TELKOM
NIKA
, Vol. 11, No. 9, September 20
13, pp.
5336
~53
4
3
ISSN: 2302-4
046
5336
Re
cei
v
ed
Jan
uary 19, 201
3
;
Revi
sed
Jun
e
11, 2013; A
c
cepted
Jun
e
21, 2013
Motor Fault Diagnosis Based on Wavelet Transform
Lijun Wang*
, Huijuan Gu
o, Shenfeng
Zhang
Dep
a
rtment of Mecha
n
ica
l
En
gin
eeri
ng, Nort
h Chi
na
Un
iver
sit
y
of W
a
ter Resourc
e
s an
d Electric Po
w
e
r,
Z
hengz
ho
u 45
001
1, Hen
an, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
w
l
jmb@
16
3.com
A
b
st
r
a
ct
T
he w
a
velet transfor
m
the
o
r
y
is used to
motor fa
ult di
agn
osis i
n
thi
s
pap
er, consi
deri
ng it
s
character
i
stics
of
multi-r
e
so
luti
o
n
a
nd str
ong
er featur
e
extracti
o
n
a
b
ility th
an Fo
urier. The
pa
pe
r
emph
asi
z
e
s
d
e
-no
i
sin
g
a
nd
eli
m
i
nati
ng th
e
sing
ul
ar
val
u
e po
int of th
e
w
a
velet trans
form
in th
e n
on-
stationary si
gn
al. And it ma
kes a detai
le
d
and in-
d
e
p
th
analys
is ab
o
u
t how
to detect the freque
ncy
compo
nents
of
w
eak si
gn
al
b
y
usi
n
g
eq
uiv
a
l
ent p
o
w
e
r
sp
ec
trum
of rec
onst
r
uction
sig
n
a
l
,
w
h
ich is
ac
quir
e
d
by usin
g the w
a
velet transform. T
h
rou
gh
the co
mpar
is
o
n
ana
lysis of the si
mu
latio
n
sign
al an
d mo
to
r
vibrati
on sig
n
a
l
’
s
exp
e
ri
me
ntal dat
a, the c
o
rresp
ond
in
g ener
gy of orig
inal s
i
gn
al
’
s
e
quiv
a
le
nt pow
e
r
spectru
m
a
nd reconstructi
ng sign
al
’
s
eq
uiva
lent
p
o
w
e
r sp
ectrum
are c
o
mp
are
d
to d
e
termine
the fa
ul
t
freque
ncy, so as to accurate
l
y
find out the
motor fa
ult.
Ke
y
w
ords
:
w
a
velet transfor
m
, motor, equ
iva
l
ent pow
er sp
e
c
trum, w
eak signa
l
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The no
rmal o
peratio
n of the motor i
s
of great
signifi
cance for the
safety, high e
fficiency,
high q
uality a
nd lo
w con
s
u
m
ption op
era
t
ion in t
he p
r
oce
s
s of man
u
facturi
ng a
n
d
pro
d
u
c
tion.
In
the operation
process, the perfo
rman
ce of the motor gra
duall
y
deteriorate
s, influenced
by
variou
s fa
ctors,
such
a
s
el
ectri
c
al,
ther
m
a
l, mech
ani
cal factors and
the su
rro
u
nding
environ
ment,
whi
c
h
ultim
a
tely ma
ke t
he m
o
tor
break do
wn.
T
hus the
re
qu
ireme
n
t of e
a
rly
diagn
osi
s
a
n
d
ea
rly wa
rni
ng is
be
comi
ng in
cre
a
si
ng
ly urgent. Th
rough
analy
s
i
s
of the vib
r
a
t
ion
colle
ction
sig
nal, the fault sou
r
ce sepa
ration and fa
ul
t pattern re
co
gnition in mot
o
r ru
nnin
g
st
ate,
the motor fa
u
l
t can b
e
dete
c
ted a
nd furt
her d
e
te
ri
orat
ion can b
e
prevent
ed, re
d
u
cin
g
produ
ct
ion
losse
s
ca
use
d
by unexpe
cted incid
ents.
The
re
sea
r
ch
of motor fau
l
t diagno
si
s t
e
ch
nolo
g
y ha
s b
egun
si
nce the 1
960
s
abro
ad;
the importa
nce of the motor equipm
ent fault
diagno
si
s tech
nique h
a
s already b
een re
co
gni
zed
in Chin
a. But until the 198
0
s
, we
are
ded
icated to th
e
study of ele
c
t
r
ical
equi
pme
n
t failure o
n
li
ne
diagn
osi
s
system. In rece
nt ten yea
r
s,
motor fa
ult
diagn
osi
s
te
chnolo
g
y ha
s
been
develo
ped
rapidly; the
a
nalytical m
o
d
e
l metho
d
, si
gnal
pro
c
e
s
si
ng a
nd exp
e
rt kno
w
le
dge
are
acco
unte
d
the main rese
arch compo
n
ents of fault d
i
agno
si
s
tech
nology, incl
u
d
ing relevant
function hi
gh
er-
orde
r stati
s
tics, spe
c
trum analysi
s
, etc.
Howeve
r, these meth
od
s are only co
n
f
ined to the fault
appe
arin
g in the stable o
p
e
ration p
r
o
c
e
ss of the
mot
o
r. People p
a
y
more and
more attentio
n to
real
-time dia
g
nosi
s
of the f
aults u
nde
r d
y
namic
co
nd
itio
n
s
s
u
c
h
as in
th
e
s
t
ar
ting
, a
c
c
e
ler
a
tion
,
bra
k
ing a
nd
other dyna
mi
c pe
riod
s of the motor.
Fo
r the failure o
f
the motor st
ator, the Hilb
ert
transfo
rm i
s
use
d
to sign
al pre
p
ro
ce
ssing, adopt
in
g
wavelet pa
cket de
com
p
o
s
ition to achi
eve
stator fault feature
s
extra
c
tion. The re
co
nstru
c
ted
coe
fficients u
s
ing
sub-ban
d no
de RMS rate
of
cha
nge a
r
e
con
s
id
ere
d
a
s
the fault feature indi
cat
o
r [1]. It adopts Park's v
e
ctor m
e
thod
to
diagn
ose mo
tor pha
se fai
l
ure, ma
ki
ng
stator thre
e
phase cu
rrent from
,,
ab
c
coordin
a
tes
down-conve
r
sion
,
dq
coo
r
di
n
a
tes. T
he ve
ctor traje
c
to
ry of no
rmal
o
peratio
n m
o
tor
ca
n o
n
ly
be clo
s
e to a circl
e
, it becomes a
n
ellip
se wh
en there are all kin
d
s
of faults [2]. But only when
the faults go to a cert
ain extent, and it will
have a certa
i
n influen
ce o
n
the trajecto
ry.
Wavelet tran
sform i
s
a ho
t technolo
g
y in sign
al pro
c
essing. Beca
use it ha
s go
od time-
freque
ncy
l
o
calizatio
n cha
r
acteri
stics, a
nd can
a
c
cu
rately gra
s
p
s
the tra
n
sie
n
t
sign
al [3], it i
s
very suitabl
e for the analy
s
is of the m
o
tor dynami
c
sign
al. The result
s of the
study show t
hat:
the wavel
e
t d
e
com
p
o
s
ition
and
re
con
s
truction
ca
n no
t only be abl
e
to remove
th
e sin
gula
r
poi
nt
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Motor Fa
ult Diagno
si
s Based on Wavele
t Transfo
rm
(Lijun Wang
)
5337
of acqui
sition
signal, a
c
cording to the hi
gh-frequ
en
cy coefficie
n
ts
also
can a
ccurately dete
r
mine
the lo
cation
of the p
o
int
appe
arin
g
si
ngula
r
valu
e
,which i
s
the
time poi
nt o
f
failure. By t
h
e
equivalent p
o
we
r sp
ect
r
u
m
of the wavelet coe
ffici
ents, useful frequ
en
cy informatio
n can
be
extracted a
n
d
thus the mot
o
r fault is dia
gno
sed a
c
curately.
2. Wav
e
let Transform
As a n
e
w
sig
nal p
r
o
c
essin
g
metho
d
[4], wavelet a
nal
ysis m
a
kes t
he vari
ous freque
ncy
comp
one
nts
decompo
se
d
into
non
-ov
e
rlap
ping
ba
nd, p
r
oviding
an
effective
way
for si
g
nal
filtering, sig
n
a
l-to-noi
se separat
ion a
n
d
feature ext
r
actio
n
. After wavelet tran
sform, the
si
gnal
cha
r
a
c
teri
stics (su
c
h a
s
sing
ular
poin
t
) have
the
same tim
e
domain l
o
cation at different
decompo
sitio
n
scale
s
, an
d remai
n
un
cha
nge
d with
original
sign
al analysi
s
freque
ncy. In the
time domain,
the total informatio
n of every decom
p
o
se
d is more
than the original sig
nal, the
time-dom
ain
cha
r
a
c
teri
stics of the orig
inal si
g
nal can be mai
n
tained; an
d h
a
s a go
od fil
t
er
cha
r
a
c
teri
stic in the freque
ncy domai
n,
each scale b
and bet
wee
n
the overlap
p
i
ng.
Figure 1 i
s
wavelet tran
sform p
r
in
cipl
e of 3
scale
s
. The
1
t
,
2
t
and
3
t
is the
approximatio
n coefficie
n
ts
and th
e
1
t
,
2
t
and
3
t
is
the
d
e
tail coefficient
s wh
en
the
scale
j = 1, 2 and 3.
And the
hz
,
2
hz
and
4
hz
is the low-pa
ss and the
g
z
,
2
g
z
and
4
g
z
is the
high-pa
ss filter. He
re:
11
22
1
33
2
1
xz
z
z
zz
z
zz
z
z
(1)
Like thi
s
co
nstantly decom
positio
n, so it can r
eali
z
e th
e sign
al multi-re
sol
u
tion de
comp
ositio
n.
Figure 1. Wa
velet Tran
sform Principl
e
2.1. Wav
e
let Trans
f
orm De-noising
Signal noi
se
redu
ction p
r
o
c
essing i
s
an i
m
porta
nt app
lication of
wa
velet analysi
s
. In th
e
pra
c
tical
en
gi
neeri
ng
appli
c
ation
s
, the
analyzed
sig
nal may
co
ntain ma
ny pe
aks o
r
m
u
tation
parts,
an
d th
e noi
se
i
s
n
o
t sm
ooth
white noi
se.
T
o
the
noi
se
redu
ction
pro
c
e
ssi
ng
of
such
sign
als, the t
r
adition
al Fo
urie
r tran
sfo
r
m analys
i
s
m
e
thod a
ppea
rs hel
ple
ss, b
e
ca
use it ca
n
not
give the cha
n
ges of the si
g
nal at a point in time.
Becau
s
e th
e
engin
e
sig
nal
of each
stat
e is p
r
e
s
ent i
n
terferen
ce,
and the inte
rf
eren
ce
affects
the e
x
traction
of chara
c
te
risti
c
para
m
et
ers o
f
fault, the in
terfere
n
ce
si
gnal
sh
ould
be
remove
d first. We can rem
o
ve the interferen
ce
sign
al by using the excellent filtering p
r
op
erti
es
of the wavele
t transform, and get the ch
ara
c
teri
stic in
formation of the different st
ates.
2.2. Wav
e
let Trans
f
orm Discontinu
o
u
s
Point De
te
ction
The a
b
ru
pt cha
nge
poin
t
of the sig
nal o
ften
co
ntains im
po
rtant informat
ion for
equipm
ent op
erating
statu
s
; they reflect the failure
ca
u
s
ed by the
crash, o
scill
atio
ns, frictio
n
an
d
stru
ctural def
ormatio
n
, etc. The abrupt
cha
nge p
o
in
t of the sign
al is al
so
kn
own a
s
sing
ular
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 533
6 – 5343
5338
point. Judgi
n
g
the
eme
r
ge
nce
mom
ent
of sin
gula
r
va
lues of
status sig
nal
to
real
ize
qua
ntitative
descri
p
tion of
the sig
nal
si
ngula
r
ity has
importa
nt
me
aning i
n
the fi
eld of si
gnal
pro
c
e
ssi
ng a
nd
fault diagno
si
s.
Gene
rally sp
eaki
ng, if the function i
s
some
whe
r
e di
scontinuity or
the derivative of a
given ord
e
r is discontinuo
u
s
; it sa
ys the function ha
s the cha
r
a
c
teri
st
ic of sing
ula
r
ity at this point
[5]. In order t
o
investigate
the relatio
n
sh
ip of
different
scale wavel
e
t transfo
rm
and si
ngul
arity,
we a
dopt the
convolutio
n
of the wavele
t transfo
rm. T
r
eated
scale
a
as inde
pen
d
ent variabl
es,
thus expressi
on of wavelet
trans
form is
as
follows
:
1
,
a
R
xt
Wf
a
x
f
t
dt
f
t
aa
aR
x
R
,
(2
)
Whe
n
wav
e
l
e
t
has
comp
a
c
t sup
p
o
r
t, i.e. there is
0
k
. If
x
k
,
0
x
. For the formula
(2), if
xt
k
a
,
0
xt
a
. Thus formula (2) e
qual
s to zero outsid
e
,
x
ak
x
a
k
. Namely:
1
,
xa
k
xa
k
xt
Wf
a
k
f
x
d
t
aa
(3)
Whe
n
0
a
, wavelet tran
sform i
s
on th
e rea
c
tion of the lo
calize
d
state
s
in the poi
nt
x
, namely
we can take advantag
e of wavelet tran
sform
to judge
the local
sing
ularity of function.
Since the Fo
urie
r tran
sform conve
r
ts the sig
nal into
a pure fre
q
u
ency dom
ain
sign
al, it
doe
s not h
a
ve time re
sol
u
tion, so th
e ch
ange
point
of
the sig
nal fre
quen
cy can't
be dete
c
ted
b
y
it. The time p
o
int of the
sig
nal mutatio
n
can
be
accu
rately dete
c
te
d by u
s
ing
wavelet tran
sfo
r
m.
Thro
ugh the
sign
al’s multi-scale an
alysi
s
, whe
n
the signal app
ea
rs mutation, the coeffici
ents of
its wavelet transfo
rm
hav
e mo
dulu
s
m
a
xima. We
can d
e
termi
n
e
the p
o
int in
t
i
me of th
e fai
l
ure
throug
h the modulu
s
max
i
ma point det
ection.
2.3.
Signal Identific
a
tion
of Wav
e
let Transform
In the actu
al
pro
b
lem, si
g
nals th
at nee
d
to be a
d
d
r
essed
are
often mixed
wit
h
othe
r
element
s, for example, the
high-f
r
eq
uen
cy inform
atio
n like
noi
se.
Gene
rally, th
e gen
eral f
a
ctors
of rea
c
ting system itself n
a
ture a
r
e oft
en so
me
sl
o
w
ly cha
ngin
g
information.
Thro
ugh
wav
e
le
t
transfo
rm,
al
ong with
the increa
sing nu
mber of
la
yers of the
wav
e
let tran
sform decomp
o
si
tion,
approximate coeffici
ents contain
le
ss
high freq
uen
cy informatio
n. With the high freq
uen
cy
comp
one
nt filtered
out ste
p
by ste
p
, th
e rem
a
in
in
g i
ngre
d
ient
s a
r
e getting
clo
s
e to the ove
r
all
trend
s of the sign
al.
2.4.
Identific
a
tion Signal Spectr
u
m Componen
t
s
In the a
c
tual
sign
al p
r
o
c
e
s
sing, th
e
sign
al
often co
nta
i
ns a
lot of
fre
quen
cy comp
onent
s.
If the accurate position of the sign
al co
mpone
nt
is need not to determine, the tradition
al Fou
r
ie
r
transfo
rm
m
e
thod i
s
ve
ry effective
to solv
e thi
s
p
r
obl
em.
Becau
s
e
after the
wavelet
decompo
sitio
n
, different scale
s
have d
i
fferent time and fre
quen
cy resol
u
tion
s, and thu
s
the
different freq
uen
cy comp
o
nents in si
gn
al can be
sep
a
rated by usi
ng wavelet d
e
com
p
o
s
ition
.
By
mean
s of eq
uivalent po
wer spe
c
trum,
most of
frequ
ency compo
n
ents contain
e
d
in the sign
al
can b
e
reveal
ed.
Then, thro
ug
h an example
,
we prove th
e possi
bility a
nd usefulne
ss of wavelet tran
sform
applie
d in pra
c
tice. Con
s
tru
c
ts a si
gnal, li
ke:
12
11
2
23
4
0.
08
c
o
s
2
1
0
c
o
s
2
0.2
c
os
2
0
.5
cos
2
xt
x
t
x
t
x
tf
t
f
t
x
tf
t
f
t
(4)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Motor Fa
ult Diagno
si
s Based on Wavele
t Transfo
rm
(Lijun Wang
)
5339
Her
e
,
1
25
f
Hz
,
2
50
f
Hz
3
150
f
Hz
and
4
300
f
Hz
.
In the simula
tion signal, random
white
noise followi
ng a no
rmal
distrib
u
tion i
s
added. Sam
p
ling fre
que
n
c
y is
200
0
H
z
. The si
mulation
sign
al and it
s po
we
r spe
c
trum
are
sh
own i
n
Figu
re 2. In the p
o
we
r spe
c
tru
m
diag
ram,
only
freque
ncy
2
f
can be se
en
.
Figure 2. Simulation Signal
and its Powe
r Spect
r
um
The sim
u
latio
n
sign
al is a
n
a
lyzed by u
s
i
ng 5
layer m
u
lti-re
sol
u
tion
throug
h the
sele
ction
of mother
wa
velet
5
db
and the
result is
sho
w
n in Fi
gure
3.
5
a
is a low freque
ncy coef
ficient of
scale 5 and
1~
5
dd
is resp
ectiv
e
ly the low frequ
en
cy co
efficient of scale
1~
5
. Due to the
sampli
ng fre
quen
cy is
200
0
H
z
, the main fre
quen
cy ban
d
range of
1~
5
dd
is respectively
500
~
1
000
H
zH
z
,
25
0
~
5
0
0
H
zH
z
,
125
H
z
~
250
H
z
62
.5
~
1
2
5
H
zH
z
and
3
2
.1
25
~
6
2.5
H
zH
z
. The
freque
ncy ba
nd ran
ge of
5
a
is
0
~
32.12
5
H
zH
z
. Clearly,
500
~
1
000
H
zH
z
doe
s not co
n
t
ain the
main
com
p
o
nents of th
e
sig
nal,
and
its m
a
in
compon
ent i
s
the
noi
se.
Becau
s
e
of
the
interferen
ce of
noise,
the developm
ent trend of
t
he signal i
s
not vi
sible. T
he d
e
velopment tre
n
d
of the signal
after wavelet
de-n
o
isi
ng is
clea
re
r.
The fre
que
ncy spe
c
trum
o
f
a sign
al m
ean
s that th
e sig
nal i
s
transfo
rme
d
from time
domain
to a freque
ncy d
o
m
a
in, and it i
s
only different
rep
r
e
s
entatio
n metho
d
of the same
kin
d
of
sign
al. But the sig
nal i
s
studied
by the
power
spe
c
trum [6, 7] tha
t
is from th
e
energy point
of
view, which
shows th
e u
n
it ban
d
sign
al
power
ch
ang
es
with f
r
eq
u
ency
co
nversion. Th
e Fo
urier
transfo
rm of the ran
dom
si
gnal do
es n
o
t exis
t; therefo
r
e, we
study its power
spe
c
trum.
Figure 3. Wa
velet Analysis of Simulation
Signal
Figure 4. Equivalent Powe
r Spectru
m
Di
agra
m
of Simulation Signal
In orde
r to ef
fectively extract the
wea
k
co
mp
one
nt of the sig
nal,
high-f
r
eq
uen
cy noi
se
comp
one
nt
1
d
is ignore
d
. The
powe
r
sp
ect
r
um of
2~
5
dd
and
5
a
are obtain
ed separately. The
power
sp
ectrum dia
g
ra
m
is no
rmali
z
e
d
to det
ermin
e
the maxim
u
m value
of
each spe
c
tru
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 533
6 – 5343
5340
Since the
sa
mpling fre
que
ncy of ea
ch scale i
s
t
he sa
me, so they h
a
ve the sam
e
freque
ncy. T
h
e
equivalent
po
wer spe
c
trum
can
be
obtai
ned by th
e
superpo
sition
of the po
we
r
spe
c
tru
m
g
r
a
ph.
The re
sult is
sho
w
n in Fig
u
re 4. The p
o
w
er
spe
c
trum
peak of
1
25
f
Hz
,
2
50
f
Hz
,
3
150
f
Hz
and
4
30
0
f
Hz
of the
si
mulation
sig
n
a
l can
clea
rly be o
b
serve
d
in Figu
re
4. T
h
rou
gh
equiv
a
lent
power
sp
ectrum of the
wa
velet tran
sform co
effi
cient
s, the
wea
k
e
r
frequ
en
cy co
mpone
nts i
n
the
origin
al sig
nal
can be d
e
tected.
3. Applicatio
n of Wav
e
let Transform in Motor Fa
ult Diagno
sis
3.1. Acquisition Parameters of Vibra
t
ion Signal
The sa
mplin
g
frequen
cy is:
20
00
s
f
Hz
(5
)
Sampling points is:
1024
N
(6
)
Acco
rdi
ng to
the Nyq
u
ist
sam
p
ling th
eore
m
, it is
kno
w
n th
at the hig
h
e
s
t standard
freque
ncy:
ma
x
/
2
10
00
s
f
fH
z
(7
)
3.2.
Fault Ide
n
tifica
tion a
nd Trea
tmen
t
Comp
ared t
he a
c
tual
measured
chara
c
te
rist
i
c
s with the
archival
cha
r
acte
ri
stic
informatio
n,
wheth
e
r the f
a
ilure
o
c
curred
can
be
d
e
termin
ed. If
no failu
re
o
c
curs,
co
ntinu
e
to
monitor the
motor
statu
s
.
If there
a
r
e
some
fault
s
,
analyze d
a
ta
and
dete
r
min
e
the
fault types.
Localized fea
t
ures of
wave
let analysi
s
m
a
ke
it wi
d
e
ly
applie
d in th
e
sig
nal p
r
ep
ro
ce
ssi
ng. Fig
u
r
e
5 is a sig
nal p
r
ep
ro
ce
ssi
ng
pro
c
e
ss.
Figure 5. Signal Prep
ro
ce
ssi
ng
3.3. Motor F
a
ults
There are various fault
s
[8, 9] in the operat
ion p
r
ocess of the moto
r and the vibration is
a wide
sp
re
ad
pheno
men
o
n
of all equip
m
ent duri
ng
operation. Th
e vibration of
the motor d
u
r
ing
operation i
s
divided
int
o
two
ki
nd
s: mechani
cal
vibration
a
nd el
ectrom
agneti
c
vibra
t
ion.
Mech
ani
cal vibration will
be ca
used
wh
en
th
e motor roto
r imbalan
ce, rolling
b
e
a
r
in
gs
abno
rmalitie
s, the plain be
aring
s
a
bno
rmal, in
stallati
on bad
and
adju
s
tment a
d
verse, amo
ng it
the rotor u
n
b
a
lan
c
e is the
most co
mmo
n. The re
a
s
o
n
that cau
s
e the roto
r imbal
ance mainly has:
rotor part
s fall off or shi
ft, rotor coil
shift or
loose due to insulation contraction,
coupli
n
g
imbalan
ce, a
s
well as th
e
coolin
g fan
and the
roto
r surfa
c
e
eve
n
ly fouling, e
t
c. The latter is
related to th
e motor assembly, such as stato
r
inte
r-tu
r
n sh
ort c
i
rcuit, bro
k
e
n
rotor ba
rs a
n
d
uneven ai
r ga
p.
4. Experimental Re
sults
and An
aly
s
is
In this
pape
r,
the mai
n
p
u
m
p moto
r of
hydr
auli
c
stat
ion i
s
taken
as
an exa
m
p
l
e. The
appli
c
ation of
wavelet tran
sform in the
motor fault d
i
agno
si
s is e
x
pound
ed. The main pu
mp
motor of hyd
r
auli
c
statio
n
is a thre
e-p
hase
asyn
ch
ronou
s moto
r, and its rate
d spe
ed i
s
1
480
RPM.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Motor Fa
ult Diagno
si
s Based on Wavele
t Transfo
rm
(Lijun Wang
)
5341
Motor fault di
agno
si
s sig
n
a
l
that is got throug
h the ex
perim
ent is p
r
epro
c
e
s
sed
a
t
first in
the pa
per.
Th
en the
wavel
e
t tran
sform
tech
niqu
e is a
pplied to
an
al
yze the
si
gna
l and
get
rid
of
the si
ngul
ar
points. By
eq
uivalent p
o
wer
sp
ectr
um of
the re
con
s
tructed
sign
al
,
three
sub
-
b
and
sign
al compo
nents
1
RP
M
,
2
RP
M
and
4
RP
M
of the fault signal
can b
e
extracted. T
he
results confirm that the
met
hod of e
quivalent po
wer
spe
c
tru
m
of the wavelet analysis
coeffici
ents
can effectively
improve th
e
freque
ncy d
o
main ali
a
si
n
g
phe
nome
n
on an
d a
c
hie
v
e
good h
a
rm
oni
c extra
c
tion result
s.
The o
r
iginal
sign
al is
sho
w
n in Fi
gure
6. It
is evident that there
are two si
ngu
lar value
points in
50
0
n
and
80
0
n
.
Figure 6. The
Original Sig
n
a
l
Figur
e7. Wav
e
let Tran
sform Results of
Origin
al Sign
al
The ori
g
inal
sign
al is do
n
e
by 5 layer decompo
sitio
n
usin
g moth
er wavelet
5
db
. The
wavelet coefficient
s are
sh
own in Fig
u
re
7. We
can fi
nd the sin
gul
ar value poi
nts are
contai
n
ed
in the d
e
tail signal
1
d
,
2
d
,
3
d
and
4
d
, an
d the
sing
ular values are in
good
ag
ree
m
ent with i
n
t
h
e
origin
al sig
nal
.
In orde
r to eliminate sin
g
u
l
arities a
nd g
e
t reco
nst
r
u
c
tion sign
al,
1
d
,
2
d
,
3
d
and
4
d
are
equal to zero
. The resultin
g sign
al wav
e
form is
a
s
shown in Figure 8. Compa
r
i
ng Figu
re 6 a
nd
Figure 8, the sing
ular valu
e point is
not
very obvious
after this met
hod.
Equivalent p
o
we
r sp
ectru
m
of recon
s
truction
si
gnal
after eliminati
ng the sing
ul
ar value
point is
sho
w
n in Figu
re 9.
By contra
st Figure
9 an
d
Figure
2, the re
sult ca
n
be got that if the
origin
al
sign
a
l
of moto
r vi
bration
i
s
a
n
a
lyzed
by
po
wer spe
c
tru
m
directly,
only
the sig
nal with
large e
n
e
r
gy can b
e
got. The ene
rgy of the we
ak
sign
al is relatively
small and its
spe
c
tru
m
is n
o
t
so
obviou
s
.
Thro
ugh
the
equivalent
po
wer spe
c
trum
of recon
s
tru
c
tion
sig
nal
a
fter the
wave
let
coeffici
ents, power sp
ectrum
pe
ak
of the wea
k
sign
al is
easi
e
r to
be
see
n
. Thi
s
p
r
ovide
s
a
ne
w
method for ex
tracting
wea
k
signal
s in the
motor fault diagno
si
s.
Figure 8. Wa
veform after
Eliminating
Singula
r
ities
Figure 9. Equivalent Powe
r Spectru
m
of
Re
con
s
tru
c
tio
n
Signal
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 9, September 201
3: 533
6 – 5343
5342
As the motor spee
d is
148
0
RP
M
, th
e motor rotation frequenc
y
is
148
0
6
0
2
4
.
7
H
z
. In
Figure 9,
19.53
H
z
is aroun
d
24
.7
H
z
. Due to the frequen
cy accu
ra
cy of
spect
r
al anal
ysis, there is
often not
pea
k in
the th
eoretical
rotation
al fr
eq
uen
cy.
If there a
r
e
o
b
vious freq
ue
ncy
comp
one
nt
in the theo
ry rotation frequ
ency
f
rang
e, then it ca
n be
consi
dered a
s
the rotation
al
freque
ncy
of the m
o
tor.
Therefore
19.53
H
z
ca
n
be
identified
the rotational
freq
uen
cy of
the m
o
tor. It
can
be
see
n
that
sig
nal en
ergy in
1
RP
M
is big
g
e
s
t, also the
sign
al
energy is big
ger i
n
2
RP
M
and in
4
RP
M
. Contra
st to the cha
r
a
c
teri
stic fre
quen
cy t
able [10] shown in Tabl
e 1, rotor un
balan
ce
and
bea
ring
l
oosene
ss
ca
n be
a
s
certai
ned
as the
fa
ult of the
mot
o
r
preli
m
ina
r
il
y. And be
ca
u
s
e
the sign
al en
ergy in
1
RP
M
is bigg
est, the rotor i
m
balan
ce i
s
deeme
d
to the main motor
failure.
Table 1. Part
Cha
r
a
c
teri
stic Fr
equ
en
cy of the Motor Vibration
Fault t
y
p
e
Characteristic fre
quenc
y
rotor un
balance
1×RPM
rotor misalignment
1×RPM,2×RPM
bearing loosenes
s
various frequenc
y doubling
fraction frequenc
y doubling
clearance vibration
0.4~0.5×RPM,1~
5×RPM
rotor an
d station
a
r
y
parts friction
higher harm
onic, Lo
w
harmonic and co
mbination
harmonic
5. Conclusio
n
Based
on
wa
velet tran
sform, the fault
d
i
agno
si
s met
hod
ha
s in
co
mparable
adv
antage
s
in the asp
e
ct
of fault diagnosi
s
. It does not nee
d the mathem
atical mod
e
l of
the object, the
cal
c
ulatio
n i
s
sm
all an
d it
s a
b
ility to ov
ercome
the
n
o
ise
is strong
. It has
goo
d
time-freque
n
c
y
locali
zation
chara
c
te
risti
c
s and it has the abilit
y to
make a
daptiv
e zoom an
d multi-re
sol
u
tion
analysi
s
of si
gnal
s.
A simulation
sign
al is take
n as an
exam
ple firstly. Through the
sele
cted moth
er
wavelet
function
5
db
, the simulatio
n
si
gnal is
analy
z
ed
with wavelet tran
sform. With the
help of the
equivalent
po
wer spe
c
trum
of the wavel
e
t coeffici
ent
s that al
rea
d
y
has
got, we ca
n verify t
h
e
effectivene
ss and p
r
a
c
tica
lity of this method in
extracting sign
al cha
r
a
c
teri
stic
freque
ncy.
T
he
wea
k
signal
in motor vib
r
ation is the
rese
arch o
b
je
ct in this p
a
per. Th
e we
ak
signal i
n
the
origin
al
sign
a
l
ha
s b
een
ef
fectively extracted
by
thi
s
method
and
i
t
s po
we
r
sp
e
c
trum
pe
ak al
so
has
bee
n cl
e
a
rly sho
w
n. Compa
r
ing
po
wer spe
c
trum
of the ori
g
ina
l
sign
al an
d e
quivalent p
o
wer
spe
c
tru
m
of the recon
s
tru
c
ted si
gnal, most of t
he sign
al frequ
e
n
cy co
ntaine
d in the origi
nal
sign
al has b
e
en accu
rately
acqui
red, especi
a
lly t
he freque
ncy of weak si
gnal th
at is not easy
to
find by ge
ne
ral metho
d
. B
y
the compa
r
ative analy
s
is of the
sig
nal
frequ
en
cy correspon
ding
to
each po
we
r
spe
c
tru
m
pe
ak a
nd p
a
rt
of the charac
teristic freq
u
ency of vib
r
a
t
ion motor, m
o
tor
fault diagno
si
s ba
sed o
n
wavelet transfo
rm ha
s bee
n accurately re
alize
d
.
Ackn
o
w
l
e
dg
ements
This wo
rk wa
s
suppo
rted by
National Nature
S
c
ien
c
e Foun
dation
of China
(51
0760
46),
Zheng
zh
ou
Measuri
ng & Control
Tech
nolo
g
y and Instrumentation
s
Key Laboratory
(121PYF
ZX1
81), Scie
nce and Te
ch
nol
ogy Re
sea
r
ch
Key Project
of the Education De
part
m
ent
Hen
an Province
(13A
510
710), Sta
r
tup
Re
sea
r
ch
F
ound
ation for High
Level
Talents of North
Chin
a Unive
r
sity of Water
Re
sou
r
ces a
nd Electri
c
Powe
r.
Referen
ces
[1]
Xu C
ao, Hu
a
x
un Z
han
g.
T
r
ansie
nt Impact
Sign
al
’
s
D
e
tec
t
ion Base
d on
W
a
velet T
r
ansformati
o
n
.
Advanc
es in In
tellig
ent S
y
ste
m
s. 2012; 13
8: 309-3
13.
[2]
Marqu
e
s Card
oso AJ. Inter-turn Stator W
i
n
d
in
g F
ault Di
a
gnos
is in T
h
ree-ph
ase Ind
u
ction Motors, b
y
Park's Vector Approach.
Jour
nals & Mag
a
z
i
nes.
199
9; 14(
3): 595-5
98.
[3]
Rachel E, Alan S.
A
W
a
vel
e
t Packet Approac
h to T
r
ansie
nt Signa
l Classific
a
tio
n
. Appl
ied a
n
d
Comp
utation
a
l
Harmon
i
c Ana
l
ysis. 1
995; 2(3
)
: 265-27
8.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Motor Fa
ult Diagno
si
s Based on Wavele
t Transfo
rm
(Lijun Wang
)
5343
[4]
K
y
us
un
g Kim. Inductio
n
Mo
tor F
ault Dia
g
nosis Bas
ed
on Ne
uro
p
red
i
ctors and W
a
velet Sig
n
a
l
Processi
ng.
Jo
urna
ls & Maga
z
i
n
e
s
. 20
02; 7(
2):201-
21
9.
[5] Jian
Cen,
Y
i
nb
oW
u.
F
eature
Extraction Met
hod for
F
ault D
i
ag
nosis
of Ma
chin
e Un
it Bas
ed o
n
W
a
ve
le
t
Sing
ular
ity Prin
ciple a
nd Immuno
logy Opti
mi
z
a
ti
on Pri
n
cip
l
e
. F
r
ontiers in Comp
uter Edu
c
ation. 20
12;
133: 42
3-4
30.
[6]
Cusido J. Fault Detection
in
Ind
u
ction
Machi
nes Usi
ng Po
w
e
r Sp
ectral De
nsit
y in W
a
vel
e
t
Decom
positi
o
n
.
Journals & M
aga
z
i
nes.
2
008
; 55(2): 633 – 6
43.
[7]
Stilia
n Stoev
a, Murad S T
aqq
ua,
Ch
eol
w
o
o
Parkb, JS Marr
onc.
On the W
a
vel
e
t Spectru
m
D
i
ag
nostic
for Hurst Par
a
meter Estimation in
the
Analys
is of In
ternet T
r
affic
. Comp
uter N
e
t
w
o
r
ks. 2
005
;
48(3):4
23-4
45.
[8]
Z
hongm
ing
Ye
, Bin W
u
.
A Re
view
on In
ducti
on Motor On
li
n
e
F
ault D
i
ag
no
sis.
T
he
T
h
ird Internati
o
n
a
l
Po
w
e
r El
ectron
ics and Moti
on
Contro
l Confer
ence Proc
ee
di
ngs. Beij
ing. 2
000;
135
3 - 13
58.
[9]
Qing
xi
n Z
han
g, Jin Li, Ha
ib
in
Li an
d Ch
on
g L
i
u.
Motor F
ault
Diag
nos
is Bas
ed o
n
W
a
vel
e
t Analys
is an
d
F
a
st F
ourier T
r
ansfor
m
.
Adva
nced Mater
i
als
Rese
arch. 20
1
1
; 301-3
03:1
4
0
1
-14
05.
[10]
Liju
n W
ang, D
ongfe
i
W
ang,
Yong
lia
ng Hu
a
ng. F
ault Diag
nosis for Moto
r Based on E
M
D Algorithm.
Journ
a
l of T
h
e
o
retica
l and A
p
plie
d Infor
m
ati
on T
e
chn
o
l
ogy
.
2012; 44(
2): 265-2
70.
Evaluation Warning : The document was created with Spire.PDF for Python.