Beruflich Dokumente
Kultur Dokumente
Doctoral thesis
Zhe Li
Zhe Li
Deep Learning Driven
Approaches for Predictive
Maintenance
NTNU
Norwegian University of Science and Technology
Thesis for the Degree of
Philosophiae Doctor
Faculty of Engineering
Department of Mechanical and Industrial
Engineering
Zhe Li
Faculty of Engineering
Department of Mechanical and Industrial Engineering
© Zhe Li
"#
$
%
"&
' %
('
)
*
+
&'
)
,
-.
.
,
/
%
#&
0
-1)
0
%!
)2
,
33(
%
,&!
(
,
-,
3
%
89 % 8%9 0
809
8)9
:5
J31+>+L>90+-363>+>/+79</+.+:>+,6/+8.=7+<>7+38>/8+8-/
1
1
"
1
1
1
'
1
;
)
1
!"
#
#
$
%
% &
'
# (
##
)%*,
#$ !
)
$
#% - "./000 ,
#' 1
0
#%
!
#3 /)$%
4
$'
<
$ '
$#
"'
"
#
"
$
%
&
$$ /'%
$% .15
'(
'
%
!
(
!
)
*
!
$' #. 3*
+
+
$3 $.1
33
%
,
Maintenance scheduling optimization
%
'
$( /)3,
4
36
% ((
%# )
5
(6
%$ /
"
,%
-
'
+
'
+
'
+
(
'
+
'!
.
!&
%% /)6*
4
6
' 7
6(
'# 7
6,
'$ 8
)
981-/: **
'% 170
*
'' 7
; *(
/0
&
1 2"
&'
+
-
&
+
( 2"
( 2"
+
'3 81-/07-/&
) (
1
3
13
4+513
(
"
'( /) #,
4
#,
3 ")) $'
3 $'
3#
$(
3$ <-1 $,
3% / %
3' 17- %$
33
%(
6
$
"
"
&
+
3( /=0/!) ''
/5(
-
).5(
$
"
&
7
3, /) 3,
<
4
36
<
46 :566
44)
6:
=6
:5>?
=4
?6
:4.
75
::B
B
77
:>
B
7C
>6D
CC
>4% '%655
>=%BD)654
>:.
65>
>>$
D365?
>?D3
65?
>A%
657
>7
65C
>CD
666
>65
D366:
>66
=566>
>64
=666>
>6:
%$
66C
2
>6>'%"D%$
64A
?6="
!6=C
?4!
6:5
?=D
3
6:4
?><
wavelet coefficients and energy
6:=
??0336:>
?A.
336:C
?7
336>5
?C)
6>6
?65
D336>6
?66
%<)6>4
?64
33.6>4
?6=
D36>4
?6:%0*"
;
6>>
?6>
"
1
6>A
?6?0
6>7
?6A
6?5
?6C3
>"
6?=
?46
;6?>
?44%0*",%)
6??
2
.
6
?4=B
%0*",%)6?A
2
460
6C
440
0;
46
4=%
=5
>6
665
>4
D33D3
%<)B664
>=66=
>:
13D66:
>>
D33D3
%<)B66?
%+,6/"+<>90.+>+:</.3->/.>2<9?12"$V764=
>A!
; 8E964:
>7D
'%"D%$647
?60
336:?
?4
6:7
?=3
6>=
?:
energy-based
D:
6>?
?>
%0*",%)
6?7
2
3
0 0
0*0 0* 0
033 0
3
3
D D
D33 D
3
3
D%$ D
%
$ 1
..33 .
.
3
3
. .
.3. . 3
.
.33 .
3
3
.33 .
3
3
.% ."
%
D3 D
3
)
)
33 3
3
!33
!
3
3
! !
*) *
)
*% *%
,% 11,%
@( @
(
'% '
%
'' '
"'
.
.
2<
% %
33. "3
3.
,%) ,% )
), )
"
)%* )
%;
*
)%$ )
%
$ 1
33 3"113
$0 $
0
) )
33
3
3
%$
%
$ 1
BD33 B
D
3
3
BD) BD
1
)
B B
;
B33 B3
3
%0* %
0
%.00 %.
0;
%$) %
"
)
% %
%<) %<)
%<)B %<)
B
0
0
<0 <
0
! !
! !
2<
.
6
3
8.9
;
; 0
1
0
;
;
1
;
1 1
1
;
;
;
0
;
6
.
6
0
4
.
6
certain domains or with prerequisites, which could provide effective guidance to
select suitable deep learning methods to implement predictive maintenance. Some
novel applications of deep learning-driven fault diagnosis and prognosis
approaches such as DNN-based degradation assessment, DBN-based error
prediction, SAE-based feature reconstruction, and LSTM-based anomaly detection
are also presented along with detailed case studies in the thesis, which demonstrate
the superiorities of deep learning in self-learning, big data analysis, fault
identification, and degradation assessment.
The framework for predictive maintenance concerning the Industry 4.0 is proposed
in the thesis to achieve accurate failure prediction, efficient maintenance
scheduling, extension of data sources and tracking degradation of equipment. As a
result, the efficiency of maintenance implementation could be enhanced. The
framework also provides an overall understanding and helpful guidance for
researchers and practitioners to implement predictive maintenance in the Industry
4.0 era.
An Industrial 4.0 scenario about the implementation of predictive maintenance for
machining centers together with a case study of DBN-based backlash error
prediction are demonstrated in the thesis. In the case study, a novel HDPS-BPSO
maintenance implementation strategy is proposed to achieve predictive
maintenance in practical applications. The numerical result in that case not only
proves the superiority of deep learning methods in knowledge discovery with
strong self-learning ability but also demonstrates the benefit of implementing
predictive maintenance compared with preventive maintenance.
An experiment of fault classification and degradation assessment for rotary
machinery is also presented in the thesis to provide a comprehensive comparison
of different types of data driven models, in which DNN-based degradation
assessment outperforms the other models and proves the advantages of deep
learning. In this case, a novel SAE-LSTM approach is also proposed for anomaly
detection to train data-driven models in an unsupervised learning environment
when the empirical knowledge is missing and all data is collected without labels.
=
.
6
!
Zhe Li, Yi Wang, and Kesheng Wang. (2017) Applying Data-driven Method Based
on Deep Belief Networks to Predict Backlash Error in a Machining center. Journal
of Intelligent Manufacturing. DOI: 10.1007/s10845-017-1380-9.
Zhe Li, Yi Wang, and Kesheng Wang. (2017) Intelligent Predictive Maintenance
for Fault Diagnosis and Prognosis in Machine Centers - Industry 4.0 scenario.
Advances in Manufacturing. https://doi.org/10.1007/s40436-017-0203-8.
Zhe Li, Yi Wang, and Kesheng Wang. (2017) A Deep Learning Driven Method for
Fault Classification and Degradation Assessment in Mechanical Equipment.
(Submitted to Computers in Industry)
Zhe Li, Yi Wang, Kesheng Wang, and Jin Yuan. (2018) A Deep Learning
Approach for Anomaly Detection based on SAE and LSTM in Mechanical
Equipment. (Submitted to Expert System with Application)
Kesheng Wang, Zhe Li, Jørgen Braaten, and Quan Yu. (2015) Interpretation and
Compensation of Backlash Error Data in Machine Centers for Intelligent
Predictive Maintenance Using ANNs. Advances in Manufacturing, 3(2), 97-104.
Zhe Li and Kesheng Wang. (2015) Comparing Different Methods of Fault
Classification in Centrifugal Pumps. WHITE Transactions on Engineering
:
.
6
Jinghui Yang, Yavor Stefanov, and Zhe Li. (2016) Applying Built-in Virtual
Personal Assistant for Educational Equipment. Atlantis Press. 2016. ISBN 978-
94-6252-243-5. Advances in Economics, Business and Management Research
(IWAMA-16).
Shewei Wang, Kesheng Wang, and Zhe Li. (2016) A Review on Data-driven
Predictive Maintenance Approach for Hydro Turbines/Generators. Atlantis Press.
2016. ISBN 978-94-6252-243-5. Advances in Economics, Business and
Management Research (IWAMA-16).
Zhe Li, Viggo Gabriel Borg Pedersen, Kesheng Wang, and Yafei He. (2017) Fault
Classification and Degradation Assessment Based on Wavelet Packet
Decomposition for Rotary Machinery. International Workshop of Advanced
Manufacturing and Automation 2017.
" #
.
6
.
4
:5
.
=
:5
.
:
;
.
>
.
?
.
A
>
.
4,
;
;
1
A2/8>2//;?3:7/8>3=J</+5.9A8LA23-27/+8=3>3=896981/<:9==3,6/>9A9<5%23=
6C>5 #
;
.9A8 %2/ 8/A =><+>/1C A+= -+66/. += J"</@/8>3@/ +38>/8+8-/L <97 >2/8 98 :6+8>
;
0
"
%
<
!
;
0
.;
G';1
456:H
;
'
A
.
4,
:5
<
1
"
*
6AA5
6A74
#
!
%
;
45
1
:5 1
"
8.%989
8%9G,
456:H
7
.
4,
-977?83-+>3988/>A9<5A2/</M38>/<8/>N
0
11
:5
G-
456=H
'
4566@
G !
456?H
/
?=><C
=?-2 += J8>/6631/8> +8?0+->?<381 =C=>/7L +8. J%
)+8?0+->?<381L
D
:5
1
:5
C
.
4,
1
G
D
456AH
:5
G%
456:H!
:5.%
1$.%
8)9
"
%
"
1
1
G !
456?H
:5
F
Cyber-Physical Systems;
Internet of Things;
Big Data & Data Mining;
Internet of Service.
46
0
:5.%
/
G)
456:H
GD
@
4566H.%
";
65
.
4,
;
/.%
)%*="
;
; G.
3
4566H>+669A=J%2381=9<!,4/->=L38>/<+->
9>2/<+8.-99:/<+>/A3>2>2/3<M=7+<>N
.%
;
)
8"
9
66
.
4,
G0
456?H
0
G%
0
456?H
!
G'
4566H
;
;
%
%
%
"
:5
J%
J
.%
1
.
$
.%
%
"
1
1
"
G'
456>H
64
.
4,
)
"
'
:5
.
=
!
$
0*
*36==5?F4565
J
;J
G.*3I.=6C4565H
F
"
J
8
J
"J9
*3 6==5?F4565
J)
J'
:5
9=9:2CJ/B/-?>/+>>2/<312>>37/L+38>/8+8-/=2+66,/:/<09<7/.
6=
.
4,
>2381=+=+-97:98/8>N=./>/<39<+>3989<-/<>+380+36?</$37:6303/.7+38>/8+8-/>+5/:6+-/
1
"
1
1
G !
456>H
44
Maintenance
Opportunity Design-Out
44)
.
G&!
456:H
6:
.
4,
(
;
;
F
.
"
/
;
;
(
F;
0 *
*3 6==5?F4565
J)
JG.*3I.=6C4565H
"
; 1
;
G
456>H
;
6>
.
4,
'
;
3
;
1
;
;
8
9; D
89
"
0
;
1
;
;
;
6?
.
4,
D
F
;
(
"
/
;
G)
6CC>H
."
;
"
0;
"
*
:5
;
"
.
F
6A
.
4,
Generating minimal interference in equipment during the operation.
Increasing availability and reliability of machines.
Extending life of equipment and processes.
$;
;
;
;
;
/
0
;
)
F
;
";
;
G#
456>H;
67
.
4,
;
;
0
/
'
;
;
";
G#
456>H
%
/
46
460
6C
.
4,
0
0
:</.3-> -97:98/8>N= ./>/<39<+>398 9< -/<>+38 0+36?</ </;?3</= /6+,9<+>/ /8138//<381 8
0;
G&
456:H
80339
0;
033
0
"
/033
G3
1
'
456>H<
033
8D339G#
4565KB
4566H
8BD339 G@ 2
456=H
8.339 G
456>H
8..339
G
456>H ,<+ G.". !
#
4554H
"
8%$)9 GB
(
456AK .". !
#
4554H
8B339 G&
4556H
8!339
G<
!
4556H
8339G-
456AH
"118339G!!
455AH
G0
456:H0
0;
8*%9 033
*%
1
"
GD
%
6CC7H *
45
.
4,
'
G
6C7AH
=3>?+>398 7+C -+?=/ >2/ J-97,38+>9<3+6 /B:69=398L A23-2 3= 38@96@/. 38 -97:?>+>398
G&
4565H$
8%<)9G
.
4566KB
(
456AH11
8,%9G%
456:H89G)
%
456=H
GD"%
&
4557H
"
44
0;
440
0;
[B.-S. Yang, Oh, et al., Predicting the operating conditions of machine T & Neuro-fuzzy
2008] System
46
.
4,
[Ghate and Dudul, Fault classification for three-phase induction CCNN
2011] motor
[G. Xiong et al., 2013] Fault diagnosis of large-scale power systems RBFNN
[C.-C. Wang and James Rotating machine fault detection SOM & LVQ
Too, 2002]
[W. Wang, 2007] Prediction of spur gear condition value one NFN
step ahead
[D.-M. Yang et al., Diagnosis of motor bearing BPNN
2002]
[Khomfoi and Tolbert, Fault diagnosis for multilevel inverter drive BPNN
2007]
[J.-l. Zhao and Zhao] Fault diagnosis for missile electronic BPNN
command system
44
.
4,
[Din and Marnerides, Power load mapping DNN
2017]
[L. Li and Dai Guilan, Fault classification for semiconductor DNN
2017] manufacturing process
[L. Wang et al., 2017] Identification for failures in wind turbine DNN
gearbox
[Jia et al., 2016] Processing massive fault data to evaluate the SAE
health condition rotating machinery
[Galloway et al., 2016]
ator by SAE
mining information from spectrograms
[Sun et al., 2016] Fault diagnosis in induction motor SAE
[Gan and Wang, 2016] Fault location and severity ranking in rolling- DBN
element bearing
[Tamilselvan and Wang, Evaluating the health state of aircraft engine DBN
2013] and electric power transformer
[H. Shao et al., 2015] Identification of rolling bearing faults DBN
[R. Zhao, Wang, et al., Prediction of tool wear in a high speed CNC LSTM
2016] machine
[de Bruin et al., 2017] Diagnosis time for railway track circuit LSTM
4=
.
4,
[Liao and Ahn, 2016] Asset health assessment LSTM
[Yuan et al., 2016] Fault diagnosis and remaining useful life LSTM
estimation for aero engine
[Malhotra et al., 2016] Prognosis for a pulveriser mill from multi- LSTM
sensor time-series data
[Aydin and Engine condition monitoring LSTM
Guldamlasioglu, 2017]
[ElSaid et al., 2016] Predicting excess vibration events in aircraft LSTM
engines
[Janssens et al., 2016] Fault detection for rotating machinery CNN
D33%<)BD33
%$)
'
/
/;?3:7/8>98/7+C0+-/-2+66/81/=9<M,9>>6/8/-5=N
"
/8>+>39890:</.3->3@/7+38>/8+8-/>9=96@/>29=/M,9>>6/8/-5=N
* +
)
G&
4:
.
4,
4565H %
;
G."&,
456=H
G"%'
456=H
G
456>H
0
@G%
456?H
'
;
;
G,.
456>H
"
;
$
"
)
G@
!
456?H
"
Deep learning has had a long and rich history, but has gone by many names
! "
!"
and waned in
popularity.
Deep learning has become more useful as the amount of available training data
has increased.
Deep learning models have grown in size over time as computer hardware and
software infrastructure for deep learning has improved.
4>
.
4,
Deep learning has solved increasingly complicated applications with increasing
accuracy over time.
0
.6C>?(%
#).
G).
455?H
>2+>A/A9?6.-98=3./<J=7+<>L9<38>/6631/8>%2/./@/69:7/8>+8.=:</+.9038>2/6+=>
/
G.
"$1
456:H
1
"
;G
4564H
0
"
8D9
'
1
D"
6C756CC6
'
/
F
!
"
G' 6CC6H %
4?
.
4,
$
G% 456>HF
;
0;
;
G#
455?H0,.G456>H
"
;
4A
.
4,
;
)
G@
!
456?H0
"
/
G
&456:H%
G456>H
1
1
"
<
G 1
4564KB
4566K
456:K%-
4565HG@'
4564K%
456=K%
1
456=H
!
G456=H
"
D3
D
1
8BD)9
D
%<)
47
.
4,
%$)
D33 D3
!
G456AH
1
0
@
!
G456?H
BD)
%<)
D33
D3
H
#
G456?H
"
0
G4564H
4C
.
4,
;
"
"
8339
8%0*9 D3 ,%)
8.339
A23-27312>3./+66C1/><3.90=97/M,9>>6/8/-5=N>2+>-98@/8>398+67/>29.=0+-/..?<381
44
4=
4=%
=5
.
4,
for image-based data. And due to the ability to learn complex and
robust representation via its convolutional layer, filters in
convolutional layers may extract local patterns in raw data and
further build complex patterns for machine health monitoring
through stacking these convolutional layers.
0
D3333
.
33
*
G
)
456AK,,
@
456AK,!
456AH
G,!
456AH
33
;
33
"
"
,
0
%
BB%<)
D33G,,
@
456AH
33
"
33
%<)
%0*
G%
456=K -
1
456?H
%0*
0
G@*'
%
455?H)
%0*
%
G#
456?H%0*
=6
.
4,
@
G456?H
%0*
%0*
A3>238>2/>?<,38/N=1/8/<+>9<0
"
%<)
33
%0*
D3
"
BD)
GD455CH,
""
"
D3
G@
!
456?H
D3
"
"
D3
D3
1
D33
%<)
D3
D33
%<)
!
G456=H
D3
"
%<)
D3
.
>
D3
"
D3
0
B33,%)
,%)
G@
456=H
G
456?H G'
456?H
; G%
% 456AH 0
-
G456?H
,%)
;
-
G456?H
,%)
.3.
,%)
8%<)B9
"
8),9
,%)
GD
456AH,%)
)
G456?H
,%)"
"
"
"
,%) "
.33
"
"
==
.
4,
4
";
0-
G456?H
.33
image-based data.
.
>
D3"
D3
D3
33 %0*
,%)
.
?33"
%0*"
,%)"
, ($
;
:5
G 1
4564KB
4566K
456:K%-
4565HG@'
4564K%
456=K%
1
456=H
G'&2
456>H
:5
=:
.
4,
33%0*D3,%)
+8. A23-2-+837:<9@/9<=96@/=97/M,9>>6/8/-5=N38:</.3->3@/7+38>/8+8-/+</
:5
-
0
< B/
@ BM #")
B . 845659 0
"
"
8:9
6?::"6?>4
0!%
%
%B8456:9
! 456: ( 0..
. .
8.$3B$,9
0
D
0 8456:9 0
H
"
#8C9:66=":644
0
0 .
.
%
) * 8456?9
$
%
&
#'(7A"C4
0 $
@
% 8456A9 ) *+$
! 456A :
.*
*
*8.***9
D
@ % -
, 2", 8456?9 ,
!
D
B
@
'845669."
#-6?6"6??
=>
.
4,
D
)
%
386CC79)
"
"
%
$ %
./
0 -#-84965C"66?
.
"$1 # 8456:9 0
F D
).0
"
#8=9A76"A7>
.*3I.=6C845659*36==5?F4565)
I
F*
%
.
,!
3
-845669D
)3
4
,
-#
84>"=79F%
.-,.
%
1B"<8456>9@
+
5
-(#6
.
%
#
18456497
!
#
F.%
#")
#"86C7A9
F
"
1
/
6C"4A
./<?38%'/<,/<>+8.+,?Q5+#
#+36A+C><+-5-3<-?3>0+?6>.3+189=3=
-88=9>4=">==
'.
#@%
$
#,)8456>90
4
,456"45C
=?
.
4,
, 845649
F
%+%
,
&8456:9,
F)
0
1
2+% 98=H:96CA"=7AF656>?6I45555555=C
11
B
% 8455=9 $"
$ +
+ %
#98?96=46"6==:
@)(
)
0 8456A9 +
: :
! 3
.
8.3.9 456A
..
2
' + 8456A9 *"
B
' 0 8456:9 :5F ' NG H
8849>?">7
*
%
0 !
D ' #
8456?9 ) *+$
" ,
,
!456?***64
."%8"%9
$
!
( 8456=9 )
!
0
B
B
%
. & , , & & B D . &
- & 8455:9
"
"
"
,#;8696AC"
67?
&-
&+
',-'
,
#8456>90
%
4%'#4C"=:
@
@%.
<)
B0
,.8456?9
=A
.
4,
/%
%
<
$
+
-(#!8')%96A4"675
@
)
!
.8456?9.
$ +
+% 9-C4"65:
@
).%
1"D)0
#8455?9%)0F
%
)
F 0
698?9>>4">?7
@
<3
%<845669.
""
"
688>9
6>>>"6>?=
@
0 )
0"
' @ 8456=9 +
!456=
0
8
9
@,@
''
''2
,%8456?9)
+
5
-(#
'
#
)
#845669
/
3F*
'
# &
&
- & 8456?9
"
&
''8>9?7>"?C4
'
' )
D
! . 845669 %
"""
;
(8659=:75"
=:C4
=7
.
4,
'
)D
,
%.#
<%'%
!
8456=9 .
:
6((8A:?696?7"6A:
'
)
$D8456>9
:5
F
),
?
'@*
%
BB8455?9B
+ '#'8>A7?9>5:">5A
'%
% #86CC?9D
+8. J981 $29<>"%/<7 /79<CL +
'9?>"A4
'
08456>9*)B*$.$$30@
'F3*!&$B )*%
?457'0<*3*!&$B 3&65567(%0
'
"%'2@
#,00
',8456=9 *
!
44 0.)
.
Q
'
! % @ ' '
2 8456:9
F
+
#68>946C6"4456
#
B %
#
845659 0
0
3
3
80339
$
$
6#8C966A5"6675
#
8456>9)
),
?
#;
#
$%
/<<D%
,)<%
<
' % 8456?9 .
0+
5
'99==6"=:>
#
0 ,
D
/8455?90
"
$ +
+% -(8A96:7="6>65
#
,&,#-2
,38456?9
F0
$ +
+% 9-=5="=6>
%
,)8455A9
0"
;
68?94C>:"4C?7
'"*
0.)
#
.D" 845649D
"
';8>9
>455">46=
.
845669D
% < )
8%<)9
+
##8?9
:45=":466
*
*
1)<
8456>9
:5
.
4,
)
A#'7"6A
10%
'@*845649
,
!
0
,
' '"@
'
) 8456:9 :5
&B
+
8:94=C
, #
'"0
&
% 8456:9 %
:5
%
4%#="7
,
!
+ 8456>9 +
!
%
'
)
.8')9456>
,,
@
-&8456A90) "
)
" '
3
30
,
,
0'"8456?9. ,
%
0
0
'
)
A
,&@,!
+0@@)
,
'8456590
$ +
+% -8794C?6"4CA6
,."&)'@#-,
,&" 8456=9
05
4
-849666"66?
,.!
-"&+!",
)
#8456A9
"
:6
.
4,
+% #'(=AA"=77
)
).2!
%,&
,!8456?9&
!
%
0
8%09
)
+
)
)8456:94
,%
)&
:
!
0.,.
)
!8456:9
!(/+
F@
Q
)
<<B
00
@<,0
%@
8456?9)
"
"C,
C,/#(8!(#6
)
, %
0
@ ) 86CC>9 %
"
"
N0D
$
#869:?"
>C
).
#)),B3
%
.*8455?90
/
=66C>>
-98:964
)/)
/3)
'
D
/
0845579
/ 0
3
3 "
'68=96644"66=6
)
! 0
)
, 8456>9 .
"
D
0
$
-849==6"=:?
)
B
&86CC59B
$ *
!
="=5
)
<
%
< 8456=9
"
$
869=>="=>C
:4
.
4,
3
1
0
'
%8456>91
,
,
!
.
3
.
' , % & @
# ' 2 . # !
B 8456?9
" F 0
E$
+
*% F+*%G-8:9?C:"A5A
BD845649
),
6
+
&-
'
2
8456:94
,
%:
!
456:
***I%.*
% %
8%9
B845659.H
F.
GH%
;88796=?="6=??
B
0
(
% ' 8456A9
"
%
$
%
/0$ + 5C>::5?46AA55675
BD@S
%
%
0845669
=
!
.
B
B D
, 3 845669
!
:=
.
4,
3
B
0 0
"
) '
0
% 845669
"
'88A97CCC"C55A
%
3 )
0" D
B
D 8456=9
,
*5+4!
456= ***
.0%
%
%
'
%0!8456A9
;
"
8
9
/
F @
%
3
B
845659
8!9
80339 "
'98?9:6?7"
:676
% #8456>9
F0:
#7>"66A
%
1),&
!
&8456=9,
!
456=***
.0%
%
%
'#
'-
2
3)8456>9B
1
$
+
-866966>554
%
% % ! !
@
B 2
&
B 8456?9 *
,
!
%
0
8%09
'68796C=5"6C:=
%
'
0)
.#@10%,<
@
,
)8456?9)
@
:
6-;8A>7A9:7:":7C
%
8456:9 11"
"
%
;(8?96:C"6>=
%
)
0
%8456?90
;
<&,(>=:"
>:6
%
@B)
10
'!8456597
B$
&
%
=.
,7
F4'GB
% ! %
% -
B &
B -
2
. 2 8456?9 0
"
"
$
8;6A6"6A7
/%B1
%0
D'
/
@8456A9B
0
%
)
%
.
%)
.
0
4 #-8796A>4"6A>A
!
8456=9
4
B+
+
##664:"6=>
:>
.
4,
# # #
0 ,. &
D
. 8456:9 0
,
!
0
(
) $
)
)
' 8456:9
1
$
68
67A"6C?
<
@
!
845569 1
,
!
0($*%.$3 4556 ***
% B
.
!
.".
#
@"845549B
'$%
0
$
#'8:947="
4C=F65654=I
F656?54::47AC=
!
.-
&
--845579
H;
$ +
+ % --879
6C76"6CC=
!
8456:9
B 8"B9
, $
-84965?"645
!
8456?9
)
8)9% H:5%
H
+ ##'8694>C"4?7
!
, - D
#
& + 8456>9
033
, $
'849CA"65:
!
!
&845649
F-"
)
8-)9
F
0
3
&
!
,-
-,'2#
,B8456A9!
#'8=96=?5"6=?7
!
! 8455A9 0
$
:?
.
4,
+
+% -#84975C"74=
!
&.!#!
&
2&8456:90
; 1'4?C"47=
2'&0
D,,#D')&B
' B 8456>9
+ '98?467964>:75?
&
B.)!,,
845569
%
% ."D
)
0
,
$
#98>9=7="=C6F65655AI556A556A56A=
&
#8456:9$ %
%7
$
$
F
#!
Q%
&
D"% 2
'
845579 B
0
--8C96A6?"6A4>
&
D"%$)"%
0..845579)
""
$ +
+
% --8>966AC"66C=
&
")%
0)
.
#845549"
;
$ +
+% #84"=9=C6":66
:A
.
4,
&
# -
&
- & 8455A9
%<)
$ +
+
% -#8>94564"454:
&B845559)
0
.
")
")
% < ),
+
H
&
)!&
,,8456?91
!
***
.0
(
% 80(%9
-
1
#B#-#-
'+.&
-)
%8456?93
:
#866"65
-
.DD.%
#
-
8456=9
"
+
--8=94?C"4A7
-
#
/)0
'B8456:9<
$
-8496>6"
6>A
-
,&
-
&
-℈?94
,
!
456? ***
.
8.9
-
!,.@.&
-
-8456790
$
#((:=C":>=
-
-!
&
!
8456=9
0
:7
.
4,
$
-8?9646="644A
-
#"
-
#"
&%::
&41
$
+
!8455?9
%!
.
.
0
-
B!
#&
B
)
8456?9$
*+$
!
456?65
.%
8.%9
-
B&
B.-)
!
@
B28456?9,
0
)
'
)F 0 % C,
C,/##-!(9(
-@
&
!.8456A903
)
.
)
D
,
0
+
-(#9
-%.+
!
2845659
,&
!
4565 ***
.
-
!.
#.
#
)&,8455=45"4:#
455=9=
,
!
#.3
3455=
-0B)+
-,8456A9
6"C
:C
.
=
:5
!
$"%
!
1
;
1 1
1
'
1
;
"
0
;
G#
,
456:H
:5
:5
1
"
:5
1
;
1
!
G@
455?H.
F;
>6
.
=
:5
;
K
; K
K
K
;
'
F
High demands on data access, data quality and data fusion from multiple sources
for data sharing and data publication. Since these sources of data often operate
in a heterogeneous environment, integration between the systems is problematic
[Aljumaili et al., 2015].
The capability to deal with industrial big data. To leverage big data, industrial
businesses need the ability to support different types of information, the
infrastructure to store massive data sets, and the flexibility to leverage the
information once collected and stored. In other words, enabling historical
analysis of critical trends to enable real-time predictive analysis [J. Liu et al.,
2007].
The prediction accuracy for predictive maintenance. The inaccurate predictive
information may result in either unnecessary maintenance, such as early
replacement of components, or production downtime because of unexpected
machine failures. Therefore, the accuracy of remaining useful life prediction,
particularly the long-term prediction, which gives sufficient time to prepare for
a maintenance operation, plays an essential role in the full realization of the
potentials of predictive maintenance [J. Liu et al., 2007].
!
"#
:5
.%
%
).%
GD
@
4566H
;
;
G.
3
4566H %
>4
.
=
:5
:5
>=
.
=
:5
;
)!N989@+8/>+6*
;
:5
J%
J
.%
1
.
$
.%
<
"
1
1
"
G'
456>H
G#
,
456:H
"
%
>>
.
=
:5
$.1
/ !0&/
!
=
)
/)!
4 -
>- !/&
1
1
1-
1
-
1-
1
1
=;
=6
:5
>?
.
=
:5
=6
;
1
%
"
; %
;
"1"
;
>A
.
=
:5
1
;
:5
"
"
<
"
"0
G%
455CH
0
G !
455=H%
" !
:5
; 8%.009
;
;
G
456:H
0
"
$
0 +::63-+>39890-69?..+>+,+=/%9.+CN=7+8?0+->?<381
GD"',
4566H0
G-
456:H0
;
"" %
&
B@%
0
B
/
>C
.
=
:5
/ G3
456:H 0
B
0B
!
B
G0
456?H
;
B
?5
.
=
:5
10 "!
Prognosis Diagnosis
1
1
/!0
/
/!0
/
10
-
1<
1
=;
=" 1
1
&
&
&
!
! 1
5)1
05)1
=4
'
/
.
?6
.
=
:5
8
9
0
@
F
;
";
.
4
46
.
?
"
1
;
%0*
;0
;
"
)
'
1
G%
%
455?H
1
G%1
"@1
456>H
3
1F
0G@/
455?H
0
?4
.
=
:5
/
G'
4566H
*
'
;
0
;
,%
""
;
G@/
455?H0
0G)
456=H
G
6CC7H)
1
F
G)
4564H
QR
"
. ¦ ·µQ R ¶¸D ¦ 4
'.
4
'
. ¦ ·Q ¸D ¦ 4
"
;
/
F
"G@#<
455?H)
"
;
'
"
$
";
%<)
3
"
!
;
"
;
"
;
"
%
/
.
4
44
0;
0
D33G#
4565KB
4566HBD33G@2
456=H.33G
456>H..33G
456>H%$)GB
(
456AK .". !
#
4554H 33 G-
456AH
G0
456:H %<) G
.
4566K B
(
456AHD3G@
!
456?K
!
456=H
8339 G#
456?K , !
456AH
?:
.
=
:5
;
G#
B455?H
?>
.
=
:5
G@
!
456?H
/
0
D
"
"
"
"
"
G,.
456>H.
G@
!
456?K
!
456=H
G,!
456AH
0
( !
)
0
1
G% ,
456:H
1
G%
4564H
1
??
.
=
:5
1
G.
456AH
8 9
;
G&
456?H %;
1
1
;
G@#<
455?H
'
GD
4565H'
"
/
3
G."'!
456:H
1
G,
4566H
G2 ,
456>H
G-
456AH
G
)
4564H
G%
456AH0
;
1
scheduling 1
?A
.
=
:5
G
456:H
1
.
>
'%"D%$
1
( !
:5
%
6/@/<+1/0?8->398+63>C909>2/<=%2/A9<.M38>/<"9:/<+>/N37:63/=
/
G.
4557H
0
;
;
!/ ($
:5
.%
?7
.
=
:5
-
0!%
%
%B8456:9
!
456:( 0..
.
.
8.$3B$,9
0
00
"%
00
)0
00
"'
%0
"0
)
0
0
"
'%8456?9(
F
0
+#8>9A5A
0
/
)!
B
8456>9)
;
0D
$
-#8=9=>7"
=A:
D
B
@
'845669."
#-6?6"6??
D0&
0 ,
)1)845659D"/
1
1
B7
4 '98C96>7:"6>C?
D#.)
)
#845659.
F"
$ ID
6869A>"7?
.
,!
3
-845669D
)3
4
,
-#
84>"=79F%
. 2"!
, 2 8456:9 D
F
->6:">4>
.%-
+2D,
,
,8456A95
,
!
456A ***
.
'
)
8.')9
<)
)
@845649
"
"/
/"
B $
8?9
4666"466A
)
%
% 8456:9 %
%.00
!0!!!
)
)8456:90
"
1
4
B+
+
#-C7"65?
@
)
!
.8456?9.
$ +
+% 9-C4"65:
@
).%
1"D)0
#8455?9%)0F
%
)
F 0
698?9>>4">?7
@/
3Q%
% 8455?90
F(
06F'G>C5H?55
@
*
)%T 0"D!/0
V )
8456:9
"
B
9=C5":55
'
#
)
#845669
/
3F*
A5
.
=
:5
'0-
%
0.
)
#8455C9B
F
%
$ +
+
% -'8=9A4:"A=C
'
)
$D8456>9
:5
F
),
?
#
B %
#
845659 0
0
3
3
80339
$
$
6#8C966A5"6675
#
,&,#-2
,38456?9
F0
$ +
+% 9-=5="=6>
#
)
)
) 8456>9 )
F
+ ';8?4:>94>>"4?5
# B
B .8455?9 &
"
/
,,
!
'"*
0.)
#
.D" 845649D
"
';8>9
>455">46=
B
86CC79@
$ '(84"=94A6"4A:
.
845669D
% < )
8%<)9
+
##8?9
:45=":466
,#8455A9
,
,
/
!
%!
.*0)
'
(
A6
.
=
:5
, #
'"0
&
% 8456:9 %
:5
%
4%#="7
,D"'-
,B,.
2"
,&",-
2845669
$
+
#98=9::C":>A
,,2,
2,845649.
"D
%. 0
+
+'#->>C">?:
,2!#-B
'&8456>9 +
4 +
$
%
,$
3
&
I
!
456> 65
.
%
*8% *9
,
B-'@1%
&
,845669
1
"
;
"
%+
4 8#864945A:"
4575
,#//
3#.
3
,#8455A9%
688?9>>7">??FFIII65656?I/ 455?6455:
,%.!!&
,)8456:90
1
F
5
'(86496=A="6=C=
)
B % .
# @
)
) 8456=9 $ /
F%%QD)
3
%#8456:9)
89
F@
!N989@+8 " /+2C <?>98 +8. !N$?663@+8 8456>9 D
F
0&
-8696"44
A4
.
=
:5
# . * ,
# !
'1 ) 8456>9 .
$ +
+% (=6?"=4>
B
0
(
% ' 8456A9
"
%
$
%
/0$ + 5C>::5?46AA55675
B
0 0
"
) '
0
% 845669
"
'88A97CCC"C55A
%
3 B
% %
0 !
) 0
D
@8456A9 0$
%
)
*; )
$
'1 ( 0 .
0
0 -8=9
6:4"6:C
% #8456>9
F0:
#7>"66A
%
'%
0
@
008456490
1
,
#88796=6="6=4C
%1
"@1
,%
;
<0
"$WY1*<
#
.
8456>9,
0
*#8:9:CC">65
%#"3@&
D"%'
"'
"%8455C9
"
'8C966C76"66CC6
%
%
%
%8455?9
8<
4C9F%
!
8456=9
4
B+
+
##664:"6=>
A=
.
=
:5
<
@
!
845569 1
,
!
0($*%.$3 4556 ***
% B
.
<
@#,'0
!D8455?9
F!
$
,
!
.".
#
@"845549B
'$%
0
$
#'8:947="
4C=F65654=I
F656?54::47AC=
!
."'
%"! 8456:9 $ 1 "/
"
0
$
-68=9?5="?6?
!
8455=9
/
8<
C=9F$%
!
,-
-,'2#
,B8456A9!
#'8=96=?5"6=?7
2'&0
D,,#D')&B
' B 8456>9
+ '98?467964>:75?
&
D"%$)"%
0..845579)
""
$ +
+
% --8>966AC"66C=
-
)
0
B
% ' 0 8456A9
/
A:
.
=
:5
"
F
+
68::C":?:
-
, , &
, D ' B , -
2 , & 8456:9 .
F
+
88496?A"67A
-0B)+
-,8456A9
6"C
A>
.
:
"
"
:5
/
3
3
)
455C"456:
@
D
#
#
3
(%
833(9
.
D
8.D9
8 9
AA
.
:
0:6
;
0
;
;
G
456>H $
; $
G';1
456:H'
G-
456:H
%
G)
)
455?H0
8 9
;
;
/
G0456=H
"
" )
;
G,
456:HD
/
;
AC
.
:
:4.
3
F
GD
456?K
456?K
456>K ,
456?K%
456>K
%
456?K(
456>H'
;
"
:5
1
/
75
.
:
.
@
/
G%
4557H
3
"
G
.4566H@
G-
456>H 0
G)
6CCCH0
11
"
G%
455>H %
1
G !
456>H$
G-
456=H
5
F5G
<
"
;
; G%
4566H<
0
;
;G
4565K
1
4566K
76
.
:
%<)G@!
456:H,%G
'
455AH
G.,
455CH
G#
4555H
7F7G
)
0
;
;
$
75E
;
G 4565H $
"
G !
456=H
,
G'
4565H
G%
456:H
G!
456=H
"
$0
F869
K
;
K849
;
8
9K8=9
8
1
9
A9<5:3/-/+8.-?>>381>996N=/.1/+837:<9@/7/8>907+-238/.=?<0+-/;?+63>C+8.+8
G .
G '80
9
74
.
:
G .
G
G D
)
0
GD
4564H
"
;
G,
456:H
G0
"%
455>H
GBY
4565H
FG
0
809
;
0
;
0
0
;
'
0
0
""
(
<00*
1";
>5'-
7=
.
:
)'- 1
0*
;
0*
%
0*<0
$0
;
0*;
$0G
455AH
7:
.
:
"
*
)
;
/
;
;
<
"
")
%
G%
455CH
;
" '
0
;
'
;@
/
;
)
7>
.
:
" '
)
1
1
G%
%
455?H
)
1
G%1
"@1
456>H
)
.
=
)
"
;
! "
" ;
%
"
%
.
4 0
;
.
>
D3"
.
?33
1
;
,%)"
"" '
$
0
0
:=
)
7?
.
:
B
B
;
:::>
B
%+1/</%+1JB
L-+8
3
>64
1
?:0%.
B
0+?6>= 7+C 2+::/8 /@+6?+>398 >37/ +8. .+>/ 38 >2/ 09<7 90 J! "!%
&%%
%L$
'*2
8.$3Z59
8$Z69
475>456?80(,Z475>6?9
6:F=>465>456?
8)*Z6:F=>0*Z465>6?9
7A
.
:
::B
B
.
B
;
77
.
:
:>
B
)
1
839
"
7C
.
:
"%
;
(
1
1
0
1
).
>
'%"D%$
"( +
"
.%
"
1
11"
)
% ;
"
'
"" ($
)
.
)
C5
.
:
)
.
=
'%"D%$
-
0
% B
/
0
0
3 8455A9 B
"
$ +
+% -#8A94C=="4C:>
0
"%
0D
)0
%0 8455>9(
0$
%
#8=9=?:"=A6
C.O8 S +<+5F=/ +8. 5O8 97,38/. 38>/6631/8> 7/>29.= ,+=/. 98
0
$
-8:9A6A"A4C
D
)*,))
0#UB
D.)@,)
D8456?90
"
.3.
4
=
$
(=:":=
C6
.
:
B@&
!.
'
+8456>90
;
$
$
';87964C?"6=5:
0 0B*B@
#,
<
* 845659 <
"
+
+
+
8=9==>"=:A
#0
#0D.B '0
3
08456?9)
"
.3.
$ +
+% >5>">45
&
#
&
, 8456>9 (
.3.
F
I%
$ +
+% (?>?"
??A
#
1
)845669)B@<
( 1
%H*) %
5
% 75% -(##
8::6"::?9F%
'
#@
*
D
#845659,
$
%=
86A="6C?9F%
#
)
(
!"#8455590
%$)
1
8=CA":5>9F%
)
'
(1
8455>9 0 3 "$
0I<
%I.
0
.
%
'
)+,+
<
$
8::="
C4
.
:
:>49F%
0 !
. ! 845669 0
"
0
,
$
6'8>"79?6>"?47
B
'
%'8455A90
)
11
"
1 +
+
#6884:94A6>"4A==
, #
'"0
&
% 8456:9 %
:5
%
4%#="7
,-!
'&8456?9:5H
)
H
,
$
,.#
-
)8455C9
4&1
&$!
%
% 3
38%33455C9
,#
B
#
%8456:9
"
0%
66?7"6AA
)
86CCC9 .
$
;869==">4
)
)B
)
8455?9
0 $
$
8C9CC?"6556
%
) B
%
<
%
3 845669 )
C=
.
:
"
+
'88:9=76C"=74?
%'
!'
/
'!
0% B
845579 @
I
4%=$
69849??5"?A>
%%,#!
2
228456>90
.3.
0$
$
;':C"?5
%1
"@1
,%
;
<0
"$WY1*<
#
.
8456>9,
0
*#8:9:CC">65
/
F % % Q
D)
%#"3@&
D"%'
"'
"%8455C9
"
'8C966C76"66CC6
%
)%
00)
3
)
'
)8456?9.
.3.
$
86?>"AC
%
) %
0 0 )
3
'
) 8456:9
.3.
0
,
$
9(8>"796576"65C5
%
%
%
%8455?9
8<
4C9F%
%
%
3)$B
D
B8455>90
H
11
+
8694?"=A
.
)
8455A9 0
C:
.
:
$ +
+
% -#869457"4==
(-%
00)
3
!
)308456>9)
.3.<
)
.
$
#=C">5
<
'0
,8456=90
" 4
B+
+
#-(
=C">5
!
.
&
. & 845649 ." 1
$
) 13 7
+
!
0
)
B
!
@&
&-
&
2+8456:9<
7983>9<381 ?=381 W =?::9<> @/->9< 7+-238/ +8. 69-+63>C :</=/<@381 :<94/->398
+
/% -(;4:"=4
!
, - D
#
& + 8456>9
033
, $
'849CA"65:
! ."!
."' .
."
%
&"% 845649
0
,
$
6;8>"79?76"?7C
!!'&
-8456=9
"
+ + 686494C6:"4C4?
-
& &
#
-
8456=9 @
"
0
,
$
686":94A>"
476
-
. 8456:9 1
,
=
!
4?
C>
.
:
..
.8456:...9
-@!
.&
%-*
@&8456>9
>"
0$
$
8;6:4"6>5
C?
.
>
*
D3
'%"D%$
G-,
456AH
* 0
@
G,
&
456=K-
4564H
"
;
G'
456>K#
.456>K)1
456?H
G.
456:K%1
"@1
456>K-
456>H%
G4557H
0
;
/
'
"
G.
456?K
0
4557K'
,
4565K
456=K%
4564H
G456=H
CA
.
>
%
G4564H
0
;
;
3
=/8=9<=+8.A3</6/==>/-2896913/=>9-+:>?</.+>++>+66=>+1/=90+:<9.?->N=630/G
456AH 9< >23= </+=98 J31 +>+L 2+= +>><+->/. 89> 986C </=/+<-2/<=N ,?> +6=9
7+8?0+->?</<=N+>>/8>398+6981A3>2>2/./@/69:7/8>90.+>+"
G#,
456>H
G
456AH
G)
456?H %
0
%<)
,%
.
G456?H
11
"
0
G4557H
,
G4565H
"
'
1
C7
.
>
G
6CC?H3
;
"
)
;
;
"
G'
,
4565H
0>6
D
"
>6D
CC
.
>
*! 1
$
21+&(3
0
"
'%
D3
!
'%
;
"1"
Parts
Par
a of currentnt 7
4
Geometric Measurement Backlash Interpretation
backlash DBN Training
rotary encoder
Backlash Prediction
linear scale
Training
T i i Training
T i i
Reference Input Current Input
Training Training
Paremeters Targets Paremeters Targets
Sample Selection Reference Diagnosis M
Measured Measured Diagnosis Backlash in all
B
Backlash Model Backlash Backlash Model position
>4% '%
655
.
>
0>4'%
;
;
;
0
%
%
"
D33%<)B
D3
$
"
BD)
D3
BD)
/
$
'%
656
.
>
"
'
455?G@*'
455?H
BD)
F
F
8
9
Once this stack of RBM is trained, it
can be used to initialize a multi-layer neural network for classification or regression [Erhan
et al., 2010].
BD
1
)
G
'
456:HBD)
>=
"
"G&
4566HA
"
A
"
"#
#
"$
$ $
"%
A
"
>=%BD)
654
.
>
u v u v
F
Kµ^ ] Z [ \¶ ¦ ; gc @ grµ^]Z[\¶
!-
;
8>=9F
; ¦ ° ° @ grµ^]Z[\¶ µ¶
^ ]
BD)
G@'4565H
;
8>>9
8>?9F
u
65=
.
>
Y
Kµ^ Z [ \¶
¦ )O C *~|| ¥ )O C *~
µ¶
YP
Y
Kµ^ Z [ \¶
¦ )C *~|| ¥ )C *~
µ¶
Y[
Y
Kµ^ Z [ \¶
¦ )O *~|| ¥ )O *~
µ ¶
Y\
!
F
!a a|
a}
0
BD)
) *~||
) *~
;
@
G
'
456:H.
8.9
G@ * ' 4554H
/
) *~
@
@
;
8>669
;
8>649
>:0
.
G.
"
'455>H
65:
.
>
B* B B
* A A A
B* B B
>:.
65>
.
>
47!
# #
47!# # 47!# #
>>$
D3
BD)
D3
>?
BD)
6+C/<09<0+?6>=-6?=>/<3810+?6>C-97:98/8>N=3./8>3>C9</@+6?+>39890:9>/8>3+60+36?</=
$D3
1D
D33
/
Potential Failures
Evaluation
>?D3
65?
.
>
65A
.
>
>A%
' '
,¹Kº ¦ ( Ac µN¶ ?N ¥ ( Ad µN¶ ?N (5.13)
K ¦ Qc µN } ¶, (5.14)
!F
Ad µN¶
K
,¹Kº
K
N }
'
N ' ;
8>6=9
657
.
>
0
>7
1
>7
!F
65C
.
>
Qd µN¶
N b
A@@?µN¶
;
1
=<>FµN¶
;
1
%
1
;
8>6A9F
'
*
98+"3/><9+<8+123%@/<>3-+67+-238381-/8>/<A3>2+-966/->39890A//5=N
A=6
>6
>6
)
Qc
665
.
>
Qd
P 3
9c
9d
9e 0
976 )
;
N %
5 0
>CD
666
.
>
%
,
;
;
8>679F
= § 0µ9c 9d 9e N 976 K = K ¶µ¶
,
66>4
!
D33%<)B
D3
>4)
*
8)*9)
%;
*8)%*9
>4
D33D3
%<)B
=5
D33
%<)B
D3
"
664
.
>
>=
)
P 3
9c
9d
9e 0
976 )
;
5 0
=<>FG<MCcygc D
P
5 ¥
=<>FG<MCdygd D
P ¥
5 ¥
66=
.
>
D3
BD)
1D3
>:
BD
1
)
>:
13D
BD)6 BD)4 BD)= BD):
,
556 556 556 556
75
45
45
>65
D3
>66
>64
=5
=6
)%*
=5
66:
.
>
=63=V7+8.
V7A3>2+8.
+=</=:/->3@/6C
-+=/>2/+B37?7"/<73==3,6/<<9<"09<>2/,+-56+=2/<<9<3==/>+=V78A//5
4C
=6
)*
;
=5
)*
>66
=5
>64
=6
66>
.
>
%<)B
D33
D33
>5,")
;
1
D33 %<)B @
0
45
45
1
>>
>>
D33D3
%<)B
:5
D3
)%*
)*
D3
/
"
D3
D3
'
D3
"
0
66?
.
>
0D33
%<)B
D3
'
/
D3
'%
GB
4566H
'%"D%$
%
%$
G&"#
4564H
GB
*
6CC>H
>6=
´´´$
Q
K ´´´$
K ´´´´$/
K
´´´$
66A
.
>
K
´´´´$3
O
´´´$
Q
´´´$
´´´$
1
/O
1
=97/ =37:6/ 7+>2/7+>3-+6 09<7?6+/ 9@/< >2/ :+<>3-6/N= :9=3>98 +8
G+&456>H
%$
>6:
667
.
>
/
C
=
>6:
%$
%$
%$
%$
/
%$
F
6
1
1
66C
.
>
v i
x i
4,
=*
1
N=:9=3>398 x i
:(
p i
>(
p g
?.
t th
(t 1)th
F
vi (t 1)
vi (t ) c1r1 p x (t) c r p
i i 2 2 g xi (t ) 8>6C9
!P
>c
>d
Lc
Ld
G56H
A(
;
F
xi (t 1) xi (t ) vi (t 1) 8>459
7
/
645
.
>
>c
>d *;
>6C
%$
' "
0
>c ¦ >d Z4
>c ¦ >d Z6:C
0*;
>6C
>c © >d
>c ¨ >d %
>c
>d >c ¤ >d ¨ G.
4554H
>c
>c >c|
>d
>d| >d
8
9G-
456:H
( .0
%$
G
455AH
6CCA
*
%$ 1
G
*
6CCAH
/+-2:+<>3-6/</:</=/8>=3>=:9=3>39838,38+<C@+6?/=A23-2+</9<+-2:+<>3-6/N=@+6?/
1
646
.
>
c
8 »O µN¶¼ ¦ ¡ µ ¶ 8>469
cf
!O µN¶
E O
´´´$µN¶
*;
8>449
JNC@LPDM@Q µN ¤ ¶Z5
!L<I?
G56H
Q µN ¤ ¶E Q ´´´$µN
¤ ¶
'
G31
"
4557H
D%$
8
69
1 !
1
1
65
5>
8
9
*;
8>4=9
£¡ µ ¶ £¡ µ ¶
g
8 »O µN¶¼ ¦ ¿N<IC »SO µN¶¼¿ ¦ £¡ µ ¶ £¡ µ ¶ 8>4=9
f
!S
644
.
>
>?
'%V7
64=
.
>
>A!
; 8E9
! ,
! ,
! ,
-t
-w
-u '
F
8! 69 %;
'%
64:
.
>
-t
-w )
+F3
5F3
.y F)
.x F.
3F.
2J<? F!
D
1F)
CF
. F
D
E
F
'%
²
. F
D
E
F
SF!
TF!
?
D
Q
Q 1
F
DAH<DIN@I<I>@K@LAJLH<I>@?DIP@@FDNC@IQ ¦68>4:9
JNC@LPDM@Q ¦
64>
.
>
F
²
-t ¦ ° ° ° 2J<? ' 1 ' Wµ. ¶ µ¶
&{ &q &y
²
. ¦ .~
DAQ ¦ NC@I? ¦
JNC@LPDM@? ¦ ?gc ¤
*;
8>4?9F
²
DA. ª .x
² ³ ²
W¬. ® ¦ «
' 5LDA.x ¨ . ª .y 8>4?9
²
5LDA. © .y
-w ¦ 3 ' ° Q µ¶
&{
-
F
!
%Q & :Q ' C ¤ 2J<? ' 1 ª 1
D
;
-u
%
S T
'%"D%$
F
1655
>55S T
64?
.
>
6:4>+4
54>5L45553$ I36>555
3$ .y 3= V7 .x 64> V71 :> C 4
1
'%"D%$
>6>
1
>7
'%"D%$
C
67
===5= 3$ 0
?
!
A 6=
6C
:A776 3$
6:>A7
3$
>6>'%"D%$
64A
.
>
>7D
'%"D%$
(
$ +
. 1
!C67 ===5=3$
*/ ($
'%
D3
'%
'%
D33
%<)B
D3
/
)
'%"D%$
-
D & 8455C9 ,
0 1
2
647
.
>
$ *-8696"64A
.
"
)0
'@8455>9 7
,, *!
0%0%
.+-
'-
@@
.
,8456:90
"
0
,
$
9686":965A"646
.
)
# 845549
"
,
869>7"A=
*
B
# 86CC>9
!
%
% ) )
'
%6CC>)'%[C>
*
B.
%&845559
!
*
.
4555
*
D&.
0)
1
"0<
D%845659
! "
N 0 $
*4 ##89?4>"??5
@
)
!
.8456?9.
$ +
+% 9-C4"65:
'@8456590
D
1
$
64C
.
>
;869C4?
'@*845549
1
:
#8796AA6"6755
'
3D+
!
&8456>9
>"
0
$
$
;#65C"66:
#
2
.B#8456>90
"
0$
$
8;6>6"6>7
# & -")
&
, & 86CC?90
.3.
0$
$
'8>9>=C">>5
#
*
B.86CCA9
,
!
% )
. 6CCA 6CCA
***
..
.
%
) 0
'
) ) 8456:9 0
/)0,0D
8D3<4
69C,
C,/#(8!'-
)0
)
%
)08455A9,
!
)
..
Q0
455A)*[5A
08456A9%
:
68A?:794=
6=5
.
>
, "
&
%"'8456=9)
"
"
"
0$
$
9(:>">4
,#
.&
-
,8456>9D
0
,
$
8#86":9??A"
?7:
, - !
&
!
8456A9 0
"
0
$
F65655AI657:>"56A"6=75"C
,'0
0
-
!8455A9011
0 ,
,
#869=C":A
,'22
@845659D
-85?9:5=
)
0 )/
- 3 8456?9
"
D
0
$
-98>965=A"65:7
31
" ' B
"%
)
)
"
) 845579
D
1
F
+0
+ 86946"=4
)1
$ 8456=9 ,
!
456=***
%
*
8%*9
6=6
.
>
B%
%)
&
1%%845669
1
F
;
0
#8696C"4?
%'
!'
/
'!
0% B
845579 @
I
4%=$
69849??5"?A>
% &
*
B . 845569 1
,
!
4556.*
.
%1
"@1
,%
;
<0
"$WY1*<
#
.
8456>9,
0
*#8:9:CC">65
%
) 3
0
D 0 845649 )
4
'(85A966?A"66A>
.#++-
)-
'
%8456A9
"
0
,
$
6"6:
&
,845669
G
H+% $ -88696:>"6>:
&"#
@ ) % # -
$ !"3 .
-"' - 845649
$ 1 B 3
(
%
$ 1
0
!B
B
*
!
!!88:9C55"
C64F65665CI4564445>=C5
-
-8456:9
$
=
$
/
1
1
% +
F1%+G!
33("
-@!
.&
%-*
@&8456>9
6=4
.
>
6==
.
?
,
$$
)
;
%
;
%
;
"
;
,
0;
;
;
;
;
G,
456=H
.
G*
456?K
456>H
%
A5HC5E
6=>
.
?
G !
@ 456=H
"
)
;
;
!
%
'
1
G& &
456>H
!
)
G,
6=?
.
?
!
33
*
0 %$
4576?
;
F
GJ%$4576?"6J456?H'
/
"
'
"
;
B
;
89%
8%9
8*)9
!"<
8!<9 0
;
;
'
6=A
.
?
;
0
"
";
!<
;
%
%
G$
4566K<
455?H
(
!
;
!
"
;
!
;
;
;
,! 5&+
!
1
!
"
"
";
G& -
456?H
8!9
H;
!
8!9G2
455=H
!
;
;
"
6=7
.
?
?6="
!
6=C
.
?
?4!
F & ;G
4557H
;
8?69
8?49
Xb µN¶ ¦ ° Cb µF¶ Xgc µI
¶
Xb µN¶ ¦ ° Cc µF¶ Xgc µI ¶
;
8?=9
8?:9
A µN¶
: µN¶
>
d
6:5
.
?
fo
A µN¶
¦ ° >
µN¶: µN¶ µ¶
go
6:6
.
?
?=D
3
?:
*
;
?>
wavelet coefficient-based and energy-based
6:4
.
?
?:!
?><
wavelet coefficients and energy
0
wavelet coefficients and energy
;
!
;
D
6:=
.
?
0
J
L
33
0
'
33
*
0D33
f : X Y
C
J
33
D33
8
9
033
??
0
D33
,
;
,_6
0
"
;
0
6::
.
?
G
456AH
Q
A
R
G@
456?H
33
/
/
/
D33
??033
6:>
.
?
33
)
"
G%
456=H
1
GB
4566H
;
@
D G4565H
5
%
33
?6 G.
456>HQ
P
S
S ¦ # Q P ¤ =
?60
33
S
¥S
*;
8?>9
5
nkljµ¶ £ g £
AµS¶ ¦ imnjµ¶ ¦ 8?>9
£ f £
G,.
4564H)
1
0
6:?
.
?
)
8
;
9G#@
456=H
.
33
33
"
33
>5 ^ >5 ^ =4 ^ =4 ^ =4
1
?4
6:A
.
?
?4
6?57
6=A?
8
9
4=4
8
9
8
59
6A5=
6>>6
8
9
?76A
6F 6>4
8
9
6A74
6>>>
8
9
4FD
44A
8
9
6A4:
6>:C
8
9
=F,
6A>
8
9
>
?A
33 /
CC7AE
6:7
.
?
?A.
33
33
==CC
/
/
54>7
8
655
1
9
6:C
.
?
?7
33
#
)
456AHD3G--
-
456AH 33.G'
456AH
D33G0
)-
456>H!
> :
?C
8)..B9
"
33
"
;
%
;
6>5
.
?
0
1
>
8=57C
=65
339?65"?6=
?65
D33
6>6
.
?
?66
%<)
?64
33.
?6=
D3
0
D3
D33
33
6>4
.
?
?=3
| |
)
% )%*µ ¶ ,*µ ¶
cb cb
0 %<)
1
%
33.
0
GD
456?H
33.
1
33.
33.
;
0
33D33
D3
D3
D33
;
)%* 0G.-
456AH D3
D3
D3.
?D3
D
;0
.
433
6>:
.
?
; %
?6:%0*"
;
6>>
.
?
6.7.1.1 Feature normalization
?:
energy-based
D:
=; = ;9 : = ;9#: = ;9$: = ;9%:
6#E'#(%, *E%63(3 *E#*336# *E# 33$$ 3E''#%$%
# 6#E33#3 *E%,%%6' *E 63 #( *E#*3($, 3E%'**%%
$ 6#E3'366 *E' # 3% *E# * '' *E# #*%, 3E%*,3%'
% 6#E33('( *E%3#(,, *E 6#,$' *E#%$3,3 3E%$$ #
' 6#E 66 ( *E' *3*% *E#*% #, *E#$,,6( 3E,%(#*'
3 6#E%$ (( *E%6%%$ *E 63 *3 *E## ', 3E3'3'$%
( 6#E%$6'# *E%,,,## *E# #( *E#*,%*$ 3E3'*'%'
, 6#E'%%3 *E%',,%$ *E#*(63 *E 6(,(' 3E'6*( 3
6 6#E'$#6% *E%,*'(6 *E#*$6 *E 6%* # 3E',,'(#
* 6$E*36, *E%%*,$' *E 3,#,$ *E ('33% 3E %'%*$
6$E*(6%, *E%(*#3# *E ((('( *E 6'' ' 3E*(36,'
# 6#E3#,6, *E' #'#$ *E# $( # *E 6% ,( 3E%'*'6,
$ 6#E(3 # *E%,(*$, *E#*$$$# *E 6 *3 3E$'($3'
% 6#E36(# *E%,#63, *E 6$$3# *E#*# (, 3E%#%#6%
' 6#E$ (%' *E' $ '$ *E#*( $$ *E##%(# 3E($('%(
3 6#E%,3%( *E%6,, $ *E#* '%# *E# $,($ 3E'66$*(
( 6$E(%%6, *E%# $#' *E (* 6$ *E 3,33( 'E%6%,$
, 6$E*(%$# *E%$,',% *E (33%% *E 6'% 3 3E '*$3
6 6#E,$6%' *E%,3'3( *E 6$#,% *E ,(*,# 3E#6$3#
#* 6#E**, , *E'' $6# *E#$$(%, *E##*'* 3E6,3 ((
6>?
.
?
/
"
1
10² %0*"
s gs
0² ¦ 8??9
s¢ gs
D
?6>
"
1
%0*"
80
/
9<
1
?6>
"
1
6>A
.
?
455AK
455AH
"
GD
456=H%0*
)
K
0
?6?0
] ¦ A` µ_ ¶8?A9
6>7
.
?
2µ_ ±
_¶ )
2µ_ ±
_¶
_
_ ±GD
456=H
`
`² 1
*;
8?C9
c
Wpr µ` `² ¶ ¦ w #w
hc 2µ_ B`³ µA` µ_ ¶998?C9
0
%
G@
456?H*
%0*"
!
"
%0*
G*
4565H
"
0.
4
%0*
;
!
%0*
1
;
%0*
,4
1
==
;
0
;
?6A
6>C
.
?
?6A
,%)
,%)
6CCA
B33
G'
% 6CCAH
,%)
F
,%)
"
G% 456>H
B33
6?5
.
?
B33
B33
"
,%)
B33
"
,%)
"
G@
456?H.
B33,%)
1
,%)
G-
456AH,%)
S T U J
B33G,
0456?H
?67
S
5
6
-gc
5
6
;
*;
8?659
S ¦ Vµ: ! ¹Cgc Q º ¤ = ¶µ¶
!V :S
S G H
Q Q
N C Q :S :T
:U :
= = = =
S T UJ
T
*;
8?669
T ¦ V¬: ! ¹Cgc Q º ¤ = ®µ¶
3
U
F
U ¦ N<IC¬: ! ¹Cgc Q º ¤ = ®µ¶
-gc
- *;
8?6=9
- ¦ S ! -gc ¤ T ! U µ¶
*;
8?6:9
6?6
.
?
J ¦ Vµ: ! ¹Cgc Q º ¤ = ¶µ¶
xt
o
ht-1 ht
Sigmoid function
tanh
tanh Tanh function
tanh
Elementwise multiply
+ Elementwise add
Ct-1 + Ct
D
B33
;
;
,%)
;
"1
"
G%
456:H
B33
'
B33
"
0
,%)
6?4
.
?
,%)
@
0
65
,%)
65C
*
>55
,%)
>"
?6C
>"
;
$E'
$
!/5=9H:
#E'
"
#
#
E' $
%
'
*E'
*
# $ % '
?6C3
>"
>"
%0*",%)
6?=
.
?
;
<
?45
?46
<
475
,%)
;
"5: 5=
%
;
!
=
=
?45.
6?:
= .
?
C
?46
;
(4 5
;
%0*"
"
1
,%)
?44
%0*",%)
0
65
0%0*"
C
;
65
,%)
C,%)
C
;
6?>
.
?
#
?44%0*",%)
%0*",%)
455
6>5
>5
?4=
%0*",%)
6??
.
?
#
$
%
'
?4=B
%0*",%)
)%*
>"
; %
/
?>
%0*",%)
;
6?A
.
?
?>
%0*",%)
6 675:67 ?> C4>
4 6AA>?> ?5 C>
= 6AA?A? >A C?>
: 6A=?A> >7 C?
> 6?AC=C 65C A6>
,9 ($
%
;
6?7
.
?
;
";
;
%!
*) !
;
;
perception to represent and interpret the signs of faults. During the experiment,
all the original features represented through WPD are used as inputs without selection.
%
"
0
33
;
0
%0*",%)
-
0
)-0%
))
)!%
0B
B
8456>9
3
?=381 (% +8. " ,+=/. 98 6+<5/N= ><+8=09<7+>398 :
#88%
.9C7="CC=FFIII65656?I/ 456>5>54?
D
.
)
- * 8456?9 '
"
66"6=
D&.
0
<8456=9B
F0
D & ,
,
' 8455A9 .
=
!
0
6?C
.
?
.
0
<,
*
0
08456>9
4F'4$
./<?38%'/<,/<>+8.+,?Q5+#
#+36A+C><+-5-3<-?3>0+?6>.3+189=3=
-88=9>4=">==
*
%@
D "' *
..
@
8456?9 . .
9;8%
.9 6:"==
FFIII65656?I/ 456>5?557
*
D&.
0)
1
"0<
D%845659
! "
N 0 $
*4 ##89?4>"??5
. D B
D , 8455=9 0
%
4
*
-8A9CA="
C74
@
@%.
<)
B0
,.8456?9
/%
%
<
$
+
-(#!8')%96A4"675
@
2
D & 845659 )
!
.0
%
@ %
B ]#%DB
% #8456?9
,%)F 0
6A5
.
?
'
0 (',."#D
'
'
.8456A9)
"
"
;
B #(8%
.9 C?"65>
FFIII65656?I/ 456A5>54C
' %
% # 86CCA9 , %" ) :
;8796A=>"6A75F6566?4I6CCAC76A=>
#
B %
#
845659 0
0
3
3
80339
$
$
6#8C966A5"6675
.20
'38456A9
"
F %
%Q >55 A
-6;849?7C"A54
,.&0D,$@D
)P
"B845649*
:
/
8C":79F%
,&,#'-
-)#8456=90
$ +
+
% '6869657"64?
,
,
0'"8456?9. ,
%
0
0
'
)
A
, #
. + 8456:9 0
$ +
+% 88696A:"67A
,.!
-"&+!",
)
#8456A9
"
+% #'(=AA"=77
)
) B
3
%
1
& 845659 .
6A6
.
?
**@
0 &
+
'85:9=C5
3
)08456>93
F
(%0
$
%845669
1
!.
)
(
B
D
3
#
)
D8456A9
*)
%<) + 0 8849 65="666
FFIII65656?I/
/456>5755>
B
* ' @ *
!
B # 86C7>9 *
B
%
'%0
D
8456:9 *
=
!
0
.
% .
0
%.
@
8455:9%
,
,
F*
% #8456>9
F0:
#7>"66A
6A4
.
?
%
0)/
-38456>9D
'
H'
869>4"?4
%&"/-
%)
."
,#" 8455A9 D
0),
$
#98696=7"6:4
% )
#
@
' @ 8456=9 7
!
!@"1&D
"
&
'%845579**@
$
#8?9?67"?4>
<
@#,'0
!D8455?9
F!
$
,
(+81 " +8. ?9 +=/. 98 >2/ -9+6 738/N= /==/8>3+6 =+0/>C 7+8+1/7/8>
0 ,
%4 #8=9?4"?7
!
-"& , .
- D 8456A9
$ +
+%
2#"--
'-."2
&
2"@8455=9H,
"
.
!
455=
)
&
")%
0)
.
#845549"
;
$ +
+% #84"=9=C6":66
&
&2--
!
)@8456>9<
1 H;
"
0+
5
''6=>5"=??
-
., +0
.8456A9)
/
6A=
.
?
-
&,D#2
'
8456?9.
**@%
D
0 )
!
: %
*
6"6:
-
-
-
# 8456A9 0
B
#(98%
.9 =C>":5A
FFIII65656?I/ 456A545:6
-
B&
B!
#
)
8456A9,
"
+#98494A=
6A:
.
A.
/
:5
;
1
1
;
1
"
:5
.
4
-+837:<9@/9<=96@/=97/M,9>>6/8/-5=N
1
6A>
.
A.
;
.
:
:5
;
1
.
>
0
D3
D33
%<)B
D3
)
'%"D%$
.
?
!
;
%
"
33
0
%0*"
,%)
6A?
.
A.
/ (
F
B
6AA