TELKOM
NIKA
, Vol.12, No
.3, Septembe
r 2014, pp. 5
33~540
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i3.76
533
Re
cei
v
ed Fe
brua
ry 23, 20
14; Re
vised
May 29, 20
14
; Accepte
d
Ju
ne 12, 201
4
Failure Mechan
ism Analysis and Failure Number
Prediction of Wind Turbine Blades
Yu Chun-y
u
, Guo Jian-
y
ing, Xin Shi-guang
High
e
r Educ
ati
ona
l Ke
y
La
bor
ator
y
for Me
as
urin
g
& Control
T
e
chnol
og
y a
nd Instrument
a
t
ion of
Heil
on
gji
a
n
g
Provinc
e
Harbi
n
Univ
er
sit
y
of Scie
nce
and T
e
chno
log
y
, Chi
n
a
e-mail: h
one
yc
hun
yu
@1
26.co
m
A
b
st
r
a
ct
Pertinent to th
e pro
b
le
ms
th
at w
i
nd turb
ine
bl
a
des
op
erat
e in c
o
mp
licat
ed co
nd
itions,
freque
nt
failur
e
s a
nd l
o
w
replace
m
ent
rate as w
e
l
l
as
ration
al
i
n
ve
nto
r
y nee
d, this
p
aper, w
e
b
u
il
d
a fault tre
e
mo
de
l
base
d
on i
n
-d
epth a
nalys
is of the failur
e
causes. As
the
mec
h
a
n
ica
l
vib
r
ation of the w
i
nd turb
in
e take
s
plac
e first on the bl
ad
es, the pap
er gives
a detai
led
ana
lys
i
s to the F
a
ilur
e
mec
h
a
n
is
m
of blad
e vibr
ati
on.
T
herefore th
e
pap
er puts for
w
ard a
dyn
a
m
i
c
predicti
on
mode
l of w
i
nd tu
rbin
e bl
ade fa
il
ure n
u
mber
ba
sed
on the
grey the
o
ry. T
he relativ
e
error b
e
tw
ee
n its pred
ictio
n
and th
e fiel
d in
vestigati
on d
a
ta is less th
an 5
%
,
me
etin
g the ac
tual ne
eds
of engi
neer
in
g an
d
verifying th
e e
ffect
iveness a
n
d
app
lic
abi
lity of the prop
ose
d
alg
o
rith
m. It is
of i
m
porta
nt e
n
g
in
eeri
ng s
i
gn
ificanc
e for
it to
provi
de
a the
o
r
e
tical fo
un
dati
o
n for the
fail
ure
ana
lysis, failur
e
researc
h
an
d
inventory l
e
vel
of w
i
nd turbine
blad
es.
Ke
y
w
ords
: w
i
nd turbi
ne, bl
a
des
,
fault nu
mber
,
grey
mod
e
l
,
failure
mec
h
a
n
is
m
1. Introduc
tion
As the
criti
c
al co
mpo
nen
t, the blade
plays a
n
imp
o
rtant
role i
n
wind
turbi
n
e
s
. As it
su
stain
s
com
p
reh
e
n
s
ive effect of centrif
ugal forc
e, fluid powe
r
, vibration, temperature differen
c
e
(therm
a
l stress) and medi
a,
it is fault-prone [1],[2].
With a great latitudinal expand, Chi
na
has
low tempe
r
at
ure in the no
rth, typhoons in the s
outh
,
wind-b
o
rn
e san
d
in the north
we
st, all of
whi
c
h a
r
e th
e prin
cip
a
l cause of freq
uent failure
s.
Acco
rdi
ng to JB/T 10
19
4-20
00, the l
i
mit
desi
gn temp
eratu
r
e of bl
ade
s ran
ge from -3
0 to +50. Howe
ver, in the northe
r
n area,
the
minimum te
mperature i
s
below
-3
0; more
over,
mo
st of the
wind tu
rbin
es use
d
in the
wind
farms in th
e early sta
ge a
r
e of foreig
n tech
nol
o
g
y which m
a
y not be entirely su
itable to Chin
a’s
natural
environment, e
s
p
e
cially the l
o
w-tem
perat
ure enviro
n
me
nt. The ind
e
termin
ate vibration
gene
rated in
stall wind turbine blad
es b
y
low
temperature will da
mage the bla
de stru
ctu
r
e an
d
thus
affect th
e turbine’
s
no
rmal
ope
ratio
n
[3]. Mo
st a
c
cide
nts
due
t
o
bla
d
e
faults are di
sa
strou
s
,
for they will cau
s
e te
rribl
e loss, and
have strong
impact on tu
rbine’
s e
c
on
omical a
nd safe
operation. Fi
eld data
sho
w
s th
at
mechani
cal vibration occu
rs first on bla
d
e
s
[4]. Accordin
gly,
analysi
s
shall
be given to
the vibration
Failure me
ch
anism
so
a
s
to put forward pro
p
o
s
al
s fo
r
avoiding or minimizi
ng
vi
bration. Wh
e
n
the
bl
ade
s
failed, ho
w to
determi
ne th
e blad
e’s failure
quantity in future by
simple
method
s an
d highe
r
a
c
cu
racy, where reasona
ble in
ventory sh
all be
kept, no excessive fund b
e
taken up, requireme
nt
s of recove
ry service b
e
me
t, operation cost
be minimi
ze
d
and
non
-pla
nned
sh
utdo
wn time
re
du
c
ed. T
h
is is
a challen
g
ing
issue
both
at
home an
d ab
road in
reg
a
rd of rea
s
ona
b
l
e reserve q
u
antity of criminal com
pon
e
n
ts.
No
w rese
arches
on bl
ade
Failure me
ch
anism
analy
s
i
s
,
predi
ction of
failure num
ber and
predi
ction of
spa
r
e p
a
rts
d
e
mand
are le
ss [5]. The
p
r
edictio
n of failure n
u
mbe
r
mostly emplo
y
s
traditional
statistics and neural
network t
heory [6
],[7].
However, i
n
t
he ma
i
n
tenance support
field
of
wind t
u
r
b
i
ne sy
st
e
m
s,
cat
a
st
rop
h
ic f
a
ilure t
o
criti
c
al components is usually a small
sam
p
le
and po
or info
rmation. Thu
s
, there are in
here
n
t defect
s
in su
ch met
hod
s.
Based o
n
the
detailed an
al
ysis of failure
fa
ctors of wi
nd turbin
e bl
ade
s, the pa
per
sets
up a fault tre
e
model of
wind turbi
ne bl
ade. Ma
king
use of the fa
ult tree, the pape
r ma
ke
s a
detailed logi
c analysis of the failure causes
an
d p
r
obe
s the vibration Fail
ure mech
ani
sm of
blade
s. The f
a
ilure
numb
e
r
of win
d
turb
ine bla
des
ca
n be rega
rd
e
d
as time
se
ries, so that the
grey alg
o
rith
m req
u
ire
s
n
o
ma
ss
data,
neither pr
ed
etermin
a
tion
of informatio
n ch
ara
c
te
rist
ics.
The pa
pe
r ad
vance
s
a m
e
thod of p
r
edi
ct
ing the failu
re
numbe
r of wi
nd turbi
ne bl
a
des
ba
sed o
n
dynamic
gre
y
model, pre
d
icting th
e fa
ilure n
u
mb
er of wind tu
rb
ine bla
d
e
s
b
y
use of
sm
all-
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 53
3 – 540
534
sampl
e
fault
data a
nd v
e
rifying the
prop
osed
al
gorithm i
n
a
c
cord
an
ce
with literature
[8]
measured
dat
a. The
re
sult
has a
4.7% relative er
ro
r
with the
mea
s
ured
data,
meeting th
e a
c
tua
l
need
s of en
ginee
ring, ve
rifying the effectivene
ss o
f
the propo
sed algo
rithm
and solvin
g
the
probl
em in predictio
n of failure nu
mbe
r
with par
tially
kno
w
n a
nd p
a
rtially unkno
wn informatio
n.
2. Blade Fau
l
t Tree Mode
l
A large
-
si
ze
d wind tu
rbi
ne syste
m
is
mainly co
mposed of
wind
whe
e
l, gearbox,
gene
rato
r, yaw sy
stem, pitch
system, b
r
akin
g sy
st
em
, lubrication
system, elect
r
i
c
al
system a
nd
frequency converter in parallel [9
],[10],
as shown in
Figure 1.
Figure 1. Reli
ability Logic
Block
Diag
ra
m of Wind Tu
rbine
Acco
rdi
ng to the statisti
cal i
n
formatio
n of bl
ade fault in
wind farms, the fault tree shall be
stru
ctured, as sho
w
n in Fig
u
re 2.
Figure 2. Blade Fault Tree
In the figu
re
, X indicates the bla
de f
ault of top
-
tree event;
X
1
is
th
e b
l
ade
mas
s
unbal
an
ce of
interme
d
iate
event,
X
2
th
e aerodynam
ic un
balan
ce,
X
3
the blad
e crack d
a
m
age,
X
4
other fault
s
;
X
11
oil l
e
a
k
of
pitch
co
ntrol valve;
X
12
freezi
ng
corro
s
io
n di
rt
;
X
21
wind
vane
inac
cu
ra
cy
;
X
22
blade co
rrosio
n stain;
X
23
control
er
ror;
X
24
bla
d
e setting an
g
l
e erro
r;
X
31
bad
pro
c
e
ssi
ng;
X
32
material
aging;
X
33
lig
htning;
X
34
i
m
pro
per tra
n
s
po
rtation
an
d in
stallation;
X
41
poor fit;
X
42
u
n
kn
own failure, and
X
43
low perfo
rma
n
ce.
The intermed
iate failure ev
ent
X
1
come
s from the logi
cal combin
ation of bottom events
X
11
and
X
12
, t
hat is
:
11
1
1
2
X
XX
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Failure Mech
anism
Analysis and Fail
ure
Num
ber Pre
d
iction of Wi
n
d
Turbi
ne …. (Yu Ch
un-yu
)
535
The interm
edi
ate failure ev
ent
X
2
comes from the logi
cal combin
ation of bottom events
X
21
,
X
22
,
X
23
and
X
24
, that i
s
:
22
1
2
2
2
3
2
4
++
X
XX
X
X
(2)
The interm
edi
ate failure ev
ent
X
3
comes from the logi
cal combin
ation of bottom events
X
31
,
X
32
,
X
33
a
nd
X
34
, that is
:
33
1
3
2
3
3
3
4
++
X
XX
X
X
(3)
The interm
edi
ate failure ev
ent
X
4
comes from the logi
cal combin
ation of bottom events
X
41
,
X
42
and
X
43
, that is
:
44
1
4
2
4
3
+
X
XX
X
(4)
3. Blade Vibration Failure
Mechanism
Analy
s
is
The
stre
ss
o
n
the
win
d
t
u
rbin
e bl
ade
s in
rotation
can
be
simpl
i
fied a
s
aero
d
ynamic
force,
centrifu
gal force and
grav
ity, expre
s
sed a
s
form
ula (5
), (6) a
n
d (7)
re
spe
c
tively.
2
2
1
=s
i
n
c
o
s
2
1
=c
o
s
s
i
n
2
XS
A
l
d
YS
A
l
d
FW
c
C
C
FW
c
C
C
(5)
In the formul
a,
is the air den
sity; W incomin
g
win
d
velocity; c chord le
ngth;
l
C
lift
c
oeffic
i
ent;
angle of
atta
ck;
d
C
coefficie
n
t of d
r
ag;
X
SA
F
ae
rodynami
c
fo
rce i
n
X
dire
ction;
YS
A
F
the aerodyn
a
m
ic force in Y
directio
n.
()
c
o
s
c
o
s
(
)
sin
c
os
sin
()
s
i
n
c
o
s
c
o
s
XS
G
i
YS
G
i
ZS
G
i
Fm
r
g
t
Fm
r
g
t
Fm
r
g
t
(6)
In the formul
a,
()
i
mr
is the co
nce
n
trated m
a
ss of blade
element
s at r;
t
the blad
e
azimuth of
rotation;
the axial inclination;
the pro
peller
pitch
angle;
X
SG
F
gravity in X
dire
ction;
YS
G
F
the
gravity in Y directio
n;
Z
SG
F
the gravity in Z direction.
R
2
r
R
2
r
=(
)
c
o
s
c
o
s
=(
)
s
i
n
c
o
s
XR
P
i
YR
P
i
Fm
r
g
r
d
r
t
Fm
r
g
r
d
r
t
(7)
In the formul
a,
X
RP
F
is the cen
t
rifugal force i
n
X directio
n;
YR
P
F
the centrifug
a
l force in
Y directio
n.
Under the action of the three forces
above,
the blade will mainly develop flap,
shimmy
and torsio
n. These thre
e mechani
ca
l vibration
s
join the ae
rodynami
c
fo
rce to
prod
uce
aero
e
la
stic problem
s. If the
i
r intera
ction i
s
wa
keni
ng, the vibration i
s
steady, or
more de
struct
ive
flutter and
ra
diation
will o
c
cur. T
he
blad
e’s a
e
ro
el
a
s
tic p
r
oble
m
s
m
a
inly involve “stall flutter” a
nd
“c
lass
ic
flutter”.
In “stall flutter”, blades will
produce “li
m
it cycle oscill
ation”
, whi
c
h will,
characteristic
of
steady vibration, and la
rge
and con
s
tant
amplitude, ul
ti
mately resul
t
in “self-limit
ed oscillatio
n
”. If
the bla
de
ha
s a
deq
uate t
o
rsi
onal
flexibility, it
will
prod
uce
“torsion
al dive
rg
ence” or “no
n
-
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 53
3 – 540
536
vibration
failu
re”,
nam
ely b
u
ckling
failure o
r
to
rs
i
onal
failure. Su
ch
flutter
can
b
e
an
alyze
d
b
y
nonlin
ear sta
b
ility
equation
s
.
X
Y
W
Figure 3. Blade Airflow An
gle of
Attack-Stalling Cha
r
acteri
stics
The
blade
section
is sho
w
n
as Fig
u
re
3.
Whe
n
ai
rflow
com
e
s from
W direct
ion, the
angle of atta
ck i
s
. When
is great, the airflow
will separat
e in flo
w
ing a
c
ro
ss t
he blad
es,
developin
g
th
e “stall
” call
e
d
in aero
d
yna
m
ics.
The blad
e wi
ll produ
ce a
lift force F
l
under the a
c
ti
on of wind
current, as sh
own in
formula (8).
2
1
=
2
ll
F
CS
v
(8)
In the formul
a,
v
is the inflo
w
co
ndition a
t
a finite distance; an
d S is
the blade
are
a
. The
relation b
e
tween
l
C
and the angle of attack
is as
sho
w
n in Figure 4.
Figure 4. Lift
Coeffici
ent of Blade
In the figure,
0
is the critical
angle of attack. From
the formula (8), the lift force of the
blade
is rel
a
ted to th
e a
n
g
le of
attack. Wh
en
0
, the li
ft increa
ses
with the
in
creasi
ng
;
whe
n
0
, the lift decrea
s
e
s
with the in
creasi
ng
. In addition, when the tip of the blade
make
s b
endi
ng motion u
p
w
ards
at a speed
relative
to the root, the variation
of the angle
of
attack will
change the lift. If
0
, the lift will
decrease, hind
ering
the blade tip bending
upward
s
; if
0
, the lift increa
ses, p
r
omotin
g
the bla
de tip
to ben
d up
ward
s a
nd a
g
g
r
avating
the blad
e vib
r
ation
so m
u
ch that the
bl
ade
will cra
c
k or ru
pture i
n
a so short
time (ten
s of
or
dozen
s of se
con
d
s).
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Failure Mech
anism
Analysis and Fail
ure
Num
ber Pre
d
iction of Wi
n
d
Turbi
ne …. (Yu Ch
un-yu
)
537
“Classic flutt
e
r”, occurring in
potential
flow, is a
sel
f
-su
s
taine
d
u
n
stea
dy oscil
l
ation. It
can result in sud
den o
s
cillation of blade
s, whi
c
h will i
n
crea
se to the extent of damage.
If the blade d
e
viates up
wa
rds f
r
om the
equilib
ri
um p
o
sition, the el
astic
re
storin
g force
will d
r
ag
it to
the e
quilib
ri
um po
sition
and th
us
an
inertia
fo
rce acts
on
the blade’
s cente
r
of
gravity. The t
o
rque of the i
nertia force t
o
the to
rsion center will reduce
the
angl
e of attack of
the
blade, and th
us produ
cin
g
an addition
al aero
d
ynami
c
force to spee
d up the blad
e’s move to the
equilibrium position. During the course, the aer
ody
namic force is an exciting force, which is
dire
ctly prop
ortional to th
e blade’
s rot
a
tion rate, a
nd inversely
propo
rtion
a
l
to the blade’s
dampin
g
force.
Therefore, th
e most effect
ive way to prev
ent blad
e
fluttering is
to move the blade’
s
cente
r
fo
rwards to
re
du
ce
the in
ertia
moment.
With field i
n
vest
igation, oth
e
r failure
s
ca
n
be
eliminated by
such mea
s
ure
s
as regu
lar inspe
c
tion
, regular m
a
i
n
tenan
ce, re
gular
clea
nin
g
,
optimizatio
n
of co
ntrol
poli
c
y an
d a
r
ith
m
etic m
e
thod
and
adju
s
tm
ent of
setting
angl
e. The
knife
scars or da
mage on th
e comp
one
n
t
surface ca
n be treate
d
by finish machi
n
ing,
with
con
s
id
eratio
n
of material st
rength a
nd te
chni
que
s.
4. Gre
y
Algo
rithm
The g
r
ey
alg
o
rithm
reg
a
rd
s rand
om q
u
antities
as grey qua
ntities,
so
that it ne
eds not
study the probability distri
bution re
gularities in data processing, but
lays particul
ar em
phasis
on
the se
arch of
laws bet
wee
n
data. Th
ro
ugh d
a
ta
pro
c
e
ssi
ng, ne
w data will
be
prod
uced, a
nd
implicit laws
of initial data will be
figured out accordi
n
gly [11],[12].
4.1 GM (1,1) Model
Given that
(0)
X
is the origin
al serie
s
,
(0)
(
0)
(0)
(
0)
=(
1
)
,
(
2
)
,
,
(
)
X
xx
x
n
(9)
For the conve
n
ien
c
e of mo
del buildin
g, a new
seri
es
(0)
Y
will be generated from every
seri
es by formula (1
0).
(0
)
(0
)
(0
)
()
()
(1)
x
i
yi
x
(10
)
In the formula,
1,
2
,
,
in
.
(0)
(
0
)
(
0
)
(
0)
=(
1
)
,
(
2
)
,
,
(
)
Yy
y
y
n
(11
)
A first accu
mulation giv
en to
(0)
Y
, the s
e
ries
(1
)
Y
of failure will b
e
gene
rated from
formula (12
)
.
(1
)
(
0
)
0
()
(
)
k
i
yk
y
i
,
1,
2
,
,
kn
(
12)
(1
)
(
1
)
(1
)
(
1
)
=(
1
)
,
(
2
)
,
,
(
)
Yy
y
y
n
(13)
The background value
will be
structured from
(1
)
Y
, that is
:
(1
)
(
1
)
(1
)
(
1
)
=(
2
)
,
(
3
)
,
,
(
)
Z
zz
z
n
(14
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 53
3 – 540
538
In the formula,
(1)
()
zk
can b
e
obta
i
ned from formula (1
1).
(1
)
(
1
)
(1
)
()
(
1
)
(
1
)
(
)
2
,
3
,
,
zk
a
y
k
a
yk
k
n
,
(1
5)
The co
rrespo
nding
whiteni
zation e
quati
on is a
s
follo
ws:
(1
)
(1
)
dY
aY
b
dt
(16)
In the formula
,
a is the developin
g
coeffi
cient,
wh
ose effective interval is (-2, 1
)
; b is the
grey varia
b
le,
which ca
n be
obtained fro
m
formula (1
7).
1
*
ˆ
,
T
TT
n
aa
b
B
B
B
Y
(1
7)
In the formula,
*
n
Y
and B are o
b
tained fro
m
formul
as (18
)
and (1
9) resp
ectively.
*
(
0)
(
0
)
(
0)
(
2
),
(3
)
,
,
(
)
T
n
Yy
y
y
n
(18)
(1
)
(
1
)
(1
)
(
1
)
(1
)
(
1
)
(1
)
(
2
)
1
2
(2
)
(
3
)
1
2
(1
)
(
)
1
2
yy
yy
B
yn
yn
(1
9)
The sol
u
tion
of the differen
t
ial equation (16) is:
(1
)
(
0
)
ˆ
(1
)
(
(
1
)
)
ak
bb
yk
y
e
aa
(20)
In the formula,
1,
2
,
,
kn
.
(1
)
ˆ
(1
)
yk
, with inverse
accumul
a
tion
, can be redu
ced to be a
s
follows:
(0)
(
1
)
(
1
)
ˆˆ
(1
)
(
1
)
y
(
)
yk
y
k
k
(21)
4.2 D
y
namic
Gre
y
Prediction Model
To pre
d
ict t
he failure nu
mber n
eed
s to build a predi
ction m
odel a
c
cordi
ng to the
contin
uou
sly
obtaine
d failu
re
numb
e
r, t
hat is,
the
l
a
test
d
a
ta se
rves as
referen
c
e point.
T
h
u
s
,
the new mo
d
e
l in the topol
ogy model
s is us
ed to ap
proximate to the actual vale.
If as many as k data are u
s
ed to build a
basi
c
predi
ction model, the
k+1 data a
n
d
above
will be predi
cted. When the k+1 data is predicted, it
shall
be com
pared
with the actual val
u
e to
deci
de
wheth
e
r it is true
o
r
false. If it’s true, it
sh
all
be ad
ded i
n
to the o
r
igin
al
se
rie
s
to form a
new se
rie
s
compo
s
ed
of k+1
data; an
d
then
th
e
k+1 data i
s
u
s
e
d
to buil
d
a
p
r
edi
ction m
o
d
e
l,
prod
uci
ng a n
e
w dynami
c
model.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Failure Mech
anism
Analysis and Fail
ure
Num
ber Pre
d
iction of Wi
n
d
Turbi
ne …. (Yu Ch
un-yu
)
539
5. Engineering Applica
t
ion Example
The pap
er ta
ke
s the fault data in Table
4-13 in the lit
eratu
r
e [8] as original d
a
ta, which is
now tab
u
late
d in Table 1, to predi
ct the failure
n
u
mbe
r
of blades in t
he next year. (cal
cul
a
ted b
y
30 days a mo
nth)
Tabel 1. Cu
m
u
lative failure
time of blades (day
s)
i
1 2
3 4 5 6 7 8
i
t
753
800
902
924
1047
1109
1151
1153
The predi
ctio
n results a
r
e tabulate
d
in Table 2.
Tabel 2. Fo
re
ca
sting re
sult
s
Future inte
rval (d
a
y
)
90
180
360
Literature (piece)
3.02
6.46
14.33
The pape
r (piece
)
2.5019
6.4333
12.5695
Actual (piece)
2
6
12
The vari
an
ce
of the actu
al
data is
0.5, i
ndicating a
hi
gh di
spe
r
si
on
degree; the
resid
ual
varian
ce is 0.
2323
2572, in
dicatin
g
a low dispe
r
si
on d
egre
e
. Their
spe
c
ific value
is C=0.4646
51,
indicating tha
t
the difference between t
he cal
c
ulat
e
d
value from the model an
d the actual value
is not so di
screte though th
e or
igin
al dat
a is very discrete.
6. Conclusio
n
s
With the win
d
turbine bl
a
des a
s
the obje
c
t and the blade’
s failure n
u
mbe
r
as th
e
conte
n
ts, the
paper b
u
ild
s a fault tree model for
wi
nd turbin
e bl
ade
s, analyzes the vibrati
o
n
fa
ilu
r
e
mec
han
is
m a
n
d
c
ons
tr
uc
ts
a
d
y
na
mic
gr
e
y
pre
d
iction mo
del
of wind turbi
ne blad
e failu
re
numbe
r b
a
se
d on
small
sa
mple data. T
he rel
a
tive
error
of the pre
d
iction
re
sults by the dyna
mic
grey p
r
edi
ctio
n model
of wi
nd turbi
ne bl
ade failu
re
n
u
mbe
r
sho
w
s that the mod
e
is of exten
s
ive
engin
eeri
ng
appli
c
ation v
a
lue, providi
ng a theo
re
t
i
cal foun
dati
on for the
rese
rve of cri
t
ica
l
comp
one
nts.
Ackn
o
w
l
e
dg
ements
This work was supp
orte
d by the Rese
arch Fo
u
ndation of
Educatio
n Burea
u
of
Heilo
ngjian
g
Province, Chi
na (G
rant No. 12511
093
).
Referen
ces
[1]
Xi
e Z
h
ij
ia
ng, Li
u Jun, T
ang Yi
ke et al. Di
ag
n
o
stic
metho
d
o
f
blad
e-to bl
ad
e d
y
n
a
mic sp
a
c
e for turb
o
machi
ner
y’s b
l
ade fau
l
ts.
Chi
nese Jo
urn
a
l o
f
Mechanic
a
l E
ngi
neer
in
g.
20
04; 40(8): 9
6
-9
9.
[2]
Xi
e Z
h
iji
an
g, T
ang
Yik
e
, Ha
n
Z
h
ih
ua.
Dia
gn
osis M
e
thod
of
D
y
n
a
mic P
a
ra
meter i
n
Bl
ad
e
-
to-bla
de f
o
r
T
u
rbo Machine
r
y
'
s Blad
e F
aul
t.
China Mech
a
n
ical E
ngi
ne
eri
n
g
. 200
3; 14(1
3
): 1081-
10
83.
[3] http://
w
w
w
.
n
e
w
ener
g
y
.
o
rg.cn/
Html/006
5/20
0
652
2_
100
94.ht
ml
[4]
QIAO Yin-hu,
Z
H
ANG Ch
un-
ya
n. Vi
brati
on
F
a
tigue
Mec
h
a
n
ism
of W
i
nd
T
u
rbines’ B
l
a
d
e
.
Mech
anic
a
l
eng
ine
e
r
. 200
7
;
(4): 55-57.
[5]
WANG Yi-pin
g
,
SUN Wen-l
e
i, ZHAO Qun. Re
se
arch o
n
D
y
namic
Ch
a
r
acteristics of
Larg
e
W
i
nd
T
u
rbine F
l
e
x
ib
l
e
Blad
es.
Machin
e Buil
di
ng &
Automation
. 2
011; 40(
5): 134
-136.
[6]
Z
HU Hu
a
y
ua
n,
Z
H
AO Jingc
h
eng,
XU
W
e
iq
i
n
. F
a
ilur
e
n
u
m
ber pr
edicti
on
of aircraft mast
er oi
l
y
p
u
mp
base
d
on SVM
regressi
on al
g
o
rithms.
Aviati
on mai
n
ten
anc
e & engi
ne
erin
g
. 2007; (4): 25
-26.
[7]
MA W
e
i-jun. F
o
recasti
ng of p
r
oduct troub
le
base
d
on a
n
i
m
prove
d
gre
y
GM(1, 1) model.
Journ
a
l of
Natura
l Scienc
e of Hei Lo
ng
Jian
g Univ
ersit
y
. 2005; 22(
3): 389-
392.
[8] Sun
Y
o
n
g
q
uan
.
Research on
system re
liability growth pr
edi
ction theory
and
m
e
thod
. Har
b
in
un
iversi
t
y
of science a
nd
techno
lo
g
y
. 2
0
11; 6.
[9]
Z
H
OU
Z
hen, MA
Dezh
on
g, YU Xia
o
y
a
ng, et
al.
A
ppl
icati
on
of fuzz
y gr
e
y
rel
a
tio
nal
a
nal
ysis
i
n
fau
l
t
tree ana
l
y
sis.
Electric mach
in
es and co
ntrol
.
2012; 1
6
(3): 6
0
-64.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 53
3 – 540
540
[10] Yang
M
i
n
g
min
g
.
F
ault
mo
de
statistic & an
alysis
and
fail
ure d
i
ag
nos
is
of larg
e-scal
e
w
i
nd turbi
n
e
.
200
9; 1.
[11]
LI Junfan
g, Z
H
ANG
Buhan,
XIE Guanglo
ng,
et al. Gre
y
pre
d
ic
tor mod
e
ls for
w
i
n
d
spee
d-
w
i
nd
po
w
e
r
pred
iction.
Po
w
e
r System Protection a
nd C
ontrol
. 20
10; 3
8
(19): 15
1-1
5
8
.
[12]
ZHU Xi
an-h
u
i
1
, CUI Shu-mei
1
, SHI Nan, et al. Gr
e
y
pred
i
c
tion mod
e
l of
motor reliab
ilit
y of electric
vehicl
e.
Electri
c
Machin
es an
d Contro
l
. 201
2; 16(8): 42-4
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.