TELKOM
NIKA
, Vol. 11, No. 4, April 2013, pp. 1902
~19
0
8
ISSN: 2302-4
046
1902
Re
cei
v
ed
Jan
uary 11, 201
3
;
Revi
sed Fe
brua
ry 12, 20
13; Accepted
February 26,
2013
Mixed Programming Realization of the EMD-WVD
Combined Method
Miaozho
ng Sun*
1
, Hong
tao Tang
2
, Yuanli Xu
3
1,2,
3
Colle
ge of Mecha
n
ica
l
En
gin
eeri
ng, T
i
anjin Un
iversit
y
of
Science & T
e
chno
log
y
No.10
38 D
agu
South Ro
ad, H
e
xi District T
i
anjin C
i
t
y
, T
i
anji
n
300
22
2, PR Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: sunmzh6
6
@
s
ina.com
A
b
st
r
a
ct
W
i
gner-Vi
lle
Di
stributio
n (W
VD) poss
e
sses
hig
h
ti
me-freq
uency c
onc
ent
ration
an
d res
o
luti
ons
,
but bri
ngs s
e
ri
ous cross-ter
m
s in
process
of
a
multi-c
o
mp
o
nent si
gn
al to
distort gre
a
tly t
he res
u
lt of ti
me-
freque
ncy a
nal
ysis. Becaus
e
of this t
he us
e
of the W
V
D is
li
mited f
o
r ma
ny ap
plic
atio
ns
. In this pap
er
a
combi
ned method
is prop
os
ed:
Empiric
a
l Mode
D
e
co
mp
ositio
n (EMD) is first used to
deco
m
p
o
se t
h
e
origi
n
a
l
si
gn
al
i
n
to a
seri
es
of
Intrinsic Mo
de
F
unc
tions
(IMF
s), then fa
lse I
M
F
s
amon
g th
em ar
e e
l
i
m
i
n
a
t
ed
accord
ing to t
he corre
latio
n
coefficie
n
ts be
tw
een each I
M
F
and the o
r
igin
al si
gna
l, W
V
D is utili
z
e
d to
ana
ly
z
e
th
e re
ma
ne
nt IMF
s
, finally
e
a
ch W
V
D is
ad
ded
li
near
ly w
i
th tog
e
ther to
reco
n
s
truct the w
hol
e
W
V
Ds of the
origi
n
a
l
sig
n
a
l
. In ord
e
r to ca
rry
out an
d va
l
i
date th
e co
mbin
ed
met
hod,
a ti
me-freq
u
e
n
c
y
ana
lytic system is desi
g
n
e
d
and real
i
z
e
d
by usi
ng MAT
L
AB and D
e
lp
hi mixed pr
ogr
amming b
a
se
d
on
COM (Com
ponent Object Model)
module
technology. This
system
is us
ed
to perfor
m
vibr
at
ion signal time-
freque
ncy a
nal
ysis of a gr
ind
i
ng
mac
h
in
e. T
he a
nalytic
r
e
s
u
lts show
val
i
d
i
ty of the co
mb
ine
d
l
meth
od
a
n
d
success of the mix
ed pr
ogra
m
mi
ng
.
Ke
y
w
ords
:
e
m
pirica
l mod
e
de
compos
ition, w
i
gn
er-vill
e di
stri
butio
n, mix
ed p
r
ogra
m
mi
ng, ti
me-fre
qu
ency
analytic system
, grinding
m
a
chine
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
In ord
e
r to ex
tract the
fault
feature
s
and
t
hen id
entify the fault p
a
ttern
s, si
gnal
a
nalysi
s
has
bee
n an i
m
porta
nt topi
c in m
e
chani
cal fault di
ag
nosi
s
resea
r
ch and
appli
c
a
t
ions. On
e of
the
most pop
ular transfo
rms
kno
w
n to scientists
a
nd engin
eers is the Fourie
r transfo
rm that
conve
r
ts a
si
gnal fro
m
the
time domai
n
to the freq
ue
ncy dom
ain.
What the
sp
e
c
trum
co
mput
ed
by usin
g the
Fouri
e
r tran
sform tell
s u
s
a
r
e
the f
r
eque
nci
e
s
containe
d in t
he enti
r
e ti
me
waveform, no
t the frequ
en
cie
s
at a
part
i
cula
r time in
stant. The
Fo
urie
r tra
n
sfo
r
m provid
es t
he
sign
al’s ave
r
age
cha
r
a
c
te
ristics a
nd
smears the
sig
nal’s lo
cal
be
havior. Thi
s
i
s
the sho
r
tco
m
ing
of the Fouri
e
r tran
sform
that has been re
cog
n
iz
e
d
for a long
time. In
the process of the
machi
n
e
r
y fault diagn
osi
s
,
the si
gnal
s u
nder con
s
id
e
r
ations
are kn
own to
be
no
n-statio
na
ry, for
whi
c
h
the sig
nal’s paramet
ers are
time-v
arying.
Fo
r
th
e sp
ectral an
alysis
of su
ch
type of signa
ls,
the joint time
-freque
ncy
an
alysis tech
niq
ue i
s
wi
dely u
s
ed.
Wig
ner–
V
ille distri
buti
on i
s
on
e of t
h
e
best kno
w
n a
nalytic tools [
1
-3].
WVD is a q
u
adrati
c
form time-fre
que
ncy distri
bution
with infinite resol
u
tion
s in both the
time and freq
uen
cy domai
n. And it sup
p
lies hi
gh
re
solution
s and
instantan
eou
s po
we
r den
sity
spe
c
tru
m
s in
the time and freque
ncy do
main. WV
D b
e
ing qu
adratic in nature in
trodu
ce
s cro
s
s-
terms be
ca
use of a
multi-compon
ent
sig
nal. The
cro
s
s-te
rm
s a
r
e t
he mai
n
o
b
st
acle
preventi
n
g
the use of th
e WVD for
many appli
c
a
t
ions. In
ord
e
r to sup
p
re
ss o
r
delete
the cro
s
s-te
rms,
Smoothed
Pseu
do
WVD
(SPWVD) a
n
d
Choi-Willia
ms
di
stri
butio
n (CWD) etc. appe
ar, u
s
i
ng
kernel wi
ndo
w function
s [4, 5]. But these metho
d
s redu
ce the time-fre
que
ncy
con
c
entratio
n
. In
orde
r to
solv
e the
WV
D
cross-term
p
r
o
b
lem, a
n
effe
ctive way i
s
prop
osed
tha
t
a rea
s
on
abl
e
decompo
sitio
n
meth
od i
s
appli
ed to
divide
a m
u
lti-co
mpo
n
e
n
t sig
nal i
n
to a
num
ber of
indep
ende
nt
comp
one
nts,
then
WV
D i
s
a
pplie
d to
analyze e
a
ch
sin
g
le
co
mp
onent, finally
all
WVDs are ad
ded linea
rly with together to
reco
nst
r
u
c
t the origi
nal si
gnal time-freq
uen
cy analysi
s
.
EMD can b
e
able to d
e
compo
s
e the
origin
al si
gna
l into a serie
s
of ind
epe
n
dent an
d lo
cal
cha
r
a
c
teri
stic time scale i
n
trinsi
c mo
de
functi
on
s(IM
Fs) th
at are
orthog
onal to
each othe
r [6].
The above
method is ju
st the com
b
i
ned metho
d
of EMD and
WVD to be
use
d
to carry
out
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mixed Prog
ra
mm
ing Reali
z
ation of the
EMD-WVD
Co
m
b
ined Meth
od (Mia
ozhon
g Sun)
1903
vibration
sig
nal time-f
req
uen
cy analy
s
is for
a g
r
i
nding
ma
chi
ne with
hig
h
noi
se. Th
ro
ugh
comp
ari
s
o
n
o
f
analytic re
sults of FFT, WVD,
SPWV
D
and the
co
mbined m
e
th
od, the com
b
i
ned
method can g
e
t best ch
ara
c
teri
stic info
rmation of time-fre
que
ncy a
nalysi
s
.
In orde
r to p
e
rform th
e a
bove seve
ral
method
s an
d displ
a
y the
i
r analytic g
r
aphi
cs in
Wind
ows Op
eration Syste
m
, a
time-fre
quen
cy syste
m
is de
signe
d and imple
m
ented by u
s
in
g
MATLAB an
d Del
phi mix
ed p
r
og
ram
m
ing ba
se
d
on COM mo
dule te
chn
o
l
ogy. Matlab
is a
comp
uting scien
c
e en
gin
eerin
g lang
u
age with hi
gh efficien
cy. It possesses such ma
ny
advantag
es t
hat othe
r la
ngua
ge
s can
’
t match in
asp
e
ct
s of
matrix algo
rit
h
m, nume
r
i
c
a
l
cal
c
ulatin
g, si
gnal
pro
c
e
s
si
ng, sy
stem i
d
entifying, co
n
t
rol en
gine
er
i
ng, ne
ural
ne
twork, g
r
ap
hi
cs
displ
a
y etc.
No
w it b
e
co
mes
an
esse
ntial tool
soft
ware fo
r rese
arche
r
s. But
Matlab i
s
we
ak i
n
informatio
n al
ternation, p
r
o
g
ram exe
c
uti
on, par
amete
r
s in
put and
output. Delp
h
i
is a po
werf
ul
obje
c
t-o
r
iente
d
langu
age
with many a
d
vantage
s
su
ch a
s
ra
pid
developm
ent, conveni
ent use,
impleme
n
t of a perfect int
e
rface etc.. But Delphi
is found so dif
f
icult to carry out numeri
c
al
cal
c
ulatin
g a
nd g
r
ap
hics
p
r
ocessin
g
tha
t
its efficie
n
cy
is
mu
ch l
o
wer th
an M
a
tla
b
’s i
n
the
s
e
two
asp
e
ct
s. If the two l
ang
ua
ges
are com
b
ined to
prog
ram, they
ca
n sh
are
an
d excha
nge ma
ny
advantag
es f
o
r ea
ch othe
r [7, 8].
2. Cross-terms of Wi
gner-Ville Distri
bution
The WV
D of a contin
uou
s
sign
al x(t) is
defined a
s
:
d
e
t
x
t
x
f
t
WVD
f
j
x
2
*
]
2
[
]
2
[
)
,
(
(1)
If
)
(
)
(
)
(
2
1
t
x
t
x
t
x
, the correspondi
ng. WV
D ca
n be re
written as :
)}
,
(
Re{
2
)
,
(
)
,
(
)
,
(
2
,
1
2
1
f
t
WVD
f
t
WVD
f
t
WVD
f
t
WVD
x
x
x
x
x
(2)
Acco
rdi
ng to
Equation
(2
), obviou
s
ly, e
x
cept
two a
u
to-term
s
WV
D also h
a
s on
e cro
s
s-
term twice the auto
-
term
. This bad
cro
s
s-te
rm
make
s time
-freque
ncy a
nalysi
s
of g
r
eat
confu
s
io
n. In
orde
r to red
u
c
e or d
e
lete the effe
ct of the cro
s
s-te
rm,
a smoothe
d wind
ow fun
c
tion
calle
d a ke
rn
el function i
s
brou
ght into
the
WVD formula, su
ch a
s
sm
oothe
d pse
udo
Wign
er-
Ville distrib
u
tion and
Choi
-William
s
dist
ri
bution an
d so
on appe
ar [9
].
3. Principle
of Empirical Mode Deco
mposition
The Empi
ri
cal Mod
e
De
comp
ositio
n
(EMD) p
r
op
o
s
ed
by
Hua
ng in
19
98 i
s
a
very
powerful si
gn
al analysi
s
tool for both linear a
nd no
nlinea
r ca
se
s. The EMD has the ability to
decompo
se
a
n
y time
seri
e
s
into
a
num
ber
of
sp
e
c
trally inde
pend
ent o
scill
atory mode
s,
call
ed
Intrinsi
c Mo
d
e
Fun
c
tion
s
(IMFs). IMFs rep
r
e
s
ent
si
mple o
s
cillat
o
ry mod
e
s e
m
bedd
ed i
n
the
sign
al. Gen
e
r
ally, IMFs a
r
e o
r
thog
ona
l to each ot
her, an
d all
the IMF’s
co
ntain only o
ne
instanta
neo
u
s
frequ
en
cy [10, 11]. The IMF sho
u
ld sa
tisfy two definitions:
(1) In
the
whole a
nalysi
s
dataset, the
num
b
e
r of
extreme a
n
d
the num
ber of ze
ro
-
cro
s
sing
s mu
st be either e
qual or diffe
r at most by on
e.
(2) At any point, the mean value of the
envelop
e defined by lo
cal maxima
and the
envelop
e defi
ned by the lo
cal minim
a
is
zero.
EMD i
s
appli
ed to
de
com
pose the
sign
al x(t)
into a seri
es
of mon
o
-comp
one
nt
Intrin
si
c
Mode Fu
nctio
n
s, and x(t)
can be written as:
)
(
)
(
)
(
1
t
r
t
imf
t
x
n
i
n
i
(3)
Her
e
)
(
t
r
n
is the resid
ue an
d re
pre
s
ent
s the mean si
gnal’
s
trend.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 1902 – 1
908
1904
4. EMD and WVD
Combined Method
In pro
c
e
ss
of EMD, there
exists ove
r
d
e
com
p
o
s
ition
owing to
so
me followi
ng
cau
s
e
s
:
errors in
the l
o
cal
average
cal
c
ul
ation,
effect
of b
o
u
ndary
re
actio
n
an
d un
dem
andin
g
sta
n
d
a
rd
for ultimate filtration etc.
Over de
co
m
positio
n lead
s to additio
n
a
l IMFs that
do not bel
on
g to
those of the
origin
al sig
n
a
l
. The additio
nal IMFs
a
r
e
called “fal
se
IMFs” [12]. In orde
r to so
lve
this problem,
this pap
er p
r
ese
n
ts a m
e
thod that ea
ch IMF is u
s
e
d
to co
rrel
a
te
with the o
r
igi
nal
sign
al to gai
n
own
co
rrelati
on coefficie
n
t. False
IMF i
s
eliminated
when its
co
rrel
ation coefficie
n
t
is very small.
In the above
formula
(3), n
IMFs are divided into
k ba
sic IMF
s
an
d m false IMF
s
due to
over de
com
p
osition. The f
o
rmul
a (3
) ca
n be written a
s
:
)
(
)
(
)
(
)
(
)
(
)
(
1
'
1
1
t
r
t
c
t
c
t
r
t
imf
t
x
n
m
l
l
k
j
j
n
n
i
i
(4)
Her
e
m
k
n
,
)
(
1
t
c
k
j
j
repre
s
ent
s the su
m of k ba
sic I
M
Fs,
)
(
1
'
t
c
m
l
l
depicts the sum of m
fals
e IMFs
and
)
(
t
r
n
is the re
si
due. Obviou
sly the correlat
ion co
efficien
ts betwe
en th
e last two
terms a
nd the
original
sign
al are ab
out zero. The fal
s
e
IMFs and the
resid
ue can
be igno
red du
e
to their m
u
ch low
effect
duri
ng the
pro
c
e
s
s of
WVD
analy
s
i
s
. The
ori
g
in
al sig
nal x(t) is
approximatel
y combine
d
with the basi
c
IMFs written a
s
:
)
(
)
(
1
t
c
t
x
k
j
j
(5)
No
w
)
(
t
c
p
(p
≤
k)
a
m
ong the k
basi
c
IMFs is used to
co
rrelate with the origin
al sig
nal
x(t). The form
ula is written as:
)
(
)
(
)
(
)]
(
)
(
[
)]
(
)
(
[
)]
(
)
(
[
)]
(
)
(
[
)]
(
)
(
[
)
(
,
,
1
,
,
2
1
1
,
p
p
p
j
p
p
p
c
c
k
p
j
j
c
c
c
c
p
k
p
p
p
k
j
j
p
c
x
R
R
R
t
c
t
c
E
t
c
t
c
E
t
c
t
c
E
t
c
t
c
E
t
c
t
x
E
R
(6)
0
)
(
,
1
,
k
p
j
j
c
c
p
j
R
,beca
u
se the different ba
si
c IMFs a
r
e orthogon
al to each oth
e
r.
Equation (6)
sho
w
s that the
cross-co
rrel
ation f
unctio
n
betwee
n
ea
ch basi
c
IMF
and the
origin
al sign
a
l
is just the self-co
r
relation
func
tion of each self basi
c
IMF. The cross-correlati
o
n
coeffici
ent be
tween the jth
basic IMF a
nd the origi
n
a
l
signal i
s
writ
ten as:
j
j
j
c
x
c
j
x
c
x
t
c
t
x
E
)]
)
(
)(
)
(
[(
,
(7)
Her
e
j
c
x
,
re
spe
c
ti
vely
repre
s
e
n
t
mean
value
of the o
r
igin
al
sig
nal
and
m
ean val
ue
of the jth
basi
c
IMF.
j
c
x
,
re
spe
c
tively
rep
r
esent vari
an
ce
s of t
he o
r
i
g
inal
sign
al a
nd varia
n
ces of the
jth ba
sic IMF
.
Obviou
sly,
1
0
,
j
c
x
. When
j
c
x
,
=0,it sh
ows th
at
the ori
g
inal
sign
al
doe
sn’t a
n
y
correl
ate
with
the jth
ba
si
c IMF. When
j
c
x
,
=1, it
depi
cts that the
ori
g
inal
sign
al
compl
e
t
e
ly
c
o
rrel
a
t
e
s
wit
h
t
he jt
h b
a
si
c I
M
F.
Thu
s
j
c
x
,
can
be u
s
e
d
to ju
dge
whi
c
h
ba
sic IMF is
con
s
id
ere
d
a
s
fal
s
e
IMF o
r
n
o
t. If ce
rtai
n false IMF
s
among
k b
a
si
c IMF
s
are
el
iminated, th
ere
exists L (L
≤
k) reman
ent ba
sic IMF
s
. The
original
sign
al is simply written as:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mixed Prog
ra
mm
ing Reali
z
ation of the
EMD-WVD
Co
m
b
ined Meth
od (Mia
ozhon
g Sun)
1905
)
(
)
(
1
t
c
t
x
L
l
l
(8)
The WV
D of the origi
nal si
gnal can re
prese
n
ts a
s
:
L
l
c
c
c
c
x
f
t
WVD
f
t
WVD
f
t
WVD
f
t
WVD
f
t
WVD
l
L
1
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
2
1
(9)
5. Realiza
t
io
n of the Tim
e
-fr
eque
nc
y
Analy
t
ic Sy
s
t
em
5.1. Mixed Programming
COM suppli
e
s with an
obje
c
t-o
r
iente
d
and expa
nded
comm
unication pro
t
ocol for
Wind
ows Op
eration Syst
em. It is a
comm
on obj
ec
t interfac
e.
Ac
c
o
rding to the interfac
e
stand
ard, a
n
y
languag
e
can
call it [13]. Firs
tly p
r
og
rammi
ng
algorith
m
s of
signal
anal
ytic
method
s m
u
st be
edited
in
to functio
n
fil
e
s
(M
files)
stead of
scri
pt files u
s
ing
th
e Matla
b
M
file
editor. T
hen
each M file i
s
tra
n
sfo
r
me
d into diffe
re
nt DLL
file
with the hel
p
of Matlab
COM
Builder in p
o
s
sessio
n of an extern
al compile
r su
ch
as Borla
nd
(3, 4, 5, 6), Microsoft Visual
Studio (5.0, 6
.
0) and Mi
cro
s
oft Visual St
udio.Net etc.
Microsoft Visual Studio 6.
0 is
cho
s
e in
this
pape
r. Finall
y
the each
DLL file
as t
y
pe libra
ry
is assembl
ed i
n
to an a
c
tive
-X co
mpon
en
t in
Delp
hi. The compon
ents
a
r
e put togeth
e
r in a d
e
si
g
ned inte
rface
of the
time-freque
ncy sy
stem.
Thro
ugh th
e
interface, the
com
pon
ents coo
p
e
r
ate
with each othe
r to inte
ra
ct. Whe
n
an
alytic
method
s a
r
e
need
ed to ex
pand, the
wh
ole sy
stem
wi
ll not be
ch
an
ged, only the
co
rre
sp
ondi
n
g
components
will be pl
aced to reas
sembl
e
the
new upgrade analytic
functions. F
u
rthermore this
system
ca
n b
e
executed in
Wi
nd
ows
se
paratin
g fro
m
the Matlab
e
n
vironm
ent. Figure 1
sho
w
s
the frame of the pro
g
ramm
ing flow
Figure 1. Frame
of Progra
mmi
ng Flow
5.2. Design
of the M
a
in Inter
f
ac
e
Figure 2 presents the mai
n
inte
rfa
c
e of
the time-freq
uen
cy analytic syste
m
. It is mad
e
up of three m
odule
s
: “inp
ut data”, “si
gnal
analysi
s
” an
d “Graphi
cs shifting wind
o
w
s”.
The firs
t module is
c
o
mbined with three s
e
c
t
ions: input sam
p
ling
data file includes file
path a
nd file
name, in
put d
a
ta len
g
th (sa
m
pling
data
d
o
t), input
sam
p
ling frequ
en
cy (Hz).
Delp
hi
us
es
thes
e data to be in
communic
a
tion with Matlab.
The
se
con
d
modul
e i
s
comp
osed
of six fun
c
tion
s: waveform
display, FF
T, WVD,
SPWVD, EM
D-WVD (th
e
combi
nation
a
l
method
of
E
M
D
and
WV
D) and
ALL
W
VD (All
WV
Ds of
true b
a
si
c IM
Fs).
Wh
en
progra
mming, t
he
six corre
s
pondi
ng A
c
tive X compo
nents that h
a
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 1902 – 1
908
1906
been
in
stalle
d ab
ove a
r
e
put into th
e
main inte
rf
a
c
e. Finally the
main
progra
m
is
co
mpile
d into
an executed file that can b
e
sep
a
rate
d from Matlab.
In the last
module, a
windo
w is
set
up co
mbin
e
d
with Delp
hi Panel a
n
d
Image
comp
one
nts. After
ce
rtain analytic
item is
cli
c
k
ed i
n
sign
al an
alysi
s
mo
dule
wit
h
a mo
use, an
analytic g
r
ap
hics will
be d
i
splaye
d in g
r
aphi
cs
shifting wi
ndo
ws.
The g
r
ap
hics can
be
shifted
whe
n
a ne
w analytic item is ch
ang
ed.
Figure 2. Main interface of syst
em
Figure 3. Rea
l
time waveform
6. An Ac
tual
Example
6.1. Signal Sampling
In orde
r to verify effectivene
ss of
th
e pro
posed
combi
ned m
e
thod an
d the time-
freque
ncy an
alytic system,
they were ev
aluated u
s
in
g
the real vibra
t
ion data mea
s
ured in a type
S3SL gri
ndin
g
ma
chin
e which
ru
ns i
n
speed
285
0r
p
m
/m (the
rota
te frequ
en
cy f0=47.5
H
z)
with
big
vib
r
ation
and noi
se.
A vertical
vibrati
on sign
al
i
s
pi
cked
up
from
its surfa
c
e
sh
ell u
s
ing
a
type
YD-1
2 accel
e
rom
e
ter. A test syste
m
consi
s
ts of
type YD-12 a
c
cele
rom
e
ters, a type DHF
-4
cha
r
ge
amplif
ier with filte
r
, a type PCI2
006 d
a
ta sa
mpling
ca
rd,
a com
pute
r
a
nd a
softwa
r
e o
f
vibration
sign
al co
ntinuo
us larg
e data
a
c
qui
siti
on
an
d processin
g
etc. Fig
u
re
3 displays a
real
time waveform with 1000
Hz sa
mpling freque
ncy an
d 1000
sampli
n
g
data dots.
6.2. Signal Analy
s
is Usin
g FFT and WVD
The a
bove
waveform i
s
an
alyzed
with t
he Fa
st F
ouri
e
r T
r
an
sfo
r
m. Figu
re
4 (Thi
s figu
re
and the follo
wing figu
re
s only displ
a
y analytic grap
hi
cs in the scope of shiftin
g
wind
ows, cutting
the other pa
rt
s of the main interface)
sho
w
s the
spe
c
trum of the wa
veform. In the figure, obvio
us
pea
k value
s
are fou
nd out
to occu
r in 1
f
0, 3f0
positio
ns of low freq
uen
cy, but bad cha
r
a
c
teri
st
ic
informatio
n a
ppea
rs i
n
the high fre
quen
cy scop
e
du
e to existen
c
e o
f
noise. From
figure 5a
WV
D
time-freq
uen
cy analytic graphi
cs, 1f0
chara
c
te
rist
i
c
freque
ncy i
s
stri
king, but
others are to
o
difficulty to distingui
sh a
s
a
result of WV
D’s
cro
s
s-te
rm leading to
wro
ng e
s
tima
tion.
0
50
100
15
0
20
0
25
0
30
0
35
0
400
45
0
500
0
20
40
60
80
10
0
12
0
14
0
16
0
F
r
equ
enc
y
f
/
H
z
A
m
pl
i
t
ude A
FF
T
A
n
a
l
y
s
is
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Mixed Prog
ra
mm
ing Reali
z
ation of the
EMD-WVD
Co
m
b
ined Meth
od (Mia
ozhon
g Sun)
1907
Figure 4. Spectrum of Waveform
Figure 5. WV
D Analytic Re
sult
6.3. Analy
s
is
of Combine
d
Method o
f
EMD and WVD
Figure 6 sho
w
s EM
D results, ea
ch WVD re
su
lt of
IMFs and
correlation
co
efficients
betwe
en ea
ch IMF and the origin
al sig
nal. The co
rrelation coefficient
s (IMF1
-
IMF7) re
sp
ectively
pre
s
ent: 0.6
0
57, 0.297
1,
0.5876, 0.2
0
78, 0.012
9,
0.0062, 0.0
0
69. Obviou
sl
y, the last th
ree
correl
ation co
efficients a
r
e
very small, to sho
w
that IMF5,IMF6 and
IMF7 are con
s
ide
r
as
“false
IMFs” a
nd th
eir correspon
ding WV
D va
lues a
r
e a
b
o
u
t zero. Of course, the WVD of the re
sidue
rep
r
e
s
entin
g the mean tre
nd of the ori
g
inal si
gnal i
s
also con
s
id
ered
as
ze
ro
. WVDs
will
be
perfo
rmed
on
ly for the fo
ur rem
ane
nt IMFs
(IMF
1
-
IMF
4
).
All
a
bove analytic re
sul
t
s
can be se
e
n
in Figure 6. F
i
nally the four WV
Ds are a
dded lin
early
together to g
e
t the all WV
Ds di
spl
a
yed
in
Figure 7.
Figure 6. EMD Re
sult
s, ea
ch WV
D re
sul
t
of IMFs, correlation
coeffi
cient
s betwee
n
each IMF and
the original
signal
Figure 7. All
WVDs Re
sul
t
Figure 8. SPWVD
Re
sult
Figure 7
com
pare
s
with Fi
gure
8
whi
c
h
sho
w
s result of smo
o
thed
pse
udo
Wig
ner-Ville
distrib
u
tion. It is o
b
viou
sly found th
at 3
f
0 ch
ar
a
c
teri
stic frequ
en
cy of Figure 7
is m
u
ch mo
re
stri
king th
an
that of Figu
re 8. Thi
s
sh
ows t
hat SP
WVD
sm
ears partly the
3f0 ch
aracte
ristic
freque
ncy
wh
ich
rep
r
e
s
ent
s certain
information of
g
r
i
nding m
a
chin
e vibration. T
here
exist
so
me
low va
rying f
r
eque
nci
e
s (le
s
s than
1f0
)
i
n
sco
pe
of
a
bout time
0.3
-
0.75
s in
Fig
u
re
7. Thi
s
show
T
i
me
t/
s
F
r
equ
enc
y
f
/
H
z
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
0
50
10
0
15
0
20
0
25
0
30
0
35
0
40
0
45
0
ti
m
e
t
/
s
f
r
equ
enc
y
f
/
H
z
S
m
oot
h P
s
eud
o W
i
gne
r
-
V
i
l
l
e
D
i
s
t
r
i
bu
t
i
on
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
0
50
10
0
15
0
20
0
25
0
30
0
35
0
40
0
45
0
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 1902 – 1
908
1908
that the grin
ding ma
chin
e
vibration sig
nal imp
lie
s some low fre
q
uen
cy inform
ation (le
ss th
an
1f0), but there don’t in Figure 8. In two
figures
, the
r
e is an obvio
us commo
n 1f0 cha
r
a
c
teristic
freque
ncy. T
he above
an
alytic re
sults
reveal t
hat S
P
WVD
supp
resse
s
the cross-te
rm of
WVD
but re
du
ce
s
the time-freq
uen
cy co
nce
n
tration to
smear
so
me i
n
formatio
n. In Figu
re7, t
h
e
analytic
re
sul
t
of the EM
D-WVD combin
ed meth
od
cl
early illu
strates th
at the
bi
gge
st propo
rt
ion
of the charact
e
risti
c
fre
que
ncy is in
the
positio
n of 1f
0, the secon
d
is in
that of
3f0, the follo
w is
in that of 7f0.
This
phen
om
ena d
e
mon
s
t
r
ates that
the
sam
p
led
sig
nal of the
gri
nding m
a
chin
e is
a mo
dulated
sign
al
with
certain
odd
fre
quen
cie
s
. A
c
cording
to th
e relational
t
heory
of
rotat
i
ng
machi
n
e
r
y fault diagno
sis [14],
this gri
nding ma
chi
n
e come
s fort
h a serio
u
s i
m
balan
ce fa
ult,
particula
rly a
bad e
c
cent
ricity of the grin
ding
whe
e
l th
at lead
s to
bi
g vibratio
n a
nd noi
se
in t
he
above rotatin
g
spe
ed.
7. Conclusio
n
s
The EM
D-WVD
combi
ned
metho
d
that
use
d
to
pe
rfo
r
m time
-fre
qu
ency
analy
s
is not o
n
ly
sup
p
re
sse
s
t
he seri
ou
s cross-te
rm
s
of WVD due
to
a multi-com
p
onent o
r
igi
nal
sig
nal b
u
t al
so
hold
s
goo
d time-fre
que
ncy
con
c
ent
ratio
n
and
re
solut
i
ons. Thi
s
m
e
thod effe
ctively overcome
s
the short
c
omi
ngs of d
e
crea
se i
n
time
-fre
quen
cy
con
c
entration
cau
s
ed
by su
ch as pseud
o WVD,
smooth
e
d
p
s
eudo
WV
D
a
nd
Choi
-Williams di
stributi
on
et
c. Th
us it is a vali
d
method
of ti
me-
freque
ncy a
n
a
lysis. T
h
ro
u
gh mixed
pro
g
rammi
ng of
Matlab an
d Delphi b
a
sed o
n
CO
M mod
u
l
e
techn
o
logy,
a novel
prog
rammin
g
tho
ught h
a
s be
en
reveale
d
in the
proce
s
s of d
e
si
gn
an
d
impleme
n
t of the time-fre
quen
cy an
alytic sy
ste
m
in
whi
c
h i
s
put
FFT, WV
D, SPWVD, EM
D-
WVD a
nd AL
LWVD fu
ncti
ons to
gethe
r. With the hel
p of the ne
w thought, the
r
e exist a l
o
t of
stron
gpoi
nts
in develo
p
m
ent of
virtual
instru
ment.
Ne
w an
alytic comp
one
nts can
be e
a
sily
upgrade
d to add in this sy
stem buildi
n
g
up a large system with m
u
ch mo
re an
alytic functio
n
s.
This time
-fre
quen
cy analy
t
ic system i
n
posse
ssion o
f
the above
method
s ha
s been a
pplie
d to
carry out the
vibration
sign
al time-fre
que
ncy analy
s
is
of a grindi
ng
machi
ne. The
analytic re
su
lts
validate the succe
ss of mix
ed pro
g
ramm
ing with COM
module tech
nology.
Referen
ces
[1]
Hua
ng Jia
n
zh
a
o
, Xi
e Jian, Li
Hon
g
cai, T
i
an Gu
i, Chen
Xi
a
obo. Self-a
da
p
t
ive decom
posi
t
ion lev
e
l de-
noisi
ng m
e
tho
d
bas
ed o
n
w
a
vel
e
t transf
o
rm.
T
E
LKOMNIKA Indon
esia J
ourn
a
l
of Electrica
l
Engi
neer
in
g
. 2012; 10(
5): 101
5-10
20.
[2]
Xi
an
g L
i
n
g
, T
ang
Guij
i, Hu
Aiju
n. Vibr
atio
n
si
gna
ls time
-freque
nc
y a
n
a
l
y
sis
an
d co
mpariso
n
for
a
rotating mac
h
i
ner
y
.
C
h
i
na jo
u
r
nal of vibr
atio
n and sh
ock
. 2
010; 12
9(2): 42
-45.
[3] Shie
Qian.
Introducti
on
to T
i
me-F
re
qu
ency
an
d W
a
v
e
let
T
r
ansforms
.
P
e
kin
g
: Ch
in
a
Machi
ne Pr
ess
.
200
5: 1-18.
[4]
Shao-
bai Z
h
a
ng, Da
n-da
n
Hua
ng. Electr
oenc
ep
hal
ogra
p
h
y
featur
e e
x
tractio
n
usi
n
g hig
h
time-
freque
nc
y
r
e
so
lutio
n
an
al
ysis.
2012; 1
0
(6): 1
415-
142
1.
[5]
Luo
Yi, Wu G
uan
gfen
g, LI
Chu
n
tian. A
p
p
l
icatio
n
of Cho
i
-Williams distri
butio
n to
el
ectrical s
i
gn
als
detectio
n
in C
O
2 arc
w
e
ld
ing
.
T
r
ansaction
o
f
the china
w
e
l
d
in
g institut
i
on.
2008; 2
9
(2): 1
01-1
03,1
07.
[6]
Keji
an Guo, Xing
ang Z
h
a
ng,
Hong
gua
ng L
i
, Guang
Men
g
. Applic
ation
of EMD method to frictio
n
sign
al proc
essi
ng.
Mecha
n
ic
al
Systems an
d Sign
al Process
i
ng
. 20
08; 22:
248-
259.
[7]
LIU Hai-
ya
n, JIANG Lin, HU K
e
. T
he Applicat
i
on of Mi
xed Pr
ogrammi
ng Me
thod Bas
ed o
n
Delp
hi a
n
d
Matlab in T
r
affi
c
Flow
Estimat
e
.
Microco
m
p
u
t
er Applic
ation
s
. 2009; 30(
6): 62-6
6
.
[8]
Cui Y
u
a
n
. Re
alizati
o
n
of me
rgin
g pr
ogram
ming
of Matl
a
b
a
nd
Del
ph.
Chin
a jour
nal
of
co
mp
uter
&
digit
a
l en
gi
neer
ing
. 20
11; 26
3(
9): 176-1
78.
[9]
Baop
ing
T
ang
, W
e
n
y
i L
i
u,
T
ao Song. W
i
nd tur
b
i
ne fa
ult di
ag
nosis
base
d
o
n
Mo
rlet
w
a
v
e
let
transformatio
n
and W
i
g
ner-Vi
l
l
e
distrib
u
tion.
Ren
e
w
able
E
n
ergy
. 201
0; 35:
2862-
28
66.
[10]
Yu Jian
g, Li Qin, Yuele
i
Z
hang, Jin
gpi
ng
W
u
.
Vibration
sign
al proc
essi
ng for gear fa
ult dia
gnos
i
s
base
d
o
n
emp
i
rical
mod
e
de
compos
ition
a
nd
non
lin
ear
b
lind
so
urce s
e
parati
on.
N
o
is
e & Vi
brati
o
n
Wo
rld
w
id
e
. 20
11: 11: 55-
61.
[11]
Q Gao, C D
uan, H F
an,
Q
Meng. Rotating mac
h
i
ne fault di
ag
nosis us
ing e
m
pirica
l mod
e
decom
positi
on.
Mechan
ical Sy
stems a
nd Sig
nal Proc
essin
g
. 2008; 22: 10
7
2–1
08
1.
[12]
Cai Y
a
n
p
in
g, L
i
Aih
ua, W
a
ng
T
ao, Yao L
ian
g
, Xu
Pi
ng
IC.
eng
ine
vibr
atio
n time- fre
q
u
e
n
c
y
a
nal
ys
is
base
d
on EMD
-
W
i
gner-Vi
lle.
Chin
a Jour
na
l of Vibratio
n En
gin
eeri
n
g
. 20
1
0
; 23(4): 43
0-4
37.
[13]
Ren
p
in
g Sh
ao,
W
entao Hu, J
i
ngmi
ng C
ao.
Gear dam
age
detectio
n
a
nd
dia
gnos
is s
y
stem bas
ed o
n
COM modul
e.
Proced
ia En
gin
eeri
n
g
. 20
11; 1
5
: 2301
–2
30
7.
[14]
W
enhu
Hu
an
g,
Son
g
b
o
Xia, Rui
y
a
n
Li
u.
Ap
plicati
o
n
to tec
hni
que
a
n
d
th
eory
of the
ma
chin
ery fa
ul
t
dia
gnos
is
. Peki
ng: Chi
na scie
n
ce press. 19
9
7
: 77-81.
Evaluation Warning : The document was created with Spire.PDF for Python.