research article process monitoring and fault diagnosis
TRANSCRIPT
Research ArticleProcess Monitoring and Fault Diagnosis forShell Rolling Production of Seamless Tube
Dong Xiao12 Xuyang Gao12 Jichun Wang3 and Yachun Mao4
1State Key Laboratory of Synthetical Automation for Process Industries Northeast University Shenyang 110004 China2Information Science and Engineering School Northeastern University Shenyang 110004 China3College of Science Liaoning Industry University Jinzhou 121000 China4College of Resources and Civil Engineering Northeastern University Shenyang 110004 China
Correspondence should be addressed to Dong Xiao xiaodongiseneueducn
Received 6 November 2014 Revised 9 January 2015 Accepted 14 January 2015
Academic Editor Sangmin Lee
Copyright copy 2015 Dong Xiao et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Continuous rolling production process of seamless tube has many characteristics including multiperiod and strong nonlinearityand quickly changing dynamic characteristics It is difficult to build its mechanism model In this paper we divide productiondata into several subperiods by 119870-means clustering algorithm combined with production process then we establish a continuousrolling production monitoring and fault diagnosis model based on multistage MPCA method Simulation experiments show thatthe rolling production process monitoring and fault diagnosis model based on multistage MPCA method is effective and it has agood real-time performance high reliability and precision
1 Introduction
The deformation process of seamless tube production canbe summarized into three stages perforation extension andfinish rolling The main purpose of the perforation processis to perforate a solid round billet into a hollow shell Themain purpose of Elongator is to further reduce the crosssection and to make the shell improve on dimensionalaccuracy surface quality and organizational performanceAfter elongator rolling steel tube is called shell whichrequires further molding in the finishing mill in order toachieve the requirements of he finished tube [1] Continuousrolling mill is an elongator which has the highest productionefficiency and superior product quality So it has been widelyapplied to the big steel mills Online monitoring of oper-ating parameters can effectively avoid accidents eliminateequipment damage save a lot of maintenance costs increaserunning time improve set utilization and reduce spare partsinventory and time Process monitoring and fault diagnosishas a very important practical significance for safe productionand scientific maintenance [2]
Reference [3] analyzed the causes of roll sticking steel andgave several methods to avoid sticking steel but did not give asticking steel monitoring method Reference [4] introducedthe rolling steel tube transverse wall and longitudinal wallthickness error monitoring method but it needs to introduceexpensive measuring instrument Reference [5] introducedthe fault diagnosis methods for DC motor of rolling rollReference [6] introduced a continuous rolling mill onlinemonitoring system based on virtual instrument technologybut it monitored the variables separately and did not considercorrelation between the variables which made some deep-seated faults difficult to bemonitored In the area of industrialprocess significant researches have been done for onlineprocess monitoring and fault diagnosis [7ndash12] Multivariatestatistical analysis techniques such as principal componentanalysis (PCA) [13ndash15] and partial least squares (PLS) [16ndash18]have long been used for detection and diagnosis of abnormaloperating situations in many industrial processes
Continuous rolling process is a batch process which hastypical multi-period and dynamic multivariate characteris-tics According to multi-period characteristics of continuous
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2015 Article ID 219710 12 pageshttpdxdoiorg1011552015219710
2 Mathematical Problems in Engineering
Displacement
21 7 8
Tim
e
Bite
stag
eSt
able
rolli
ng
stage
Stee
l le
avin
g sta
ge
Perio
d
0A
C
Dd
c
B
c998400
d998400
D998400
t1
t2
t3
A998400 aa998400
B998400
C998400
b998400 b
Figure 1 Time and displacement of rolling tube process
rolling process it can be divided into bite stage stable rollingstage and steel leaving stage The production data has thefollowing characteristics
(1) Large Amount of Data High Dimension and Strong Cou-pling Seamless tube rolling process is a computer supervisoryand computer control industrial process It need regularlyacquire system state variables and equipment status to be usedfor display and control Each cycle of seamless tube rollingproduction process will produce tens of thousands of processdata After accumulating the amount of data is enormousAt the same time the behavior of the rolling process is theoutcome of combining action of many variable factors whichhave a strong coupling relationship
(2) Industrial Noise andUncertainty Because systemworks incomplex environments the output signal of electronic sens-ing devices is susceptible to noise And it is also vulnerable tothe uncertainties
(3) Multimode Data is a reflection of the system statechanging The data are not only in normal working statebut also in various kinds of abnormal state and fault stateThe former is the main part The latter has a relatively smallamount of data but it is indispensable in the knowledgediscovery and data mining
(4) Staircase Distribution In the biting stage the tube entersrolls from the first to the eighth and in the steel leaving stagethe situation is in turn So the data is staircase distribution
In this paper by using 119870-means clustering algorithmcombinedwith production process we establish a continuousrolling production process monitoring and fault diagnosismodel based onmultistageMPCAmethod Firstly accordingto the production process we divide production data intothree subperiods including bite stage stable rolling stage andsteel leaving stage Secondly we use clustering algorithm tofurther divide those larger changing stages A satisfactory
effect is difficult to be obtained if using clustering algorithmto classify the production data alone We classify it combinedwith production process Finally we get a satisfactory moni-toring result Online monitoring and fault diagnosis systemnot only can online monitor equipment but also can carryout fault alarm and diagnosis timely During the monitoringprocess it does not require the dedicated testers and doesnot need professional and technical personnel to make ananalysis and judgment
2 Analysis of Factors Affecting the ContinuousRolling Production
In order to build the monitoring and fault diagnosis modelwe analyze the influencing factors for continuous rollingproduction firstly
As shown in Figure 1 the continuous rolling process canbe divided into three stages
(1) Bite StageAs shown in Figure 1 the head of steel tubemoves from point a1015840 to point b1015840 and the tail movesfrom point A1015840 to point B1015840
(2) Stable Rolling Stage As shown in Figure 1 the head ofsteel tube moves from point b1015840 to point c1015840 and the tailmoves from point B1015840 to point C1015840
(3) Steel Leaving Stage As shown in Figure 1 the head ofsteel tube moves from point c1015840 to point d1015840 and the tailmoves from point C1015840 to point D1015840
Factors affecting seamless steel rolling production processare mainly the following roller rotational speed roller inputcurrent and roller torque
(1) Roller Rotational Speed In continuous rolling process theroller rotational speed is very important If the speed is toofast the surface of shell cannot be completely eliminated
Mathematical Problems in Engineering 3
Batches
Time
Variable1
1
X
K
J
J
2J KJ
I
I
X
Figure 2 Arrangement and decomposition of a three-way array byMPCA
In addition if the speed is too slow the consumption ofelectricity and other energy will increase
(2) Roller Torque The roller torque is one of the controlledvariables It is controlled by the input current The torquedirectly affects the quality of steel tube If torque is too small itcannot completely remove the blank of tube surface If torqueis too large shell will deform under the pressure Reasonablycontrolled torque of the rolling process is an important factorwhich cannot be ignored In order to save energy resourcesduring the wait state the torque will be reduced When thetube is rolled the torque increases When the tube leaves theroll the torque decreases
(3) Roller Input Current Roller current is one of the mostimportant control variables in the process of rolling It isan important factor to control torque and rotational speedBecause the sensitivity of the current is higher when thepierced tube enters the stand mill current increases quicklyThen the roller rotational speed and roller torque increase inorder to roll smoothly Current and torque are proportionalto the relationship The current increases when the steel tubegets into the rollThen it tends to a stable stateWhen the tubeleaves the roll current drops rapidly So correctly controllingthe size and direction of current is an important part of theprocess of rolling and the current is themain control variablein the rolling process
3 Multiway Principal Component Analysis
Batch processes are repetitive production process Theirdata sets have one more dimension than the continuousproduction process data set We can use three-dimensionaldata matrix 119883 = (119868 times 119869 times 119870) instead of the batch processdata collection where the three dimensions 119868 119869 and 119870respectively represent the batch number of samples numberof process variables and the number of measuring points ineach operation [19 20] MPCA will unfold119883 = (119868 times 119869 times119870) in
such a way as to put each of its vertical slices (119868 times 119869) side byside to the right which start with the one corresponding tothe first time interval The resulting two-dimensional matrixhas dimensions119883(119868 times 119869119870) [21 22] (Figure 2)
After three-dimensional data matrix 119883 is expanded intotwo-dimensional data it will be decomposed into a seriesof principal components consisting of score vectors 119905
119895and
loading matrices 119875119895 together with a residual matrix 119864 by the
principles of PCA The MPCA model can be written as
119883 =
119896
sum
119895=1
119905119895otimes 119901119895+ 119864 (1)
where the score vectors 119905119895is related only to batches and the
loading matrices 119875119895is related to variables and their time
variation The noise or residual part 119864 is as small as possiblein a least squares sense
4 Establish a Multiperiod ContinuousRolling Production Process MonitoringModel Based on 119870-Means and MPCA
41 Collecting the Production Data Based on actual produc-tion data characteristics of seamless steel rolling process inthis paper we use ibaAnalyzer software to collect 20 dataunder normal production condition As shown in Table 1 weselect 24 production process indicator variables to establish amonitoring model In this paper the three-dimensional datamatrix is 119883(119868 times 119869 times 119870) The three dimensions respectivelyrepresent the batch number of samples number of processvariables and the number of measuring points in eachoperation (119868 = 20 119869 = 24 119870 = 400) In order to obtainvertical data slice 119883
119896(119868 times 119869) we cut the three-dimensional
matrix along the direction of the third dimension Thusduring a period we can get 400 time slice matrixes By usingPCA for the 400 two-dimensional time slice matrix we got400 load matrixes and got the whole model load matrix bytaking an average
42 Online Monitoring Based on PCA Method The wholePCA model can be defined as follows
119875lowast=1
400
400
sum
119896=1
119875119896 (2)
where 119896 is the number of load matrixThe number of principal components 119860lowast can be calcu-
lated by the cumulative contribution rate In this paper weset 119860lowast = 6
The cumulative contribution rate =sum119860lowast
119895=1120582lowast
119895
trace (119878lowast)ge 90 (3)
where 119878lowast = diag(1205821 1205822 120582
119869) is an eigenvalues diagonal
matrix of the matrix119883The whole PCA load matrix 119875lowast is divided into two parts
the main component space 119875lowast(24 times 6) and the residual
4 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40014
145
15
155
16
165
Time
T2
(a) 1198792 control limits
0 50 100 150 200 250 300 350 40020
40
60
80
100
120
140
Time
SPE
(b) SPE control limits
Figure 3 Seamless tube rolling process 1198792 control limits and SPE control limits
Table 1 Measured variables for seamless steel rolling productionprocess
Symbol Variables Unit1 1st roller speed rs2 1st roller current A3 1st roller torque Nlowastm4 2nd roller speed rs5 2nd roller current A6 2nd roller torque Nlowastm7 3rd roller speed rs8 3rd roller current A9 3rd roller torque Nlowastm10 4th roller speed rs11 4th roller current A12 4th roller torque Nlowastm13 5th roller speed rs14 5th roller current A15 5th roller torque Nlowastm16 6th roller speed rs17 6th roller current A18 6th roller torque Nlowastm19 7th roller speed rs20 7th roller current A21 7th roller torque Nlowastm22 8th roller speed rs23 8th roller current A24 8th roller torque Nlowastm
space 119875lowast
(24 times 18) Similarly eigenvalue diagonal matrix 119878lowast
is correspondingly divided into two parts 119878lowast and 119878lowast
= 119883 (119875lowast
)119879
119883 = (119875
lowast
)119879
119864 = 119883 minus119883
(4)
In current online monitoring and fault diagnosis judgingwhether statistics1198792 and SPE are over the limit is usually usedto determine whether faults happen The control limits of 1198792approximately obey 119865 distribution
119863 sim
119860lowast(1198992minus 1)
119899 (119899 minus 119860lowast)119865119860lowast119899minus119860lowast120572 (5)
where 119870 = 400 and 119899 is the number of samples data formodeling The control limits of 1198792 is shown in Figure 3(a)For residual subspace SPE
119896of the PCAmodel approximately
obey 1205942 distribution [23 24] at time 119896
SPE119896120572= 1198921198961205942
ℎ119896120572
119892119896=
V119896
2119898119896
ℎ119896=2 (119898119896)2
V119896
(6)
where 119892 is a constant ℎ is the freedom degree of the 1205942distribution and V
119896and 119898
119896are respectively the mean and
variance of the square prediction error at time 119896 SPE controllimits are shown in Figure 3(b)
In order to monitor the rolling process first we needto obtain the measured data of new production periodstandardize the new data calculate the main component andthe prediction error of the data by formula (7) and checkwhether the 1198792 and SPE are beyond their own control limitIf the statistic exceeds the control limit it indicates that afault may occur at the time We now should analyze possiblecauses of the failure by variable 1198792 and SPE time-varyingcontribution plots and exclude or isolate the faults Consider
119905 = 119909119875lowast
119890 = 119909 (119868 minus 119875lowast
(119875lowast
)119879
)
1198792= 119905119879(119878lowast
)minus1
119905
SPE = 119890119890119879
(7)
Mathematical Problems in Engineering 5
0 50 100 150 200 250 300 350 40002468
1012141618
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 4 1198792 and SPE plot for MPCA monitoring results of the normal rolling process
As shown in Figure 4 under normal conditions moni-toring plot of SPE has an obvious alarm phenomenon duringthe first 50 sampling times and the last 50 sampling timesWecan explain the reason for alarm in the process of monitoringcombining with the rolling process The first 50 samples arein the bite stage where current speed and torque suddenlychange due to steel tube enter The last 50 samples are inthe steel leaving stage where speed current and torquedecrease sharply because of tube leaving Because of suddenchanges of variables standardized data still remain largedeviations Therefore PCA monitoring model appears alarmphenomenon So it is necessary to establish a multistageMPCA monitoring model
43 Build a Multistage MPCA Monitoring Model accordingto Production Process According to the process rollingproduction process can be divided into three stages bitestage stable rolling stage and steel leaving stage Then thispaper established a multistage MPCA monitoring modelaccording to the three stages
As shown in Figure 5 the situation of the steel leavingstage has improved But alarm phenomenon still exists andthe monitoring effect still needs to be improved So thebite stage and steel leaving stage should be further divided119870-means and Fuzzy 119862-means (FCM) clustering algorithmare commonly used FCM algorithm does not consider anyinformation related to the image space continuity so it ishighly sensitive to noise However 119870-means algorithm issimple and fast Particularly when dealing with the large datasets it has a very high efficiency So this paper chooses 119870-means clustering algorithm as a segmentation method
44 119870-Means and Multistage MPCA Combined to Build aMonitoring Model 119870-means algorithm is to cluster 119899 objectsbased on attributes into 119870(119870 lt 119899) partitions It assigns eachobject to the cluster which has the nearest center The centeris defined as the average of all the objects in the cluster whichstarts from a set of random initial centers The main steps of119870-means clustering algorithm is as follows
(1) Set up the cluster number 119870(2) Directly generate119870 random points as cluster centers(3) Assign each other points to the nearest cluster center(4) Recalculate the new cluster centers after new points
are clustered into the clusters(5) Repeat 3 and 4 until cluster centers do not change
According to process and the 119870-means algorithm forsegmentation number of principal components retained ineach substage PCA model is 119860lowast
119888 which can be obtained by
formula (3) The whole PCA load matrix 119875lowast119888is divided into
two parts the main component space 119875lowast119888and the residual
space 119875lowast
119888 which can be obtained by formula (4) Similarly
eigenvalue diagonalmatrix 119878lowast119888is correspondingly divided into
two parts 119878lowast119888and 119878
lowast
119888 SPE control limits can be obtained by
formula (4) 1198792 control limits is defined as follows
119863119896sim
119860lowast
119888(1198992
stage 119888 minus 1)
119899stage 119888 (119899stage 119888 minus 119860lowast
119888)
119865119860lowast
119888119899stage 119888minus119860
lowast
119888120572 (8)
As shown in Figure 6 when the number of stages is 7(bite stage is divided into three stages no segmentation stablerolling stage and steel leaving stage is divided into threestages) 1198792 and SPE are beyond their own control limits Themodel has a good performance in monitoring
According to the contrast of the foregoing analysis wecan easily conclude that seamless tube continuous rollingproduction process has too many characteristics includ-ing multiperiod strong nonlinearity and quickly changingdynamic characteristics It is hard tomonitor production pro-cess by the traditional MPCA method In this paper we usemultistage MPCA method according to production processand clustering algorithm This method can solve nonlinearproblem of seamless tube rolling production process andimprove the accuracy of online monitoring At the sametime the method has a strong guiding significance for theproduction
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Mathematical Problems in Engineering
Displacement
21 7 8
Tim
e
Bite
stag
eSt
able
rolli
ng
stage
Stee
l le
avin
g sta
ge
Perio
d
0A
C
Dd
c
B
c998400
d998400
D998400
t1
t2
t3
A998400 aa998400
B998400
C998400
b998400 b
Figure 1 Time and displacement of rolling tube process
rolling process it can be divided into bite stage stable rollingstage and steel leaving stage The production data has thefollowing characteristics
(1) Large Amount of Data High Dimension and Strong Cou-pling Seamless tube rolling process is a computer supervisoryand computer control industrial process It need regularlyacquire system state variables and equipment status to be usedfor display and control Each cycle of seamless tube rollingproduction process will produce tens of thousands of processdata After accumulating the amount of data is enormousAt the same time the behavior of the rolling process is theoutcome of combining action of many variable factors whichhave a strong coupling relationship
(2) Industrial Noise andUncertainty Because systemworks incomplex environments the output signal of electronic sens-ing devices is susceptible to noise And it is also vulnerable tothe uncertainties
(3) Multimode Data is a reflection of the system statechanging The data are not only in normal working statebut also in various kinds of abnormal state and fault stateThe former is the main part The latter has a relatively smallamount of data but it is indispensable in the knowledgediscovery and data mining
(4) Staircase Distribution In the biting stage the tube entersrolls from the first to the eighth and in the steel leaving stagethe situation is in turn So the data is staircase distribution
In this paper by using 119870-means clustering algorithmcombinedwith production process we establish a continuousrolling production process monitoring and fault diagnosismodel based onmultistageMPCAmethod Firstly accordingto the production process we divide production data intothree subperiods including bite stage stable rolling stage andsteel leaving stage Secondly we use clustering algorithm tofurther divide those larger changing stages A satisfactory
effect is difficult to be obtained if using clustering algorithmto classify the production data alone We classify it combinedwith production process Finally we get a satisfactory moni-toring result Online monitoring and fault diagnosis systemnot only can online monitor equipment but also can carryout fault alarm and diagnosis timely During the monitoringprocess it does not require the dedicated testers and doesnot need professional and technical personnel to make ananalysis and judgment
2 Analysis of Factors Affecting the ContinuousRolling Production
In order to build the monitoring and fault diagnosis modelwe analyze the influencing factors for continuous rollingproduction firstly
As shown in Figure 1 the continuous rolling process canbe divided into three stages
(1) Bite StageAs shown in Figure 1 the head of steel tubemoves from point a1015840 to point b1015840 and the tail movesfrom point A1015840 to point B1015840
(2) Stable Rolling Stage As shown in Figure 1 the head ofsteel tube moves from point b1015840 to point c1015840 and the tailmoves from point B1015840 to point C1015840
(3) Steel Leaving Stage As shown in Figure 1 the head ofsteel tube moves from point c1015840 to point d1015840 and the tailmoves from point C1015840 to point D1015840
Factors affecting seamless steel rolling production processare mainly the following roller rotational speed roller inputcurrent and roller torque
(1) Roller Rotational Speed In continuous rolling process theroller rotational speed is very important If the speed is toofast the surface of shell cannot be completely eliminated
Mathematical Problems in Engineering 3
Batches
Time
Variable1
1
X
K
J
J
2J KJ
I
I
X
Figure 2 Arrangement and decomposition of a three-way array byMPCA
In addition if the speed is too slow the consumption ofelectricity and other energy will increase
(2) Roller Torque The roller torque is one of the controlledvariables It is controlled by the input current The torquedirectly affects the quality of steel tube If torque is too small itcannot completely remove the blank of tube surface If torqueis too large shell will deform under the pressure Reasonablycontrolled torque of the rolling process is an important factorwhich cannot be ignored In order to save energy resourcesduring the wait state the torque will be reduced When thetube is rolled the torque increases When the tube leaves theroll the torque decreases
(3) Roller Input Current Roller current is one of the mostimportant control variables in the process of rolling It isan important factor to control torque and rotational speedBecause the sensitivity of the current is higher when thepierced tube enters the stand mill current increases quicklyThen the roller rotational speed and roller torque increase inorder to roll smoothly Current and torque are proportionalto the relationship The current increases when the steel tubegets into the rollThen it tends to a stable stateWhen the tubeleaves the roll current drops rapidly So correctly controllingthe size and direction of current is an important part of theprocess of rolling and the current is themain control variablein the rolling process
3 Multiway Principal Component Analysis
Batch processes are repetitive production process Theirdata sets have one more dimension than the continuousproduction process data set We can use three-dimensionaldata matrix 119883 = (119868 times 119869 times 119870) instead of the batch processdata collection where the three dimensions 119868 119869 and 119870respectively represent the batch number of samples numberof process variables and the number of measuring points ineach operation [19 20] MPCA will unfold119883 = (119868 times 119869 times119870) in
such a way as to put each of its vertical slices (119868 times 119869) side byside to the right which start with the one corresponding tothe first time interval The resulting two-dimensional matrixhas dimensions119883(119868 times 119869119870) [21 22] (Figure 2)
After three-dimensional data matrix 119883 is expanded intotwo-dimensional data it will be decomposed into a seriesof principal components consisting of score vectors 119905
119895and
loading matrices 119875119895 together with a residual matrix 119864 by the
principles of PCA The MPCA model can be written as
119883 =
119896
sum
119895=1
119905119895otimes 119901119895+ 119864 (1)
where the score vectors 119905119895is related only to batches and the
loading matrices 119875119895is related to variables and their time
variation The noise or residual part 119864 is as small as possiblein a least squares sense
4 Establish a Multiperiod ContinuousRolling Production Process MonitoringModel Based on 119870-Means and MPCA
41 Collecting the Production Data Based on actual produc-tion data characteristics of seamless steel rolling process inthis paper we use ibaAnalyzer software to collect 20 dataunder normal production condition As shown in Table 1 weselect 24 production process indicator variables to establish amonitoring model In this paper the three-dimensional datamatrix is 119883(119868 times 119869 times 119870) The three dimensions respectivelyrepresent the batch number of samples number of processvariables and the number of measuring points in eachoperation (119868 = 20 119869 = 24 119870 = 400) In order to obtainvertical data slice 119883
119896(119868 times 119869) we cut the three-dimensional
matrix along the direction of the third dimension Thusduring a period we can get 400 time slice matrixes By usingPCA for the 400 two-dimensional time slice matrix we got400 load matrixes and got the whole model load matrix bytaking an average
42 Online Monitoring Based on PCA Method The wholePCA model can be defined as follows
119875lowast=1
400
400
sum
119896=1
119875119896 (2)
where 119896 is the number of load matrixThe number of principal components 119860lowast can be calcu-
lated by the cumulative contribution rate In this paper weset 119860lowast = 6
The cumulative contribution rate =sum119860lowast
119895=1120582lowast
119895
trace (119878lowast)ge 90 (3)
where 119878lowast = diag(1205821 1205822 120582
119869) is an eigenvalues diagonal
matrix of the matrix119883The whole PCA load matrix 119875lowast is divided into two parts
the main component space 119875lowast(24 times 6) and the residual
4 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40014
145
15
155
16
165
Time
T2
(a) 1198792 control limits
0 50 100 150 200 250 300 350 40020
40
60
80
100
120
140
Time
SPE
(b) SPE control limits
Figure 3 Seamless tube rolling process 1198792 control limits and SPE control limits
Table 1 Measured variables for seamless steel rolling productionprocess
Symbol Variables Unit1 1st roller speed rs2 1st roller current A3 1st roller torque Nlowastm4 2nd roller speed rs5 2nd roller current A6 2nd roller torque Nlowastm7 3rd roller speed rs8 3rd roller current A9 3rd roller torque Nlowastm10 4th roller speed rs11 4th roller current A12 4th roller torque Nlowastm13 5th roller speed rs14 5th roller current A15 5th roller torque Nlowastm16 6th roller speed rs17 6th roller current A18 6th roller torque Nlowastm19 7th roller speed rs20 7th roller current A21 7th roller torque Nlowastm22 8th roller speed rs23 8th roller current A24 8th roller torque Nlowastm
space 119875lowast
(24 times 18) Similarly eigenvalue diagonal matrix 119878lowast
is correspondingly divided into two parts 119878lowast and 119878lowast
= 119883 (119875lowast
)119879
119883 = (119875
lowast
)119879
119864 = 119883 minus119883
(4)
In current online monitoring and fault diagnosis judgingwhether statistics1198792 and SPE are over the limit is usually usedto determine whether faults happen The control limits of 1198792approximately obey 119865 distribution
119863 sim
119860lowast(1198992minus 1)
119899 (119899 minus 119860lowast)119865119860lowast119899minus119860lowast120572 (5)
where 119870 = 400 and 119899 is the number of samples data formodeling The control limits of 1198792 is shown in Figure 3(a)For residual subspace SPE
119896of the PCAmodel approximately
obey 1205942 distribution [23 24] at time 119896
SPE119896120572= 1198921198961205942
ℎ119896120572
119892119896=
V119896
2119898119896
ℎ119896=2 (119898119896)2
V119896
(6)
where 119892 is a constant ℎ is the freedom degree of the 1205942distribution and V
119896and 119898
119896are respectively the mean and
variance of the square prediction error at time 119896 SPE controllimits are shown in Figure 3(b)
In order to monitor the rolling process first we needto obtain the measured data of new production periodstandardize the new data calculate the main component andthe prediction error of the data by formula (7) and checkwhether the 1198792 and SPE are beyond their own control limitIf the statistic exceeds the control limit it indicates that afault may occur at the time We now should analyze possiblecauses of the failure by variable 1198792 and SPE time-varyingcontribution plots and exclude or isolate the faults Consider
119905 = 119909119875lowast
119890 = 119909 (119868 minus 119875lowast
(119875lowast
)119879
)
1198792= 119905119879(119878lowast
)minus1
119905
SPE = 119890119890119879
(7)
Mathematical Problems in Engineering 5
0 50 100 150 200 250 300 350 40002468
1012141618
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 4 1198792 and SPE plot for MPCA monitoring results of the normal rolling process
As shown in Figure 4 under normal conditions moni-toring plot of SPE has an obvious alarm phenomenon duringthe first 50 sampling times and the last 50 sampling timesWecan explain the reason for alarm in the process of monitoringcombining with the rolling process The first 50 samples arein the bite stage where current speed and torque suddenlychange due to steel tube enter The last 50 samples are inthe steel leaving stage where speed current and torquedecrease sharply because of tube leaving Because of suddenchanges of variables standardized data still remain largedeviations Therefore PCA monitoring model appears alarmphenomenon So it is necessary to establish a multistageMPCA monitoring model
43 Build a Multistage MPCA Monitoring Model accordingto Production Process According to the process rollingproduction process can be divided into three stages bitestage stable rolling stage and steel leaving stage Then thispaper established a multistage MPCA monitoring modelaccording to the three stages
As shown in Figure 5 the situation of the steel leavingstage has improved But alarm phenomenon still exists andthe monitoring effect still needs to be improved So thebite stage and steel leaving stage should be further divided119870-means and Fuzzy 119862-means (FCM) clustering algorithmare commonly used FCM algorithm does not consider anyinformation related to the image space continuity so it ishighly sensitive to noise However 119870-means algorithm issimple and fast Particularly when dealing with the large datasets it has a very high efficiency So this paper chooses 119870-means clustering algorithm as a segmentation method
44 119870-Means and Multistage MPCA Combined to Build aMonitoring Model 119870-means algorithm is to cluster 119899 objectsbased on attributes into 119870(119870 lt 119899) partitions It assigns eachobject to the cluster which has the nearest center The centeris defined as the average of all the objects in the cluster whichstarts from a set of random initial centers The main steps of119870-means clustering algorithm is as follows
(1) Set up the cluster number 119870(2) Directly generate119870 random points as cluster centers(3) Assign each other points to the nearest cluster center(4) Recalculate the new cluster centers after new points
are clustered into the clusters(5) Repeat 3 and 4 until cluster centers do not change
According to process and the 119870-means algorithm forsegmentation number of principal components retained ineach substage PCA model is 119860lowast
119888 which can be obtained by
formula (3) The whole PCA load matrix 119875lowast119888is divided into
two parts the main component space 119875lowast119888and the residual
space 119875lowast
119888 which can be obtained by formula (4) Similarly
eigenvalue diagonalmatrix 119878lowast119888is correspondingly divided into
two parts 119878lowast119888and 119878
lowast
119888 SPE control limits can be obtained by
formula (4) 1198792 control limits is defined as follows
119863119896sim
119860lowast
119888(1198992
stage 119888 minus 1)
119899stage 119888 (119899stage 119888 minus 119860lowast
119888)
119865119860lowast
119888119899stage 119888minus119860
lowast
119888120572 (8)
As shown in Figure 6 when the number of stages is 7(bite stage is divided into three stages no segmentation stablerolling stage and steel leaving stage is divided into threestages) 1198792 and SPE are beyond their own control limits Themodel has a good performance in monitoring
According to the contrast of the foregoing analysis wecan easily conclude that seamless tube continuous rollingproduction process has too many characteristics includ-ing multiperiod strong nonlinearity and quickly changingdynamic characteristics It is hard tomonitor production pro-cess by the traditional MPCA method In this paper we usemultistage MPCA method according to production processand clustering algorithm This method can solve nonlinearproblem of seamless tube rolling production process andimprove the accuracy of online monitoring At the sametime the method has a strong guiding significance for theproduction
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 3
Batches
Time
Variable1
1
X
K
J
J
2J KJ
I
I
X
Figure 2 Arrangement and decomposition of a three-way array byMPCA
In addition if the speed is too slow the consumption ofelectricity and other energy will increase
(2) Roller Torque The roller torque is one of the controlledvariables It is controlled by the input current The torquedirectly affects the quality of steel tube If torque is too small itcannot completely remove the blank of tube surface If torqueis too large shell will deform under the pressure Reasonablycontrolled torque of the rolling process is an important factorwhich cannot be ignored In order to save energy resourcesduring the wait state the torque will be reduced When thetube is rolled the torque increases When the tube leaves theroll the torque decreases
(3) Roller Input Current Roller current is one of the mostimportant control variables in the process of rolling It isan important factor to control torque and rotational speedBecause the sensitivity of the current is higher when thepierced tube enters the stand mill current increases quicklyThen the roller rotational speed and roller torque increase inorder to roll smoothly Current and torque are proportionalto the relationship The current increases when the steel tubegets into the rollThen it tends to a stable stateWhen the tubeleaves the roll current drops rapidly So correctly controllingthe size and direction of current is an important part of theprocess of rolling and the current is themain control variablein the rolling process
3 Multiway Principal Component Analysis
Batch processes are repetitive production process Theirdata sets have one more dimension than the continuousproduction process data set We can use three-dimensionaldata matrix 119883 = (119868 times 119869 times 119870) instead of the batch processdata collection where the three dimensions 119868 119869 and 119870respectively represent the batch number of samples numberof process variables and the number of measuring points ineach operation [19 20] MPCA will unfold119883 = (119868 times 119869 times119870) in
such a way as to put each of its vertical slices (119868 times 119869) side byside to the right which start with the one corresponding tothe first time interval The resulting two-dimensional matrixhas dimensions119883(119868 times 119869119870) [21 22] (Figure 2)
After three-dimensional data matrix 119883 is expanded intotwo-dimensional data it will be decomposed into a seriesof principal components consisting of score vectors 119905
119895and
loading matrices 119875119895 together with a residual matrix 119864 by the
principles of PCA The MPCA model can be written as
119883 =
119896
sum
119895=1
119905119895otimes 119901119895+ 119864 (1)
where the score vectors 119905119895is related only to batches and the
loading matrices 119875119895is related to variables and their time
variation The noise or residual part 119864 is as small as possiblein a least squares sense
4 Establish a Multiperiod ContinuousRolling Production Process MonitoringModel Based on 119870-Means and MPCA
41 Collecting the Production Data Based on actual produc-tion data characteristics of seamless steel rolling process inthis paper we use ibaAnalyzer software to collect 20 dataunder normal production condition As shown in Table 1 weselect 24 production process indicator variables to establish amonitoring model In this paper the three-dimensional datamatrix is 119883(119868 times 119869 times 119870) The three dimensions respectivelyrepresent the batch number of samples number of processvariables and the number of measuring points in eachoperation (119868 = 20 119869 = 24 119870 = 400) In order to obtainvertical data slice 119883
119896(119868 times 119869) we cut the three-dimensional
matrix along the direction of the third dimension Thusduring a period we can get 400 time slice matrixes By usingPCA for the 400 two-dimensional time slice matrix we got400 load matrixes and got the whole model load matrix bytaking an average
42 Online Monitoring Based on PCA Method The wholePCA model can be defined as follows
119875lowast=1
400
400
sum
119896=1
119875119896 (2)
where 119896 is the number of load matrixThe number of principal components 119860lowast can be calcu-
lated by the cumulative contribution rate In this paper weset 119860lowast = 6
The cumulative contribution rate =sum119860lowast
119895=1120582lowast
119895
trace (119878lowast)ge 90 (3)
where 119878lowast = diag(1205821 1205822 120582
119869) is an eigenvalues diagonal
matrix of the matrix119883The whole PCA load matrix 119875lowast is divided into two parts
the main component space 119875lowast(24 times 6) and the residual
4 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40014
145
15
155
16
165
Time
T2
(a) 1198792 control limits
0 50 100 150 200 250 300 350 40020
40
60
80
100
120
140
Time
SPE
(b) SPE control limits
Figure 3 Seamless tube rolling process 1198792 control limits and SPE control limits
Table 1 Measured variables for seamless steel rolling productionprocess
Symbol Variables Unit1 1st roller speed rs2 1st roller current A3 1st roller torque Nlowastm4 2nd roller speed rs5 2nd roller current A6 2nd roller torque Nlowastm7 3rd roller speed rs8 3rd roller current A9 3rd roller torque Nlowastm10 4th roller speed rs11 4th roller current A12 4th roller torque Nlowastm13 5th roller speed rs14 5th roller current A15 5th roller torque Nlowastm16 6th roller speed rs17 6th roller current A18 6th roller torque Nlowastm19 7th roller speed rs20 7th roller current A21 7th roller torque Nlowastm22 8th roller speed rs23 8th roller current A24 8th roller torque Nlowastm
space 119875lowast
(24 times 18) Similarly eigenvalue diagonal matrix 119878lowast
is correspondingly divided into two parts 119878lowast and 119878lowast
= 119883 (119875lowast
)119879
119883 = (119875
lowast
)119879
119864 = 119883 minus119883
(4)
In current online monitoring and fault diagnosis judgingwhether statistics1198792 and SPE are over the limit is usually usedto determine whether faults happen The control limits of 1198792approximately obey 119865 distribution
119863 sim
119860lowast(1198992minus 1)
119899 (119899 minus 119860lowast)119865119860lowast119899minus119860lowast120572 (5)
where 119870 = 400 and 119899 is the number of samples data formodeling The control limits of 1198792 is shown in Figure 3(a)For residual subspace SPE
119896of the PCAmodel approximately
obey 1205942 distribution [23 24] at time 119896
SPE119896120572= 1198921198961205942
ℎ119896120572
119892119896=
V119896
2119898119896
ℎ119896=2 (119898119896)2
V119896
(6)
where 119892 is a constant ℎ is the freedom degree of the 1205942distribution and V
119896and 119898
119896are respectively the mean and
variance of the square prediction error at time 119896 SPE controllimits are shown in Figure 3(b)
In order to monitor the rolling process first we needto obtain the measured data of new production periodstandardize the new data calculate the main component andthe prediction error of the data by formula (7) and checkwhether the 1198792 and SPE are beyond their own control limitIf the statistic exceeds the control limit it indicates that afault may occur at the time We now should analyze possiblecauses of the failure by variable 1198792 and SPE time-varyingcontribution plots and exclude or isolate the faults Consider
119905 = 119909119875lowast
119890 = 119909 (119868 minus 119875lowast
(119875lowast
)119879
)
1198792= 119905119879(119878lowast
)minus1
119905
SPE = 119890119890119879
(7)
Mathematical Problems in Engineering 5
0 50 100 150 200 250 300 350 40002468
1012141618
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 4 1198792 and SPE plot for MPCA monitoring results of the normal rolling process
As shown in Figure 4 under normal conditions moni-toring plot of SPE has an obvious alarm phenomenon duringthe first 50 sampling times and the last 50 sampling timesWecan explain the reason for alarm in the process of monitoringcombining with the rolling process The first 50 samples arein the bite stage where current speed and torque suddenlychange due to steel tube enter The last 50 samples are inthe steel leaving stage where speed current and torquedecrease sharply because of tube leaving Because of suddenchanges of variables standardized data still remain largedeviations Therefore PCA monitoring model appears alarmphenomenon So it is necessary to establish a multistageMPCA monitoring model
43 Build a Multistage MPCA Monitoring Model accordingto Production Process According to the process rollingproduction process can be divided into three stages bitestage stable rolling stage and steel leaving stage Then thispaper established a multistage MPCA monitoring modelaccording to the three stages
As shown in Figure 5 the situation of the steel leavingstage has improved But alarm phenomenon still exists andthe monitoring effect still needs to be improved So thebite stage and steel leaving stage should be further divided119870-means and Fuzzy 119862-means (FCM) clustering algorithmare commonly used FCM algorithm does not consider anyinformation related to the image space continuity so it ishighly sensitive to noise However 119870-means algorithm issimple and fast Particularly when dealing with the large datasets it has a very high efficiency So this paper chooses 119870-means clustering algorithm as a segmentation method
44 119870-Means and Multistage MPCA Combined to Build aMonitoring Model 119870-means algorithm is to cluster 119899 objectsbased on attributes into 119870(119870 lt 119899) partitions It assigns eachobject to the cluster which has the nearest center The centeris defined as the average of all the objects in the cluster whichstarts from a set of random initial centers The main steps of119870-means clustering algorithm is as follows
(1) Set up the cluster number 119870(2) Directly generate119870 random points as cluster centers(3) Assign each other points to the nearest cluster center(4) Recalculate the new cluster centers after new points
are clustered into the clusters(5) Repeat 3 and 4 until cluster centers do not change
According to process and the 119870-means algorithm forsegmentation number of principal components retained ineach substage PCA model is 119860lowast
119888 which can be obtained by
formula (3) The whole PCA load matrix 119875lowast119888is divided into
two parts the main component space 119875lowast119888and the residual
space 119875lowast
119888 which can be obtained by formula (4) Similarly
eigenvalue diagonalmatrix 119878lowast119888is correspondingly divided into
two parts 119878lowast119888and 119878
lowast
119888 SPE control limits can be obtained by
formula (4) 1198792 control limits is defined as follows
119863119896sim
119860lowast
119888(1198992
stage 119888 minus 1)
119899stage 119888 (119899stage 119888 minus 119860lowast
119888)
119865119860lowast
119888119899stage 119888minus119860
lowast
119888120572 (8)
As shown in Figure 6 when the number of stages is 7(bite stage is divided into three stages no segmentation stablerolling stage and steel leaving stage is divided into threestages) 1198792 and SPE are beyond their own control limits Themodel has a good performance in monitoring
According to the contrast of the foregoing analysis wecan easily conclude that seamless tube continuous rollingproduction process has too many characteristics includ-ing multiperiod strong nonlinearity and quickly changingdynamic characteristics It is hard tomonitor production pro-cess by the traditional MPCA method In this paper we usemultistage MPCA method according to production processand clustering algorithm This method can solve nonlinearproblem of seamless tube rolling production process andimprove the accuracy of online monitoring At the sametime the method has a strong guiding significance for theproduction
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40014
145
15
155
16
165
Time
T2
(a) 1198792 control limits
0 50 100 150 200 250 300 350 40020
40
60
80
100
120
140
Time
SPE
(b) SPE control limits
Figure 3 Seamless tube rolling process 1198792 control limits and SPE control limits
Table 1 Measured variables for seamless steel rolling productionprocess
Symbol Variables Unit1 1st roller speed rs2 1st roller current A3 1st roller torque Nlowastm4 2nd roller speed rs5 2nd roller current A6 2nd roller torque Nlowastm7 3rd roller speed rs8 3rd roller current A9 3rd roller torque Nlowastm10 4th roller speed rs11 4th roller current A12 4th roller torque Nlowastm13 5th roller speed rs14 5th roller current A15 5th roller torque Nlowastm16 6th roller speed rs17 6th roller current A18 6th roller torque Nlowastm19 7th roller speed rs20 7th roller current A21 7th roller torque Nlowastm22 8th roller speed rs23 8th roller current A24 8th roller torque Nlowastm
space 119875lowast
(24 times 18) Similarly eigenvalue diagonal matrix 119878lowast
is correspondingly divided into two parts 119878lowast and 119878lowast
= 119883 (119875lowast
)119879
119883 = (119875
lowast
)119879
119864 = 119883 minus119883
(4)
In current online monitoring and fault diagnosis judgingwhether statistics1198792 and SPE are over the limit is usually usedto determine whether faults happen The control limits of 1198792approximately obey 119865 distribution
119863 sim
119860lowast(1198992minus 1)
119899 (119899 minus 119860lowast)119865119860lowast119899minus119860lowast120572 (5)
where 119870 = 400 and 119899 is the number of samples data formodeling The control limits of 1198792 is shown in Figure 3(a)For residual subspace SPE
119896of the PCAmodel approximately
obey 1205942 distribution [23 24] at time 119896
SPE119896120572= 1198921198961205942
ℎ119896120572
119892119896=
V119896
2119898119896
ℎ119896=2 (119898119896)2
V119896
(6)
where 119892 is a constant ℎ is the freedom degree of the 1205942distribution and V
119896and 119898
119896are respectively the mean and
variance of the square prediction error at time 119896 SPE controllimits are shown in Figure 3(b)
In order to monitor the rolling process first we needto obtain the measured data of new production periodstandardize the new data calculate the main component andthe prediction error of the data by formula (7) and checkwhether the 1198792 and SPE are beyond their own control limitIf the statistic exceeds the control limit it indicates that afault may occur at the time We now should analyze possiblecauses of the failure by variable 1198792 and SPE time-varyingcontribution plots and exclude or isolate the faults Consider
119905 = 119909119875lowast
119890 = 119909 (119868 minus 119875lowast
(119875lowast
)119879
)
1198792= 119905119879(119878lowast
)minus1
119905
SPE = 119890119890119879
(7)
Mathematical Problems in Engineering 5
0 50 100 150 200 250 300 350 40002468
1012141618
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 4 1198792 and SPE plot for MPCA monitoring results of the normal rolling process
As shown in Figure 4 under normal conditions moni-toring plot of SPE has an obvious alarm phenomenon duringthe first 50 sampling times and the last 50 sampling timesWecan explain the reason for alarm in the process of monitoringcombining with the rolling process The first 50 samples arein the bite stage where current speed and torque suddenlychange due to steel tube enter The last 50 samples are inthe steel leaving stage where speed current and torquedecrease sharply because of tube leaving Because of suddenchanges of variables standardized data still remain largedeviations Therefore PCA monitoring model appears alarmphenomenon So it is necessary to establish a multistageMPCA monitoring model
43 Build a Multistage MPCA Monitoring Model accordingto Production Process According to the process rollingproduction process can be divided into three stages bitestage stable rolling stage and steel leaving stage Then thispaper established a multistage MPCA monitoring modelaccording to the three stages
As shown in Figure 5 the situation of the steel leavingstage has improved But alarm phenomenon still exists andthe monitoring effect still needs to be improved So thebite stage and steel leaving stage should be further divided119870-means and Fuzzy 119862-means (FCM) clustering algorithmare commonly used FCM algorithm does not consider anyinformation related to the image space continuity so it ishighly sensitive to noise However 119870-means algorithm issimple and fast Particularly when dealing with the large datasets it has a very high efficiency So this paper chooses 119870-means clustering algorithm as a segmentation method
44 119870-Means and Multistage MPCA Combined to Build aMonitoring Model 119870-means algorithm is to cluster 119899 objectsbased on attributes into 119870(119870 lt 119899) partitions It assigns eachobject to the cluster which has the nearest center The centeris defined as the average of all the objects in the cluster whichstarts from a set of random initial centers The main steps of119870-means clustering algorithm is as follows
(1) Set up the cluster number 119870(2) Directly generate119870 random points as cluster centers(3) Assign each other points to the nearest cluster center(4) Recalculate the new cluster centers after new points
are clustered into the clusters(5) Repeat 3 and 4 until cluster centers do not change
According to process and the 119870-means algorithm forsegmentation number of principal components retained ineach substage PCA model is 119860lowast
119888 which can be obtained by
formula (3) The whole PCA load matrix 119875lowast119888is divided into
two parts the main component space 119875lowast119888and the residual
space 119875lowast
119888 which can be obtained by formula (4) Similarly
eigenvalue diagonalmatrix 119878lowast119888is correspondingly divided into
two parts 119878lowast119888and 119878
lowast
119888 SPE control limits can be obtained by
formula (4) 1198792 control limits is defined as follows
119863119896sim
119860lowast
119888(1198992
stage 119888 minus 1)
119899stage 119888 (119899stage 119888 minus 119860lowast
119888)
119865119860lowast
119888119899stage 119888minus119860
lowast
119888120572 (8)
As shown in Figure 6 when the number of stages is 7(bite stage is divided into three stages no segmentation stablerolling stage and steel leaving stage is divided into threestages) 1198792 and SPE are beyond their own control limits Themodel has a good performance in monitoring
According to the contrast of the foregoing analysis wecan easily conclude that seamless tube continuous rollingproduction process has too many characteristics includ-ing multiperiod strong nonlinearity and quickly changingdynamic characteristics It is hard tomonitor production pro-cess by the traditional MPCA method In this paper we usemultistage MPCA method according to production processand clustering algorithm This method can solve nonlinearproblem of seamless tube rolling production process andimprove the accuracy of online monitoring At the sametime the method has a strong guiding significance for theproduction
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 5
0 50 100 150 200 250 300 350 40002468
1012141618
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 4 1198792 and SPE plot for MPCA monitoring results of the normal rolling process
As shown in Figure 4 under normal conditions moni-toring plot of SPE has an obvious alarm phenomenon duringthe first 50 sampling times and the last 50 sampling timesWecan explain the reason for alarm in the process of monitoringcombining with the rolling process The first 50 samples arein the bite stage where current speed and torque suddenlychange due to steel tube enter The last 50 samples are inthe steel leaving stage where speed current and torquedecrease sharply because of tube leaving Because of suddenchanges of variables standardized data still remain largedeviations Therefore PCA monitoring model appears alarmphenomenon So it is necessary to establish a multistageMPCA monitoring model
43 Build a Multistage MPCA Monitoring Model accordingto Production Process According to the process rollingproduction process can be divided into three stages bitestage stable rolling stage and steel leaving stage Then thispaper established a multistage MPCA monitoring modelaccording to the three stages
As shown in Figure 5 the situation of the steel leavingstage has improved But alarm phenomenon still exists andthe monitoring effect still needs to be improved So thebite stage and steel leaving stage should be further divided119870-means and Fuzzy 119862-means (FCM) clustering algorithmare commonly used FCM algorithm does not consider anyinformation related to the image space continuity so it ishighly sensitive to noise However 119870-means algorithm issimple and fast Particularly when dealing with the large datasets it has a very high efficiency So this paper chooses 119870-means clustering algorithm as a segmentation method
44 119870-Means and Multistage MPCA Combined to Build aMonitoring Model 119870-means algorithm is to cluster 119899 objectsbased on attributes into 119870(119870 lt 119899) partitions It assigns eachobject to the cluster which has the nearest center The centeris defined as the average of all the objects in the cluster whichstarts from a set of random initial centers The main steps of119870-means clustering algorithm is as follows
(1) Set up the cluster number 119870(2) Directly generate119870 random points as cluster centers(3) Assign each other points to the nearest cluster center(4) Recalculate the new cluster centers after new points
are clustered into the clusters(5) Repeat 3 and 4 until cluster centers do not change
According to process and the 119870-means algorithm forsegmentation number of principal components retained ineach substage PCA model is 119860lowast
119888 which can be obtained by
formula (3) The whole PCA load matrix 119875lowast119888is divided into
two parts the main component space 119875lowast119888and the residual
space 119875lowast
119888 which can be obtained by formula (4) Similarly
eigenvalue diagonalmatrix 119878lowast119888is correspondingly divided into
two parts 119878lowast119888and 119878
lowast
119888 SPE control limits can be obtained by
formula (4) 1198792 control limits is defined as follows
119863119896sim
119860lowast
119888(1198992
stage 119888 minus 1)
119899stage 119888 (119899stage 119888 minus 119860lowast
119888)
119865119860lowast
119888119899stage 119888minus119860
lowast
119888120572 (8)
As shown in Figure 6 when the number of stages is 7(bite stage is divided into three stages no segmentation stablerolling stage and steel leaving stage is divided into threestages) 1198792 and SPE are beyond their own control limits Themodel has a good performance in monitoring
According to the contrast of the foregoing analysis wecan easily conclude that seamless tube continuous rollingproduction process has too many characteristics includ-ing multiperiod strong nonlinearity and quickly changingdynamic characteristics It is hard tomonitor production pro-cess by the traditional MPCA method In this paper we usemultistage MPCA method according to production processand clustering algorithm This method can solve nonlinearproblem of seamless tube rolling production process andimprove the accuracy of online monitoring At the sametime the method has a strong guiding significance for theproduction
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 40002468
101214161820
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 5 1198792 and SPE plot for multistage MPCA monitoring results of the normal rolling process
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(a) 1198792 monitoring plot
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot
Figure 6 1198792 and SPE plot for 7 stages monitoring results of the normal rolling process
Through the experiment we find that the number ofsegments is larger and the precision is higher But at thesame time it easily causes misjudgment In this paper afterseveral experiments we finally decided to select 119870 = 7which both can be very good for rolling production processmonitoring and avoid misjudgment Compared with thesituation without segmentation segmentation model canmore accurately judge running state of rolling process whichhas a positive meaning for actual production
5 Fault Diagnosis
By monitoring the test method based on statistic can onlymonitorwhether faults occur and the approximate time of theoccurrence but it could not determine the source of the faultThe method of contribution plot provides possibilities ofdetermining the fault sources It can reflect the contributionto the statistics from variables at each moment
For the main component and residual subspace there aretwo contribution plots that can be used for fault diagnosismdash1198792 contribution plot and SPE contribution plot
The contribution to the 119886th principal component 119905119886from
119895th process variable 119909119895can be defined as follows
119862119905119886119909119895
=
119909119895119901119895119886
119905119886
(119886 = 1 119860 119895 = 1 119898) (9)
The contribution to the statistic SPE from the 119895th processvariables is
119862SPE119909119895
=
(119909119895minus 119909119895)2
SPE
(10)
In order to verify the performance of the multistagemonitoring model this paper introduces two typical faultdata for monitoring and fault diagnosis
Fault 1 1st roller speed fault from75th to 125th sampling timethe roller speed is 0
Fault 2 1st roller current fault from 70th to 130th samplingtime the roller current is 0
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 7
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 7 1198792 monitoring plots of fault 1
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 8 SPE monitoring plots of fault 1
51 Fault Diagnosis for the 1st Roller Speed As shown inFigure 7 for fault 1 monitoring plots of SPE have an obviousalarm phenomenon which 1198792 monitoring plots do nothave For comparison 1198792 contribution plot are still drawntogether with the SPE contribution plot In order to diagnosethe cause of the fault this paper respectively drew maincomponent contribution plots 1198792 contribution plots andSPE contribution plots of 60th 120 and 240th samplingtime As shown in Figure 8(a) MPCA model does not havean obvious alarm phenomenon in the fault time As shownin Figure 8(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 9 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the firstprincipal component from process variables According tocontribution rate from each variable to1198792 and SPE this paper
studied and determined the fault sources They are shown inFigures 10 and 11
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 10does not detect the fault variable As shown in Figure 11 inthe fault time the first variable (1st roller speed) has largercontribution rate to the first principal component From theresults we can see that SPE monitoring plots have an obviousalarm phenomenon According to the SPE contribution ratewe can diagnose the fault So the proposedmethod is correct
52 Fault Diagnosis for the 1st Roller Current As shownin Figure 12 for fault 2 monitoring plots of SPE have anobvious alarm phenomenon which 1198792 monitoring plots donot have For comparison 1198792 contribution plots are stilldrawn together with the SPE contribution plot In order todiagnose the cause of the fault this paper respectively drawsmain component contribution plots 1198792 contribution plotsand SPE contribution plots of 50th 125 and 250th samplingtime As shown in Figure 13(a) MPCA model does not have
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Mathematical Problems in Engineering
1 2 3 4 50
00501
01502
02503
03504
04505
Principal components
Con
trib
utio
n
(a) Time = 60
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(b) Time = 120
1 2 3 4 50
0102030405060708
Principal components
Con
trib
utio
n
(c) Time = 240
Figure 9 Principal component contribution plots of fault 1
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 60
0 5 10 15 20 25
001020304
Variables
Con
trib
utio
n
minus04
minus03
minus02
minus01
(b) Time = 120
0 5 10 15 20 25
0
05
1
15
2
Variables
Con
trib
utio
n
minus05
minus1
(c) Time = 240
Figure 10 1198792 contribution plots of fault 1
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 9
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 60
0 5 10 15 20 250
002004006008
01012014016018
02
Variables
Con
trib
utio
n
(b) Time = 120
0 5 10 15 20 250
005
01
015
02
025
03
035
Variables
Con
trib
utio
n
(c) Time = 240
Figure 11 SPE contribution plots of fault 1
0 50 100 150 200 250 300 350 40002468
10121416
Time
Control limits
T2
T2
(a) 1198792 monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
Time
Control limits
T2
T2
(b) 1198792 monitoring plot for 7 stages MPCA
Figure 12 1198792 monitoring plots of fault 2
an obvious alarm phenomenon in the fault time As shownin Figure 13(b) multistage MPCA model can quickly andaccurately detect the fault
As shown in Figure 14 the contribution rate of eachprincipal component is not the same at different time Nearthe fault time the first principal component contributionrate was larger Away from the fault time the first principalcomponent contribution rate is less than the second principalcomponent This paper analyzes contribution rate to the first
principal component from process variables According tocontribution rate of each variable this paper studies anddetermines the fault sources They are shown in Figures 15and 16
Because 1198792 monitoring plots do not have an obviousalarm phenomenon 1198792 contribution plot shown in Figure 15does not detect the fault variable As shown in Figure 16 in thefault time the second variable (1st roller current) has largercontribution rate to the first principal component From
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Mathematical Problems in Engineering
0 50 100 150 200 250 300 350 4000
20
40
60
80
100
120
140
Time
SPE
SPEControl limits
(a) SPE monitoring plot for MPCA
0 50 100 150 200 250 300 350 4000
102030405060708090
100
Time
SPE
SPEControl limits
(b) SPE monitoring plot for 7 stages MPCA
Figure 13 SPE monitoring plots of fault 2
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(a) Time = 50
1 2 3 4 50
01020304050607
Principal components
Con
trib
utio
n08
(b) Time = 125
1 2 3 4 50
01
02
03
04
05
06
07
Principal components
Con
trib
utio
n
(c) Time = 250
Figure 14 Principal component contribution plots of fault 2
the results we can see that SPEmonitoring plots have an obvi-ous alarm phenomenon According to the SPE contributionrate we can diagnose the fault So the proposed method iscorrect
6 Conclusions
According to strong nonlinearity and dynamic property ofthe seamless tube continuous rolling production process this
paper divides production data into subperiods by 119870-meansclustering algorithm combined with production processThen we establish a continuous rolling production processmonitoring and fault diagnosis model based on multistageMPCAmethodThe results have shown that themodel devel-oped in this paper has better performances in monitoringand fault diagnosis Meanwhile the proposed method can beextended to the other industry processes
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 11
0 5 10 15 20 25
0
02
04
06
08
Variables
Con
trib
utio
n
minus06
minus04
minus02
(a) Time = 50
0 5 10 15 20 25
0005
01015
02025
03
Variables
Con
trib
utio
n
minus015
minus01
minus005
(b) Time = 125
0 5 10 15 20 25
0
05
1
Con
trib
utio
n
Variables
minus05
(c) Time = 250
Figure 15 1198792 contribution plots of fault 2
0 5 10 15 20 250
002
004
006
008
01
012
Variables
Con
trib
utio
n
(a) Time = 50
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(b) Time = 125
0 5 10 15 20 250
002004006008
01012014016
Variables
Con
trib
utio
n
(c) Time = 250
Figure 16 SPE contribution plots of fault 2
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by National Natural Science Foun-dation of China (Grant no 61203214) Provincial Science andTechnology Department of Education projects the generalproject (L2013101) and National Natural Science Foundationof China (Grant nos 41371437 61403277 61304121 6147307261203103 and 61374146)
References
[1] D Xiao J C Wang and H X Tian ldquoQuality prediction andcontrol of reducing pipe based on EOS-ELM-RPLS mathemat-ics modeling methodrdquo Journal of Applied Mathematics vol2014 Article ID 298218 13 pages 2014
[2] D Levesque S E Kruger G Lamouche et al ldquoThickness andgrain size monitoring in seamless tube-making process usinglaser ultrasonicsrdquoNDTampE International vol 39 no 8 pp 622ndash626 2006
[3] Y Lv Y R Li and H G Wang ldquoOn-line monitoring and faultdiagnosis for mill drivesrdquoHeavyMachinery vol 2002 no 2 pp56ndash59 2002
[4] M Reggio F McKenty L Gravel J Cortes G Morales andM-A Ladron de Guevara ldquoComputational analysis of theprocess for manufacturing seamless tubesrdquo Applied ThermalEngineering vol 22 no 4 pp 459ndash470 2002
[5] H Jin Fault Diagnosis of Large DC Motor Strip Mill Universityof Science and Technology of China 2000
[6] L Lei and H L Zhang ldquoContinuous rolling-tube unit on-linesupervision system based on virtual instrument technologyrdquoNatural Science Journal of Xiangtan University vol 26 no 1 pp102ndash105 2004
[7] C H Zhao and F R Gao ldquoFault-relevant Principal ComponentAnalysis (FPCA) method for multivariate statistical modelingand process monitoringrdquo Chemometrics and Intelligent Labora-tory Systems vol 133 no 1 pp 1ndash12 2014
[8] J Yu ldquoLocal and nonlocal preserving projection for bearingdefect classification and performance assessmentrdquo IEEE Trans-actions on Industrial Electronics vol 59 no 5 pp 2363ndash23762012
[9] B Zhang C Sconyers C Byington R Patrick M E Orchardand G Vachtsevanos ldquoA probabilistic fault detection approachapplication to bearing fault detectionrdquo IEEE Transactions onIndustrial Electronics vol 58 no 5 pp 2011ndash2018 2011
[10] S Yin S X Ding A Haghani H Hao and P Zhang ldquoA com-parison study of basic data-driven fault diagnosis and processmonitoring methods on the benchmark Tennessee Eastmanprocessrdquo Journal of Process Control vol 22 no 9 pp 1567ndash15812012
[11] S X Ding ldquoData-driven design of monitoring and diagnosissystems for dynamic processes a review of subspace techniquebased schemes and some recent resultsrdquo Journal of ProcessControl vol 24 no 2 pp 431ndash449 2014
[12] Y Lei J Lin Z He and M J Zuo ldquoA review on empiricalmode decomposition in fault diagnosis of rotating machineryrdquo
Mechanical Systems and Signal Processing vol 35 no 1-2 pp108ndash126 2013
[13] S Bedoui R FalehH Samet andAKachouri ldquoElectronic nosesystem and principal component analysis technique for gasesidentificationrdquo in Proceedings of the 10th International Multi-Conference on Systems Signals amp Devices (SSD rsquo13) pp 1ndash6IEEE Hammamet Tunisia March 2013
[14] G Georgoulas M O Mustafa I P Tsoumas et al ldquoPrincipalComponent Analysis of the start-up transient and HiddenMarkov Modeling for broken rotor bar fault diagnosis inasynchronous machinesrdquo Expert Systems with Applications vol40 no 17 pp 7024ndash7033 2013
[15] M Evans and J Kennedy ldquoIntegration of adaptive neuro fuzzyinference systems and principal component analysis for thecontrol of tertiary scale formation on tinplate at a hot millrdquoExpert Systems with Applications vol 41 no 15 pp 6662ndash66752014
[16] C M Ringle M Sarstedt R Schlittgen and C R TaylorldquoPLS path modeling and evolutionary segmentationrdquo Journal ofBusiness Research vol 66 no 9 pp 1318ndash1324 2013
[17] E M Abdel-Rahman O Mutanga J Odindi E Adam AOdindo and R Ismail ldquoA comparison of partial least squares(PLS) and sparse PLS regressions for predicting yield of Swisschard grown under different irrigation water sources usinghyperspectral datardquo Computers and Electronics in Agriculturevol 106 pp 11ndash19 2014
[18] B Lu I Castillo L Chiang and T F Edgar ldquoIndustrial PLSmodel variable selection using moving window variable impor-tance in projectionrdquo Chemometrics and Intelligent LaboratorySystems vol 135 no 15 pp 90ndash109 2014
[19] N Lu F Gao and F Wang ldquoSub-PCA modeling and on-linemonitoring strategy for batch processesrdquoAIChE Journal vol 50no 1 pp 255ndash259 2004
[20] X-T Doan R Srinivasan P M Bapat and P P WangikarldquoDetection of phase shifts in batch fermentation via statisti-cal analysis of the online measurements a case study withrifamycin B fermentationrdquo Journal of Biotechnology vol 132 no2 pp 156ndash166 2007
[21] M Emparan R Simpson S Almonacid A Teixeira and AUrtubia ldquoEarly recognition of problematic wine fermentationsthrough multivariate data analysesrdquo Food Control vol 27 no 1pp 248ndash253 2012
[22] W Dong Y Yao and F Gao ldquoPhase analysis and identificationmethod formultiphase batch processes with partitioningmulti-way principal component analysis (MPCA) modelrdquo ChineseJournal of Chemical Engineering vol 20 no 6 pp 1121ndash11272012
[23] PNomikos and J FMacGregor ldquoMulti-way partial least squaresin monitoring batch processesrdquo Chemometrics and IntelligentLaboratory Systems vol 30 no 1 pp 97ndash108 1995
[24] Y S Qi P Wang X J Gao and Y J Gong ldquoBatch processmonitoring and fault diagnosis based on improved multi-wayprincipal component analysisrdquo CIESC Journal vol 85 no 1 pp82ndash93 2009
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of