TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 12, Decembe
r
2014, pp. 81
5
2
~ 816
0
DOI: 10.115
9
1
/telkomni
ka.
v
12i12.64
44
8152
Re
cei
v
ed
Jun
e
30, 2014; Revi
sed Septe
m
ber
27, 201
4; Acce
pted
Octob
e
r 19, 2
014
The Automatic Recognition of Large Ball Valve Sealing
Bolt Based on Digital Image
Song Qingjun*
1,2
, Xiao X
i
ngming
1
, Jiang Haiy
an
2
, Zhao
Xiegua
ng
2
1
School of Mec
han
ical a
nd El
ectrical En
gin
e
e
rin
g
, Chin
a U
n
iversit
y
of Min
i
ng & T
e
chnol
o
g
y
,
Xuz
h
o
u
, Jian
g
s
u 221
00
8, Chi
n
a
2
School of T
a
i-an, Shan
do
ng
Univers
i
t
y
of Scienc
e & T
e
ch
nol
og
y,
T
a
i-an, Shand
ong, 27
10
19, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: qjson
g
7
6
@1
26.com
A
b
st
r
a
ct
In this
pa
per,
w
e
ado
pted
th
e ar
ea
fill
ing
method
of th
e
mathe
m
atic
al
morph
o
lo
gy to
fill
the
ho
l
e
s
of the nut. In add
ition, the b
i
mod
a
l metho
d
and
mult
i-thr
e
shol
d metho
d
w
e
re combin
e
d
for the imag
e
seg
m
e
n
tatio
n
. F
u
rthermore, t
he i
m
age
ed
ge
w
a
s detected
by
mathe
m
atic
al mor
pho
lo
gy alg
o
rith
m.
F
i
na
lly,
the an
gl
e b
e
tw
een th
e b
o
lt a
n
d
man
i
pu
lator
w
a
s calcul
ated
usin
g the r
o
tat
i
on c
onv
ersio
n
matrix. W
i
th t
h
e
system
atic
err
o
r and correlat
i
on c
oeffi
cient, the calc
ulat
ed angle was ver
i
fied. Experim
e
ntal res
u
lts show
that the meth
o
d
can
protect the ed
ge int
egri
t
y of
the nut image, w
i
th fast proc
essi
ng spe
e
d
and stron
g
a
n
ti-
nois
e
abi
lity. The w
o
rk in this paper
prov
id
e
s
a theoretical
basis for the
a
u
tomatic reco
g
n
itio
n in the lar
g
e
ball v
a
lve seali
ng bolt system
.
Ke
y
w
ords
:
mathem
atic
al morphology,
struct
ure elem
ent, angle rec
ognition,
systematic
error, c
o
rrelation
coefficient
Copy
right
©
2014 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
In the turbin
e inlet syste
m
, the large ball valv
es a
r
e install
ed in a high altitude with ma
nual
operation fo
r
the gap
adju
s
tment in
process. The ve
ry narro
w spa
c
e
accretes th
e labo
r inten
s
ity
of worke
r
s.
Therefore, th
e gap a
d
ju
sting time is
lo
nger. Mo
re
over, it is very
inconve
n
ient
in
prod
uctio
n
wi
th man
ual
op
eration.
In o
r
der to imp
r
ov
e the
wo
rking
co
ndition
s,
we
develo
p
e
d
a
large b
a
ll valve seali
ng bolt
auto-adj
ustin
g
system
. Th
e system i
s
mainly comp
osed of the orbit
determi
nation
device, trav
elling me
ch
a
n
ism, mani
p
u
lator, sen
s
o
r
s a
nd the
control u
n
it. The
three
-
dime
nsi
onal
comp
uter a
s
semble
model i
s
sh
own i
n
Figu
re 1. The fiel
d installatio
n
is
sho
w
n i
n
Fig
u
re
2, Figu
re
2(a
)
i
s
the i
n
itial st
ate
of the ma
nipul
ator, Fig
u
re
2(b
)
is its
wo
rkin
g
state.
(1) Ball valve shell; (2) Seal
ing bolt; (3) S
ensor
; (4) Orbit determination
device; (5) Travelling
mech
ani
sm; (6) Mani
pulato
r
Figure 1. 3-di
mensi
on mo
d
e
l
of the auto-adju
s
ting sy
stem
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Autom
a
tic Re
co
gnition
of Large Ball
Valve Sealin
g Bolt Based
on Digital
…
(SONG Qi
ngju
n
)
8153
(a)
(b)
(1) Ball valve shell; (2
) Seal
ing bolt; (3) M
anipul
ator
Figure 2. The
field installati
on pictu
r
e
The orbit det
ermin
a
tion de
vice is in
stall
ed on
the o
u
ter of the larg
e ball valve, whi
c
h is
positio
ned
b
y
a larg
e int
e
rnal
gea
r d
r
ive. The
se
nso
r
s are u
s
ed for
se
ndi
ng the p
o
siti
on
informatio
n of the seali
n
g bolt to the cont
rolle
r. The ball v
a
lve seali
n
g
gap is a
d
j
u
sted
automatically by three m
anipul
ators u
n
iformly di
stributed i
n
the
orbit dete
r
mination
dev
ice.
Ho
wever, the
r
e i
s
a p
r
obl
e
m
that the he
xagon
s of
the
bolt nut an
d
manipul
ator
u
s
ually mi
smat
ch
(se
e
Fi
gu
re
3). Th
erefore
,
how to
re
co
gnize the
an
gle
θ
bet
wee
n
the
bolt
an
d ma
nipulato
r
i
s
critical for the
auto-adj
ustin
g
system.
(a) Po
sition o
f
the bolt; (b) Initial position
of the manipulator
Figure 3. Ske
t
ch of the inst
allation po
sition
It is more e
a
s
er fo
r the
co
mputer
and h
i
gh-p
e
rfo
r
ma
nce im
agin
g
equipm
ent fo
r the bolt
positio
n re
co
gnition ba
sed
on the digita
l image proc
essing a
naly
s
is. Th
e ch
aracteri
stics of
the
image de
scri
bed with mat
hematical me
thod must
be
analyzed in
the recognitio
n
. A method of
recogni
zin
g
t
he a
ngle
θ
b
e
twee
n the
b
o
lt and
mani
p
u
lator i
s
pre
s
ented
with g
e
o
metri
c
al fe
ature
extraction a
n
d
grap
h co
nversi
on matrix,
achi
evin
g accurate po
sitio
n
ing mani
pul
ators.
2. Image Processing
2.1. Image Preproc
essin
g
In ord
e
r to
eli
m
inate the
useless info
rma
t
ion,
the color image i
s
co
n
v
erted to g
r
ay
image
throug
h the
weig
hted ave
r
age m
e
thod
[1, 2]. Additionally, the h
o
les in th
e n
u
t cha
r
a
c
ters are
filled to maint
a
in the int
egrity of the informati
on. Du
ri
ng
the bina
ry
conv
e
r
sion,
sin
c
e the
pe
ak
valley of the
nut ima
ge
histogram
is un
obviou
s
, not
obviou
s
, Son
k
a
et al. [3]
p
r
opo
se
d that
it is
difficult to obtain satisfa
c
tory re
sults only
using
the single-t
h
re
shol
d ima
ge seg
m
ent
ation
method. Thu
s
, the bimod
a
l method a
nd multi-th
re
shol
d metho
d
are
combi
ned for the i
m
ag
e
segm
entation
.
Suppose that
G
(
x
,
y
)
is the grey imag
e,
Z
(
i
,
j
)
is the gray value of the point
(
i
,
j
)
,
Z
ma
x
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 12, Decem
ber 20
14 : 8152 – 81
60
8154
is the p
e
a
k
valley of the hi
stogra
m
, which
is defined a
s
the grey t
h
re
shol
d of
image
segm
entation
,
Z
min
is the smallest g
r
ay thre
shol
d, then
Z
(
i
,
j
)
can be
expre
s
sed a
s
:
(
1
)
The six
edg
e points of
the multi-thre
sh
ol
d
prelimina
r
y pro
c
e
ssi
ng
i
m
age are
a
s
sumed
to
be the
se
ed,
i.e., the gro
w
th
starting
point in
th
e
backg
rou
nd
region [4]. If the adj
acent
gray
differen
c
e
(4
-con
ne
cted) is less tha
n
a
given thre
sho
l
d, the pixel
s
are t
houg
ht
of belon
ging
to
the sam
e
im
age regio
n
a
nd are me
rg
ed into on
e
region. Oth
e
rwise, they belong to diffe
rent
image region
s. Set S is the se
ed poin
t
in
the (k-1
)-th filled re
g
i
on, the coo
r
dinate
s
colle
ct
ion
in the k-th filling
can be defined as:
(2)
Whe
r
e
C
(
i
,
j
)
i
s
the
gray filled
conditio
n
, and
F
K
-1
is the coo
r
dina
tes
colle
ction
in the
(
k
-1
)-
th
filling.
If the gray value of S is
, and the adj
a
c
e
n
t gray value to the seed S is
,
the gray co
nformity formul
a
can b
e
expressed a
s
follo
ws:
(
3
)
So that,
(
4
)
Whe
r
e
M
is a
con
s
tant value of grey co
nformity.
The preproce
ssi
ng imag
es
are sho
w
n in
Figure 4.
(a) Gray image; (b) Image of
area filling;
(c) bi
nary im
age
Figure 4. The
prep
ro
ce
ssi
n
g
image
s
2.2. Image Edge De
tec
t
io
n
There are m
any tradition
al edge
dete
c
tion meth
od
s, su
ch a
s
S
obel op
erator, Robe
rt
operator, Pre
w
itt operator,
Log ope
rato
r and
Can
n
y, base
d
on
calcul
ating the
differen
c
e in
a
small lo
cal area of the image. Zhou et al. and He
et
al. [5, 6] proved t
hat these operators a
r
e
more
sen
s
itive to n
o
ise, furthe
rmo
r
e t
he n
o
ise
will be
strength
ened
in th
e
edge
dete
c
ti
on.
Ho
wever, the
mathematical morp
holo
g
y
edge det
e
c
tion operator mainly usin
g morp
holo
g
i
c
al
gradi
ent doe
s not strength
e
n
the noise, a
l
though it is
also se
nsitive to the noise.
Hua
ng et al. [7]
pointed o
u
t mathemati
c
al
morp
hology
is a no
nlin
e
a
r filtering
m
e
thod, in whi
c
h the
stru
ct
ure
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Autom
a
tic Re
co
gnition
of Large Ball
Valve Sealin
g Bolt Based
on Digital
…
(SONG Qi
ngju
n
)
8155
element
s wit
h
ce
rtain fo
rms i
s
u
s
ed t
o
mea
s
u
r
e a
nd extra
c
t th
e image
edg
e for a
nalysi
s
an
d
target re
co
gni
tion. Therefo
r
e,
the stru
cture element
s
are the basi
c
m
o
rph
o
logi
cal f
a
ctors [8].
2.2.1. Structure Elements
Tang
et al.
a
nd Sun
et
al.
[9,10] after
a
nalyzin
g diffe
rent
eleme
n
t, pointe
d
o
u
t the multi-
stru
cture ele
m
ents
and m
u
lti-scal
e
ele
m
ents
edge
detectio
n
alg
o
rithm, while,
the algo
rith
m wa
s
not expre
s
se
d in an exact
numbe
r and
used in t
he
engin
eeri
ng. By compari
n
g the test re
sult
and th
e op
eration time
a
m
ong
structu
r
al el
ement
s
wi
th the
scal
e of 3, 5,
an
d 7, the
scal
e of
stru
ctural ele
m
ents i
s
d
e
signed
as 5 a
nd 3. Th
e
structu
r
e
elem
ents a
r
e
sh
o
w
n in
Figu
re
5,
whe
r
ea
s the
black box de
monst
r
ate
s
t
he origin of st
ructural elem
e
n
ts.
(a) 6
0
° st
ru
cture ele
m
ent
s
(b) 1
20°
stru
cture elem
ents
(c) 90°
stru
ct
ure ele
m
ent
s
Figure 5. Structure el
eme
n
ts
2.2.2. Edge Detec
t
ion Al
gorithm
of M
a
thema
t
ical
Morphology
Gon
z
ale
z
et
al. and P
r
att
et al. [1, 2] d
e
mon
s
trate
d
the two
ba
sic arithm
etic o
p
e
ration
s,
i.e., erosi
on
and dil
a
tion,
in the mat
h
e
m
atical m
o
rp
hology. Th
e
ero
s
ion
an
d
dilation a
r
e
n
o
t
inverse b
u
t concatenate
d
operation i
n
edge
dete
c
ti
on [11]. In
o
r
de
r to
ma
ke
full u
s
e
of the
advantag
es
of different structu
r
al ele
m
ent
s, the morp
holo
g
y edge dete
c
ti
on algo
rithm
was
prop
osed ba
sed on
th
e multi-shap
e/scale struct
u
r
al elem
ents.
With the
defi
n
ition that
denote
s
th
e boun
dary of the
inp
u
t
ima
ge
G
, and
repre
s
e
n
ts a grou
p
of stru
ctural
element
s, the edge dete
c
ti
on ope
rato
r is written a
s
:
(
5
)
Whe
r
e
n
indi
cate
s the
scale, and
n
= 3
is cho
s
e
n
. The stru
ctura
l
element
s
of
B
i
are given
in
Figure 5.
indicate
s that G is dilated by
B
, and
.
indica
tes that
G is eroded b
y
B
, and
.
The result of the edg
e d
e
tection
sh
o
w
s th
at the
prop
osed al
g
o
rithm
can
a
c
curately
detect the
p
o
sitioni
ng det
ails of the e
dge
s, ma
ki
n
g
the imag
e
outline
s
cle
a
r, co
mplete
and
coh
e
re
nt (Fig
ure 6
)
.
Figure 6. Re
sult of the edge detectio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 12, Decem
ber 20
14 : 8152 – 81
60
8156
3. Conv
ersion of the
Coo
r
dinate
The
co
ordi
na
te ori
g
in
of th
e imag
e i
s
usually set at
(
0
,
0
)
.
Ho
wev
e
r, in
ou
r
re
cogni
zing
work the
co
ordin
a
te sy
stem mu
st be
conve
r
s
ed t
o
the middl
e
of the imag
e as th
e o
r
i
g
in.
Suppo
se th
at the ima
ge
scale
after
pre
p
ro
ce
ssi
ng i
s
,
P
0
and
P
denote
the
coordi
nate
matrices befo
r
e a
nd afte
r t
he conve
r
si
o
n
, re
spe
c
tivel
y
. The coordi
nate
conve
r
si
on mo
del i
s
as
follows
:
(
6
)
Whe
r
e
T
is th
e conve
r
si
on
coo
r
din
a
te m
a
trix,
After c
o
ordinate c
onvers
i
on, the
c
o
ordinate is
s
u
bjec
ted to:
(
7
)
4. Recog
n
ition of the Sh
ape Fea
t
ure
Points of the
Nut
4.1. Extrac
tion of the Im
age Ce
ntr
e
For the
object image, the centre of the shape
can b
e
obtaine
d with
the following
expre
ssi
on b
e
low.
(
8
)
4.2. Extrac
tion of the Sh
ape Fea
t
ure
Points
The feature shape p
o
ints o
f
the nut are defi
ned a
s
A, B, C, D, E and F. The distance
DI
from the edg
e point to the centre ca
n be
expresse
d a
s
:
(
9
)
Whe
r
e
O
i
s
the centre point
of the nut,
M
is the point
on the edg
e o
f
the shape.
Acco
rdi
ng to
the characte
ristics of the
hex
ago
n, the
distan
ce
s
(A
O, BO, CO,
DO, EO
and FO
) a
r
e t
he greate
s
t a
nd eq
ual. Th
e set of the
g
r
eate
s
t dista
n
c
e
s
is
define
d
as th
e feat
ure
point matrix
DI
max
with the
following form as
:
(
1
0
)
Whe
r
e the fe
ature poi
nt A must be in th
e first quad
ra
nt, i.e., both
x
A
and
y
A
are positive.
The ce
nter a
nd sh
ape feat
ure poi
nts a
r
e
sho
w
n in Fig
u
re 7.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Autom
a
tic Re
co
gnition
of Large Ball
Valve Sealin
g Bolt Based
on Digital
…
(SONG Qi
ngju
n
)
8157
(a)
(b)
Figure 7. Illustration of the feature p
o
ints
A thorou
gh
study of the im
age
s ab
out b
o
lt at diffe
ren
t
position
s
was
carrie
d ou
t. It is found that
two ca
se
s de
cide the di
re
ction and angl
e of manipula
t
or
rotation. One case is that there is o
n
ly
one featu
r
e
p
o
int (A) in th
e first
qua
dra
n
t (Fig
ure
7(a)), th
e othe
r is that th
ere
are
two fe
ature
points (A a
n
d
F) in the first quad
rant (Fi
g
ure 7 (b)).
5. Recog
n
ition of An
gle bet
w
e
e
n Ma
nipulator an
d Bolt
The initial p
o
s
ition of the
manipul
ator i
s
fix
ed (Fi
gure 3(a
)),
while
the po
sition
of the bolt
(Figu
r
e 3
(
b
))
is ra
ndom. T
he mani
pulat
or ne
ed
s to rotate an an
gl
e
θ
to fit the nut of the bol
t.
DI
ma
x
'
is defined
as the feature point matri
x
after rotating,
θ
is
the rotating
angle, the rot
a
ting formul
a is expre
s
sed
as Equatio
n (11).
(
1
1
)
Whe
r
e
F
is
the rotation convers
i
on matrix,
.
Equation (11) can be
rea
r
range
d as:
(
1
2
)
Since the fe
ature p
o
int in the first q
uadr
ant dete
r
mine
s the a
ngle an
d direction of
rotation, point
A is cho
s
en t
o
cal
c
ulate th
e
rotation an
g
l
e. Equation (13) can be d
e
duced to:
(
1
3
)
Acco
rdi
ng to Equation (13), we obtain th
e rotation an
g
l
e as follo
ws:
If there is
on
ly one featu
r
e point in th
e firs
t q
uad
ra
nt as Fi
gure
7(a
)
, the ma
nipulato
r
rotates (60
-
θ
)°
in
the counterclo
c
kwi
s
e dire
ction.
If
t
here
are
t
w
o fe
ature
p
o
ints i
n
the
first
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 12, Decem
ber 20
14 : 8152 – 81
60
8158
quad
rant an
d
the absci
ssa value of the featur
e p
o
int A is the biggest a
s
Figure 7(b
)
, the
manipul
ator rotates
θ
° in th
e clo
c
kwi
s
e d
i
rectio
n.
6. Discussio
n
To judge th
e
accuracy of
rotational a
ngle, a mea
s
ureme
n
t in mathemati
c
s must be
establi
s
h
ed.
denote
s
the
g
r
ay matrix
of im
age profile
after
rotation with
the grey
value
g
(
i
,
j
), and
is the
gray m
a
trix of
stan
dard ima
ge p
r
ofile
wit
h
the g
r
ey val
ue
h
(
i
,
j
) .
Th
e a
c
c
u
ra
cy
of reco
gnition
is norm
a
lize
d
as the corre
l
ation co
effici
ent
R
[12, 13].
(
1
5
)
Whe
r
e cov
(
G
,
H
) is th
e gra
y
covaria
n
ce
of rotating a
n
d
stan
dard i
m
age
s,
an
d
are
th
e stand
ard deviation,
and
are th
e gray avera
ge. Th
e
grea
ter
is, the m
o
re
accurate the rotational an
gl
e is. Whe
n
R
= 1, the rotati
onal an
gle is
compl
e
tely accurate.
It has bee
n g
enerally acce
pted that the
meas
uri
ng a
c
curacy of digi
tal image p
r
o
c
e
ssi
ng
can
achieve
0.01 pixel
s
[1
4]. However,
Hub
e
rt et
al.
and Schreier
et al. [15, 16] indicated the
r
e
exist sy
stem
atic e
rro
rs du
e to sub
-
pixe
l re
co
n
s
tru
c
ti
on, illumin
a
tion inten
s
ity chang
es, the
r
mal
noise, enviro
n
mental fa
cto
r
s, an
d
other factors. In order to im
p
r
ov
e the re
cog
n
i
t
ion robu
stne
ss
for redu
cing
the
systemati
c
e
r
ror,
we
a
nalyze
d
Equ
a
t
ion
(1
5) and
derived
the n
e
w expre
s
sio
n
s
of correlatio
n coeffici
ent
as
follows
.
(
1
6
)
Whe
r
e
is the weightin
g co
efficient,
.
an
d
are the gre
y
average an
d
grey value of image ed
ge p
i
xels, respe
c
tively.
l
is the
maximum gray. If
; if
.
In orde
r to quantitatively analyze the accura
cy of rotational an
gl
e and to imp
r
ove the
measurement
preci
s
io
n, the syst
em
atic
error is d
e
fine
d as:
(
1
7
)
Whe
r
e, the cl
ose
r
to ze
ro g
Î[0,1] is, the
highe
r the accuracy is.
The b
o
lts
wit
h
several
dif
f
erent a
ngle
s
we
re i
dentifi
ed a
nd te
ste
d
u
s
ing th
e
image
pro
c
e
ssi
ng m
e
thod
s ab
ove
.
The sy
stem
atic e
rro
r
a
n
d
the rotatio
n
error
of every
edge
pixel a
r
e
cal
c
ulate
d
a
c
cording to
Eq
uation (17
)
. T
he imag
e u
s
e
d
wa
s 1
28x1
28 pixel in
si
ze, whe
r
e
60x
6
0
= 360
0 pixe
l points
were cal
c
ulate
d
. As see
n
from Figu
re (8), the larg
e error m
a
i
n
ly
con
c
e
n
trate
s
in the
middl
e
part
of the
im
age
due
to th
e bolt
craft h
o
le. Th
e e
r
ror in th
e e
dge
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Autom
a
tic Re
co
gnition
of Large Ball
Valve Sealin
g Bolt Based
on Digital
…
(SONG Qi
ngju
n
)
8159
very small, p
r
oving that th
e algo
rithm i
s
effici
ent on
the recognitio
n
of edg
es a
nd can m
eet
the
requi
rem
ent of the bolt rotation mea
s
u
r
ement.
(a) S
c
atter p
s
eudo
colo
r m
ap of error
(b) 3
D
s
u
rfa
c
e of erro
r
Figure 8. Analysis dia
g
ra
m of measu
r
em
ent error
In addition, the correl
ation
coeffici
ent
wa
s cal
c
ul
ated, sho
w
n in Ta
b
l
e 1.
Table 1. Re
sults analy
s
is i
n
formatio
n
For both
clo
c
kwi
s
e a
nd co
unterclo
c
kwi
s
e rotation
s, the co
rrelation
coefficie
n
t R is very
clo
s
e to 1. The com
puting
time is 29s
with the
syst
ematic e
rro
r
< 0.02.
This
prove
s
that the
algorith
m
can
accu
rately id
entify the ch
a
r
acte
ri
stic p
a
rameters of th
e se
aled
bolt, and th
en the
manipul
ator
can be exa
c
tly position
ed.
7. Conclusio
n
This p
ape
r pre
s
ent
s an
algorithm o
f
reco
gni
zing
the position
and angl
e of bolt.
Additionally, we propo
se
an image m
easure
m
ent
method
for engin
eeri
ng appli
c
ation. The
expre
ssi
on of
the image g
r
ay correlatio
n
with weig
hte
d
coeffici
ent is put forwa
r
d
in this pap
er.
More
over, th
e system
atic
error
wa
s de
rived from
the
cla
ssi
cal m
e
asu
r
em
ent error. Th
e sy
stem
not only ha
s
good
rob
u
stn
e
ss, but also
ensu
r
e
s
the
pro
c
e
ssi
ng
spe
ed with
certain fea
s
ibil
ity.
Ran
dom bolt
position
s
we
re created
wi
th variable
a
ngle
s
to examine differen
t
situations. The
experim
ental
re
sult
s in
T
a
i-shan
p
u
m
ped-sto
r
ag
e
power statio
n
sh
ow t
hat t
he al
go
rithm
ha
s
good d
enoi
si
ng ability and
edge
con
n
e
c
tivity for t
he blurred b
u
t not too co
mp
lex image
s. The
system can a
c
curately re
cogni
ze
the po
sition of the bolt. With the Scatter map
and 3D surf
ace
of error, the
a
c
cura
cy of b
o
l
t rotation
wa
s verifi
ed,
pro
v
ing that the
autom
atic re
cognition
of large
ball valve sea
ling bolt ca
n be well a
c
hie
v
ed.
Ackn
o
w
l
e
dg
ements
We
would li
ke to thank t
he Tai-sh
an
pumpe
d-storage po
we
r station for hel
p in this
r
e
sear
ch.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 12, Decem
ber 20
14 : 8152 – 81
60
8160
Referen
ces
[1]
RC Gonza
l
ez,
RE W
oods. Digita
l
Image
Proce
ssin
g
Using MAT
L
AB. M. Publish
i
ng H
ouse o
f
Electron
ics Ind
u
str
y
, Beij
in
g. 201
2.
[2]
W
K
Pratt, Digital Image Proc
e
ssing. M. Acad
emic Press, Ne
w
York. 1
991.
[3]
M Sonka, V
Hl
avac, R Bo
yl
e.
Image pr
oces
sing,
a
n
a
l
y
s
is,
and m
a
chi
ne v
i
sion. M. T
homson Asi
a
Pt
e
Led. 20
02.
[4]
Li
X, G W
e
i,
XD Bia
n
. Res
ear
ch on
det
ection
of la
ne
base
d
on mac
h
i
ne vis
i
on.
J. Jo
urna
l
of South
eas
t
Univers
i
ty (Eng
lish Ed
ition).
2
004: 17
6 -17
9
.
[5]
JX Z
h
o
u
, GP W
u
, et al. A Combin
ed e
d
ge det
ecti
ng
base
d
on me
dia
n
f
ilter and
mathematics
mo
rp
ho
l
ogy
.
J.
Modern C
o
mp
uter
. 2005; (2):
73-75.
[6]
KX
He, Q W
ang. Ima
ge e
dge
detecti
on
base
d
o
n
n
o
n
-sub-s
a
mpl
e
d
contour-
l
et tr
ansform a
n
d
mathematic
al morph
o
lo
g
y
.
J
.
Journa
l of Sou
t
heast Univ
ersi
ty (English Ed
ition).
20
12; (28)
: 445 -449.
[7]
JL H
uan
g, H
Z
ou. An
im
age
edg
e
detectio
n
metho
d
of mul
t
i-structure e
l
e
m
ents a
n
d
mul
t
iscale
bas
e
d
on morp
hol
og
y.
J. Microelectronics & Co
mpu
t
er
. 2009; (26):
149-1
53.
[8]
PB Ramkuma
r
, KV Pramod. Conve
x
g
e
o
m
etr
y
a
nd ma
thematica
l
mo
rpho
log
y
i
n
a
gener
aliz
e
d
structure. J
. Internati
ona
l Jour
nal of
Co
mpute
r
Applic
ations
.
201
0; (6): 1-6.
[9]
YH T
ang, HZH Lu. Morph
o
l
ogic
a
l e
dge
detectio
n
alg
o
r
ithm base
d
o
n
multi-structu
r
e eleme
n
ts
compo
und filt
e
r
.
J.
Geomantic
s and Infor
m
ati
on
Scie
nce of
W
uhan U
n
iver
sity
. 2012; (37)
: 50-53.
[10]
JX
Sun,
DB G
u
, et a
l
. A m
u
lti
-
scale
ed
ge
de
te
ction al
gorith
m
bas
ed on
w
a
vel
e
t
d
o
mai
n
vector
h
i
dd
e
n
Markova tree
mode
l.
J. Patte
rn Recognition
.
2004; (7): 13
1
5
-13
24.
[11]
YT
Z
hang, Z
K
Yin, JY W
ang. A ne
w
im
a
ge ed
ge d
e
tec
t
ion schem
e. J. Journal of Electron
ics &
Information T
e
chno
log
y
. 2
008
; (30): 1295-1
2
99.
[12]
W
J
Maclean,
JK T
s
otsos.
Fa
st pattern re
cogn
ition
usi
n
g norm
a
liz
ed
gre
y
-scal
e
cor
r
elati
on i
n
a
py
ra
mi
d
i
m
a
ge re
p
r
e
s
en
ta
ti
on.
J. Machine Vi
sion a
nd Ap
plic
ations
. 20
08; (
19): 163-
17
9.
[13]
JY Yan, JG Yang. Ap
plic
atio
n of s
y
nt
hetic
gre
y
corre
latio
n
theor
y o
n
thermal p
o
int o
p
timizatio
n
for
machi
ne tool th
ermal error c
o
mpens
ation.
J. Int J Adv Manuf T
e
chnol
. 20
0
9
; (43): 112
4-1
132.
[14]
HW
W
ang, YL Kang, HP Xie
.
Advance in d
i
gital s
peckl
e correlati
on met
hod a
nd its ap
plicati
on.
J.
Advanc
es in M
e
chanics.
20
05
; (35): 195-20
3
.
[15]
W
Hubert, Schreier, A Micha
e
l. S
y
stem
atic erro
r in d
i
g
i
tal
imag
e correl
a
ti
on du
e to un
d
e
r match
e
d
subset sha
pes.
J. Experiment
al Mech
anics
.
200
2; (42): 303
-310.
[16]
HW
Schreier,
MA Sotton, et al
. S
y
stemati
c
errors in di
g
i
tal ima
ge corr
elati
on ca
used
b
y
i
n
tensit
y
interp
olati
on.
J. Opt Eng.
2002
; (36): 2361-2
3
61.
Evaluation Warning : The document was created with Spire.PDF for Python.