a
r
c
h
f
or
f
ood
us
ing
their
s
e
ns
e
of
s
mell
.
{
=
_
+
=
_
+
(
2
)
-
S
tep
3.
S
ince
the
pos
it
ion
of
the
f
ood
is
unknown
,
f
ir
s
t
of
a
ll
,
the
dis
tanc
e
f
r
om
the
or
igi
n
(
Dis
t
i
)
is
e
s
ti
mate
d,
a
nd
then
the
tas
te
c
onc
e
ntr
a
ti
on
judgm
e
nt
va
lue
(
S
i
)
is
c
a
lcula
ted,
whic
h
is
the
inve
r
s
e
of
the
dis
tanc
e
.
=
(
2
+
2
)
;
=
1
⁄
(
3
)
-
S
tep
4.
S
ubs
ti
tut
ing
the
tas
te
c
onc
e
ntr
a
ti
on
judg
men
t
va
lue
(
S
i
)
int
o
the
tas
te
c
onc
e
ntr
a
ti
on
judg
ment
f
unc
ti
on
(
or
c
a
ll
e
d
f
it
ne
s
s
f
unc
ti
on)
,
s
o
that
the
tas
te
c
onc
e
ntr
a
ti
on
(
S
mell
i
)
of
the
indi
vidual
pos
it
ion
of
the
f
r
uit
f
ly
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
A
nalog
c
ir
c
uit
faul
t
diagnos
is
v
ia
F
OA
-
L
SS
V
M
(
Yu
W
e
nx
in
)
253
S
m
e
l
l
=
(
)
(
4
)
-
S
tep
5.
F
ind
the
highes
t
-
dos
e
f
r
uit
f
ly
in
thi
s
popul
a
ti
on
(
maxi
mum
va
lue)
[
]
=
m
a
x
(
S
m
e
l
l
)
(
5
)
-
S
tep
6.
M
a
int
a
in
the
be
s
t
s
mell
c
onc
e
ntr
a
ti
on
va
lu
e
a
nd
x
,
y
c
oor
dinate
;
the
D
r
os
ophil
a
s
wa
r
ms
will
de
tec
t
thi
s
pos
it
ion
a
nd
f
ly
towa
r
ds
it
.
{
=
_
=
(
)
_
=
(
)
(
6
)
-
S
tep
7.
P
e
r
f
or
m
it
e
r
a
ti
ve
opti
mi
z
a
ti
on,
r
e
pe
a
t
s
tep
s
2
-
6
a
nd
de
ter
mi
ne
whe
ther
the
tas
te
c
onc
e
ntr
a
ti
on
is
be
tt
e
r
than
that
in
the
pr
e
vious
it
e
r
a
ti
o
n;
if
s
o,
go
t
o
s
tep
6
.
3.
L
E
AST
S
QUAR
E
S
S
UP
P
ORT
VE
CT
OR
M
AC
HI
NE
S
T
he
s
uppor
t
ve
c
to
r
mac
hine
(
S
VM
)
[
30
,
31]
map
s
the
s
a
mpl
e
s
pa
c
e
to
a
high
-
dim
e
ns
ional
or
e
ve
n
inf
ini
te
-
dim
e
ns
ional
f
e
a
tur
e
s
pa
c
e
thr
ough
a
non
-
li
ne
a
r
mapping,
s
o
that
the
non
-
li
ne
a
r
ly
s
e
pa
r
a
b
le
pr
oblem
in
the
or
igi
na
l
s
a
mpl
e
s
pa
c
e
is
tr
a
ns
f
or
med
int
o
a
l
in
e
a
r
ly
s
e
pa
r
a
b
le
pr
oblem
in
the
f
e
a
tur
e
s
pa
c
e
.
S
ta
r
t
ing
f
r
om
the
mac
hine
lea
r
ning
los
s
f
unc
ti
on
,
S
uyke
ns
e
t
a
l.
p
r
opos
e
d
a
lea
s
t
s
qu
a
r
e
s
s
uppor
t
ve
c
tor
mac
hine
(
L
S
S
VM
)
[
32
]
,
whic
h
us
e
s
the
s
e
c
o
nd
no
r
m
in
the
objec
ti
ve
f
unc
ti
on
of
it
s
opti
mi
z
a
ti
on
pr
oblem
.
T
he
e
qua
li
ty
c
ons
tr
a
int
c
ondit
ion
is
us
e
d
ins
tea
d
of
the
inequa
li
ty
c
ons
tr
a
int
c
ondit
ion
in
the
S
VM
s
tanda
r
d
a
lgor
it
hm,
s
o
that
the
op
ti
mi
z
a
ti
on
p
r
oblem
of
the
L
S
S
VM
meth
od
be
c
omes
a
s
olut
ion
o
f
a
s
e
t
of
li
ne
a
r
e
qua
ti
ons
obtaine
d
by
Kuhn
-
T
uc
ke
r
c
ondit
ion.
T
his
make
s
it
pos
s
ibl
e
to
r
e
duc
e
the
c
omput
a
ti
ona
l
c
ompl
e
xit
y,
incr
e
a
s
e
the
ge
ne
r
a
li
z
a
ti
on
a
bil
i
ty
a
nd
the
s
olut
ion
s
pe
e
d
wh
e
n
the
e
xtr
e
me
c
ondit
ions
a
r
e
met,
a
nd
it
c
a
n
be
e
f
f
e
c
ti
ve
ly
a
ppli
e
d
to
pa
tt
e
r
n
r
e
c
ognit
ion
a
nd
f
unc
ti
on
e
s
ti
mation.
I
n
L
S
S
VM
,
the
r
e
gr
e
s
s
ion
is
e
xpr
e
s
s
e
d
a
s
given
be
low:
m
in
,
,
(
,
)
=
1
2
⁄
‖
‖
2
+
2
∑
2
=
1
(
7
)
.
.
=
(
)
+
+
,
=
1
,
2
,
…
,
1
,
whe
r
e
is
the
r
e
gular
iza
ti
on
pa
r
a
mete
r
,
de
ter
m
in
ing
the
tr
a
de
of
f
be
twe
e
n
the
f
it
ti
ng
e
r
r
o
r
mi
nim
iza
ti
on
a
nd
s
moot
hne
s
s
,
a
nd
is
e
r
r
o
r
va
r
iable
.
T
he
L
a
gr
a
ngian
e
qu
a
ti
on
is
de
f
ined
a
s
f
oll
ows
:
(
,
,
,
)
=
(
,
)
−
∑
[
(
)
+
+
−
]
=
1
(
8
)
o
pti
mi
z
e
(
8)
,
w
e
ge
t
the
opti
mal
s
olut
ion
o
f
the
f
ol
lowing
c
ondit
ions
:
{
=
∑
(
)
=
1
,
∑
−
0
=
1
,
=
=
(
)
+
+
(
9
)
o
mi
tt
ing
a
nd
lea
ds
to
the
Ka
r
us
h
–
Kuhn
–
T
uc
ke
r
(
KK
T
)
c
ondit
ions
:
[
]
=
[
0
Ω
+
/
]
[
0
]
(
10
)
w
he
r
e
=
[
1
,
2
,
…
,
]
,
=
[
1
,
2
,
…
,
]
,
=
[
1
,
1
,
…
,
1
]
,
Ω
(
(
)
(
)
)
×
is
ke
r
ne
l
f
unc
ti
on
matr
ix
,
r
e
pr
e
s
e
nts
the
identit
y
matr
ix
.
De
f
ine
(
,
)
=
(
)
(
)
,
whic
h
is
s
a
ti
s
f
ied
with
M
e
r
c
e
r
’
s
c
ondit
ion.
I
n
the
pa
e
r
,
W
e
c
hoos
e
Ga
us
s
i
a
n
R
a
dial
B
a
s
is
F
unc
ti
on
(
R
B
F
)
a
s
the
ke
r
ne
l
f
unc
ti
on
,
a
s
is
mea
n
e
d
in:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
251
-
257
254
(
,
)
=
e
xp
{
−
|
−
|
2
2
}
(
11
)
whe
r
e
int
r
oduc
e
s
a
pos
it
ive
r
e
a
l
numbe
r
,
take
n
int
o
a
c
c
ount
a
s
the
ke
r
ne
l
f
unc
ti
on
.
S
o,
the
f
ol
lowing
r
e
lations
hip
is
f
ound
a
s
the
f
inal
r
e
s
ult
:
(
)
=
∑
(
,
)
+
=
1
(
12
)
I
t
s
hould
be
note
d
that
the
pe
r
f
o
r
manc
e
of
the
L
S
S
VM
model
is
s
igni
f
ica
ntl
y
a
f
f
e
c
ted
by
the
ke
r
ne
l
f
unc
ti
on
width
c
oe
f
f
icie
nt
a
nd
the
r
e
gular
iza
ti
on
f
a
c
tor
,
the
width
of
the
R
B
F
is
a
f
f
e
c
ted
by
the
wi
dth
σ,
a
nd
the
c
ompl
e
xit
y
a
nd
punis
hment
a
r
e
a
f
f
e
c
ted
by
.
4.
F
OA
-
L
S
S
VM
AL
GO
RI
T
HM
I
n
thi
s
s
e
c
ti
on,
a
mong
the
methods
pr
opos
e
d
in
thi
s
pa
pe
r
,
the
opti
mi
z
a
ti
on
of
the
L
S
S
VM
c
las
s
if
ier
by
F
OA
is
s
hown
a
s
f
oll
ow
s
:
-
No
.
1.
L
e
t
us
a
s
s
ume
the
maximum
number
of
it
e
r
a
ti
ons
(
maxge
n)
,
population
s
ize
(
s
ize
pop
)
,
a
nd
we
a
ls
o
c
a
n
r
a
ndom
ly
e
mer
ge
a
f
r
ui
t
f
ly
s
wa
r
m’
s
s
tar
ti
ng
po
s
it
ion
(
I
nit
X_a
xis
,
I
nit
Y_a
xis
)
in
or
de
r
to
c
r
e
a
te
r
a
ndom
f
li
ght
dis
tanc
e
(
FR
).
-
No
.
2.
S
ur
ppos
e
ge
n
=
0
,
it
’
s
a
s
s
igned
that
e
a
c
h
f
r
uit
f
ly
(
F
l
y
i
)
r
e
s
pe
c
ti
ve
ly
looks
f
or
f
ood
towa
r
d
a
r
a
ndom
di
r
e
c
ti
on,
a
nd
it
goe
s
f
or
a
r
a
ndom
a
mount
of
d
is
tanc
e
.
(
,
:
)
=
_
+
×
−
(
,
:
)
=
_
+
×
−
(
13
)
,
a
r
e
C
ons
tants
whic
h
c
a
n
be
s
e
lec
ted.
-
No
.
3.
C
a
lcula
te
the
dis
tanc
e
of
the
ini
ti
a
l
pos
it
io
n
Dis
t
i
,
then
we
c
a
n
de
ter
mi
ne
the
va
l
ue
of
the
s
mell
c
onc
e
ntr
a
ti
on
S
i
.
P
r
ogr
a
m
Dis
t
i
whic
h
is
de
not
ed
by
(
D
(
i
,
1
)
,
D
(
i
,
2
)
)
,
s
o
we
ha
ve
:
(
,
1
)
=
√
(
,
1
)
2
+
(
,
1
)
2
(
,
2
)
=
√
(
,
2
)
2
+
(
,
2
)
2
(
14
)
L
e
t
(
,
1
)
=
1
(
,
1
)
⁄
,
(
,
2
)
=
1
(
,
2
)
⁄
(
15)
s
o,
we
c
a
n
ge
t
the
c
onc
lus
ion
that
is
r
e
pr
e
s
e
nted
b
y
(
(
,
1
)
,
(
,
2
)
)
.
L
e
t’
s
put
int
o
the
model
of
L
S
S
VM
.
W
e
a
s
s
ume
=
×
(
,
1
)
,
2
=
(
,
2
)
,
whe
r
e
is
C
ons
tant
whic
h
c
a
n
be
s
e
lec
ted.
[
,
]
a
r
e
P
a
r
a
mete
r
s
of
L
S
S
V
M
,
whic
h
c
a
n
be
r
e
pr
e
s
e
nt
e
d
by
[
(
,
1
)
,
(
,
2
)
]
.
As
the
r
e
s
ult
o
f
c
las
s
if
ica
ti
ons
,
the
s
mell
c
onc
e
ntr
a
ti
o
n
c
a
n
be
c
a
lcula
ted
,
whic
h
is
us
e
d
to
be
the
mea
n
s
qua
r
e
e
r
r
or
(
R
M
S
E
)
in
or
de
r
to
mea
s
ur
e
the
p
r
e
dicte
d
a
nd
a
c
tual
va
lue.
is
a
s
a
mpl
e
c
a
pa
c
it
y,
i
s
a
mea
s
ur
e
d
va
lue,
a
nd
̂
is
a
pr
e
dictive
va
lue.
=
√
1
∑
(
−
̂
)
2
=
1
(
16
)
-
No.
4.
S
ur
ppos
e
d
ge
n
=
ge
n
+
1
,
a
c
c
or
ding
to
(
13)
-
(
1
5)
it
e
r
a
ti
ons
,
a
nd
put
the
va
lue
of
it
e
r
a
ti
ons
int
o
L
S
S
VM
model.
T
he
r
e
a
f
ter
,
c
a
lcula
te
t
he
s
mell
c
onc
e
ntr
a
ti
o
n.
-
No.
5
.
W
he
n
ge
n
r
e
a
c
he
s
the
maximum
it
e
r
a
ti
ons
,
it
c
a
n
de
c
ide
to
s
top.
T
he
n,
we
will
ha
ve
the
be
s
t
model
that
mee
ts
L
S
S
VM
model
pa
r
a
mete
r
s
.
Othe
r
wis
e
,
we
will
r
e
tu
r
n
to
N
o
.
2
.
-
No.
6
.
W
e
ge
t
the
opti
mi
z
e
d
pa
r
a
mete
r
s
,
a
nd
we
e
s
tablis
h
F
OA
-
L
S
S
VM
models
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
A
nalog
c
ir
c
uit
faul
t
diagnos
is
v
ia
F
OA
-
L
SS
V
M
(
Yu
W
e
nx
in
)
255
5.
I
L
L
UST
RA
T
I
VE
E
XA
M
P
L
E
T
h
e
S
a
ll
e
n
-
Ke
y
is
t
e
s
ted
a
s
a
l
owp
a
s
s
f
il
te
r
c
ir
c
u
it
t
o
v
e
r
i
f
y
e
f
f
e
c
ti
ve
ne
s
s
a
n
d
c
o
r
r
e
c
t
ne
s
s
i
n
th
is
s
e
c
ti
on
.
T
h
e
r
e
s
is
t
o
r
s
a
n
d
c
a
pa
c
i
to
r
s
a
r
e
a
s
s
u
me
d
t
o
m
e
e
t
5%
t
o
le
r
a
n
c
e
s
r
e
s
pe
c
t
ive
l
y
.
T
he
S
a
ll
e
n
-
Ke
y
ba
nd
pa
s
s
f
i
l
t
e
r
i
n
F
i
g
u
r
e
1
un
de
r
C
1
,
C
2
,
R
2
a
n
d
R
3
va
r
y
wi
th
in
the
i
r
t
o
le
r
a
nc
e
s
.
N
F
r
e
p
r
e
s
e
n
ts
n
on
-
f
a
ul
t
c
l
a
s
s
.
T
he
nor
mal
va
lues
f
or
e
a
c
h
c
omponent
a
r
e
s
hown
in
T
a
ble
1
.
R
e
s
is
tor
s
a
nd
c
a
pa
c
it
or
s
ha
ve
5%
tol
e
r
a
nc
e
s
r
e
s
pe
c
ti
ve
ly.
E
ve
r
y
no
r
mal
va
lue
is
:
1
=
5
,
2
=
5
,
1
=
1
Ω
,
2
=
3
Ω
,
3
=
2
Ω
,
4
=
5
=
4
Ω
.
He
r
e
,
we
s
uppos
e
r
e
s
is
tor
s
a
nd
c
a
pa
c
it
or
s
in
thi
s
int
e
r
va
l
[
(
50%
,
95%
)
∪
(
105%
,
150%
)
]
(
is
the
r
e
g
ular
va
lue)
.
T
he
n
f
a
ult
s
c
a
n
be
c
las
s
if
ied
to
8
f
a
ult
pa
tt
e
r
n
:
C
1
↑
,
C
1
↓
,
C
2
↑
,
C
2
↓
,
R
2
↑
,
R
2
↓
,
R
3
↑
,
R
3
↓
.
I
n
th
is
wa
y,
tr
a
ini
ng
a
nd
tes
t
s
a
mpl
e
s
ge
ne
r
a
ted
a
f
ter
pr
e
pr
oc
e
s
s
ing
c
a
n
be
tr
a
ined
a
nd
tes
ted
a
f
ter
F
OA
-
L
S
S
VM
opti
mi
z
a
ti
on
.
T
h
e
s
ingl
e
f
a
ult
c
a
tegor
ies
a
nd
the
nomi
na
l
a
nd
f
a
ult
c
omponent
va
lues
f
o
r
the
S
a
ll
e
n
-
Ke
y
ba
ndpa
s
s
f
il
ter
a
r
e
li
s
ted
in
T
a
ble
1.
F
igur
e
1.
S
a
ll
e
n
-
Ke
y
ba
ndpa
s
s
f
il
ter
T
a
ble
1.
S
ingl
e
f
a
ult
c
las
s
e
s
a
nd
the
nomi
na
l
a
nd
f
a
ult
y
c
omponent
va
lues
F
a
ul
t
c
ode
F
a
ul
t
c
l
a
s
s
N
or
ma
l
F
a
ul
ty
va
lu
e
1
C
1↑
5nF
7.5nF
2
C
1↓
5nF
2.5nF
3
C
2↑
5nF
7.5nF
4
C
2↓
5nF
2.5nF
5
R
2↑
3kΩ
4.5kΩ
6
R
2↓
3kΩ
1.5kΩ
7
R
3↑
2kΩ
3kΩ
8
R
3↓
2kΩ
1kΩ
9
NF
-
-
W
e
c
a
r
r
y
out
50
ti
mes
M
onte
C
a
r
lo
a
na
lys
is
to
th
e
diagnos
is
c
ir
c
uit
by
P
S
p
ice
10.
5
s
of
twa
r
e
whe
r
e
the
a
c
quis
it
ion
va
lue
of
the
output
volt
a
ge
Vout
a
s
s
our
c
e
da
ta,
thr
ough
Ha
a
r
wa
ve
let
tr
a
ns
f
or
m
a
nd
f
a
ult
da
ta
whic
h
we
r
e
obtaine
d
450
s
a
mpl
e
s
,
of
whic
h
40
%
wi
ll
be
us
e
d
a
s
tr
a
ini
ng
da
ta
s
a
mpl
e
,
60
%
o
f
the
da
ta
a
s
a
tes
t
s
a
mpl
e
s
.
W
e
s
uppos
e
the
f
r
uit
f
ly
population
is
100,
a
nd
the
number
o
f
it
e
r
a
ti
ons
is
30
s
teps
,
f
li
e
s
s
wa
r
m
or
igi
na
l
pos
i
ti
on
is
a
r
a
ndom
ge
ne
r
a
tor
by
mat
lab
r
a
nds
f
unc
ti
on.
Af
ter
the
s
im
ulation,
F
OA
-
L
S
S
VM
opti
mi
z
a
ti
on
it
e
r
a
ti
on
c
onve
r
ge
nc
e
is
s
hown
in
F
ig
ur
e
2
.
W
e
c
a
n
s
e
e
that
F
OA
-
L
S
S
VM
opti
mi
z
a
ti
on
it
e
r
a
ti
on
s
teps
c
onve
r
ge
to
0.
02,
whe
n
the
it
e
r
a
ti
on
s
tep
is
be
twe
e
n
1109
a
nd
2699
unde
r
the
loca
l
opti
mal
a
r
e
a
.
Ac
c
or
ding
to
s
e
ve
r
a
l
tes
t
s
of
the
downw
a
r
d
tr
e
nd
,
we
ha
ve
f
ound
that
the
opti
mal
it
e
r
a
ti
on
c
a
n
c
onve
r
ge
0
to
4000
s
teps
.
F
inally,
we
ha
ve
obtaine
d
the
opti
mal
pa
r
a
mete
r
s
(
s
e
e
T
a
ble
2)
.
As
is
s
e
e
n
f
r
om
T
a
ble
3,
f
ive
kin
ds
o
f
f
a
ult
modes
c
a
n
diagnos
e
the
c
or
r
e
c
t
one
s
.
T
he
n,
NF,
C
1↓,
R
2↓
f
a
ult
modes
of
7
tes
ted
s
a
mpl
e
da
ta
a
r
e
diagnos
e
d
uns
uc
c
e
s
s
f
ul.
Us
ing
the
opt
im
ize
d
pa
r
a
mete
r
s
is
c
l
a
s
s
if
ied,
diagnos
ti
c
a
c
c
ur
a
c
y
of
S
a
ll
e
n
-
Ke
y
ba
nd
-
pa
s
s
f
il
ter
c
ir
c
uit
is
97.
04
%
by
F
OA
-
LS
S
VM
methods
.
LM
124
+
3
-
2
V+
4
V-
11
OU
T
1
R2
3k
R5
4k
R4
4k
R3
2k
C1
5n
C2
5n
out
R1
1k
V1
1Vac
0Vdc
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
251
-
257
256
F
igur
e
2.
T
r
a
ini
ng
e
r
r
or
of
e
a
c
h
it
e
r
a
ti
on
T
a
ble
2.
S
ingl
e
f
a
ult
diagnos
ti
c
tes
t
pa
r
a
mete
r
r
e
s
u
lt
s
S
iz
e
pop
ma
xge
n
ga
mbe
s
t
s
ig
be
s
t
100
30
2.0521
0.0806
X
_a
xi
s
Y
_a
xi
s
[
-
6.4958,
-
6.9183]
[
10.3066,10.4556]
T
a
ble
3.
S
a
ll
e
n
-
Ke
y
ba
nd
-
pa
s
s
f
il
ter
c
ir
c
uit
s
ingl
e
f
a
ult
diagnos
is
NF
C
1↑
C
1↓
C
2↑
C
2↓
R
2↑
R
2↓
R
3↑
R
3↓
NF
27
1
2
C
1↑
30
C
1↓
2
27
(
I
nf
)
C
2↑
1
30
C
2↓
30
R
2↑
30
R
2↓
1
28
R
3↑
30
R
3↓
30
6.
CONC
L
USI
ON
I
n
thi
s
pa
pe
r
,
the
us
e
of
F
OA
ha
s
a
good
gl
oba
l
s
e
a
r
c
hing
a
bil
it
y
li
nke
d
with
L
S
S
VM
in
pa
tt
e
r
n
r
e
c
ognit
ion
of
s
upe
r
ior
pe
r
f
or
manc
e
.
W
e
pr
e
s
e
nt
F
OA
-
L
S
S
VM
model
c
ir
c
uit
f
a
ult
diagnos
is
a
s
the
S
a
ll
e
n
-
Ke
y
ba
nd
-
pa
s
s
f
il
ter
,
whic
h
s
hows
tha
t
the
a
lgor
it
hm
obvious
ly
im
pr
ove
s
the
a
c
c
ur
a
c
y
of
f
a
ult
diagnos
is
a
nd
r
e
c
ognit
ion
o
f
f
a
ult
s
.
T
his
s
hows
tha
t
the
method
is
a
n
e
f
f
ica
c
ious
a
nd
r
e
li
a
ble
method
f
or
f
a
ult
diagnos
is
of
a
na
log
c
ir
c
uit
s
.
RE
F
E
RE
NC
E
S
[1
]
G
.
Fed
i
,
R.
G
i
o
mi
,
S.
Man
et
t
i
,
et
a
l
.
,
“
A
s
y
m
b
o
l
i
c
ap
p
ro
ach
fo
r
t
e
s
t
a
b
i
l
i
t
y
ev
al
u
at
i
o
n
i
n
fa
u
l
t
d
i
a
g
n
o
s
i
s
o
f
n
o
n
l
i
n
ear
an
al
o
g
ci
rc
u
i
t
s
,”
IE
E
E
I
n
t
e
r
n
a
t
i
o
n
a
l
S
ym
p
o
s
i
u
m
o
n
Ci
r
c
u
i
t
s
&
S
ys
t
e
m
s
,
2
0
0
2
.
[2
]
J
.
Pan
g
an
d
J
.
St
arz
y
k
,
“
Fau
l
t
D
i
a
g
n
o
s
i
s
i
n
Mi
x
e
d
-
Si
g
n
a
l
L
o
w
T
es
t
ab
i
l
i
t
y
Sy
s
t
em
,”
A
n
a
l
o
g
In
t
e
g
r
a
t
e
d
Ci
r
c
u
i
t
s
a
n
d
S
i
g
n
a
l
P
r
o
ce
s
s
i
n
g
,
v
o
l
.
28
,
n
o
.
2
,
p
p
.
1
6
1
-
1
7
2
,
2
0
0
1
.
[3
]
F.
G
ras
s
o
,
A
.
L
u
ch
et
t
a
,
S.
Man
et
t
i
,
et
a
l
.
,
“
Sy
mb
o
l
i
c
T
ech
n
i
q
u
es
fo
r
t
h
e
Se
l
ect
i
o
n
o
f
T
e
s
t
Fre
q
u
e
n
ci
e
s
i
n
A
n
a
l
o
g
Fau
l
t
D
i
a
g
n
o
s
i
s
,”
A
n
a
l
o
g
In
t
eg
r
a
t
ed
Ci
r
cu
i
t
s
a
n
d
S
i
g
n
a
l
P
r
o
ce
s
s
i
n
g
,
v
o
l
.
40
,
n
o
.
3
,
p
p
.
2
0
5
-
2
1
3
,
2
0
0
4
.
[4
]
F.
G
ras
s
o
,
A
.
L
u
ch
et
t
a,
S.
Man
et
t
i
,
et
al
.
,
“
A
Met
h
o
d
fo
r
t
h
e
A
u
t
o
mat
i
c
Sel
ect
i
o
n
o
f
T
es
t
Freq
u
e
n
ci
e
s
i
n
A
n
al
o
g
Fau
l
t
D
i
a
g
n
o
s
i
s
,
”
IE
E
E
Tr
a
n
s
a
c
t
i
o
n
s
o
n
In
s
t
r
u
m
e
n
t
a
t
i
o
n
a
n
d
M
ea
s
u
r
em
e
n
t
,
v
o
l
.
5
6
,
n
o
.
6
,
p
p
.
2
3
2
2
-
2
3
2
9
,
2
0
0
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
A
nalog
c
ir
c
uit
faul
t
diagnos
is
v
ia
F
OA
-
L
SS
V
M
(
Yu
W
e
nx
in
)
257
[5
]
M
.
T
ad
eu
s
i
e
w
i
cz
,
A
.
K
u
cz
y
ń
s
k
i
,
an
d
S
.
H
ał
g
as
,
“
Ca
t
a
s
t
ro
p
h
i
c
Fau
l
t
D
i
a
g
n
o
s
i
s
o
f
a
Cert
a
i
n
C
l
as
s
o
f
N
o
n
l
i
n
ear
A
n
a
l
o
g
Ci
rcu
i
t
s
,”
Ci
r
c
u
i
t
s
S
ys
t
em
s
a
n
d
S
i
g
n
a
l
P
r
o
ces
s
i
n
g
,
v
o
l
.
3
4
,
n
o
.
2
,
p
p
.
3
5
3
-
3
7
5
,
2
0
1
4
.
[6
]
Y
u
W
.
X
.
,
Su
i
Y
.
,
W
an
g
J
.,
“
T
h
e
Fau
l
t
s
D
i
ag
n
o
s
t
i
c
A
n
a
l
y
s
i
s
fo
r
A
n
a
l
o
g
Ci
rc
u
i
t
Bas
ed
o
n
FA
-
TM
-
E
L
M,
”
Jo
u
r
n
a
l
o
f
E
l
ec
t
r
o
n
i
c
Tes
t
i
n
g
-
Th
e
o
r
y
a
n
d
A
p
p
l
i
ca
t
i
o
n
s
,
v
o
l
.
3
2
,
n
o
.
4
,
p
p
.
4
5
9
-
4
6
5
,
2
0
1
6
.
[7
]
M.
K
h
an
l
ari
a
n
d
M.
E
h
s
an
i
an
,
“
A
n
Im
p
ro
v
ed
K
FCM
Cl
u
s
t
er
i
n
g
Met
h
o
d
U
s
ed
f
o
r
Mu
l
t
i
p
l
e
Fau
l
t
D
i
a
g
n
o
s
i
s
o
f
A
n
a
l
o
g
Ci
rcu
i
t
s
,
”
Ci
r
c
u
i
t
s
,
S
y
s
t
e
m
s
,
an
d
S
i
g
n
a
l
P
r
o
ce
s
s
i
n
g
,
v
o
l
.
3
6
,
n
o
.
9
,
p
p
.
3
4
9
1
-
3
5
1
3
,
2
0
1
7
.
[8
]
X
.
T
an
g
,
A
.
X
u
,
R.
Li
,
e
t
al
.
,
“
Si
m
u
l
a
t
i
o
n
-
Ba
s
ed
D
i
ag
n
o
s
t
i
c
M
o
d
e
l
fo
r
A
u
t
o
mat
i
c
T
e
s
t
a
b
i
l
i
t
y
A
n
a
l
y
s
i
s
o
f
A
n
a
l
o
g
Ci
rcu
i
t
s
,
”
IE
E
E
Tr
a
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
er
-
A
i
d
ed
D
e
s
i
g
n
o
f
In
t
eg
r
a
t
ed
C
i
r
c
u
i
t
s
a
n
d
S
ys
t
em
s
,
v
o
l
.
3
7
,
n
o
.
7
,
pp.
1
4
8
3
-
1
4
9
3
,
2
0
1
8
.
[9
]
M.
T
ad
eu
s
i
e
w
i
cz
a
n
d
S.
H
al
g
as
,
“
A
Me
t
h
o
d
f
o
r
L
o
ca
l
Paramet
r
i
c
Fau
l
t
D
i
a
g
n
o
s
i
s
o
f
a
Bro
a
d
Cl
a
s
s
o
f
A
n
a
l
o
g
In
t
e
g
rat
e
d
Ci
rcu
i
t
s
,
”
IE
E
E
Tr
a
n
s
a
c
t
i
o
n
s
o
n
I
n
s
t
r
u
m
e
n
t
a
t
i
o
n
a
n
d
M
e
a
s
u
r
e
m
en
t
,
v
o
l
.
6
7
,
n
o
.
2
,
p
p
.
3
2
8
-
3
3
7
,
2
0
1
8
.
[1
0
]
Y
.
X
i
a
o
an
d
Y
.
He
,
“
A
n
o
v
el
a
p
p
r
o
ach
f
o
r
an
a
l
o
g
fau
l
t
d
i
a
g
n
o
s
i
s
b
as
e
d
o
n
n
e
u
ral
n
et
w
o
r
k
s
a
n
d
i
mp
r
o
v
e
d
k
er
n
el
PCA
,
”
Neu
r
o
co
m
p
u
t
i
n
g
,
v
o
l
.
7
4
,
n
o
.
7
,
p
p
.
1
1
0
2
-
1
1
1
5
,
2
0
1
1
.
[1
1
]
Y
.
L
o
n
g
,
Y
.
X
i
o
n
g
,
Y
.
He
,
et
a
l
.
,
“
A
n
ew
s
w
i
t
ch
e
d
cu
rr
en
t
ci
rc
u
i
t
fau
l
t
d
i
ag
n
o
s
i
s
ap
p
ro
ac
h
b
a
s
ed
o
n
p
s
eu
d
o
ra
n
d
o
m
t
es
t
an
d
p
rep
r
o
ces
s
b
y
u
s
i
n
g
en
t
ro
p
y
a
n
d
H
aar
w
av
e
l
e
t
t
ra
n
s
f
o
rm
,
”
A
n
a
l
o
g
In
t
e
g
r
a
t
e
d
Ci
r
c
u
i
t
s
a
n
d
S
i
g
n
a
l
P
r
o
ces
s
i
n
g
,
v
o
l
.
9
1
,
n
o
.
3
,
p
p
.
4
4
5
-
4
6
1
,
2
0
1
7
.
[1
2
]
F.
A
mi
n
i
an
a
n
d
M.
A
mi
n
i
a
n
,
“
Fau
l
t
D
i
a
g
n
o
s
i
s
o
f
A
n
a
l
o
g
C
i
rcu
i
t
s
U
s
i
n
g
Ba
y
es
i
an
N
eu
ra
l
N
et
w
o
r
k
s
w
i
t
h
W
av
el
et
T
ran
s
fo
rm
as
Prep
r
o
ces
s
o
r
,
”
Jo
u
r
n
a
l
o
f
E
l
ec
t
r
o
n
i
c
Tes
t
i
n
g
:
Th
eo
r
y
a
n
d
A
p
p
l
i
ca
t
i
o
n
s
,
v
o
l
.
1
7
,
n
o
.
1
,
p
p
.
29
-
36
,
2
0
0
1
.
[1
3
]
P.
D
.
Q
i
n
,
Y.
P.
Q
i
n
,
an
d
A.
H
.
Z
h
an
g
,
“
A
n
Im
p
ro
v
ed
Fau
l
t
s
D
i
a
g
n
o
s
i
s
St
ra
t
eg
y
o
f
A
n
al
o
g
C
i
rcu
i
t
O
n
l
i
n
e
U
s
i
n
g
Ke
rn
e
l
Fu
zzy
C
-
Mean
s
fo
r
t
h
e
U
n
k
n
o
w
n
Fau
l
t
,
”
A
d
va
n
c
ed
M
a
t
er
i
a
l
s
R
e
s
ea
r
ch
,
v
o
l
.
7
1
2
-
7
1
5
,
n
o
.
2
,
p
p
.
1
9
5
2
-
1
9
5
5
,
2
0
1
3
.
[1
4
]
B.
Can
n
as
,
A
.
Fa
n
n
i
,
S.
Man
et
t
,
e
t
al
.
,
“
N
e
u
ral
n
e
t
w
o
rk
-
b
a
s
ed
a
n
al
o
g
fa
u
l
t
d
i
a
g
n
o
s
i
s
u
s
i
n
g
t
e
s
t
a
b
i
l
i
t
y
a
n
al
y
s
i
s
,
”
Neu
r
a
l
Co
m
p
u
t
i
n
g
&
A
p
p
l
i
ca
t
i
o
n
s
,
v
o
l
.
1
3
,
n
o
.
4
,
p
p
.
288
-
2
9
8
,
2
0
0
4
.
[1
5
]
M.
A
.
El
-
G
amal
an
d
M.
D.
A
.
Mo
h
ame
d
,
“
E
n
s
emb
l
e
s
o
f
N
eu
ral
N
et
w
o
r
k
s
fo
r
Fa
u
l
t
D
i
a
g
n
o
s
i
s
i
n
A
n
al
o
g
Ci
r
cu
i
t
s
,
”
Jo
u
r
n
a
l
o
f
E
l
ect
r
o
n
i
c
Tes
t
i
n
g
:
T
h
eo
r
y
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
3
,
n
o
.
4
,
p
p
.
3
2
3
-
3
3
9
,
2
0
0
7
.
[1
6
]
A
.
K
av
i
t
h
ama
n
i
,
V
.
Ma
n
i
k
an
d
an
,
a
n
d
N
.
D
ev
ara
j
an
,
“
S
o
ft
Fa
u
l
t
Cl
a
s
s
i
fi
ca
t
i
o
n
o
f
A
n
a
l
o
g
Ci
rcu
i
t
s
U
s
i
n
g
N
et
w
o
rk
Paramet
ers
an
d
N
e
u
ral
N
e
t
w
o
rk
s
,
”
Jo
u
r
n
a
l
o
f
E
l
ec
t
r
o
n
i
c
Tes
t
i
n
g
,
v
o
l
.
2
9
,
n
o
.
2
,
p
p
.
2
3
7
–
2
4
0
,
2
0
1
3
.
[1
7
]
F.
G
ras
s
o
,
A
.
L
u
ch
et
t
a
,
an
d
S.
Man
et
t
i
, “
A
n
ew
mu
l
t
i
-
v
a
l
u
e
d
n
eu
ra
l
n
e
t
w
o
rk
fo
r
t
h
e
ex
t
ract
i
o
n
o
f
l
u
m
p
ed
mo
d
e
l
s
o
f
an
al
o
g
ci
rc
u
i
t
s
,
”
A
n
a
l
o
g
In
t
eg
r
a
t
ed
Ci
r
cu
i
t
s
a
n
d
S
i
g
n
a
l
P
r
o
ces
s
i
n
g
,
v
o
l
.
7
3
,
n
o
.
1
,
p
p
.
13
-
20
,
2
0
1
2
.
[1
8
]
A.
A.
A.
M.
A
mi
ru
d
d
i
n
,
H
.
Z
ab
i
ri
,
S.
A
.
A
.
T
aq
v
i
,
et
al
.
,
“
N
eu
ra
l
n
et
w
o
r
k
ap
p
l
i
ca
t
i
o
n
s
i
n
fau
l
t
d
i
ag
n
o
s
i
s
an
d
d
e
t
ect
i
o
n
:
an
o
v
er
v
i
e
w
o
f
i
mp
l
emen
t
at
i
o
n
s
i
n
en
g
i
n
eer
i
n
g
-
re
l
at
e
d
s
y
s
t
em
s
,
”
Neu
r
a
l
Co
m
p
u
t
i
n
g
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
.
pp.
1
–
26
,
2
0
1
8
.
[1
9
]
S.
K
.
J
es
w
a
l
an
d
S.
Ch
ak
ra
v
ert
y
, “
Recen
t
D
e
v
el
o
p
m
en
t
s
an
d
A
p
p
l
i
ca
t
i
o
n
s
i
n
Q
u
a
n
t
u
m
N
e
u
ral
N
e
t
w
o
rk
:
A
Rev
i
e
w
,
”
A
r
c
h
i
ve
s
o
f
C
o
m
p
u
t
a
t
i
o
n
a
l
M
e
t
h
o
d
s
i
n
E
n
g
i
n
eer
i
n
g
,
p
p
.
1
-
15
,
2
0
1
8
.
[2
0
]
Q
Ma
,
Y
.
He
,
an
d
F.
Z
h
o
u
,
“
A
n
e
w
d
eci
s
i
o
n
t
ree
a
p
p
ro
ac
h
o
f
s
u
p
p
o
r
t
v
ec
t
o
r
mach
i
n
e
f
o
r
a
n
al
o
g
ci
rc
u
i
t
f
au
l
t
d
i
a
g
n
o
s
i
s
,
”
A
n
a
l
o
g
In
t
e
g
r
a
t
e
d
Ci
r
cu
i
t
s
a
n
d
S
i
g
n
a
l
P
r
o
ce
s
s
i
n
g
,
v
o
l
.
8
8
,
n
o
.
3
,
p
p
.
4
5
5
-
4
6
3
,
2
0
1
6
.
[2
1
]
J.
Z.
So
n
g
,
C.
Y
.
G
u
o
,
an
d
H.
S.
L
i
u
,
“
Sel
ect
i
v
e
SV
M
E
n
s
em
b
l
e
Bas
e
o
n
Cl
u
s
t
er
i
n
g
A
n
a
l
y
s
i
s
A
p
p
l
y
fo
r
A
n
a
l
o
g
Ci
rcu
i
t
Fau
l
t
D
i
a
g
n
o
s
i
s
w
i
t
h
Smal
l
Samp
l
e
s
,
”
A
p
p
l
i
ed
M
ech
a
n
i
c
s
a
n
d
M
a
t
e
r
i
a
l
s
,
v
o
l
.
3
8
0
-
3
8
4
,
p
p
.
8
4
1
-
8
4
5
,
2
0
1
3
.
[2
2
]
C.
Z
h
an
g
,
Y
.
He
,
L
.
Y
u
a
n
,
et
a
l
.
,
“
A
N
o
v
el
A
p
p
ro
ac
h
f
o
r
D
i
a
g
n
o
s
i
s
o
f
A
n
a
l
o
g
Ci
rc
u
i
t
Fau
l
t
b
y
U
s
i
n
g
G
MK
L
-
S
V
M
an
d
PSO
,
”
Jo
u
r
n
a
l
o
f
E
l
ec
t
r
o
n
i
c
Tes
t
i
n
g
,
v
o
l
.
3
2
,
n
o
.
5
,
p
p
.
5
3
1
-
5
4
0
,
2
0
1
6
.
[2
3
]
D
.
G
rzech
ca
an
d
J
.
Ru
t
k
o
w
s
k
i
,
“
Fau
l
t
d
i
ag
n
o
s
i
s
i
n
an
a
l
o
g
e
l
ect
r
o
n
i
c
ci
rc
u
i
t
s
-
t
h
e
S
V
M
ap
p
r
o
ac
h
,
”
M
et
r
o
l
o
g
y
a
n
d
M
ea
s
u
r
em
e
n
t
S
y
s
t
e
m
s
,
v
o
l
.
1
6
,
n
o
.
4
,
p
p
.
5
8
3
–
5
9
8
,
2
0
0
9
.
[2
4
]
C.
W
.
Hsu
an
d
C.
J
.
L
i
n
,
“
A
co
mp
ar
i
s
o
n
o
f
met
h
o
d
s
fo
r
mu
l
t
i
-
c
l
as
s
s
u
p
p
o
rt
v
ec
t
o
r
mach
i
n
e
s
,
”
IE
E
E
Tr
a
n
s
a
ct
i
o
n
s
o
n
Neu
r
a
l
Net
w
o
r
k
s
,
v
o
l
.
1
3
,
n
o
.
2
,
p
p
.
4
1
5
–
4
2
5
,
2
0
0
2
.
[2
5
]
Y
.
X
u
e
-
L
o
n
g
,
an
d
S.
W
e
i
,
“
A
p
p
l
i
ca
t
i
o
n
o
f
D
BN
i
n
A
n
al
o
g
C
i
rcu
i
t
Fau
l
t
D
i
a
g
n
o
s
i
s
,
”
M
i
c
r
o
e
l
ect
r
o
n
i
c
s
&
Co
m
p
u
t
er
,
2
0
1
6
.
[2
6
]
G.
H
.
Z
h
ao
,
an
d
W.
Y
.
L
iu
,
“
A
n
o
v
e
l
a
p
p
r
o
ach
f
o
r
an
a
l
o
g
ci
rc
u
i
t
fau
l
t
d
i
a
g
n
o
s
i
s
b
as
e
d
o
n
D
ee
p
Bel
i
ef
N
e
t
w
o
rk
,
”
M
ea
s
u
r
em
e
n
t
,
v
o
l
.
1
2
1
,
p
p
.
1
7
0
-
1
7
8
,
2
0
1
8
.
[2
7
]
Y
.
J
i
n
g
,
P.
Q
i
a
n
g
,
a
n
d
P.
H
o
n
g
b
i
n
g
, “
Fa
u
l
t
d
i
a
g
n
o
s
i
s
me
t
h
o
d
o
f
an
a
l
o
g
c
i
rcu
i
t
s
b
a
s
ed
o
n
t
em
p
o
ra
l
C
N
N
,
”
E
l
ec
t
r
o
n
i
c
M
ea
s
u
r
em
e
n
t
Tech
n
o
l
o
g
y
,
v
o
l
.
4
2
,
n
o
.
5
,
p
p
.
1
2
8
-
1
3
2
,
2
0
1
9
.
[2
8
]
W.
T
.
Pan
,
“
A
N
ew
Fru
i
t
Fl
y
O
p
t
i
m
i
zat
i
o
n
A
l
g
o
ri
t
h
m:
T
ak
i
n
g
t
h
e
Fi
n
a
n
ci
a
l
D
i
s
t
res
s
Mo
d
e
l
as
an
E
x
am
p
l
e
,
”
Kn
o
wl
e
d
g
e
B
a
s
ed
S
ys
t
em
s
,
v
ol
.
26
,
p
.
69
-
7
4
,
2
0
1
1
.
[2
9
]
J
.
Z
h
o
u
,
C.
W
a
n
g
,
a
n
d
B.
H
e
, “
Fo
reca
s
t
i
n
g
mo
d
el
v
i
a
L
SSV
M
w
i
t
h
mi
x
ed
k
ern
e
l
an
d
FO
A
,
”
Co
m
p
u
t
e
r
E
n
g
i
n
ee
r
i
n
g
a
n
d
A
p
p
l
i
ca
t
i
o
n
s
,
v
o
l
.
51
,
n
o
.
4
,
p
p
.
1
3
3
-
1
3
7
,
2
0
1
5
.
[3
0
]
H
.
Y
u
n
y
a
n
,
P.
Mi
n
fan
g
,
T
.
C
h
en
g
l
a
i
,
an
d
T
.
H
u
,
“
A
n
a
l
o
g
C
i
rcu
i
t
Fa
u
l
t
D
i
a
g
n
o
s
i
s
U
s
i
n
g
Mu
l
t
i
-
w
a
v
el
e
t
T
ran
s
f
o
rm
an
d
S
V
M
,
”
D
i
g
i
t
a
l
M
a
n
u
f
a
c
t
u
r
i
n
g
a
n
d
A
u
t
o
m
a
t
i
o
n
(
ICD
M
A
)
,
2
0
1
2
T
h
i
r
d
I
n
t
er
n
a
t
i
o
n
a
l
C
o
n
f
er
e
n
ce
,
G
u
i
L
i
n
,
p
p
.
2
1
4
-
2
1
7
,
2
0
1
2
.
[3
1
]
A
.
Z
h
an
g
an
d
C.
Ch
e
n
,
“
Fau
l
t
d
i
a
g
n
o
s
i
s
b
a
s
ed
s
emi
-
s
u
p
e
rv
i
s
ed
g
l
o
b
a
l
L
SSV
M
f
o
r
an
a
l
o
g
ci
r
cu
i
t
,
”
M
e
ch
a
t
r
o
n
i
c
s
a
n
d
Co
n
t
r
o
l
(ICM
C)
,
2
0
1
4
In
t
er
n
a
t
i
o
n
a
l
Co
n
f
e
r
en
ce
,
J
i
n
z
h
o
u
,
p
p
.
7
4
4
-
7
4
8
,
2
0
1
4
.
[3
2
]
Su
y
k
e
n
s
,
J
.
A
.
K
.
,
an
d
J
.
V
an
d
ew
a
l
l
e.
"
L
eas
t
Sq
u
ares
Su
p
p
o
r
t
V
ec
t
o
r
Mach
i
n
e
C
l
as
s
i
f
i
ers
.
"
Neu
r
a
l
P
r
o
ce
s
s
i
n
g
Let
t
e
r
s
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
2
9
3
-
3
0
0
,
1
9
99
.
Evaluation Warning : The document was created with Spire.PDF for Python.