Research ArticleDiagnosis of Elevator Faults with LS-SVM Based onOptimization by K-CV

Zhou Wan1 Shilin Yi1 Kun Li1 Ran Tao2 Min Gou1 Xinshi Li2 and Shu Guo2

1Kunming University of Science and Technology Kunming Yunnan 650500 China2Yunnan Special Equipment Safety Inspection and Research Institute Kunming Yunnan China

Correspondence should be addressed to Kun Li ghfighter163com

Received 30 May 2015 Revised 10 November 2015 Accepted 15 November 2015

Academic Editor William Sandham

Copyright copy 2015 Zhou Wan 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

Several common elevator malfunctions were diagnosed with a least square support vector machine (LS-SVM) After acquiringvibration signals of various elevator functions their energy characteristics and time domain indicators were extracted bytheoretically analyzing the optimal wavelet packet in order to construct a feature vector of malfunctions for identifying causesof the malfunctions as input of LS-SVM Meanwhile parameters about LS-SVM were optimized by K-fold cross validation (K-CV) After diagnosing deviated elevator guide rail deviated shape of guide shoe abnormal running of tractor erroneous ropegroove of traction sheave deviated guide wheel and tension of wire rope the results suggested that the LS-SVM based on K-CVoptimization was one of effective methods for diagnosing elevator malfunctions

1 Introduction

With the development of modern society elevators haveachieved rapid development as tools for transporting thingsup and down inside high-rise buildings While enjoyingexpress elevator services people have proposed more rigor-ous requirements for taking elevators comfortably and safely[1] Major indicators [2] impacting if elevators can be takenwith comfort include vibration of elevation compartmentsnoise temperature decoration and starting and brakingcharacteristics among which the vibration is a major indica-tor for evaluating whether elevators can be taken comfortably[3] Under normal circumstances passengers will not feeluncomfortable in taking an elevator in case of relativelysmaller vibration amplitude However passengers will have afeeling of great discomfort when the vibration reaches certainvalue or the frequency is up to a level to which people aresensitive [4] To guarantee passengersrsquo physical and mentalhealth as well as personal safety it is of great significance forpassengers to take elevators more comfortably and safely byreducing vibration finding out vibration source predictingmalfunctions and recovering them promptly [5]

In case of malfunctions vibration signals in elevators willbecome remarkably nonstationarywith noiseWavelet packetanalysis has been widely recognized and applied in diagnos-ing failures of machines especially in processing transientsignals because it is highly effective for localizing timefrequency and has unique advantages in processing time-varying signals When an elevator malfunctions changes tothe vibration signals contain abundant information aboutcharacteristics of malfunctions [6ndash8] In combination withnoise elevator vibration signals were decomposed at differentfrequency bands by wavelet packet transform in order todetermine the energy of signals distributed in each sub-space The energy distribution was then compared withthat of vibration signals in normally running system toexact information about characteristics of elevator malfunc-tions [9] Besides kurtosis of vibration signals in 119885-axisand peak-to-peak values of vibration signals in 119883119884-axeswere taken as characteristic parameters in time domainIn this way vibration characteristics could be extracted formalfunctions

Support vectormachine (SVM) has been extensively usedand developed in the field of fault diagnosis Tang et al

Hindawi Publishing CorporationJournal of Electrical and Computer EngineeringVolume 2015 Article ID 935038 8 pageshttpdxdoiorg1011552015935038

2 Journal of Electrical and Computer Engineering

[10] diagnosed faults of rotating machines by LIttlewood-Paley wavelet support vector machines (LPWSVM) Shuixiaet al [11] achieved promising outcomes in classifying elevatormalfunctions by diagnosing faults with kernel principal com-ponent analysis (KPCA) and SVM Li et al [12] put forwarda model for fault diagnosis based on a genetic algorithmfor hierarchically optimizing least squares support vectormachines (LS-SVM) which could increase the precisionof LS-SVM for predicting faults and improve self-adaptivediagnosis Widodo and Yang [13] diagnosed faults of asyn-chronousmotors by extracting nonlinear characteristics withSVM and achieved ideal results In this paper an attempt wasmade to optimize parameters about SVM (mainly referring topenalty parameter 119888 and kernel function 119892) via cross valida-tion by integrating theories on optimal wavelet packet withSVM so as to effectively avoid overfitting and underfittingIn other words vibration signals were analyzed accordingto theories on optimal wavelet packet By using SVM asclassifier energy characteristics and time domain indicatorsof vibration signals were extracted to construct fault featurevectors They were adopted as input of SVM and elevatormalfunctions were classified with a well-trained SVM

2 On Safety for TakingElevators and Mechanism

21 On Safety for Taking Elevators Elevator compartmentsmay vibrate vertically or horizontally [1]Major causes of theirvertical vibration include vibration incurred by running atractor vibration of gearmeshing inside a speed reducermis-alignment between axes of worm gear reducer and tractionmotor and insecure fastening of tractor base with the bearingbeam and nonuniform load of wire rope All of these factorsimpact vertical vibration of elevator compartments Horizon-tal vibration of elevator compartments is mainly caused by[14] vibration resulting from guiding system out of toleranceperpendicularity of guide rails out of tolerance of distancebetween guide rails and out of tolerance between local gapof guide rails and steps at joints partial distortion of guiderails and straightness error of working surface of guide railsThese factors possibly lead compartments to horizontallyvibrate in the course of up-and-down motion Additionallystatic balance of compartments is influenced by deviatedsuspension center and eccentric load of compartments As aconsequence horizontal vibration is affected

22 On Vibration and Mechanism of Elevator Compart-ments Although there are numerous causes of compartmentvibration after analysis and investigation the main causes[15ndash18] are concluded to be deviated guide rail deviatedshape of guide shoe abnormal running of tractor erroneousrope groove of traction sheave deviated guide wheel andnonuniform tension of wire rope and so on

221 Deviated Guide Rail Elevator compartment runs byattachment to guide rail so horizontal vibration of an elevatoris directly impacted by its guide rail Asymmetric guide railgroove undesirable perpendicularity of guide rail widenedor narrowed distance between guide rails and fastened boltsof guide rail brackets or clips and so on cause horizontalvibration of compartments

222 Deviated Shape of Guide Shoe Attaching to guiderails guide shoe is used to limit obliquity and horizontaldisplacement It will be ineffective for reducing vibrationif guide rails are adjusted to be excessively tight when anelevator is running Consequently the resistance for runningcompartments gets higher and leads to vibration Besides theguide shoes will become inelastic once the gap between themis regulated to be too large and thereby cause compartmentsto vibrate during working

223 Abnormal Running of Tractor High-speed shaft-drivenimbalance and misaligned vibration of coupling are majorcauses of tractor vibration As a tractor is rotating at a highspeed the pulse becomes an excitation source for diagnosingcompartments Errors will be caused to rope grooves andresult in inaccurate dynamic balance after the tractor is wornfor a long period In addition vibrator compartments willvibrate during running provided that speed measurementencoder is not well connected with motor or misaligned

224 Deviated Guide Wheel The deviation of guide wheelfrom the perpendicular line will be higher than 2mm whenthe wheel is empty or fully loaded The protection shallmeet given requirements when suspended traction wheel orsprocket wheel is used To be exact geneva wheel shall not beseriously and nonuniformly worn to a level that its shape ischanged or else compartments will vibrate abnormally

225 Nonuniform Tension of Wire Rope Nonuniform ten-sion of wire rope leads to unequal specific pressure ontraction rope inside the grove of pulley wheel Compartmentwill abnormally vibrate due to relative slip of rope caused byjoint difference because the rope of traction wheel is wornmore quickly in case of great force [14]

3 Signal Processing and Analysis

In this paper data on elevators provided by Yunnan SpecialEquipment Safety Inspection and Research Institute wereanalyzed The data was acquired by EVA-625 elevator testerBecause the data measured about elevator running werelimited for better diagnosis simulative samples were com-bined with actual samples of malfunctions to establish asample database necessary for the training of support vectormachines Firstly the Matlab extracts the data distributioncharacteristics (almost the same as authentic samples) ofauthentic fault samples as well as their mean value vari-ance standard deviation extremum moment skewness andkurtosis based on which it randomly generates 100 groupsof simulative samples Integrating simulative samples withauthentic fault samples it establishes the data base needed forsupport vector machine (SVM) 70 out of those samples areused for training and another 70 for testing

31 EVA-625 Elevator Tester

311 EVA-625 Introduction Aimed to record the vibrationsand noises of elevators EVA-625 elevator tester is a par-ticularly designed high-precision recorder of acceleration

Journal of Electrical and Computer Engineering 3

(s)0 5 10 15 20 25 30 35 40 45 50

Soun

d (d

BA) 100

80

60

40

X

100

50

0

minus50

Y

100

50

0

minus50

Z

100

50

0

minus50

minus0823 84249Units (mg) File 881IM1PLVE3 011354 062204

Run sound 479 L Aleq 449 Max sound 681

Max PkPk 196 A95 86 0-Pk 98

Max PkPk 188 A95 82 0-Pk 118

Max PkPk 188 A95 106 0-Pk 829

1 2 3

600

75

minus75

75

minus75

75

minus75

4

Calculate moving RMS of XY Z channels and display

Figure 1 Screenshot of variation analysis tools

Figure 2 EVA-625 control and constituent parts

and noises and the EVA system can quantize the data ofacceleration as well as noises In short if it is put in anelevator to rise and fall it can record the overall situationof time-variable operating state and noises of the elevatorDownloading the recorded information on the PC you canuse the apparatus-matched variation analysis tools softwarefor analysis just as in Figure 1

312 EVA-625 Principles and Structure The basic structureof EVA-625 is comprised of a Triaxial Accelerometer Package

a microphone interior circuits (digit simulation) a LCD a4-key keyboard a startstop switch and batteries (Figure 2)

Triaxial Accelerometer Package It is the acceleration(vibration) sensor of EVA-625 system

Axes of Sensitivity The axes of sensitivity of accelera-tion module includes 119883- 119884- and 119885-axis The 119883-axistests vibration of forward and backward while the 119884-axis measures vibration of sideways and the 119885-axisthat of vertical dimensions

4 Journal of Electrical and Computer Engineering

The Microphone It is used to collect noises whilethe sound track is designed at a restrictive quicklyreactive and fidelity RMS volume to collect thevolume which reaches the level of noises

The Tachometer The accessory allows for real timemeasurement and operation of escalator handrailsstep speed braking length and its intervals

The LCD it is made up of 4 lines and 20 characters

The Keyboard It allows users to set and modifysome necessary parameters while operating EVA-625without PC

The StartStop Switch It equals the ENT and ESCbutton to reduce useless vibrations when beginningrecord model

32 Wavelet Packet Analysis Being capable of providing anaccurate analytical method for signals it divides frequencybands into multilevels further resolving the high frequencyparts that are not dispersed by multiresolution analysis anddecomposing the frequency domain of the primitive intosimilar frequency band In case of the sampling frequencybeing recognized and adequate levels being resolved the fre-quency domain of each frequency band is restricted in a smallscope so that similar and super-low frequencies will be indifferent domains In consequences it analyzes signals moredelicately to provide more efficient approaches for extractingcharacteristics of signals As an international recognizedhigh technology to acquire and process information waveletpacket analysis is excellently applied in many fields suchas voices images graphs communications earthquakesbiomedicines mechanical vibrations and computer visions

33 The Energy Feature Extraction In case of malfunctionsabundant information about features of malfunctions wouldbe reflected from energy changes of vibration signals Signalswere decomposed by wallet packet transform at differentbands to determine the energy of signals distributed ineach subspace Then the energy distribution was comparedwith the situation when the system was normally runningin order to extract information about features of elevatormalfunctions Features were exacted according to the stepsas follows [9]

(1) First of all signals of vertical vibration accelerationwere decomposed with a 4-layer db 6-wavelet packetand the optimal wavelet packet tree was obtainedpursuant to the standard of minimum Shannonentropy as shown in Figure 3 Next the waveletpackets corresponding to the first two nodes on the4th layer the 2nd node on the 3rd layer the 2nd nodeon the 2nd layer and the 2nd node on the 1st layerwere selected to compose the optimal wavelet packetbase of signals

(2) Signals were reconstructed with the wavelet packetbases corresponding to nodes (4 0) (4 1) (3 1) (2 1)and (1 1) 119878

40was a reconstructed signal of node (4 0)

Tree decomposition

(0 0)

(1 0) (1 1)

(2 0) (2 1)

(3 0) (3 1)

(4 0) (4 1)

Figure 3 Optimal wavelet packet tree of vertical accelerationsignals

the rest could be deduced by analogy so the overallsignal reconstruction was as follows

119878 = 11987840

+ 11987841

+ 11987831

+ 11987821

+ 11987811 (1)

(3) The energy of each reconstructed was determinedand for 119878

40 the energy was calculated as follows

11986440

= int1003816100381610038161003816100381611987840

(119905)210038161003816100381610038161003816119889119905 =

119899

sum

119896=0

10038161003816100381610038161198831198961003816100381610038161003816

2

(2)

where 119883119896(119896 = 0 1 119899) represents the range of 119878

40

at discrete points and the energy at rest bands couldbe calculated in the same way

(4) Based on abundant information about malfunctionsreflected from energy of all signals at different bandsthe normalized feature vector was constructed asfollows

119879 = [11986440

11986411986441

11986411986431

11986411986421

11986411986411

119864] (3)

where 119864 = (|11986440|2+ |11986441|2+ |11986431|2+ |11986421|2+ |11986411|2)12

34 Time Domain Analysis and Feature Extraction Theelevator vibration incurred during running may reflect basicattributes of an elevator Changes to operating state of anelevator may cause changes to time signals of vibration andparameters described by time domain of signals Vibrationacceleration signals of a compartment were acquired from119883 119884 and 119885 directions in case of malfunctioning with anEVA-625 elevator tester As remarkable changes occurredto vibration acceleration in the vertical direction when anelevatormalfunctions abnormal vibrationwill also take placehorizontally Therefore kurtosis of vibration accelerationsignals in the 119885 direction and peak-to-peak values of the

Journal of Electrical and Computer Engineering 5

signals in 119883 and 119884 directions were used as time domainparameters

According to a given set of data on discreet vibrationsignals the119870 (Kurtosis) [19] was determined as follows

119870 =1

119873

119873

sum

119894=1

(119909119894minus 119909

120590119905

)

4

(4)

where 119909119894is signal value 119909wasmean signal value119873 indicated

sampling length and 120590119905represented standard deviation

4 Basic Principles of Least Squares SupportVector Machines

Support vector machines [20ndash22] construct statistical learn-ing theories according to principles of structural risk mini-mization when only a limited amount of samples is availableTo be specific input vectors were firstly projected from origi-nal space to high-dimension feature space through nonlinearmapping in order to construct the optimal decision functionin this high-dimension space based on principle of structuralrisk minimization In this process the calculation got lesscomplicated as the dot product of the high-dimension featurespace was replaced by kernel function of the original spaceConvex quadratic programming problem was solved by thesupport vector machine that the extreme values obtainedcould be guaranteed to be the global optimal solution In thisway the deficiency that neural networks easily became thesmallest within local areas was overcome

Having extended the standard SVM the least squaresSVM is proposed by J A K SuyKens and J Vandewalleacquiring outstanding achievements in pattern recognitionand nonlinear function fitting The differences between LS-SVM and standard SVM lie in that the loss function of anoptimized object is represented by 2-norm error inequalityconstraints of SVM are replaced by equality constraints andconvergence rate is increased

Assuming that the training set is (1199091 1199101) (119909

119899 119910119899)

where 119899 is total number of samples 119909119894isin 119877119899 is the 119894th sample

input and 119910119894isin minus1 1 is desired output of the 119894th sample the

function of linear regression was calculated as follows

119910 (119909) = 120596119879119883 + 119887 (5)

where 119883 = (1199091 1199092 119909

119899) is sample input 120596 = (120596

1 1205962

120596119899) is a weighted coefficient of LS-SVM and 119887 is a thresholdAccording to criteria of structural risk minimization

(SRM) [23ndash25] optimized problem could be converted into

min 1

21205962=

1

2119888

119899

sum

119894=1

1205852

119894 (6)

The constraint is as follows

119910119894= ⟨120596 sdot 119883⟩ + 119887 + 120585

119894 (7)

where 119888 is a tolerable penalty coefficient (119888 gt 0) for controllingthe degree of penalties on samples beyond the calculationerror 120585

119894is a relaxation factor and ⟨sdot⟩ is a mapping function of

kernel space

By introducing a Lagrange function the model of regres-sion function was obtained for LS-SVM according to KKToptimality conditions as follows

119891 (119909) =

119899

sum

119894=1

120572119894119870(119883119883

119894) + 119887 (8)

where 120572119894is a Lagrange multiplier and 119870(119883119883

119894) is a kernel

function for calculating the inner product of sample dataIn using a regression model with LS-SVM it is necessaryto properly set a coefficient of tolerable penalty and select asuitable kernel function

41 Classification and Analysis of LS-SVM In this paperfeature vectors of energy distribution of elevator vibrationacceleration signals including 119864

40 11986441 11986431 11986421 and 119864

11

were selected Besides time domains of vibration accelerationsignals of a compartment in119883119884 and119885 directions were usedinvolving gradient in the119885 direction and peak-to-peak valuesin 119883 and 119884 directions Furthermore extreme values of noisemeasured by a noise sensor was considered so malfunctionsymptoms composed by these nine characteristic values wereadopted as parameters input into SVM in order to classifytraining samples There are 70 groups training samples Atlast tested samples were classified and identified with awell-trained classifier Malfunctions were defined as followsnormal state = 1 deviation of elevator guide rail = 2 deviationof shape of guide show = 3 error of abnormal running oftractor = 4 error of rope groove of traction sheave = 5deviation of guide wheel = 6 and nonuniformity of tensionof wire rope = 7

42 Parameter Optimization for LS-SVM In the process ofrealizing LS-SVM two parameters should be determinedincluding kernel function 119892 and penalty factor 119888 Parameterswere optimized with K-CV while the processes for searchingand optimizing optimal parameters were shown in Figures 4and 5 as follows Firstly parameters were sketchily searchedwithin 2minus10sim210 with coarse mesh as shown in Figure 4Then optimal parameters 119888 and 119892 were found to rangewithin 2minus2sim210 and 2minus10sim20 Subsequently parameters wereprecisely searched within the sphere to find out the optimalparameter with K-CV method (Figure 5)

The searching results suggested that LS-SVM could real-ize the optimal precision for identifying faults when 119888 is 05and 119892 equals 1

Figure 1 offers part of the recognition results of 70 groupstest samples about normal state of elevators and 6 kindsof fault types and the recognition efficiency is 927 Itmay be clearly seen from the table that the normal state ofelevators deviated elevator guide rail deviated shape of guideshoe abnormal running of tractor erroneous rope groove oftraction sheave deviated guide wheel and tension of wirerope were successfully identified with a LS-SVM classifieroptimized through cross validation (Table 1)

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

2 Journal of Electrical and Computer Engineering

[10] diagnosed faults of rotating machines by LIttlewood-Paley wavelet support vector machines (LPWSVM) Shuixiaet al [11] achieved promising outcomes in classifying elevatormalfunctions by diagnosing faults with kernel principal com-ponent analysis (KPCA) and SVM Li et al [12] put forwarda model for fault diagnosis based on a genetic algorithmfor hierarchically optimizing least squares support vectormachines (LS-SVM) which could increase the precisionof LS-SVM for predicting faults and improve self-adaptivediagnosis Widodo and Yang [13] diagnosed faults of asyn-chronousmotors by extracting nonlinear characteristics withSVM and achieved ideal results In this paper an attempt wasmade to optimize parameters about SVM (mainly referring topenalty parameter 119888 and kernel function 119892) via cross valida-tion by integrating theories on optimal wavelet packet withSVM so as to effectively avoid overfitting and underfittingIn other words vibration signals were analyzed accordingto theories on optimal wavelet packet By using SVM asclassifier energy characteristics and time domain indicatorsof vibration signals were extracted to construct fault featurevectors They were adopted as input of SVM and elevatormalfunctions were classified with a well-trained SVM

2 On Safety for TakingElevators and Mechanism

21 On Safety for Taking Elevators Elevator compartmentsmay vibrate vertically or horizontally [1]Major causes of theirvertical vibration include vibration incurred by running atractor vibration of gearmeshing inside a speed reducermis-alignment between axes of worm gear reducer and tractionmotor and insecure fastening of tractor base with the bearingbeam and nonuniform load of wire rope All of these factorsimpact vertical vibration of elevator compartments Horizon-tal vibration of elevator compartments is mainly caused by[14] vibration resulting from guiding system out of toleranceperpendicularity of guide rails out of tolerance of distancebetween guide rails and out of tolerance between local gapof guide rails and steps at joints partial distortion of guiderails and straightness error of working surface of guide railsThese factors possibly lead compartments to horizontallyvibrate in the course of up-and-down motion Additionallystatic balance of compartments is influenced by deviatedsuspension center and eccentric load of compartments As aconsequence horizontal vibration is affected

22 On Vibration and Mechanism of Elevator Compart-ments Although there are numerous causes of compartmentvibration after analysis and investigation the main causes[15ndash18] are concluded to be deviated guide rail deviatedshape of guide shoe abnormal running of tractor erroneousrope groove of traction sheave deviated guide wheel andnonuniform tension of wire rope and so on

221 Deviated Guide Rail Elevator compartment runs byattachment to guide rail so horizontal vibration of an elevatoris directly impacted by its guide rail Asymmetric guide railgroove undesirable perpendicularity of guide rail widenedor narrowed distance between guide rails and fastened boltsof guide rail brackets or clips and so on cause horizontalvibration of compartments

222 Deviated Shape of Guide Shoe Attaching to guiderails guide shoe is used to limit obliquity and horizontaldisplacement It will be ineffective for reducing vibrationif guide rails are adjusted to be excessively tight when anelevator is running Consequently the resistance for runningcompartments gets higher and leads to vibration Besides theguide shoes will become inelastic once the gap between themis regulated to be too large and thereby cause compartmentsto vibrate during working

223 Abnormal Running of Tractor High-speed shaft-drivenimbalance and misaligned vibration of coupling are majorcauses of tractor vibration As a tractor is rotating at a highspeed the pulse becomes an excitation source for diagnosingcompartments Errors will be caused to rope grooves andresult in inaccurate dynamic balance after the tractor is wornfor a long period In addition vibrator compartments willvibrate during running provided that speed measurementencoder is not well connected with motor or misaligned

224 Deviated Guide Wheel The deviation of guide wheelfrom the perpendicular line will be higher than 2mm whenthe wheel is empty or fully loaded The protection shallmeet given requirements when suspended traction wheel orsprocket wheel is used To be exact geneva wheel shall not beseriously and nonuniformly worn to a level that its shape ischanged or else compartments will vibrate abnormally

225 Nonuniform Tension of Wire Rope Nonuniform ten-sion of wire rope leads to unequal specific pressure ontraction rope inside the grove of pulley wheel Compartmentwill abnormally vibrate due to relative slip of rope caused byjoint difference because the rope of traction wheel is wornmore quickly in case of great force [14]

3 Signal Processing and Analysis

In this paper data on elevators provided by Yunnan SpecialEquipment Safety Inspection and Research Institute wereanalyzed The data was acquired by EVA-625 elevator testerBecause the data measured about elevator running werelimited for better diagnosis simulative samples were com-bined with actual samples of malfunctions to establish asample database necessary for the training of support vectormachines Firstly the Matlab extracts the data distributioncharacteristics (almost the same as authentic samples) ofauthentic fault samples as well as their mean value vari-ance standard deviation extremum moment skewness andkurtosis based on which it randomly generates 100 groupsof simulative samples Integrating simulative samples withauthentic fault samples it establishes the data base needed forsupport vector machine (SVM) 70 out of those samples areused for training and another 70 for testing

31 EVA-625 Elevator Tester

311 EVA-625 Introduction Aimed to record the vibrationsand noises of elevators EVA-625 elevator tester is a par-ticularly designed high-precision recorder of acceleration

Journal of Electrical and Computer Engineering 3

(s)0 5 10 15 20 25 30 35 40 45 50

Soun

d (d

BA) 100

80

60

40

X

100

50

0

minus50

Y

100

50

0

minus50

Z

100

50

0

minus50

minus0823 84249Units (mg) File 881IM1PLVE3 011354 062204

Run sound 479 L Aleq 449 Max sound 681

Max PkPk 196 A95 86 0-Pk 98

Max PkPk 188 A95 82 0-Pk 118

Max PkPk 188 A95 106 0-Pk 829

1 2 3

600

75

minus75

75

minus75

75

minus75

4

Calculate moving RMS of XY Z channels and display

Figure 1 Screenshot of variation analysis tools

Figure 2 EVA-625 control and constituent parts

and noises and the EVA system can quantize the data ofacceleration as well as noises In short if it is put in anelevator to rise and fall it can record the overall situationof time-variable operating state and noises of the elevatorDownloading the recorded information on the PC you canuse the apparatus-matched variation analysis tools softwarefor analysis just as in Figure 1

312 EVA-625 Principles and Structure The basic structureof EVA-625 is comprised of a Triaxial Accelerometer Package

a microphone interior circuits (digit simulation) a LCD a4-key keyboard a startstop switch and batteries (Figure 2)

Triaxial Accelerometer Package It is the acceleration(vibration) sensor of EVA-625 system

Axes of Sensitivity The axes of sensitivity of accelera-tion module includes 119883- 119884- and 119885-axis The 119883-axistests vibration of forward and backward while the 119884-axis measures vibration of sideways and the 119885-axisthat of vertical dimensions

4 Journal of Electrical and Computer Engineering

The Microphone It is used to collect noises whilethe sound track is designed at a restrictive quicklyreactive and fidelity RMS volume to collect thevolume which reaches the level of noises

The Tachometer The accessory allows for real timemeasurement and operation of escalator handrailsstep speed braking length and its intervals

The LCD it is made up of 4 lines and 20 characters

The Keyboard It allows users to set and modifysome necessary parameters while operating EVA-625without PC

The StartStop Switch It equals the ENT and ESCbutton to reduce useless vibrations when beginningrecord model

32 Wavelet Packet Analysis Being capable of providing anaccurate analytical method for signals it divides frequencybands into multilevels further resolving the high frequencyparts that are not dispersed by multiresolution analysis anddecomposing the frequency domain of the primitive intosimilar frequency band In case of the sampling frequencybeing recognized and adequate levels being resolved the fre-quency domain of each frequency band is restricted in a smallscope so that similar and super-low frequencies will be indifferent domains In consequences it analyzes signals moredelicately to provide more efficient approaches for extractingcharacteristics of signals As an international recognizedhigh technology to acquire and process information waveletpacket analysis is excellently applied in many fields suchas voices images graphs communications earthquakesbiomedicines mechanical vibrations and computer visions

33 The Energy Feature Extraction In case of malfunctionsabundant information about features of malfunctions wouldbe reflected from energy changes of vibration signals Signalswere decomposed by wallet packet transform at differentbands to determine the energy of signals distributed ineach subspace Then the energy distribution was comparedwith the situation when the system was normally runningin order to extract information about features of elevatormalfunctions Features were exacted according to the stepsas follows [9]

(1) First of all signals of vertical vibration accelerationwere decomposed with a 4-layer db 6-wavelet packetand the optimal wavelet packet tree was obtainedpursuant to the standard of minimum Shannonentropy as shown in Figure 3 Next the waveletpackets corresponding to the first two nodes on the4th layer the 2nd node on the 3rd layer the 2nd nodeon the 2nd layer and the 2nd node on the 1st layerwere selected to compose the optimal wavelet packetbase of signals

(2) Signals were reconstructed with the wavelet packetbases corresponding to nodes (4 0) (4 1) (3 1) (2 1)and (1 1) 119878

40was a reconstructed signal of node (4 0)

Tree decomposition

(0 0)

(1 0) (1 1)

(2 0) (2 1)

(3 0) (3 1)

(4 0) (4 1)

Figure 3 Optimal wavelet packet tree of vertical accelerationsignals

the rest could be deduced by analogy so the overallsignal reconstruction was as follows

119878 = 11987840

+ 11987841

+ 11987831

+ 11987821

+ 11987811 (1)

(3) The energy of each reconstructed was determinedand for 119878

40 the energy was calculated as follows

11986440

= int1003816100381610038161003816100381611987840

(119905)210038161003816100381610038161003816119889119905 =

119899

sum

119896=0

10038161003816100381610038161198831198961003816100381610038161003816

2

(2)

where 119883119896(119896 = 0 1 119899) represents the range of 119878

40

at discrete points and the energy at rest bands couldbe calculated in the same way

(4) Based on abundant information about malfunctionsreflected from energy of all signals at different bandsthe normalized feature vector was constructed asfollows

119879 = [11986440

11986411986441

11986411986431

11986411986421

11986411986411

119864] (3)

where 119864 = (|11986440|2+ |11986441|2+ |11986431|2+ |11986421|2+ |11986411|2)12

34 Time Domain Analysis and Feature Extraction Theelevator vibration incurred during running may reflect basicattributes of an elevator Changes to operating state of anelevator may cause changes to time signals of vibration andparameters described by time domain of signals Vibrationacceleration signals of a compartment were acquired from119883 119884 and 119885 directions in case of malfunctioning with anEVA-625 elevator tester As remarkable changes occurredto vibration acceleration in the vertical direction when anelevatormalfunctions abnormal vibrationwill also take placehorizontally Therefore kurtosis of vibration accelerationsignals in the 119885 direction and peak-to-peak values of the

Journal of Electrical and Computer Engineering 5

signals in 119883 and 119884 directions were used as time domainparameters

According to a given set of data on discreet vibrationsignals the119870 (Kurtosis) [19] was determined as follows

119870 =1

119873

119873

sum

119894=1

(119909119894minus 119909

120590119905

)

4

(4)

where 119909119894is signal value 119909wasmean signal value119873 indicated

sampling length and 120590119905represented standard deviation

4 Basic Principles of Least Squares SupportVector Machines

Support vector machines [20ndash22] construct statistical learn-ing theories according to principles of structural risk mini-mization when only a limited amount of samples is availableTo be specific input vectors were firstly projected from origi-nal space to high-dimension feature space through nonlinearmapping in order to construct the optimal decision functionin this high-dimension space based on principle of structuralrisk minimization In this process the calculation got lesscomplicated as the dot product of the high-dimension featurespace was replaced by kernel function of the original spaceConvex quadratic programming problem was solved by thesupport vector machine that the extreme values obtainedcould be guaranteed to be the global optimal solution In thisway the deficiency that neural networks easily became thesmallest within local areas was overcome

Having extended the standard SVM the least squaresSVM is proposed by J A K SuyKens and J Vandewalleacquiring outstanding achievements in pattern recognitionand nonlinear function fitting The differences between LS-SVM and standard SVM lie in that the loss function of anoptimized object is represented by 2-norm error inequalityconstraints of SVM are replaced by equality constraints andconvergence rate is increased

Assuming that the training set is (1199091 1199101) (119909

119899 119910119899)

where 119899 is total number of samples 119909119894isin 119877119899 is the 119894th sample

input and 119910119894isin minus1 1 is desired output of the 119894th sample the

function of linear regression was calculated as follows

119910 (119909) = 120596119879119883 + 119887 (5)

where 119883 = (1199091 1199092 119909

119899) is sample input 120596 = (120596

1 1205962

120596119899) is a weighted coefficient of LS-SVM and 119887 is a thresholdAccording to criteria of structural risk minimization

(SRM) [23ndash25] optimized problem could be converted into

min 1

21205962=

1

2119888

119899

sum

119894=1

1205852

119894 (6)

The constraint is as follows

119910119894= ⟨120596 sdot 119883⟩ + 119887 + 120585

119894 (7)

where 119888 is a tolerable penalty coefficient (119888 gt 0) for controllingthe degree of penalties on samples beyond the calculationerror 120585

119894is a relaxation factor and ⟨sdot⟩ is a mapping function of

kernel space

By introducing a Lagrange function the model of regres-sion function was obtained for LS-SVM according to KKToptimality conditions as follows

119891 (119909) =

119899

sum

119894=1

120572119894119870(119883119883

119894) + 119887 (8)

where 120572119894is a Lagrange multiplier and 119870(119883119883

119894) is a kernel

function for calculating the inner product of sample dataIn using a regression model with LS-SVM it is necessaryto properly set a coefficient of tolerable penalty and select asuitable kernel function

41 Classification and Analysis of LS-SVM In this paperfeature vectors of energy distribution of elevator vibrationacceleration signals including 119864

40 11986441 11986431 11986421 and 119864

11

were selected Besides time domains of vibration accelerationsignals of a compartment in119883119884 and119885 directions were usedinvolving gradient in the119885 direction and peak-to-peak valuesin 119883 and 119884 directions Furthermore extreme values of noisemeasured by a noise sensor was considered so malfunctionsymptoms composed by these nine characteristic values wereadopted as parameters input into SVM in order to classifytraining samples There are 70 groups training samples Atlast tested samples were classified and identified with awell-trained classifier Malfunctions were defined as followsnormal state = 1 deviation of elevator guide rail = 2 deviationof shape of guide show = 3 error of abnormal running oftractor = 4 error of rope groove of traction sheave = 5deviation of guide wheel = 6 and nonuniformity of tensionof wire rope = 7

42 Parameter Optimization for LS-SVM In the process ofrealizing LS-SVM two parameters should be determinedincluding kernel function 119892 and penalty factor 119888 Parameterswere optimized with K-CV while the processes for searchingand optimizing optimal parameters were shown in Figures 4and 5 as follows Firstly parameters were sketchily searchedwithin 2minus10sim210 with coarse mesh as shown in Figure 4Then optimal parameters 119888 and 119892 were found to rangewithin 2minus2sim210 and 2minus10sim20 Subsequently parameters wereprecisely searched within the sphere to find out the optimalparameter with K-CV method (Figure 5)

The searching results suggested that LS-SVM could real-ize the optimal precision for identifying faults when 119888 is 05and 119892 equals 1

Figure 1 offers part of the recognition results of 70 groupstest samples about normal state of elevators and 6 kindsof fault types and the recognition efficiency is 927 Itmay be clearly seen from the table that the normal state ofelevators deviated elevator guide rail deviated shape of guideshoe abnormal running of tractor erroneous rope groove oftraction sheave deviated guide wheel and tension of wirerope were successfully identified with a LS-SVM classifieroptimized through cross validation (Table 1)

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

Journal of Electrical and Computer Engineering 3

(s)0 5 10 15 20 25 30 35 40 45 50

Soun

d (d

BA) 100

80

60

40

X

100

50

0

minus50

Y

100

50

0

minus50

Z

100

50

0

minus50

minus0823 84249Units (mg) File 881IM1PLVE3 011354 062204

Run sound 479 L Aleq 449 Max sound 681

Max PkPk 196 A95 86 0-Pk 98

Max PkPk 188 A95 82 0-Pk 118

Max PkPk 188 A95 106 0-Pk 829

1 2 3

600

75

minus75

75

minus75

75

minus75

4

Calculate moving RMS of XY Z channels and display

Figure 1 Screenshot of variation analysis tools

Figure 2 EVA-625 control and constituent parts

and noises and the EVA system can quantize the data ofacceleration as well as noises In short if it is put in anelevator to rise and fall it can record the overall situationof time-variable operating state and noises of the elevatorDownloading the recorded information on the PC you canuse the apparatus-matched variation analysis tools softwarefor analysis just as in Figure 1

312 EVA-625 Principles and Structure The basic structureof EVA-625 is comprised of a Triaxial Accelerometer Package

a microphone interior circuits (digit simulation) a LCD a4-key keyboard a startstop switch and batteries (Figure 2)

Triaxial Accelerometer Package It is the acceleration(vibration) sensor of EVA-625 system

Axes of Sensitivity The axes of sensitivity of accelera-tion module includes 119883- 119884- and 119885-axis The 119883-axistests vibration of forward and backward while the 119884-axis measures vibration of sideways and the 119885-axisthat of vertical dimensions

4 Journal of Electrical and Computer Engineering

The Microphone It is used to collect noises whilethe sound track is designed at a restrictive quicklyreactive and fidelity RMS volume to collect thevolume which reaches the level of noises

The Tachometer The accessory allows for real timemeasurement and operation of escalator handrailsstep speed braking length and its intervals

The LCD it is made up of 4 lines and 20 characters

The Keyboard It allows users to set and modifysome necessary parameters while operating EVA-625without PC

The StartStop Switch It equals the ENT and ESCbutton to reduce useless vibrations when beginningrecord model

32 Wavelet Packet Analysis Being capable of providing anaccurate analytical method for signals it divides frequencybands into multilevels further resolving the high frequencyparts that are not dispersed by multiresolution analysis anddecomposing the frequency domain of the primitive intosimilar frequency band In case of the sampling frequencybeing recognized and adequate levels being resolved the fre-quency domain of each frequency band is restricted in a smallscope so that similar and super-low frequencies will be indifferent domains In consequences it analyzes signals moredelicately to provide more efficient approaches for extractingcharacteristics of signals As an international recognizedhigh technology to acquire and process information waveletpacket analysis is excellently applied in many fields suchas voices images graphs communications earthquakesbiomedicines mechanical vibrations and computer visions

33 The Energy Feature Extraction In case of malfunctionsabundant information about features of malfunctions wouldbe reflected from energy changes of vibration signals Signalswere decomposed by wallet packet transform at differentbands to determine the energy of signals distributed ineach subspace Then the energy distribution was comparedwith the situation when the system was normally runningin order to extract information about features of elevatormalfunctions Features were exacted according to the stepsas follows [9]

(1) First of all signals of vertical vibration accelerationwere decomposed with a 4-layer db 6-wavelet packetand the optimal wavelet packet tree was obtainedpursuant to the standard of minimum Shannonentropy as shown in Figure 3 Next the waveletpackets corresponding to the first two nodes on the4th layer the 2nd node on the 3rd layer the 2nd nodeon the 2nd layer and the 2nd node on the 1st layerwere selected to compose the optimal wavelet packetbase of signals

(2) Signals were reconstructed with the wavelet packetbases corresponding to nodes (4 0) (4 1) (3 1) (2 1)and (1 1) 119878

40was a reconstructed signal of node (4 0)

Tree decomposition

(0 0)

(1 0) (1 1)

(2 0) (2 1)

(3 0) (3 1)

(4 0) (4 1)

Figure 3 Optimal wavelet packet tree of vertical accelerationsignals

the rest could be deduced by analogy so the overallsignal reconstruction was as follows

119878 = 11987840

+ 11987841

+ 11987831

+ 11987821

+ 11987811 (1)

(3) The energy of each reconstructed was determinedand for 119878

40 the energy was calculated as follows

11986440

= int1003816100381610038161003816100381611987840

(119905)210038161003816100381610038161003816119889119905 =

119899

sum

119896=0

10038161003816100381610038161198831198961003816100381610038161003816

2

(2)

where 119883119896(119896 = 0 1 119899) represents the range of 119878

40

at discrete points and the energy at rest bands couldbe calculated in the same way

(4) Based on abundant information about malfunctionsreflected from energy of all signals at different bandsthe normalized feature vector was constructed asfollows

119879 = [11986440

11986411986441

11986411986431

11986411986421

11986411986411

119864] (3)

where 119864 = (|11986440|2+ |11986441|2+ |11986431|2+ |11986421|2+ |11986411|2)12

34 Time Domain Analysis and Feature Extraction Theelevator vibration incurred during running may reflect basicattributes of an elevator Changes to operating state of anelevator may cause changes to time signals of vibration andparameters described by time domain of signals Vibrationacceleration signals of a compartment were acquired from119883 119884 and 119885 directions in case of malfunctioning with anEVA-625 elevator tester As remarkable changes occurredto vibration acceleration in the vertical direction when anelevatormalfunctions abnormal vibrationwill also take placehorizontally Therefore kurtosis of vibration accelerationsignals in the 119885 direction and peak-to-peak values of the

Journal of Electrical and Computer Engineering 5

signals in 119883 and 119884 directions were used as time domainparameters

According to a given set of data on discreet vibrationsignals the119870 (Kurtosis) [19] was determined as follows

119870 =1

119873

119873

sum

119894=1

(119909119894minus 119909

120590119905

)

4

(4)

where 119909119894is signal value 119909wasmean signal value119873 indicated

sampling length and 120590119905represented standard deviation

4 Basic Principles of Least Squares SupportVector Machines

Support vector machines [20ndash22] construct statistical learn-ing theories according to principles of structural risk mini-mization when only a limited amount of samples is availableTo be specific input vectors were firstly projected from origi-nal space to high-dimension feature space through nonlinearmapping in order to construct the optimal decision functionin this high-dimension space based on principle of structuralrisk minimization In this process the calculation got lesscomplicated as the dot product of the high-dimension featurespace was replaced by kernel function of the original spaceConvex quadratic programming problem was solved by thesupport vector machine that the extreme values obtainedcould be guaranteed to be the global optimal solution In thisway the deficiency that neural networks easily became thesmallest within local areas was overcome

Having extended the standard SVM the least squaresSVM is proposed by J A K SuyKens and J Vandewalleacquiring outstanding achievements in pattern recognitionand nonlinear function fitting The differences between LS-SVM and standard SVM lie in that the loss function of anoptimized object is represented by 2-norm error inequalityconstraints of SVM are replaced by equality constraints andconvergence rate is increased

Assuming that the training set is (1199091 1199101) (119909

119899 119910119899)

where 119899 is total number of samples 119909119894isin 119877119899 is the 119894th sample

input and 119910119894isin minus1 1 is desired output of the 119894th sample the

function of linear regression was calculated as follows

119910 (119909) = 120596119879119883 + 119887 (5)

where 119883 = (1199091 1199092 119909

119899) is sample input 120596 = (120596

1 1205962

120596119899) is a weighted coefficient of LS-SVM and 119887 is a thresholdAccording to criteria of structural risk minimization

(SRM) [23ndash25] optimized problem could be converted into

min 1

21205962=

1

2119888

119899

sum

119894=1

1205852

119894 (6)

The constraint is as follows

119910119894= ⟨120596 sdot 119883⟩ + 119887 + 120585

119894 (7)

where 119888 is a tolerable penalty coefficient (119888 gt 0) for controllingthe degree of penalties on samples beyond the calculationerror 120585

119894is a relaxation factor and ⟨sdot⟩ is a mapping function of

kernel space

By introducing a Lagrange function the model of regres-sion function was obtained for LS-SVM according to KKToptimality conditions as follows

119891 (119909) =

119899

sum

119894=1

120572119894119870(119883119883

119894) + 119887 (8)

where 120572119894is a Lagrange multiplier and 119870(119883119883

119894) is a kernel

function for calculating the inner product of sample dataIn using a regression model with LS-SVM it is necessaryto properly set a coefficient of tolerable penalty and select asuitable kernel function

41 Classification and Analysis of LS-SVM In this paperfeature vectors of energy distribution of elevator vibrationacceleration signals including 119864

40 11986441 11986431 11986421 and 119864

11

were selected Besides time domains of vibration accelerationsignals of a compartment in119883119884 and119885 directions were usedinvolving gradient in the119885 direction and peak-to-peak valuesin 119883 and 119884 directions Furthermore extreme values of noisemeasured by a noise sensor was considered so malfunctionsymptoms composed by these nine characteristic values wereadopted as parameters input into SVM in order to classifytraining samples There are 70 groups training samples Atlast tested samples were classified and identified with awell-trained classifier Malfunctions were defined as followsnormal state = 1 deviation of elevator guide rail = 2 deviationof shape of guide show = 3 error of abnormal running oftractor = 4 error of rope groove of traction sheave = 5deviation of guide wheel = 6 and nonuniformity of tensionof wire rope = 7

42 Parameter Optimization for LS-SVM In the process ofrealizing LS-SVM two parameters should be determinedincluding kernel function 119892 and penalty factor 119888 Parameterswere optimized with K-CV while the processes for searchingand optimizing optimal parameters were shown in Figures 4and 5 as follows Firstly parameters were sketchily searchedwithin 2minus10sim210 with coarse mesh as shown in Figure 4Then optimal parameters 119888 and 119892 were found to rangewithin 2minus2sim210 and 2minus10sim20 Subsequently parameters wereprecisely searched within the sphere to find out the optimalparameter with K-CV method (Figure 5)

The searching results suggested that LS-SVM could real-ize the optimal precision for identifying faults when 119888 is 05and 119892 equals 1

Figure 1 offers part of the recognition results of 70 groupstest samples about normal state of elevators and 6 kindsof fault types and the recognition efficiency is 927 Itmay be clearly seen from the table that the normal state ofelevators deviated elevator guide rail deviated shape of guideshoe abnormal running of tractor erroneous rope groove oftraction sheave deviated guide wheel and tension of wirerope were successfully identified with a LS-SVM classifieroptimized through cross validation (Table 1)

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

4 Journal of Electrical and Computer Engineering

The Microphone It is used to collect noises whilethe sound track is designed at a restrictive quicklyreactive and fidelity RMS volume to collect thevolume which reaches the level of noises

The Tachometer The accessory allows for real timemeasurement and operation of escalator handrailsstep speed braking length and its intervals

The LCD it is made up of 4 lines and 20 characters

The Keyboard It allows users to set and modifysome necessary parameters while operating EVA-625without PC

The StartStop Switch It equals the ENT and ESCbutton to reduce useless vibrations when beginningrecord model

32 Wavelet Packet Analysis Being capable of providing anaccurate analytical method for signals it divides frequencybands into multilevels further resolving the high frequencyparts that are not dispersed by multiresolution analysis anddecomposing the frequency domain of the primitive intosimilar frequency band In case of the sampling frequencybeing recognized and adequate levels being resolved the fre-quency domain of each frequency band is restricted in a smallscope so that similar and super-low frequencies will be indifferent domains In consequences it analyzes signals moredelicately to provide more efficient approaches for extractingcharacteristics of signals As an international recognizedhigh technology to acquire and process information waveletpacket analysis is excellently applied in many fields suchas voices images graphs communications earthquakesbiomedicines mechanical vibrations and computer visions

33 The Energy Feature Extraction In case of malfunctionsabundant information about features of malfunctions wouldbe reflected from energy changes of vibration signals Signalswere decomposed by wallet packet transform at differentbands to determine the energy of signals distributed ineach subspace Then the energy distribution was comparedwith the situation when the system was normally runningin order to extract information about features of elevatormalfunctions Features were exacted according to the stepsas follows [9]

(1) First of all signals of vertical vibration accelerationwere decomposed with a 4-layer db 6-wavelet packetand the optimal wavelet packet tree was obtainedpursuant to the standard of minimum Shannonentropy as shown in Figure 3 Next the waveletpackets corresponding to the first two nodes on the4th layer the 2nd node on the 3rd layer the 2nd nodeon the 2nd layer and the 2nd node on the 1st layerwere selected to compose the optimal wavelet packetbase of signals

(2) Signals were reconstructed with the wavelet packetbases corresponding to nodes (4 0) (4 1) (3 1) (2 1)and (1 1) 119878

40was a reconstructed signal of node (4 0)

Tree decomposition

(0 0)

(1 0) (1 1)

(2 0) (2 1)

(3 0) (3 1)

(4 0) (4 1)

Figure 3 Optimal wavelet packet tree of vertical accelerationsignals

the rest could be deduced by analogy so the overallsignal reconstruction was as follows

119878 = 11987840

+ 11987841

+ 11987831

+ 11987821

+ 11987811 (1)

(3) The energy of each reconstructed was determinedand for 119878

40 the energy was calculated as follows

11986440

= int1003816100381610038161003816100381611987840

(119905)210038161003816100381610038161003816119889119905 =

119899

sum

119896=0

10038161003816100381610038161198831198961003816100381610038161003816

2

(2)

where 119883119896(119896 = 0 1 119899) represents the range of 119878

40

at discrete points and the energy at rest bands couldbe calculated in the same way

(4) Based on abundant information about malfunctionsreflected from energy of all signals at different bandsthe normalized feature vector was constructed asfollows

119879 = [11986440

11986411986441

11986411986431

11986411986421

11986411986411

119864] (3)

where 119864 = (|11986440|2+ |11986441|2+ |11986431|2+ |11986421|2+ |11986411|2)12

34 Time Domain Analysis and Feature Extraction Theelevator vibration incurred during running may reflect basicattributes of an elevator Changes to operating state of anelevator may cause changes to time signals of vibration andparameters described by time domain of signals Vibrationacceleration signals of a compartment were acquired from119883 119884 and 119885 directions in case of malfunctioning with anEVA-625 elevator tester As remarkable changes occurredto vibration acceleration in the vertical direction when anelevatormalfunctions abnormal vibrationwill also take placehorizontally Therefore kurtosis of vibration accelerationsignals in the 119885 direction and peak-to-peak values of the

Journal of Electrical and Computer Engineering 5

signals in 119883 and 119884 directions were used as time domainparameters

According to a given set of data on discreet vibrationsignals the119870 (Kurtosis) [19] was determined as follows

119870 =1

119873

119873

sum

119894=1

(119909119894minus 119909

120590119905

)

4

(4)

where 119909119894is signal value 119909wasmean signal value119873 indicated

sampling length and 120590119905represented standard deviation

4 Basic Principles of Least Squares SupportVector Machines

Support vector machines [20ndash22] construct statistical learn-ing theories according to principles of structural risk mini-mization when only a limited amount of samples is availableTo be specific input vectors were firstly projected from origi-nal space to high-dimension feature space through nonlinearmapping in order to construct the optimal decision functionin this high-dimension space based on principle of structuralrisk minimization In this process the calculation got lesscomplicated as the dot product of the high-dimension featurespace was replaced by kernel function of the original spaceConvex quadratic programming problem was solved by thesupport vector machine that the extreme values obtainedcould be guaranteed to be the global optimal solution In thisway the deficiency that neural networks easily became thesmallest within local areas was overcome

Having extended the standard SVM the least squaresSVM is proposed by J A K SuyKens and J Vandewalleacquiring outstanding achievements in pattern recognitionand nonlinear function fitting The differences between LS-SVM and standard SVM lie in that the loss function of anoptimized object is represented by 2-norm error inequalityconstraints of SVM are replaced by equality constraints andconvergence rate is increased

Assuming that the training set is (1199091 1199101) (119909

119899 119910119899)

where 119899 is total number of samples 119909119894isin 119877119899 is the 119894th sample

input and 119910119894isin minus1 1 is desired output of the 119894th sample the

function of linear regression was calculated as follows

119910 (119909) = 120596119879119883 + 119887 (5)

where 119883 = (1199091 1199092 119909

119899) is sample input 120596 = (120596

1 1205962

120596119899) is a weighted coefficient of LS-SVM and 119887 is a thresholdAccording to criteria of structural risk minimization

(SRM) [23ndash25] optimized problem could be converted into

min 1

21205962=

1

2119888

119899

sum

119894=1

1205852

119894 (6)

The constraint is as follows

119910119894= ⟨120596 sdot 119883⟩ + 119887 + 120585

119894 (7)

where 119888 is a tolerable penalty coefficient (119888 gt 0) for controllingthe degree of penalties on samples beyond the calculationerror 120585

119894is a relaxation factor and ⟨sdot⟩ is a mapping function of

kernel space

By introducing a Lagrange function the model of regres-sion function was obtained for LS-SVM according to KKToptimality conditions as follows

119891 (119909) =

119899

sum

119894=1

120572119894119870(119883119883

119894) + 119887 (8)

where 120572119894is a Lagrange multiplier and 119870(119883119883

119894) is a kernel

function for calculating the inner product of sample dataIn using a regression model with LS-SVM it is necessaryto properly set a coefficient of tolerable penalty and select asuitable kernel function

41 Classification and Analysis of LS-SVM In this paperfeature vectors of energy distribution of elevator vibrationacceleration signals including 119864

40 11986441 11986431 11986421 and 119864

11

were selected Besides time domains of vibration accelerationsignals of a compartment in119883119884 and119885 directions were usedinvolving gradient in the119885 direction and peak-to-peak valuesin 119883 and 119884 directions Furthermore extreme values of noisemeasured by a noise sensor was considered so malfunctionsymptoms composed by these nine characteristic values wereadopted as parameters input into SVM in order to classifytraining samples There are 70 groups training samples Atlast tested samples were classified and identified with awell-trained classifier Malfunctions were defined as followsnormal state = 1 deviation of elevator guide rail = 2 deviationof shape of guide show = 3 error of abnormal running oftractor = 4 error of rope groove of traction sheave = 5deviation of guide wheel = 6 and nonuniformity of tensionof wire rope = 7

42 Parameter Optimization for LS-SVM In the process ofrealizing LS-SVM two parameters should be determinedincluding kernel function 119892 and penalty factor 119888 Parameterswere optimized with K-CV while the processes for searchingand optimizing optimal parameters were shown in Figures 4and 5 as follows Firstly parameters were sketchily searchedwithin 2minus10sim210 with coarse mesh as shown in Figure 4Then optimal parameters 119888 and 119892 were found to rangewithin 2minus2sim210 and 2minus10sim20 Subsequently parameters wereprecisely searched within the sphere to find out the optimalparameter with K-CV method (Figure 5)

The searching results suggested that LS-SVM could real-ize the optimal precision for identifying faults when 119888 is 05and 119892 equals 1

Figure 1 offers part of the recognition results of 70 groupstest samples about normal state of elevators and 6 kindsof fault types and the recognition efficiency is 927 Itmay be clearly seen from the table that the normal state ofelevators deviated elevator guide rail deviated shape of guideshoe abnormal running of tractor erroneous rope groove oftraction sheave deviated guide wheel and tension of wirerope were successfully identified with a LS-SVM classifieroptimized through cross validation (Table 1)

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

Journal of Electrical and Computer Engineering 5

signals in 119883 and 119884 directions were used as time domainparameters

According to a given set of data on discreet vibrationsignals the119870 (Kurtosis) [19] was determined as follows

119870 =1

119873

119873

sum

119894=1

(119909119894minus 119909

120590119905

)

4

(4)

where 119909119894is signal value 119909wasmean signal value119873 indicated

sampling length and 120590119905represented standard deviation

4 Basic Principles of Least Squares SupportVector Machines

Support vector machines [20ndash22] construct statistical learn-ing theories according to principles of structural risk mini-mization when only a limited amount of samples is availableTo be specific input vectors were firstly projected from origi-nal space to high-dimension feature space through nonlinearmapping in order to construct the optimal decision functionin this high-dimension space based on principle of structuralrisk minimization In this process the calculation got lesscomplicated as the dot product of the high-dimension featurespace was replaced by kernel function of the original spaceConvex quadratic programming problem was solved by thesupport vector machine that the extreme values obtainedcould be guaranteed to be the global optimal solution In thisway the deficiency that neural networks easily became thesmallest within local areas was overcome

Having extended the standard SVM the least squaresSVM is proposed by J A K SuyKens and J Vandewalleacquiring outstanding achievements in pattern recognitionand nonlinear function fitting The differences between LS-SVM and standard SVM lie in that the loss function of anoptimized object is represented by 2-norm error inequalityconstraints of SVM are replaced by equality constraints andconvergence rate is increased

Assuming that the training set is (1199091 1199101) (119909

119899 119910119899)

where 119899 is total number of samples 119909119894isin 119877119899 is the 119894th sample

input and 119910119894isin minus1 1 is desired output of the 119894th sample the

function of linear regression was calculated as follows

119910 (119909) = 120596119879119883 + 119887 (5)

where 119883 = (1199091 1199092 119909

119899) is sample input 120596 = (120596

1 1205962

120596119899) is a weighted coefficient of LS-SVM and 119887 is a thresholdAccording to criteria of structural risk minimization

(SRM) [23ndash25] optimized problem could be converted into

min 1

21205962=

1

2119888

119899

sum

119894=1

1205852

119894 (6)

The constraint is as follows

119910119894= ⟨120596 sdot 119883⟩ + 119887 + 120585

119894 (7)

where 119888 is a tolerable penalty coefficient (119888 gt 0) for controllingthe degree of penalties on samples beyond the calculationerror 120585

119894is a relaxation factor and ⟨sdot⟩ is a mapping function of

kernel space

By introducing a Lagrange function the model of regres-sion function was obtained for LS-SVM according to KKToptimality conditions as follows

119891 (119909) =

119899

sum

119894=1

120572119894119870(119883119883

119894) + 119887 (8)

where 120572119894is a Lagrange multiplier and 119870(119883119883

119894) is a kernel

function for calculating the inner product of sample dataIn using a regression model with LS-SVM it is necessaryto properly set a coefficient of tolerable penalty and select asuitable kernel function

41 Classification and Analysis of LS-SVM In this paperfeature vectors of energy distribution of elevator vibrationacceleration signals including 119864

40 11986441 11986431 11986421 and 119864

11

were selected Besides time domains of vibration accelerationsignals of a compartment in119883119884 and119885 directions were usedinvolving gradient in the119885 direction and peak-to-peak valuesin 119883 and 119884 directions Furthermore extreme values of noisemeasured by a noise sensor was considered so malfunctionsymptoms composed by these nine characteristic values wereadopted as parameters input into SVM in order to classifytraining samples There are 70 groups training samples Atlast tested samples were classified and identified with awell-trained classifier Malfunctions were defined as followsnormal state = 1 deviation of elevator guide rail = 2 deviationof shape of guide show = 3 error of abnormal running oftractor = 4 error of rope groove of traction sheave = 5deviation of guide wheel = 6 and nonuniformity of tensionof wire rope = 7

42 Parameter Optimization for LS-SVM In the process ofrealizing LS-SVM two parameters should be determinedincluding kernel function 119892 and penalty factor 119888 Parameterswere optimized with K-CV while the processes for searchingand optimizing optimal parameters were shown in Figures 4and 5 as follows Firstly parameters were sketchily searchedwithin 2minus10sim210 with coarse mesh as shown in Figure 4Then optimal parameters 119888 and 119892 were found to rangewithin 2minus2sim210 and 2minus10sim20 Subsequently parameters wereprecisely searched within the sphere to find out the optimalparameter with K-CV method (Figure 5)

The searching results suggested that LS-SVM could real-ize the optimal precision for identifying faults when 119888 is 05and 119892 equals 1

Figure 1 offers part of the recognition results of 70 groupstest samples about normal state of elevators and 6 kindsof fault types and the recognition efficiency is 927 Itmay be clearly seen from the table that the normal state ofelevators deviated elevator guide rail deviated shape of guideshoe abnormal running of tractor erroneous rope groove oftraction sheave deviated guide wheel and tension of wirerope were successfully identified with a LS-SVM classifieroptimized through cross validation (Table 1)

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

6 Journal of Electrical and Computer Engineering

Table 1 Results for identifying elevator faults with a LS-SVM optimized by cross validation

Faults of testedsamples

Input of SVM

OutputEnergy characteristics Time Domain characteristics

Noise extremes(dB)119864

4011986441

11986431

11986421

11986411

119896

Peak-to-peakvalues in119883

direction (ms2)

Peak-to-peakvalues in 119884

direction (ms2)Normal state (1) 0089 0167 0124 0118 0068 0889546 0009 001 479 1Normal state (2) 0093 0173 0136 0121 0079 0796675 0009 0012 471 1Deviation ofguide rail (1) 0044 0242 0276 0171 0026 4139485 0191 0024 581 2

Deviation ofguide rail (2) 0041 0241 0285 0166 0024 3247673 0178 0025 595 2

Deviation ofshape of guideshoe (1)

056 0294 0028 0031 0012 3173433 0021 002 556 3

Deviation ofshape of guideshoe (2)

0534 0295 0027 0034 0018 4272636 0025 0019 552 3

Abnormalrunning of tractor(1)

0545 0294 0028 0038 0022 2380825 0021 002 564 4

Abnormalrunning of tractor(2)

0582 0273 0026 0036 002 3567978 0025 0018 559 4

Error of ropegroove of tractionsheave (1)

0814 0031 0022 0022 0024 3677453 0024 0022 571 5

Error of ropegroove of tractionsheave (2)

079 0026 0021 0024 0022 2824331 0023 0022 564 5

Deviation ofguide wheel (1) 002 0212 011 0066 0105 3187661 0124 01 538 6

Deviation ofguide wheel (2) 0022 0207 0108 0056 0105 2095107 0115 01 542 6

Uniformity oftension on wirerope (1)

012 0525 0028 0118 0022 2109227 002 0019 556 7

Uniformity oftension on wirerope (2)

0124 0566 0028 0109 0021 3099955 0023 002 559 7

79

79

79 79

835

835

835 835

88

88

88 88

925

925

925 925

97

97

97

97

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10minus10

minus8minus6minus4minus2

02468

10

minus10 minus8 minus6 minus4 minus2 0 2 4 6 8 10

minus10minus8minus6minus4minus2024681030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

Best c = 014359 g = 0027205 CVAccuracy = 975 Best c = 014359 g = 0027205 CVAccuracy = 975

log 2

g

log 2g

log 2clog 2c

Figure 4 Roughly selected parameters

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

Journal of Electrical and Computer Engineering 7

754763772

78179799

80881782683584485386287188889898907916

925

934943

952

952

961

961

961

97

97

9797

979

The parameter selection result of SVC (contour plot)[GridSearchMethod]

minus2 0 2 4 6 8 10minus10

minus9minus8minus7minus6minus5minus4minus3minus2minus1

0

minus2 0 2 4 6 8 10

minus10minus9minus8minus7minus6minus5minus4minus3minus2minus1030405060708090

100

The parameter selection result of SVC (3D view)[GridSearchMethod]

Accu

racy

()

log 2

g

log 2g

log 2c

log 2c

Best c = 05 g = 1 CVAccuracy = 100 Best c = 05 g = 1 CVAccuracy = 100

Figure 5 Precisely selected parameters

5 Conclusions

(1) The basic idea of wavelet packet analysis is to cen-tralize information and energy to find out laws fromdetails and provide a more precise method for signalanalysis Thus elevator faults were diagnosed accord-ing to theory on optimal wavelet packet with the leastsquares support vector machine

(2) Feature vectors of energy distributed by elevatorvibration signals and feature vectors of faults con-structed by time-domain parameters were used asinput of LS-SVM Elevator malfunctions were clas-sified and identified with a well-trained LS-SVMclassifier Parameters of LS-SVM were optimized viacross validation in order that the classifier couldclassify LS-SVMmost precisely

(3) Experimental results suggested that this method waseffective for identifying faults of elevators As anadvancedway for intelligently diagnosing faults it hasgood prospects for application in condition monitor-ing and fault diagnosis

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the science and technology planproject of national bureau of quality inspection (2013QK104)and the science and technology plan project of qual-ity and technical supervision bureau in Yunnan province(2013ynzjkj02)

References

[1] L Li A Study on the Measurement System and Technology ofFault Diagnosis for Elevators Tianjin University Tianjin China2002

[2] S Jie ldquoThe research on the influence of the elevatorrsquos comfort-able sensationrdquo China Elevator vol 1 pp 43ndash44 1991

[3] Z Changming ldquoThe evaluation method on the comfortablesensation of elevator vibrationrdquo Construction Mechanizationvol 6 pp 29ndash32 1988

[4] Y Fashan Study on Elevator Intelligent Diagnosis System BasedMulti-sensor Information Fusion Technology Xinjiang Univer-sity Xinjiang China 2013

[5] Z Tongbo ldquoThe research on the problem of elevator vibrationrdquoGuide of Sci-Tech Magazine vol 26 pp 269ndash279 2010

[6] T Baoguo Y Yunlong and Z Jiping ldquoAeroengine fault diagno-sis based on wavelet transform and neural networkrdquo Journal ofNaval Aeronautical and Astronautical Uniersity vol 24 no 3pp 241ndash244 2009

[7] T Zan M Wang G Li and R-Y Fei ldquoApplication of waveletpacket and fuzzy ART neural network to vibration exceptionmonitoring for grinding processrdquo Journal of Beijing Universityof Technology vol 34 no 7 pp 678ndash707 2008

[8] Y Yu and DWan ldquoSimulation research on an approach of faultdetection usingwavelet neural networksrdquoComputer Simulationvol 17 no 1 pp 24ndash27 2000

[9] Y Li and H-H Zhang ldquoIntelligent elevator detecting systembased on neural networkrdquo Journal of Beijing University ofTechnology vol 36 no 4 pp 440ndash444 2010

[10] B-P Tang F Li and R-X Chen ldquoFault diagnosis based onLittlewood-Paley wavelet support vector machinerdquo Journal ofVibration and Shock vol 30 no 1 pp 128ndash131 2011

[11] H Shuixia Z Guangming andQ Chunling ldquoThe elevator faultdiagnosis system based on KPCA and SVMrdquoMachinery Designamp Manufacture vol 1 pp 196ndash198 2010

[12] F Li B-P Tang and W-Y Liu ldquoFault diagnosis based onleast square support vector machine optimized by geneticalgorithmrdquo Journal of Chongqing University (Natural ScienceEdition) vol 33 no 12 pp 14ndash20 2010

[13] AWidodo andB-S Yang ldquoSupport vectormachine inmachinecondition monitoring and fault diagnosisrdquo Mechanical Systemsand Signal Processing vol 21 no 6 pp 2560ndash2574 2007

[14] W Di A study on the multi-parameter embedded type on-site measurement system for elevator rails [MS thesis] TianjinUniversity Tianjin China 2003

[15] C Li The Analysis on the Dynamic Characteristics of ElevatorSystem and the Research on the Method of Fault DiagnosisZhejiang University Hangzhou China 1997

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

8 Journal of Electrical and Computer Engineering

[16] C Xiang ldquoThe analysis on the vibration of elevator operationrdquoChina High-Tech Enterprises vol 21 pp 60ndash62 2012

[17] L Zhuocheng ldquoThe analysis on the causes and measures ofelevator car vibrationrdquo China New Technologies and Productsvol 23 pp 113ndash115 2012

[18] S Dianbin ldquoThe discussion on the reason and solution of eleva-tor shakingrdquo Heilongjiang Science and Technology Informationvol 12 pp 45ndash46 2011

[19] Y Yunlong ldquoFault diagnosis of rolling bearing based onkurtosis-wavelet packet analysisrdquo New Technology amp New Pro-cess vol 5 pp 43ndash46 2008

[20] V Vapnik The Nature of Statistical Learning Theory SpringerNew York NY USA 2nd edition 1998

[21] B Jiang W Liu Z Dai and H Wang ldquoWorking conditionbased optimization framework for operational patterns and itsapplication in petrochemical industryrdquo CIESC Journal vol 63no 12 pp 3978ndash3984 2012

[22] L Ma H Zhang H Zhou and Q He ldquoFlow regime identifi-cation of oil-water two-phase flow based on HHT and SVMrdquoJournal of Chemical Industry and Engineering vol 58 no 3 pp617ndash622 2007 (Chinese)

[23] S Tao D Chen and W Hu ldquoGradient algorithm for selectinghyper parameters of LSSVM in process modelingrdquo Journal ofChemical Industry and Engineering vol 58 no 6 pp 1514ndash15172007 (Chinese)

[24] Q-M Zhang and H-J Liu ldquoApplication of LS-SVM in classifi-cation of power quality disturbancesrdquo Proceedings of the ChineseSociety of Electrical Engineering vol 28 no 1 pp 106ndash110 2008

[25] Q Wang and X Tian ldquoSoft sensing based on KPCA andLSSVMrdquo CIESC Journal vol 62 no 10 pp 2813ndash2817 2011

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of