TELKOM
NIKA
, Vol. 11, No. 12, Decem
ber 20
13, pp.
7548
~75
5
4
e-ISSN: 2087
-278X
7548
Re
cei
v
ed
Jun
e
28, 2013; Revi
sed Aug
u
st
15, 2013; Accepted Sept
em
ber 2, 201
3
Analysis of Impact Fatigue Life for Valve Leaves in
Small Hermetic Reciprocating Compressors
Dong Zh
ang,
Ming Xu*, Jiang-Ping Gu
, Yue-Jin Hu
ang and Xi Shen
Mecha
n
ica
l
En
gin
eeri
ng Bu
ild
ing, Z
hao
hu
i C
a
mpus, Z
hej
ia
ng Un
iversit
y
o
f
T
e
chnolo
g
y
,
18 Ch
aoW
a
ng
Roa
d
, Han
g
zh
ou 31
00
14, P. R. Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: photoma
n
2
0
04@
126.com
A
b
st
r
a
ct
Impact f
a
tigu
e
l
i
fe of v
a
lve
l
e
a
v
es h
a
s gr
eat
i
n
flue
nce
on
e
n
e
rgy s
a
vin
g
pe
rforma
nce
an
d
lifeti
m
e
of smal
l h
e
r
m
etic rec
i
proc
ati
ng c
o
mpress
o
r
s. This
p
aper
pres
ente
d
a
test system th
at inte
nd
ed t
o
ana
lysis a
nd
eval
uate of i
m
p
a
ct fatigu
e
life of va
lve
leav
es use
d
in s
m
al
l h
e
rmetic r
e
cipr
oc
atin
g
compress
ors. Firstly, an ince
ntive syste
m
w
a
s desig
ned
to simulate r
e
al w
o
rk cond
iti
on for valv
e le
af.
Then, a
d
a
ta
a
c
quisiti
on
system w
a
s b
u
ilt t
o
co
llect
th
e s
oun
d si
gn
al w
h
ile
va
lve
le
af
w
a
s bei
ng
un
d
e
r
test. Sim
u
lt
aneously, the syst
em
c
o
uld control the worki
ng
state of incentive system
s
o
that test could
be
termi
nate
d
aut
omatica
lly onc
e fa
tigue w
a
s detected. Fin
a
l
l
y, fati
gue dete
c
tion system
w
a
s design
ed
to
detect fatig
u
e
of valve
leaf.
Fatigue
dete
c
tion w
a
s th
e
key po
int of
this test system. Fast Four
ier
T
r
ansform (F
F
T
) and W
a
ve
le
t Packet T
r
ans
form (W
PT
) w
e
re a
ppl
ie
d to
ana
ly
z
e
so
un
d
sign
al, b
o
th o
f
which were effective in detec
t
i
ng the dam
age through analy
z
i
ng. Fa
cts showed that
the test system
provi
ded a fe
a
s
ible
appr
oac
h to evalu
a
te i
m
p
a
ct fatigue lif
e for valve l
eaf manufactur
i
n
g
.
Ke
y
w
ords
: Valve leaf, Im
pact Fatigue, Fa
tigue detect, FFT,
Wavelet Packet
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 re
cent yea
r
s, e
nergy sh
ortage
proble
m
is b
e
comi
ng worse
an
d wo
rse. Mo
re and
more p
eople
care abo
ut en
ergy savin
g
p
e
rform
a
n
c
e o
f
house
hold a
pplian
c
e
s
, like refrig
erato
r
s,
air co
ndition
s and so on. Usually, a refrige
r
ator
can
work for mo
re than eight years. With t
he
increme
n
t of
the working
time, the
refri
gerat
in
g p
e
rf
orma
nce
will deg
rad
e
g
r
a
dually [1]. Th
e
main re
ason
is that inlet valve leaf of the
co
mpressor was b
e
ing
fatigued. Th
e small h
e
rm
etic
reci
procating
com
p
re
sso
r
is the
co
re
com
pon
ent
of hou
seh
o
ld
refrig
erator,
whi
c
h h
a
s two
valves, an inlet valve and an exhau
st valve. As wo
rk condition a
n
d
stru
ctur
e of the valve leaves
are
differe
nt, the inl
e
t va
lve leave
s
a
r
e m
o
re
likel
y to be fatig
ue. Thi
s
will
re
sult i
n
th
e
refrig
eratin
g perfo
rman
ce
degradatio
n and mo
re en
ergy co
nsum
ption. If the fatigue proble
m
of
inlet valve le
af ca
n b
e
o
p
t
imized
by va
lve leaf
ma
n
u
factures,
en
ergy saving
perfo
rman
ce
of
refrigerators will be improv
ed obviously.
Duri
ng work con
d
ition, valve leaves are
mainly subje
c
t to bending
stre
sse
s
and
impact
stre
sse
s
[2]. High b
endi
ng
stren
g
th ste
e
l is ad
opted
by valve leaf manufactu
res in o
r
de
r to
prevent be
nd
ing fatigue failure. At present, most
high-qu
ality valve leaf manufactures a
dop
t
s
t
r
i
p s
t
ee
l imp
o
r
te
d fr
om Sw
itz
e
r
l
a
n
d
and
Ja
p
a
n
.
Bu
t
h
o
w
to
c
h
oo
se
s
t
ee
l
w
h
ic
h c
a
n
be
ar
long
time impacting is still a problem, because there
i
s
no
existing equipment to evaluate the impact
fatigue life.
Valve leaf i
s
a lig
ht an
d
thin el
asto
mer,
whe
r
e
i
n
ce
ntive met
hod i
s
not a
v
ailable.
Senso
r
s
can
not be
set
o
n
valve leaf.
In order
to
solve thi
s
p
r
oblem, n
o
n
c
ontact i
n
ce
ntive
system i
s
de
sign
ed, non
contact me
asurem
ent
is a
pplied a
nd d
e
tection m
e
thod
with so
u
n
d
sign
al model
s used is a
d
opted. This
pape
r pres
en
ts an accel
e
rated test system with bot
h
hard
w
a
r
e an
d
software whi
c
h contai
ns
a
non
c
o
n
t
act incentive
system, a
data a
c
qui
siti
on
system, a
co
ntrol mo
dule
use
d
to
control the
wo
rkin
g state
of incentive syste
m
and
a fatig
u
e
detectio
n
system to evaluate the valve leaf’s
impa
ct fatigue life. The non
co
n
t
act ince
ntive
system
simul
a
tes the be
ha
vior of
the valve leaf during
work conditi
on so that the
test has mo
re
pra
c
tical signi
fican
c
e.
Fast Fouri
e
r
Tran
sf
orm method
and Wavel
e
t Packet Tran
sform method
are
ap
plied
to an
alyze
so
und
sig
nal.
Whe
n
fatigu
e
dete
c
tion
sy
stem
dete
c
te
d the
differe
n
c
e
with so
und
si
gnal from g
o
od valve, incentive sy
ste
m
will be terminated by the control mo
dule,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Analysis of I
m
pact Fatigu
e Life for Valve Lea
ve
s in Sm
all Herm
etic… (Do
ng Zh
a
ng)
7549
time will be reco
rde
d
and
a warning will
be create
d
to remind the
operator. Thi
s
test system
is
automatic a
n
d
controllabl
e
.
2. Impact Fatigue Life Tes
t
Sy
stem
Testing th
e valve leaf dire
ctly while
co
mpre
ss
or i
s
workin
g is ve
ry inco
nvenie
n
t. Firstly,
unde
r no
rmal
work conditi
on, valve leaf has a
very long life, usu
a
lly 8 years
or even mo
re.
Secon
d
ly, Structure of
co
mpre
ssor is compa
c
t an
d t
o
tally-enclosed, so it is
very diffic
u
lt to
set
up sensors. Thirdly, noi
se from
mechanical structure will affect
test if acoustical detecti
ng
method is a
p
p
lied [3].
The te
st syst
em is
built to
get impa
ct fatigue
lifetime
of the valve l
eaf und
er
ce
rtain test
con
d
ition. Va
lve leaf is subje
c
ted to
perio
di
cal im
pact. The p
r
i
n
cipl
e is
sim
u
lating the real
behavio
r of valve leaf an
d
detectin
g
the
fatigue of va
l
v
e leaf. Lifetime of sa
me t
y
pe valve leaf
is
affected by t
h
ree fa
cto
r
s:
work freque
n
c
y, im
pact ve
locity whi
c
h
determi
ned b
y
air pre
s
su
re,
and tem
perature [2]. Thi
s
pape
r mai
n
ly con
c
e
r
n
s
a
b
out wo
rk fre
q
uen
cy and
i
m
pact vel
o
cit
y
.
Tempe
r
atu
r
e
factor
can b
e
inclu
ded b
y
just pu
tting
the non
cont
act incentive
system into
a
thermo
stat.
2.1. Hard
w
a
r
e
Archi
t
ec
tu
re
This test
syst
em in
clud
es
a no
nconta
c
t
ince
nt
ive sy
stem
sho
w
n
in
Figure 1. A
fixture i
s
desi
gne
d to fix valve plate and valve lea
f. Compresse
d
air i
s
ado
pted inste
ad of
refrig
era
n
t in
comp
re
ssor
as working m
edium to kee
p
the test saf
e
and conve
n
ient. A high spe
ed solen
o
i
d
valve which
can wo
rk at the spee
d of 0-280
Hz
is
con
t
rolled by function gen
erator to gene
rat
e
pulse ai
r flow. High
sp
eed
sole
noid val
v
e wo
rk
at voltage of
24
VDC, b
u
t fun
c
tion g
ene
rat
o
r
can
not p
r
ovi
de 24V volta
ge pul
se,
so
a solid
stat
e rel
a
y and
a 24V
DC
po
wer source
are
need
ed. Pulse air flow im
pactin
g
valve leaf to simul
a
te openi
ng
and cl
osi
ng o
f
the valve [4].
Wo
rkin
g fre
q
uen
cy can
be
cha
nged
usi
ng fun
c
tion g
enerator. Pre
s
sure
of the a
i
r so
urce
sho
u
ld
be stable. Co
mpre
ssed
air pre
s
sure can be
me
asure
d
with a
p
r
e
s
sure meter
an
d
pressu
re ca
n
be reg
u
lated
with a pre
s
su
re re
gulating
valve so
that the pre
s
sure can b
e
set in
the test. The
sho
r
t pi
pe
conne
ct the
h
i
gh
spe
ed
solenoi
d valve
with th
e fixture
should
b
e
a
s
sho
r
t a
s
possibl
e b
e
cause lo
ng
pipe
will b
e
result i
n
a
tte
nuation
of th
e airflo
w. T
o
red
u
ce n
o
ise
prod
uced by
the system
itself, some
spo
nge
s
a
r
e
put unde
r fixture. And fixture sh
ould
be
stron
g
to avoid resona
nce.
Figure 1. Non
c
onta
c
t Ince
n
t
ive System and Microp
hon
e
Data acqui
sition system co
nt
ains
a high sen
s
itive
mi
cropho
ne
whi
c
h is u
s
ed to
convert
the soun
d p
r
essure
sign
al
to ele
c
tri
c
al
sign
al
when
valve leaf i
s
workin
g. Thi
s
microp
hon
e
has
wide fre
que
n
c
y respon
se,
high se
nsitivity, and
wide d
y
namic range
, lower ba
ckg
r
oun
d noi
se so
that the
sou
nd d
a
ta can
be
colle
cte
d
mo
re a
c
cu
rately which
is
very important to fatigue
detectio
n
sy
stem.
An
indu
strial co
mput
er with Data Acqui
sition Card (DA
Q
card) colle
cts
th
e
electri
c
al
sig
nal. The high
est sam
p
ling
rate is 50
0K
Hz, an
d the resol
u
tion is 1
2
-bit. Beca
use
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 754
8 – 7554
7550
there
are
no
power
su
pp
ly ports
on
the DAQ
ca
rd, a p
o
wer sou
r
ce that
match
e
s th
e
microph
one
i
s
ne
ce
ssa
r
y.
The
mi
crop
hone output port con
n
e
c
ts
to DAQ
ca
rd with
an
I/
O
con
n
e
c
tor b
o
a
rd. To
prote
c
t the mi
cro
p
hone from d
u
s
t, a cove
r
sh
ould b
e
set u
p
. Microph
one
sho
u
ld be
pla
c
ed n
e
a
r
valve leaf to get the be
st
effect
. But do not put the micro
phon
e in fron
t
of valve leaf
because pul
se ai
r flow will
affect the result.
Control mod
u
le contain
s
an isolated o
u
tput
ca
rd in
stalled in
co
mputer
and
a rela
y
board. Isolat
ed output ca
rd can be
co
ntrolled by
compute
r
outp
u
t 1 or 0 state. Relay boa
rd
trigge
red
by i
s
olate
d
o
u
tpu
t
ca
rd
co
ntrol
the
state
of
high
sp
eed
solenoi
d valve
.
This mo
dul
e
make
s the te
st system a
n
automation
system.
2.2. Soft
w
a
r
e
Archi
t
ec
tu
re
Isolated o
u
tp
ut card an
d DAQ card m
anufa
c
ture
r p
r
ovide d
r
ivers and ba
si
c ex
ample
s
for different
develop
envi
r
onm
ents, li
ke, VC,
Delphi, LABVIEW, VB. In this paper, data
acq
u
isitio
n a
nd
cont
rol
module
software i
s
pro
g
ramm
ed
in
graphi
cal
langu
age
na
me
d
LABVIEW developed by National Inst
rument. It’s ea
sy to create a
graphical operation interface
with it. LABVIEW has a great conveni
ence in
developing data acqui
sition
sof
t
ware
and
has
been
widely u
s
ed in me
asu
r
eme
n
t field.
LABVIEW program for this
tes
t
s
y
s
t
em ma
inly
has
three modules
:
data acquis
i
tion
module,
sign
al pro
c
e
s
sin
g
modul
e, control m
odul
e. Data a
c
q
u
isition m
o
d
u
le an
d cont
rol
module
ca
n be program
med ba
se o
n
VI examples
p
r
ovide
d
by Data Acq
u
isition
Ca
rd
an
d
Is
olated
output c
a
rd manufac
t
ure.
LABVIEW has
great advantages
in meas
urement field.
But
in si
gnal
p
r
o
c
essing
field, i
t
s fun
c
tion
is limit
ed. MAT
L
AB is po
we
rful in n
u
me
ri
cal
cal
c
ul
atio
n
and h
a
ve all
kind
s
of too
l
box ap
plied
in all
kind
s
of fields. LA
BVIEW provi
de a VI n
a
m
e
d
MATLAB scri
pt node
to cal
l
MATLAB progra
m
p
r
o
c
e
ssi
ng
sign
al d
a
ta. Thu
s
, sig
nal p
r
o
c
e
ssin
g
module
is pro
g
ramm
ed
wit
h
MATLAB
called by
MAT
L
AB script
no
de. By this
way, this sy
stem
can
com
b
ine
the advantages of LABVIEW and MA
T
L
AB makes it acquire
and process signal
data very well
.
Figure 2. LABVIEW Program of the Tes
t
Sys
t
em
3. Method
Analy
s
is and
Ev
aluate
As no
rmal lif
etime of valve leaf is
usua
lly ov
er eight
years. In
ord
e
r to evalu
a
te impa
ct
fatigue lifetim
e, accel
e
rate
test meth
od i
s
a
dopte
d
. As n
o
tice
d a
b
o
v
e, the impa
ct fatigue life i
s
affected
by t
h
ree
fa
ctors, impa
ct
frequ
ency, im
pa
ct velocity a
n
d
tempe
r
atu
r
e.
Thi
s
in
ce
ntive
system
can
set freque
ncy
with functio
n
gene
rato
r an
d cha
nge im
p
a
ct velocity b
y
changi
ng th
e
air p
r
e
s
sure.
In test, fre
q
u
ency
and
air
pre
s
sure
a
r
e
set mu
ch
hig
her th
an th
ey are
un
der th
e
norm
a
l wo
rk
con
d
ition to a
c
celerate the fatigue pro
c
e
ss [5].
Fatigue
dete
c
tion i
s
the
mo
st imp
o
rtant
tech
nol
ogy in
this
system,
and
sig
nal
pro
c
e
ssi
ng al
gorithm pl
ays a very important role
in d
e
tecting fault
accurately. Duri
ng the la
st
decade
s, ma
ny peopl
e wo
rk
on d
e
tectin
g and
diag
no
sing fa
ults a
n
d
have find
o
u
t som
e
ways
to dete
c
t fault
s
whi
c
h
have
bee
n
widely
use
d
in
me
ch
anical p
r
od
ucts fault di
agn
osin
g, such a
s
fan, engine, air com
p
ressor [6].There
are several
kinds of fault detectio
n
method
s: detecti
on
with limit ch
ecking
or tre
nd che
cki
ng,
detectin
g
wi
th sign
al mo
dels
dete
c
tio
n
with p
r
o
c
e
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Analysis of I
m
pact Fatigu
e Life for Valve Lea
ve
s in Sm
all Herm
etic… (Do
ng Zh
a
ng)
7551
model
s u
s
ed
and d
e
tectio
n
with multivariate data
a
nal
ysis [7]. At prese
n
t, metho
d
s that b
a
sed
on math
emati
c
al
sign
al a
n
d
process m
o
dels to ge
nerate fault sym
p
toms li
ke
sp
ectru
m
a
nalysis
and wavelet analysi
s
are the most freq
uently-u
sed
method
s [8].
So, both Fou
r
ier T
r
an
sform
and Wavelet Tran
sfo
r
m ha
ve been tried
in this syste
m
to analyze th
e sou
nd si
gn
al.
3.1. Discre
t
e
Fourier Tra
n
sform (DFT
) and Fas
t
F
ourier Tran
s
f
orm (FFT
)
The
Di
screte
Fouri
e
r Tran
sform i
s
th
e m
o
st im
porta
nt tran
sform a
n
d ha
s
bee
n
widely
use
d
in many
pra
c
tical
app
lication
s
. In d
i
gital si
gn
al p
r
ocessin
g
, the function i
s
t
r
an
sform tim
e
based
data i
n
to fre
quen
cy based
data
,
su
ch
as th
e
pressu
re of sou
nd wave, electroma
gne
tic
wave. In data
comp
re
ssi
on
, it is used to
filter
usele
ss data to redu
ce the
size of data [9]. The
DFT i
s
also
used to
eff
i
ciently
solve
pa
rtia
l diffe
rential
equ
ations,
and
to
pe
rform
oth
e
r
operation
s
su
ch a
s
co
nvolu
t
ions
or m
u
ltiplying larg
e integers.
The
seq
uen
ce of N-pe
riod
ic is tran
sfo
r
med into
an
N-p
e
ri
odi
c se
quen
ce
of co
mplex
numbe
rs accordin
g to the DFT form
ula:
2
11
00
(
)
()
()
NN
jn
k
jn
k
N
nn
Xk
x
n
e
x
n
e
(
1
)
It describe
s
the Discrete
Fourie
r Tra
n
sform of a
n
N-peri
odi
c sequ
en
ce,
X(k)
is
coeffici
ents of
discrete fre
q
uen
cy comp
o
nents.
This is th
e inverse DF
T (IDFT):
2
11
00
11
(
)
()
()
NN
j
nk
jn
k
N
nn
xn
X
k
e
X
k
e
NN
(2)
This notatio
n use
s
j
for the imagina
ry uni
t,
n
and
k
for indices that ru
n from
0
to
N–1
.
N
is the length
of sample d
a
ta.
Discrete F
o
u
r
ier T
r
a
n
sfo
r
m
is ap
plied
to
analy
z
e
sou
nd d
a
ta
acqu
ired
from
val
v
e leaf
becau
se sou
nd sig
nal dat
a acq
u
ire
d
by compute
r
is t
i
me domai
n, it’s hard to fin
d
the sympto
m
and d
e
tect [4
]. Spectral
an
alysis i
s
th
e
pro
c
e
s
s
of id
entifying co
m
pone
nt freq
u
enci
e
s i
n
sig
nal
data. Fo
r di
screte
data,
the
comp
utational
ba
sis
of
sp
ect
r
al a
n
a
l
ysis i
s
th
e d
i
screte
Fou
r
i
e
r
transfo
rm (DFT).
Th
e DF
T
tran
sforms time
domai
n
data into f
r
eq
uen
cy dom
ai
n data.
With t
h
is
method, th
e
sound
si
gnal
d
a
ta a
r
e
ob
served in f
r
eq
ue
ncy d
o
main
a
nd
cha
r
a
c
teri
stic f
r
eq
uen
cy
can
be fo
un
d. Fre
quen
cy
com
pon
ents ca
n be
an
alyzed. But th
ere i
s
a p
r
o
b
lem
with DFT
method, wh
e
n
pro
c
e
ssi
ng
signal d
a
ta, calculat
ing
mount is too
huge, co
mp
uter can’t do
it
effec
t
ively.
FFT is an efficient metho
d
for computin
g the DFT. Whe
n
FFT al
gorithm is a
p
p
lied,
there i
s
a
differen
c
e
betwe
en the
wind
o
w
len
g
th
an
d
the tran
sform
length. Th
e
wind
ow l
ength
is the length of input data vector. It is determi
n
ed by DAQ ca
rd an
d the way data transfo
rm. As
test system d
e
tecting in re
al time, data trans
l
a
tion an
d cal
c
ulatio
n sho
u
ld be ra
pid. DMA dat
a
transfo
rm me
thod is appli
e
d to transform data at
the best spe
ed. The tran
sform length is the
length of th
e
output, the
computed
DF
T. The
ex
ecu
t
ion time of
a
n
FFT
algo
rit
h
m de
pen
ds
o
n
the tran
sform length. It is most effective wh
en
the transfo
rm length is
a power of
two.
Acco
rdi
ng to Nyqui
st Sampling The
o
re
m, samp
le freque
ncy sho
u
ld be g
r
eate
r
than twi
c
e
as
much a
s
sig
n
a
l freque
ncy. Revolu
tio
n
of freque
ncy sp
ectru
m
is
fs
/N
. After severa
l experiment
s,
in ord
e
r to a
c
hieve the ap
prop
riate
rev
o
lution an
d calcul
ation sp
eed, sa
mple
rate is
set to 10
KHz, and tra
n
sform length
is set as 1
0
0
00. T
he revol
u
tion of frequ
ency is 1
H
z. FFT algo
rith
m
will pad
s or chop
s the inpu
t to achieve the desi
r
e
d
transfo
rm lengt
h. It hardly affects the
spee
d
becau
se FFT
algorithm in
MATLAB is hi
gh efficient.
With FFT al
gorithm a
ppli
ed, valve leaves
are te
sted und
er the
conditio
n
of impact
freque
ncy i
s
100
Hz and
compresse
d
ai
r p
r
e
s
sure
is
0.1MPa. Because it is visu
alize
d
to
cent
e
r
the spec
trogram at
0, ‘ffts
hift’ func
tion in MA
TLAB is used to rearranges
the
output from FFT
with a
circula
r
shift to pro
duce a 0
-
cen
t
ered
spe
c
tro
g
ram. Expe
ri
ment re
sult
s
sho
w
a
s
follo
w
two figure
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 754
8 – 7554
7552
Figure 3. Sound Signal fro
m
Good Valv
e
Leaf and F
r
e
quen
cy Spect
r
um
Figure 4. Sound Signal fro
m
Fatigue Va
lve
Leaf and F
r
e
quen
c
y Spect
r
um
The two figures ab
ove cle
a
rly sho
w
the
di
fference be
tween
signal
from goo
d valve leaf
and
sign
al fro
m
fatigue val
v
e leaf. These two fre
que
ncy spe
c
trum
s are differen
t, and the mo
st
differen
c
e is the amplitude at cha
r
a
c
teristic
frequ
ency. Figu
re
3 sho
w
s th
e amplitude
at
cha
r
a
c
teri
stic frequ
en
cy is 156.9. Fig
u
re 4
sho
w
s th
e amplitud
e
at cha
r
a
c
teri
stic freq
uen
cy
is
128.8. Thu
s
, a value as th
reshold
can
be set, if
the amplitude at
cha
r
a
c
teri
stic freq
uen
cy i
s
greate
r
than t
h
re
shol
d that sho
w
s
the valve leaf under
test is goo
d
. Otherwise, the valve leaf is
fatigue. Program will termin
ate the sole
n
o
id va
lve, record the time a
n
d se
nd a wa
rning.
3.2. Wav
e
let Trans
form a
nd Wav
e
let Packe
t Tran
s
form
Fouri
e
r t
r
an
sform ta
ke
s
a si
gnal i
n
t
he time
do
main a
nd transfo
r
m
s
it i
n
to the
freque
ncy
do
main, whe
r
e t
he Fo
uri
e
r t
r
a
n
sform
result
rep
r
e
s
ent
s
th
e freq
uen
c
y
compon
ents of
the si
gnal.
O
n
ce
the
sign
a
l
is tran
sform
ed into
t
he fre
quen
cy do
ma
in, all info
rma
t
ion ab
out tim
e
will be lost, o
n
ly frequen
cy
remain
s [10]. In contra
st to the Fourie
r transfo
rm, wav
e
let transfo
r
m
enabl
es an
alysis
of data
a
t
multiple leve
ls of
re
solutio
n
. Wh
en
wav
e
let tran
sfo
r
m
is
appli
ed to
a
sign
al in
the
time dom
ain,
the
re
sult i
s
a two
-
dim
e
n
s
ional, time
-scale d
o
main
a
nalysi
s
of th
e
sign
al.
While wavelet
trans
fo
rm pr
o
v
id
es
fl
exib
le time–freq
u
ency resolution, it suffers
from a
relatively low resol
u
tion i
n
the high-freque
ncy re
gi
on. The wa
velet packet
method is
a
gene
rali
zatio
n
of
wavelet
decompo
sitio
n
that offers
a ri
che
r
sign
a
l
analy
s
is [1
1]. It decom
p
o
s
es
the sign
als int
o
different fre
quen
cy ran
g
e
s
and all
o
ws
extraction of f
eature
s
relati
ng to quality.
The
Wavelet
Packet tran
sf
orm h
a
s bee
n proven
very
useful
for
an
alysis
of si
gn
als a
n
d
has
been
su
cce
ssfully ap
pl
ied to dete
c
t machi
ne f
ault
.
The pro
c
e
d
u
re i
s
split the app
roximati
on
coeffici
ent vector into two parts, a ve
ctor of
app
roxi
mation coefficient
s and a
vector of det
ail
coeffici
ents
can be obtai
n, both at
a co
arser
scale. Then next st
ep, usin
g the
same
way a
s
in
approximatio
n vector splitting to decom
pose both th
e vector of a
pproxim
ation
coeffici
ents a
n
d
the vector
of detail co
effici
ents. Thi
s
offers th
e
ri
chest analysis.
Here is
a level-3 decompositi
on
pro
c
ed
ure figure [12].
Figure 5. De
compo
s
ition in
Level 3
0
100
200
300
40
0
500
60
0
700
800
900
1000
-0
.
4
-0
.
2
0
0.
2
0.
4
I
n
pu
t
si
gn
al
A
m
pl
i
t
ud
e
-
5000
-
4000
-
300
0
-
2000
-
1
000
0
100
0
2000
3
000
4000
5000
0
50
100
150
200
F
r
equ
en
c
y
(
H
z
)
A
m
p
lit
u
d
e
C
har
ac
t
e
r
i
s
t
i
c
F
r
eque
nc
y
100
H
z
A
m
pl
i
t
ude 15
6.
9
0
100
20
0
300
40
0
500
600
700
800
90
0
1000
-0
.
4
-0
.
2
0
0.
2
0.
4
I
n
pu
t
s
i
gn
a
l
A
m
pl
i
t
ude
-5
0
0
0
-4
0
0
0
-3
0
0
0
-2
0
0
0
-1
0
0
0
0
1
000
20
00
30
00
40
00
50
00
0
50
10
0
15
0
F
r
e
q
ue
nc
y
(
H
z
)
Am
p
l
i
t
u
d
e
Ch
ar
ac
t
e
r
i
s
t
i
c
F
r
eque
nc
y
100
Hz
A
m
pl
i
t
ude 128.
8
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Analysis of I
m
pact Fatigu
e Life for Valve Lea
ve
s in Sm
all Herm
etic… (Do
ng Zh
a
ng)
7553
Con
s
id
erin
g the cal
c
ul
atio
n amount
will incr
ease ra
p
i
dly with the decompo
sitio
n
level
goe
s high
er, this pa
per
cho
s
e level-3 de
comp
ositio
n to ensure thi
s
test system runs flue
ntly.
After the d
e
compo
s
ition, t
here
will
be
eight
coeffici
ents, whi
c
h can
b
e
cal
c
ula
t
ed
the
band
ene
rgy
of every frequ
ency
band
an
d con
s
tru
c
t a
feature ve
cto
r
with th
at. Fro
m
co
mpa
r
in
g
the feature vector of soun
d signal
with feature ve
ctor of sound
si
g
nal acq
u
ired from goo
d valve
leaf, the state of valve leaf
durin
g test ca
n be estimate
d [13].
Followi
ng two
figures
sho
w
result
s of usi
ng
wavelet p
a
cket tran
sform. Apparentl
y
, the
feature ve
cto
r
is
differe
nt in vecto
r
com
pone
nts’
m
a
g
n
itude whi
c
h mean
s
the b
and
frequ
en
cy
energy is
different. Thi
s
can
be th
e
basi
s
fo
r
jud
g
ing
state of
valve leaf
comp
ari
ng th
e
differen
c
e
with the soun
d
from a g
ood
valve leaf [14]. A thresh
old is
set, if the differen
c
e
excee
d
thre
shold, the syst
em w
ill re
min
d
operator th
e valve leaf
is bro
k
en. The
n
, the lifetime of
the valve leaf is re
corded.
Figure 6. Sound Signal fro
m
Good Valv
e Leaf
and Featu
r
e
Vector
Figure 7. Sound Signal fro
m
Fatigue Va
lve
Leaf and Fe
a
t
ure Vecto
r
4. Conclusio
n
In orde
r to a
nalyze a
nd e
v
aluate the i
m
pact fatigu
e life of valve leaf used i
n
small
herm
e
tic re
ci
pro
c
ating
co
mpre
ssors, a
n
evaluat
ion
system ba
se
d on real wo
rk conditio
n
of
valve leaf i
s
desi
gne
d, wh
ich i
s
mad
e
u
p
of
a
n
u
mb
e
r
of
devices i
n
clu
de
high
speed
solen
o
i
d
valve, fixture, function g
e
nerato
r
, com
p
re
ssed
ai
r sou
r
ce
to si
mulate
the o
pen
a
nd clo
s
e
behavio
r of
valve leaf. So
und
sig
nal
da
ta is
ac
qui
red
by u
s
ing
a
sensitive mi
cropho
ne, p
o
wer
source for mi
crophone, dat
a acqui
sition
card
(DAQ card) and LABVIEW
program
.
Sound signal
data is
proc
es
sed
with MA
TLAB program whic
h is
c
a
lled by MATLAB s
c
ript node in LABVIEW.
To dete
c
t the
fatigue valve leaf, both FF
T tran
sf
orm
a
nd Wavelet Packet tran
sfo
r
m are
prove
n
feasibl
e
. But
in fact, soun
d pre
s
sure
si
gnal is
un
sta
b
le, beca
u
se
amplitude of
characte
ri
stic
freque
ncy i
s
sen
s
itive with
the pre
s
su
re
of
comp
re
ssed air
so
urce
. Fatigue det
ecting
with F
FT
may be less
reliabl
e. Mea
n
whil
e, fatigue detectin
g
with Wavel
e
t Packet Tran
sform b
a
se o
n
band
en
ergy
ha
s a
hig
h
e
r
accu
ra
cy. Thi
s
te
st system p
r
ovid
es
a
way f
o
r valve
lea
f
manufa
c
tures to estimate impact fa
tigue
life and has
pra
c
tical valu
e.
Ackn
o
w
l
e
dg
ements
This
study i
s
partially
su
p
ported
by th
e fi
nan
cial
suppo
rt from
t
he
Nation
al
Nature
Scien
c
e Fou
ndation of Chin
a unde
r grant No
. 5107
6143. T
he autho
rs
also g
r
atefull
y
ackno
w
le
dge
the helpful
comment
s an
d su
gge
stio
n
s
of the
reviewe
r
s,
whi
c
h
have imp
r
ov
ed
the pre
s
entati
on.
Referen
ces
[1]
Ku B, Park J, H
w
ang Y, et al.
Perfor
ma
n
c
e eval
uati
on
of the en
ergy efficiency of
Crank-Driv
e
n
compress
or a
nd l
i
ne
ar co
mpr
e
ssor for
a ho
use
hol
d refrig
erator
.
Internatio
na
l
Compr
e
ssor
Engi
neer
in
g C
onfere
n
ce. Pur
due. 20
10: 1-8.
0
100
20
0
300
40
0
50
0
60
0
70
0
80
0
90
0
1000
-0.
4
-0.
2
0
0.
2
0.
4
I
n
pu
t
si
gn
al
A
m
p
lit
u
d
e
1
2
3
4
5
6
7
8
0
1
2
3
F
e
at
u
r
e V
e
c
t
or
A
m
p
lit
u
d
e
0
10
0
20
0
300
40
0
50
0
600
700
80
0
900
100
0
-0
.4
-0
.2
0
0.
2
0.
4
I
n
pu
t
si
g
n
al
A
m
p
lit
u
d
e
1
2
3
4
5
6
7
8
0
1
2
3
F
e
at
u
r
e V
e
c
t
or
Am
p
l
i
t
u
d
e
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 754
8 – 7554
7554
[2]
Altunl
u A C, Ismail L, Emre O, et al.
Impact fatigu
e char
acteristics of
valv
e leav
es for sma
ll h
e
rmeti
c
reciproc
atin
g c
o
mpress
ors
. 2
0
th Intern
ation
a
l C
o
mpress
or
Engi
ne
erin
g C
onfere
n
ce. W
e
st Lafa
y
ette,
India
na, USA. 201
0: 1-8.
[3]
Michel
e L, And
r
ea C.
Imp
a
ct fatigu
e on sucti
on valv
e ree
d
: new
experi
m
en
tal appr
oac
h
. International
Compress
or E
ngi
neer
in
g Con
f
erence. W
e
st Lafa
y
ette. 20
0
4
: 1-7.
[4]
Altunl
u A C, L
a
zog
l
u I, Oguz
E, et al. An i
n
vest
igati
on on the
imp
a
ct
fatigue
c
har
acteri
stics of valv
e
leav
es for sm
all
hermetic r
e
ciproc
ating c
o
mp
ressors i
n
a ne
w
a
u
toma
ted test s
y
ste
m
.
F
a
tigue &
F
r
acture of Engin
eeri
ng Mate
rials & ST
RU
. 201
2; 35(9): 82
6-84
1.
[5]
Martins JA, Leite DL, Martins
LG
. Analysis o
f
set of valve from
her
metic r
e
cipr
ocatin
g co
mpr
e
ssors
und
er accel
e
r
a
ted Life test (alt) by altern
ativ
e
machi
n
e
in re
plac
e
m
e
n
t of th
e
usua
l co
mpresso
r
benc
h test
. Internati
ona
l Com
p
ressor En
gin
e
e
rin
g
Confer
en
ce. Purdue. 2
0
08: 1-10.
[6]
Xu
X, W
a
n
g
S
,
Liu J, et al. F
an F
ault Di
a
gnos
is Base
d
on W
a
vel
e
t Pa
cket and Sam
p
le Entr
op
y.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(6): 3
451-
346
2.
[7]
Isermann
R. F
ault-d
i
ag
nos
is
s
y
stems: a
n
i
n
troducti
o
n
fro
m
fault d
e
tecti
on to fa
ult to
le
rance. N
e
w
York: Spring
er, 2006.
[8]
Li H, Yin Y. Bearin
g F
aul
t Diagn
osis B
a
sed o
n
La
pl
ace W
a
vel
e
t T
r
ansform.
TE
LKOMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(8): 2
139-
215
0.
[9]
Oppen
he
im A
V
, W
illsk
y
AS, Hamid
NS. Si
g
nals
an
d s
y
ste
m
s 2. Prentic
e
-
Hall
Eng
l
e
w
o
o
d
Cl
iffs, NJ.
198
3.
[10]
Burrus CS, Gopi
nath RA, Guo Ha
i-tao, et
al. Introductio
n
to
w
a
ve
lets and
w
a
ve
let transforms:
a
primer, 23. Pre
n
tice ha
ll Up
pe
r Saddl
e River.
1998.
[11]
GAO Robert-
x. W
a
velets: th
e
o
r
y
an
d
app
lic
ations
for ma
n
u
facturin
g.
Spr
i
ng
er Sci
ence+
Busin
e
ss
Medi
a.
201
1: 69-80.
[12]
Z
a
rei J, Posht
an J. Be
arin
g f
ault d
e
tectio
n
usin
g
w
a
v
e
l
e
t packet transf
o
r
m
of ind
u
ctio
n
motor stator
current.
T
r
ibol
o
g
y Internatio
na
l
. 2007; 4
0
(5): 763-
769.
[13]
Hasa
n O, Lo
p
a
ro K
A, Disc
enzo
F
M. Onlin
e
track
i
ng
of be
arin
g
w
e
ar usi
n
g
w
a
v
e
let p
a
cket
decom
positi
on and prob
ab
ilisti
c
model
in
g:
A method for
be
arin
g pro
g
n
o
sti
cs.
Journa
l of
Soun
d a
n
d
Vibrati
o
n
. 20
07
; 302(4): 95
1-9
61.
[14]
Sun Z
,
Chan
g
CC. Stru
ctural dama
ge ass
e
ssment base
d
on
w
a
v
e
let p
a
cket transform.
Journa
l of
Structural Eng
i
neer
ing
. 2
002;
128(
10): 13
54-
136
1.
Evaluation Warning : The document was created with Spire.PDF for Python.