TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 13, No. 1, Janua
ry 201
5, pp. 101 ~
105
DOI: 10.115
9
1
/telkomni
ka.
v
13i1.669
5
101
Re
cei
v
ed Se
ptem
ber 16, 2014; Revi
se
d No
vem
ber
12, 2014; Accepted Decem
ber 8, 201
4
An Abnormal Signal Diagnostic Emendation Technique
Resear
ch
Tiejun Cao
Schoo
l of Information Sci
enc
e and En
gi
neer
ing, Hu
na
n Internatio
nal Ec
on
omics Univ
ersit
y
,
Cha
ngsh
a
, Chi
na, postco
de: 410
20
5
email: matl
ab_
b
y
s
j
@1
26.com
A
b
st
r
a
ct
T
h
is pap
er intr
oduc
ed a sort
of abn
ormity d
a
ta metho
d
. W
e
start w
i
th analysin
g
an
exa
m
p
l
e first
and c
a
rry thro
ugh th
eory d
e
duci
ng o
n
ce
more a
nd l
i
st
thi
s
meth
od
arith
m
etic ste
p
a
n
d
appl
icati
on fie
l
d
finally. With the development
information and cont
rol technology, we need m
o
r
e
and more im
port
ant and
immine
nt to di
agn
ose a
b
n
o
r
m
ity d
a
ta an
d
eli
m
i
nate
th
e
m
. T
he q
ual
ity of me
as
uri
n
g
data is i
m
pro
v
ed
avail
a
b
l
y by the app
licati
on of
data e
m
en
dati
on techn
i
q
ue.
Ke
y
w
ords
: ab
nor
mal si
gn
al, dia
gnos
is, proc
ess error, alg
o
r
i
thm, inf
o
rmati
on, control
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
From the info
rmation the
o
ry, control the
o
ry
point of view, the pro
d
u
c
tion process can be
rega
rd
ed a
s
an inform
atio
n flow. Produ
ction proce
ss with a large
amount of proce
s
s inform
ation,
reflect ch
ang
es
i
n
the sta
t
us
of crafts and
equip
m
e
n
t, reflectin
g
the vari
ou
s
asp
e
ct
s of th
e
intera
ction a
n
d
asso
ciation
implies the
regula
r
ity
of produ
ction opti
m
izati
on. Pro
c
e
ss info
rmat
ion
is mai
n
ly refle
c
ted i
n
a
larg
e num
be
r of
state d
a
ta, it i
s
p
r
o
c
e
s
s co
ntrol a
nd
pro
c
e
s
s optimi
z
a
t
io
n
basi
s
. Ho
wev
e
r, data colle
ction an
d flow of the pr
o
c
ess, often asso
ciated
with
gro
ss e
r
ror d
a
ta,
su
ch
as sen
s
ors,
switche
s
, and
record
er of the
failu
re, artifici
al log
s
or
ran
dom
da
ta entry
errors.
No
matter
what ki
nd
of d
a
ta p
r
o
c
e
ssin
g
, data
ar
e required to
re
flect the
obje
c
tive reality
as
possibl
e, accurate, reliabl
e
and
compl
e
te. If the num
ber of true fal
s
e
cou
n
t, the re
sults
not o
n
ly
doe
s not make se
nse, bu
t also cau
s
e
errors due to
false inform
ation in deci
s
ion-m
a
ki
ng a
n
d
control. So,
wheth
e
r it i
s
manual
processing, or co
mput
er proce
ssi
ng, d
a
ta
must b
e
id
en
tified,
the first sente
n
ce of its aut
henticity, namely, f
ault diagno
si
s and
error corre
c
tion. In view of the
theory in this area more and more attention to
workers an
d the actual
pro
d
u
c
ers, the pap
er
whe
r
e the f
ault error by
regressio
n
analysi
s
, an
abno
rmal d
i
agno
stic m
e
thods d
a
ta, and
descri
b
e
s
the
application o
f
the method.
2. Regre
ssio
n
Analy
s
is o
f
Da
ta fr
om the Excep
tio
n
Instanc
e
a
bout Inte
rfe
r
ence
Reg
r
e
ssi
on
analysi
s
is
an effective
and practi
cal meth
od
of mass modelin
g.
Identification
of
outlie
rs
in meta-d
ata
a
p
p
roa
c
h
i
s
tha
t
peopl
e ofte
n d
r
a
w
two y
uan
scatter p
l
ot,
then the hu
man eye to observe the
discrete
stat
e, but in the multi-dimen
s
ion
a
l data, usin
g
observation method
s
can not
ident
ify abnormal data
.
Some peopl
e say that re
gre
ssi
on an
al
ysis
can be p
r
ed
icted and a
c
tual values
of the abs
ol
ute differen
c
e (error) to determi
ne. This
con
c
lu
sio
n
may not be rea
s
on
able. Con
s
ide
r
a few e
x
amples, the
data in Table
1.
Between y a
nd x is a
s
su
med a line
a
r relation
shi
p
. The lea
s
t square metho
d
with
reg
r
e
ssi
on, a straig
ht line L
1
:
x
y
08146
.
0
06833
.
0
(see Fig
u
re 1)
The
return v
a
lue p
o
ints y
i
and re
sidu
als r
i
a
r
e li
st
ed in
Table
1, the return
of the
remai
n
ing sta
ndard
deviati
on
σ
1
=
1.55.
As the la
rg
e
s
t ab
solute v
a
lue of
re
sidu
al | r
max
| = 2.09,
resi
dual
stan
dard d
e
viatio
n doe
s not excee
d
twice T
herefo
r
e, in a
c
cord
an
ce wit
h
usu
a
l pra
c
ti
ce,
the data can not believe to be outlie
rs, but if you look carefully you will find the data in Figure 1
except for the
first six points, other point
s in gene
ra
l i
n
a straight li
ne. Theref
ore
,
point 6 is likely
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 1, Janua
ry 2015 : 101 –
105
102
to be a
n
a
b
n
o
rmal
point, if
the first 6
re
moved, wi
th
the othe
r five
points with
a
straig
ht line, t
hen
get L2 in the Figure 1:
x
y
97767
.
0
87333
.
1
Table 1.
Th
e origin
al data
and re
gressio
n
results
N
x
y
L1
y
r
L2
y r
1 -4
2
．
48
0.39 2.09
2.04
0.44
2 -3
0
．
73
0.31 0.42
1.06
-0.33
3 -2
-0.04
0.23
-0.27
0.08
-0.12
4 -1
-1.44
0.15
-1.59
-0.9
-0.54
5 0
-1.32
0.07
-1.39
-1.87
0.55
6 10
0
-0.75
0.75
-11.64
(11.64)
Residual standar
d
deviation
σ
1
=1.55
σ
2
=0.55
|r
ma
x
|/
σ
1.35
1.00
Figure 1. Experime
n
tal dat
a and re
gression re
sults
Far from
a
straight lin
e L
2
and
L1,
and
L2 a
n
d
the va
st maj
o
rity of
points a
r
e ve
ry clo
s
e;
and L1 i
s
a l
o
t of points
with a large
r
g
ap. From
thi
s
example, lea
s
t-squa
re
s re
gre
ssi
on
can
be
s
e
en in at leas
t two weakness
es
:
1)
The individ
u
a
l
data point
s
may be a g
r
eat
influen
ce
on the reg
r
e
ssi
on results,
its
perfo
rman
ce
is the
poi
nt o
f
no
return a
nd p
a
rtici
pat
e in thi
s
poin
t
to pa
rticip
ate in
retu
rn, t
he
large
r
differe
nce b
e
twe
en
the results ob
tained, point
6 above, is th
is point.
2)
anomali
e
s in
the data valu
e, not necessarily
with the
size of the re
sidu
als fou
n
d
out,
in the example anomaly of the first six points of re
sid
uals is q
u
ite small (0.7
5),
while the larg
est
resi
dual
s i
n
point 1
(r
1
=
2.09 ) are n
o
t outliers; if
be
cau
s
e
of
its large
and
su
sp
ect it i
s
the
resi
dual
outli
ers,
even
we
ed out, it will
be worse th
an the L
1
re
gre
ssi
on
re
sults. Thu
s
, in
the
least-sq
ua
re
s regressio
n
u
s
ing the
re
sid
uals to
dete
r
mine the si
ze
anomaly is
unreli
able, ev
en
mislea
ding.
Rea
s
o
n
s fo
r these problem
s, mainly due
to
least-sq
u
a
re
s metho
d
to minimize t
he su
m
of squa
red resid
ual
s, it is equally treat all
by requiring that arro
gan
ce may be close to the
reg
r
e
ssi
on li
n
e
every
point
, whe
n
the
d
a
ta contain
s
outliers, du
e
to the mi
nim
u
m
squa
re
s i
t
is
"equal" to th
e reg
r
e
s
sion
results influ
e
n
ce
d by it, but can
not gi
ve more
accurate
reg
r
e
s
sion
equatio
n. In this reg
r
e
ssi
o
n
equ
ation to
get on b
a
se
out of the re
sidual n
a
ture i
s
not
reliabl
e. In
addition, th
e
reg
r
e
ssi
on
of ob
se
rved val
ues
for
the
d
epen
dent va
ri
able f
r
om th
e
varia
b
les x a
nd
y of two parts, while the y comp
one
nt of the re
si
dual i
s
the differen
c
e bet
wee
n
its retu
rn valu
e,
the x-comp
o
nent
which
doe
s not ful
l
y reflect
so
me of the
factors. Bel
o
w
we a
naly
z
e
theoreti
c
ally.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Abnorm
a
l Signal Dia
g
n
o
stic Em
enda
tion Tech
niqu
e Re
sea
r
ch (Tiejun Cao
)
103
3. Theore
t
ic
al Analy
s
is o
f
Gene
rated
Gross Error
Linea
r mod
e
l:
n
i
x
y
i
i
i
,
,
2
,
1
'
In which
x 'is the tra
n
sp
ose of x, (x
i
,y
i
) is ob
served d
a
ta
,
'
2
1
1
)
,
,
,
(
,
,
m
i
n
i
R
y
R
x
is reg
r
e
s
si
on co
efficien
ts to be estimated,
i
is
Ran
dom e
rro
r.
m
n
ij
in
i
i
i
in
i
i
i
x
x
x
x
x
y
y
y
y
)
(
)'
,
,
,
(
,
)'
,
,
,
(
2
1
2
1
Obtaine
d
by the method of
least
squa
re
s regressio
n
, result
s are:
y
x
x
x
'
1
'
)
(
ˆ
(
1
)
Hy
y
x
x
x
x
y
'
1
'
)
(
ˆ
ˆ
(
2
)
Among:
n
n
ij
n
n
j
i
h
x
x
x
x
x
x
x
x
H
)
(
]
)
(
[
)
(
1
'
'
1
'
(3)
H i
s
the
col
u
mn vecto
r
x i
s
a
linea
r
su
bsp
a
ce
span
ned
by the p
r
oje
c
tion
mat
r
ix; also
k
n
own as
the "hat matrix" (Hat matrix),
the first point
of the resid
u
a
l
L is:
l
j
j
lj
l
ll
n
j
j
lj
l
l
l
l
y
h
y
h
y
h
y
y
y
r
)
1
(
ˆ
1
If the observ
ed value of y, yi is an abno
rmal value, a
s
suming it'
s
n
o
rmal i
s
y
i
*,And y
i
by
y
i
* with gro
ss error
⊿
y
i
, that is
, y
i
= y
i
* +
⊿
y
i,
, if there is
no g
r
o
s
s erro
r,
the n
o
rmal
situatio
n is
the i-th re
sidu
al:
i
j
j
ij
i
ii
i
y
h
y
h
r
*
*
)
1
(
But
i
j
j
ij
i
i
ii
i
y
h
y
y
h
r
)
)(
1
(
*
As the gro
s
s error to chan
ge the i-th re
sidual, then:
i
ii
i
i
y
h
r
r
)
1
(
*
Thus,
whil
e t
he g
r
o
s
s erro
r
⊿
y
i
i
s
generally relative
ly large, but if
h
ii
cl
ose to 1
(note
hii
is the
proj
ect
i
on of the
di
agon
al
matri
x
element
s, therefo
r
e, 0
≤
h
ii
≤
1
)
. Th
e co
rrespon
d
i
ng
norm
a
l r
i
*
and
r
i
differe
nce
is
not la
rg
e.
Therefore,
in
the i-th
data
on the
g
r
o
s
s
error,
and
from
that point
of
the resi
dual
s ri
not
sho
w
n
,
due
to g
r
o
s
s e
r
rors
⊿
y
i
firs
t k
(k
≠
i) point re
sidual
s
cha
nge
s are:
)
(
)
1
(
*
i
k
y
h
r
r
i
ki
k
k
Therefore,
if
h
ki
relatively large, but the gro
s
s erro
rs in t
he k reflecte
d points o
n
the
resi
dual
s,
which
mea
n
s
that, with th
e re
sid
ual v
a
lue of
r
i
to
determine t
he
size of t
h
e
corre
s
p
ondin
g
poi
nt value
of y
i
is
not a
b
norm
a
l i
s
n
o
t
reliabl
e, be
ca
use
re
sid
ual
cha
nge
not
o
n
ly
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 1, Janua
ry 2015 : 101 –
105
104
the y compo
nent of gro
s
s erro
rs, but al
so to the x compon
ent, (h
ij
is compl
e
tel
y
determined
by
the x).
From
the a
n
a
l
ysis
of data
on the
retu
rn
of t
he i-th
eff
e
ct
size
clo
s
e
to the
co
rrespondi
ng
h
ii
, h
ii
is larg
e, this effect
may be la
rge
;
generally re
ferre
d to h
ii
b
i
g point of "l
everag
e poi
n
t
s"
(Leve
r
ag
e po
int) or th
e p
o
tential impa
ct point
s. Ge
nerally h
ii
co
nsid
ere
d
the
best in
0.2
the
followin
g
as
well, if possi
b
l
e, should b
e
at
least 0.5 or less. In the example ab
o
v
e h
66
= 0.936 is
very close to 1, so this is a
great imp
a
ct
on the
retu
rn,
but also in th
e x data, it do
es st
ray far.
4. Error Diag
nosis Algorithm
Based
on
the
above
discu
ssi
on a
nd
an
alysis, if
we
capture
to n
m
-
dime
nsi
onal
data
set
grou
p x
= (x
1
, x
2
, ..
., x
n
) '= (x
ij
)
n × m
, to
diagn
ose wh
ether th
e faul
t whi
c
h e
r
ror data, then
the
followin
g
algo
rithm step
s:
STEP1: c
a
lc
ulate x'x
STEP2: find the inverse m
a
trix (x'x)
-1
of
x'x
STEP3: In addition to abno
rmal value
s
to determi
ne a
cut boun
dary
value F (0 <F <1);
STEP4: c
a
lc
ulation
α
i
= x
i
'(x'x)
-1
x
i
, i =
1,2, ..., n, If
α
i
<F, then xi is t
he no
rmal
nu
mber
of
points; if
α
i
≥
F, then xi outliers.
STEP5: If on
e pa
rent
of t
he n
e
w data
set
z, the
n
th
e calculation
α
=z
'(
x'x)
-1
z,,With the
rule
s distin
gui
sh STEP4 z i
s
abn
orm
a
l d
a
ta or no
rmal
data.
In ord
e
r to im
prove th
e
co
mputing
accu
racy
of the
d
a
ta, the the
o
r
iginal
data
set X ca
n
be normali
ze
d to cente
r
.
5. Conclusio
n
This meth
od can b
e
use
d
to diagno
se a
nd rem
o
ve abnormal data
,
use data to improv
e
the reliability, availability and effectivene
ss.
Trou
ble
s
ho
oting: If our sig
n
a
l detectio
n
mean
s well, we a
r
e fa
ced
with a dem
an
d for the
part (su
c
h
a
s
indu
strial
proce
s
s control
)
in n
o
rm
al a
nd ab
no
rmal
data a
r
e
coll
ected
and
m
u
st
have, on
the
se
ction th
rou
gh the
alg
o
rit
h
m to
determ
i
ne the
failure
α
i
co
rrespon
ding
i
n
terval, put
(x
'x
)
-1
into the
regi
ste
r
h
a
s
been
quite
g
ood, o
n
-lin
e i
n
formatio
n a
nd d
a
ta for th
e dete
c
tion
of
Z,
cal
c
ulate
d
α
=z
'(
x'x)
-1
z; Accordin
g to the size of
α
determine the s
e
verity of failure, such as
equipm
ent p
e
rform
a
n
c
e, t
he imple
m
en
tation of
the online real
-ti
m
e
qu
antitative
analysi
s
and
fault alarm.
Planning
and
Statistics Ad
ministration:
plant proje
c
t manag
eme
n
t, statistical re
porting
and de
ci
sion
-makin
g sh
oul
d be ba
sed o
n
incomi
ng a
nd outgoi
ng
material
s an
d
energy devi
c
es
corre
c
t mea
s
urem
ents.
Ho
wever, they a
r
e with
a ra
nd
om erro
r, and
even some g
r
oss e
r
ror
dat
a
also m
a
ke m
anag
ers a
r
e
not able to
grasp th
e true
eco
nomi
c
be
nefits of the f
a
ctory.
Diagn
osi
s
and
co
rrectio
n
u
s
ing
the
d
a
ta flow an
d
temperat
ure
measurement
s, in
order to
obtain
reliab
l
e
data mana
ge
ment.
Signal tra
c
ki
ng process:
t
he use of o
n
line dia
gno
sis
and
co
rrection te
ch
ni
que
s to
analyze p
r
o
c
ess d
a
ta, so
tracking
the
o
peratio
nal
sta
t
us of
equi
pm
ent an
d devi
c
es, tre
n
d
s
a
n
d
interferen
ce, identify the error or e
quipm
ent failure sit
uation.
Process cont
rol
a
nd
o
p
timization: a pro
c
e
s
s sim
u
lation, adva
n
ce
d optimi
z
ation an
d
diagn
osti
c cal
i
bration
software used
i
n
combinatio
n
to
provid
e a
reli
able
pro
c
e
s
s
optimizatio
n.
In
pro
c
e
ss
control application
s
, the implem
entation
of diagno
stic correction softwa
r
e automati
c
a
lly,
real time a
ccess to process data, a
nd a
v
erage ti
m
e
to do the
cal
c
ulation an
d correctio
n
process
to obtain co
n
s
iste
nt data. Can al
so b
e
corre
c
ted a
n
d
the data in
put pro
c
e
s
s simulation p
r
o
g
ram
with the
late
st eco
nomi
c
d
a
ta, co
mputin
g, si
mul
a
tion and optimization
of ope
rati
ng
p
a
ra
meters
to achieve o
p
timal eco
n
o
m
ic re
sult
s. After opt
imization of pro
c
ess co
ntrol p
a
ram
e
ter val
ues
recomme
nde
d for the new set point, wh
ich is t
he onli
ne feedba
ck control syste
m
. Optimizati
on
can
be
offline, online,
an
d it ca
n con
s
titute
a co
ntrol syste
m
opt
imize
r
an
d o
n
line
clo
s
ed
-l
oop
sy
st
em.
Referen
ces
[1]
Z
hao W
e
i, S
un J
i
an
g. Pra
i
sed;
a
ne
w
d
y
nam
ic
pr
oc
ess d
a
ta c
o
rr
ection;
Co
ntro
l T
heor
y a
n
d
Appl
icatio
ns; 1
999; 04
33-3
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Abnorm
a
l Signal Dia
g
n
o
stic Em
enda
tion Tech
niqu
e Re
sea
r
ch (Tiejun Cao
)
105
[2]
Gang R
ong
W
ang
Xi. Mea
s
urin
g instrum
ents of a
si
ng
le n
ode
gross
error id
entific
ation m
e
tho
d
;
Che
m
ic
al En
gi
neer
ing
. 2
000;
51(1): 41-
45.
[3]
Rae
put Bi
ng-
Z
hen, Ch
en M
i
ng.
Data
coor
d
i
nati
on
and
gro
ss error d
e
tecti
on of sy
nchro
n
i
z
e
d r
obust
estimation method
. T
s
inghu
a Univers
i
t
y
(N
atural Sci
ence);
200
0; 40(2): 25
-30.
[4
]
Bo
L
i
, Bi
ng
-Zhe
n
,
C
h
e
n
Yo
ng
Mi
ng
pu
t th
e Ea
st se
cti
o
n
.
Fa
u
l
t de
te
cti
o
n b
a
s
ed
o
n
ti
me
re
du
nd
anc
y
error correcti
o
n
coefficient; T
s
i
ngh
ua U
n
ivers
i
t
y
; 200
0; 40(1
0
)
: 27-32.
[5]
Bo Li, Bi
ng-Z
h
en, Ch
en Y
o
n
g
Min
g
. East
section; fa
ult d
e
tection
bas
ed
on time r
e
d
u
nda
nc
y erro
r
correctio
n coef
ficient; T
s
inghu
a Univ
ersit
y
; 2
000; 40(
10): 33
-35.
[6] YUAN
Yong-root
. Che
m
ic
al
p
r
ocess d
a
ta co
rrection tec
h
n
i
ques fro
m
theo
retical r
e
searc
h
to pr
actic
a
l
convers
i
on
. C
h
ina Institute
of Process S
y
ste
m
s Engin
eer
in
g S
y
stems E
n
gin
eeri
ng C
o
m
m
ittee of the
first academic
confere
n
ce pr
o
c
eed
ings; Ji
na
n. 1991; 1
0
: 11
0-11
4.
[7]
Xu C
h
a
o
, Pan
Z
haoho
ng.
C
h
e
m
ic
al cons
is
tency of the data corre
ctio
n process; Ch
i
n
a Institute of
Process Syste
m
s En
gi
neer
in
g Systems E
n
gin
eeri
ng C
o
mmittee
of the
first academ
ic
confere
n
ce
proce
edi
ngs; Ji
nan, 19
91; 10:
118-
120.
[8] Chen
Jie.
Mat
h
e
m
atic
al Mo
d
e
lin
g ra
pid
det
ection
of che
m
ic
al pr
ocess d
a
t
a in the the
o
r
e
tical b
a
sis of
gross errors
; chemic
al proc
es
s in, the T
h
ird
Annu
al Me
etin
g Procee
di
ngs.
1990; 1
20-1
2
4
[9]
Chen Jie. C
he
mic
a
l pr
ocess
data i
n
the r
a
p
i
d d
e
tection
of
gross errors
. Mathematic
al simulati
on of
chemic
al proc
e
ss third Annu
al
Meeting Proc
e
edi
ngs. 19
90; 125-
129
[10]
Li-li
LIU the
Li
-
w
e
i
,
XIONG, Yan-
ye. T
he a
bnorm
a
l si
gna
l
detectio
n
met
hod
base
d
o
n
the fat tail
distrib
u
tion.
Jo
urna
l of Nanj
in
g Univ
ersity (Natural Sci
ence)
.
2011; (1).
[11]
XIA Y
a
-qi
n
, C
U
I Xia
o
-
y
an,
LI Jun-z
h
i, CH
E
N
W
e
i-sh
en
g,
LIU Ch
en
g-
ya
n
.
T
he stud
y of t
he
earthq
uak
e
the previ
ous a
bnorm
a
l aco
u
s
t
ic signa
ls. Beij
ing U
n
ivers
i
t
y
of
T
e
chnol
og
y.
2011; (3).
[12]
Z
hang
Xian
g, Z
hang Ji
an
q
i
, Qin Han
lin,
Liu D
e
li
an. Anoma
l
y
dete
c
tion
usi
ng
m
u
lti-reso
luti
o
n
decom
positi
on
of h
y
persp
ectral ima
ges.
Infrared a
nd L
a
ser
Engin
eeri
ng.
2
011; (3).
[13]
Liu
Z
hanf
eng,
Si Ji
ngp
in
g, Li
ang
Ho
ng
bo.
base
d
on v
i
br
ation
sig
n
a
l
a
nal
ysis
e
n
g
i
ne
c
y
li
nder
w
a
l
l
clear
ance a
b
n
o
rmal di
ag
nosi
s
. S
mall
i
n
tern
al
co
mb
ustion eng
ine and mo
torcycle.
201
1; (2).
[14]
YAN Ji-
hon
g,
W
A
NG W
e
i L
U
L
e
i. Aut
o
ma
ticall
y
up
date
the fa
ult
dia
g
n
o
s
is mo
del
b
a
sed
on
artifici
a
l
immune alg
o
rit
h
m.
Computer Integrated
M
a
n
u
facturin
g Systems
. 2
011; (4).
[15]
HU Su-
y
un, E
Jia-qi
an
g, GONG Jin-ke. Di
e
s
el co
l
d
start the proc
ess of fuzz
y
log
i
c infer
ence
unus
ua
l
dia
gnos
is.
Jour
nal of Hu
na
n U
n
iversity (Nat
ur
al Scie
nce).
20
11; (11).
[16]
WU Din
gha
i, ZHANG Pei
lin,
REN Guo
q
u
an,
XU Ch
ao, F
A
N Ho
ng
bo, b
a
s
ed o
n
th
e Ba
ye
s classifi
er of
a h
y
pers
pher
e and the d
i
esel e
ngi
ne
anom
al
y dete
c
tion ap
plic
ati
ons.
Journ
a
l of
Mechan
ical
Engi
neer
in
g
. 2011; ( 6).
Evaluation Warning : The document was created with Spire.PDF for Python.