Research Article A Hybrid Domain Degradation Feature Extraction Method for Motor Bearing Based on Distance Evaluation Technique

Transcription

1 Hindawi International Journal of Rotating Mahinery Volume 7, Artile ID 6754, pages Researh Artile A Hybrid Domain Degradation Feature Extration Method for Motor Bearing Based on Distane Evaluation Tehnique Baiyan Chen, Hongru Li, He Yu, and Yukui Wang Mehanial Engineering College, Shijiazhuang 53, China Air Fore Logistis College of PLA, Xuzhou, China Correspondene should be addressed to Hongru Li; lihr68@sohu.om Reeived 4 November 6; Revised 8 Deember 6; Aepted 9 January 7; Published 4 January 7 Aademi Editor: Dong Wang Copyright 7 Baiyan Chen et al. This is an open aess artile distributed under the Creative Commons Attribution Liense, whih permits unrestrited use, distribution, and reprodution in any medium, provided the original work is properly ited. The vibration signal of the motor bearing has strong nonstationary and nonlinear harateristis, and it is arduous to aurately reognize the degradation state of the motor bearing with traditional single time or frequeny domain indexes. A hybrid domain feature extration method based on distane evaluation tehnique (DET) is proposed to solve this problem. Firstly, the vibration signal of the motor bearing is deomposed by ensemble empirial mode deomposition (EEMD). The proper intrinsi mode funtion (IMF) omponent that is the most sensitive to the degradation of the motor bearing is seleted aording to the sensitive IMF seletion algorithm based on the similarity evaluation. Then the distane evaluation fator of eah harateristi parameter is alulated by the DET method. The differential method is used to extrat sensitive harateristi parameters whih ompose the harateristi matrix. And then the extrated degradation harateristi matrix is used as the input of support vetor mahine (SVM) to identify the degradation state. Finally, It is demonstrated that the proposed hybrid domain feature extration method has higher reognition auray and shorter reognition time by omparative analysis. The positive performane of the method is verified.. Introdution Rolling bearing is one of the most paramount parts of the motor, whih an support rotor and diretly affet the operation of the motor [, ]. The degraded bearing an give rise to a deline in the performane of the motor and even lead to the improper work of the entire system [3]. Therefore, it is of great signifiane and pratial value to extrat degradation feature. As plenty of state information is hidden in the motor bearing vibration signal, the analysis of vibration signal is one of the main methods to extrat degradation feature. The degraded bearing itself easily trigger nononentri fault of the motor stator and rotor. Consequently, the motor bearing vibration signals show strong nonstationary and nonlinear harateristis [4, 5]. The diffiulty and omplexity of extrating effetive degradation feature is greatly inreased. Degradation state reognition of the bearing onsists of three steps, inluding information aquisition, feature extration, and pattern reognition [6, 7]. Aordingly, it is the key of degradation state reognition to extrat degradation feature aurately and effetively. EEMD method is espeially appliable to the analysis of nonstationary and nonlinear signals, and the method has a wide range of appliations in many fields [8, 9]. Žvokelj et al. [] utilized EEMD method to proess aousti emission (AE) signal of slewing bearing, and the mixed domain fault feature of eah IMF omponent was extrated through prinipal omponent analysis (PCA). Lei et al. [] ombined EEMD and wavelet neural network (WNN) to ahieve loomotive roller bearing fault diagnosis. Jiang et al. [8] proposed an improved EEMD method with multiwavelet paket, and it was applied to rotating mahinery multifault diagnosis. It is bound to be refleted in a ertain IMF omponent when the bearing fault appears. The degradation harateristis will be more obvious after the amplifiation of IMF. Therefore, it an ahieve a better feature extration effet based on EEMD. Single domain feature is arduous to fully and aurately desribe the different degrees of degradation in omplex mehanial systems []. A method based on a hybrid domain feature extration is proposed to make full use of time domain, frequeny domain, and time-frequeny domain harateristi information.

2 International Journal of Rotating Mahinery Thefeaturefusionmethodanmakefulluseofthefeature information of multiple domains, whih an reflet the degradation state of the equipment more omprehensively. Yu[3]fusedmultipledomainfeatureswithloalitypreserving projetions (LPP) for bearing performane degradation assessment. Chen et al. [4] utilized Loal tangent spae alignments(ltsa)tofusemultipledomainfeaturesforrolling bearing multifault diagnosis study. However, the fused feature does not enjoy a lear physial meaning, and it an not explain the speifi performane of the feature as the degenerative state hanges. The extrated feature values often appear in the following two onditions [5]. One is the harateristi parameter whih has no relation with the lassified target, and the other is the redundant harateristi parameter whih has high orrelation with other harateristi parameters. Feature seletion is to selet the most effetive features from a set of features to redue the dimension of the feature spae [6]. Compared with feature fusion method, the advantage of using feature seletion methods for dimension redution is that the feature is not transformed and still maintains the original physial meaning [7 9]. As a lassial algorithm for feature seletion, DET an redue the time of degradation statereognitionandensuretheaurayofreognitionat thesametime. Based on the above analysis, motor bearing hybrid domain degradation feature extration based upon EEMD and DET method is proposed in this paper. Firstly, EEMD method is used to deompose the vibration signal of the motor bearing, and the sensitive omponents are seleted aording to the sensitive fators. Hybrid domain degradation features are onstruted, and sensitive omponents areseletedbydetmethod.finally,thedegradationstate reognition rate obtained by SVM is used to verify the auray of the feature extration. The paper is organized as follows: Setion presents the EEMD method. And the seletion of the sensitive IMF omponent is also given in this setion. Setion 3 presents a mathematial analysis in detail to selet the most effetive featuresbasedondetmethod.setion4presentsafavorable disussion and analysis of the experimental results by omparingtheproposedmethodwithsingledomaindegradation feature extration and high dimension degradation feature extration. Finally, our onlusions are provided in Setion 5.. Algorithm of EEMD.. EEMD Method. A new noise-assisted data analysis method was proposed to overome the problem of mode mixing in EMD []. It is EEMD whih defines the true IMF omponents as the mean of an ensemble of trials. Eah trial onsists of the deomposition results of the signal adding a white noise of finite amplitude. The algorithm is desribed as follows []. () Add white noise sequene to the signal x(t). x (t) =x(t) +n(t), () where x (t) represents the noise-added signal and n(t) indiates the added white noise. () x (t) is deomposed into IMFs aording to the EMD method. (3) Repeat steps () and () by adding a different white noise sequene. (4) The average value of the orresponding IMFs is obtained... Seletion Algorithm Based on the Similarity Evaluation. The signal is deomposed into a group of IMFs by EEMD. A portion of the IMFs are sensitive omponents losely related to the degradation state, while the others are degenerate independent or noise interfering omponents. Consequently, it is essential to hoose sensitive IMFs whih are losely related to the bearing degradation before evaluating the sensitivity of eah harateristi parameter. It an improve the auray of the degradation feature extration by ignoring other irrelevant IMFs []. A sensitive IMF seletion algorithm based on similarity evaluation is adopted in this paper. The speifi algorithm of sensitive IMF seletion algorithm[3]isasfollows. () Calulate orrelation oeffiients u n between signal x(t) and its IMF n,wheren=,,...,n. () Calulate orrelation oeffiients β n between the IMF n of signal x(t) and signal x(t) olleted under normal operating onditions. (3) The orrelation oeffiient of bearing degradation is defined as η n =β n u n, () where n=,,...,n. (4) DefineandalulateIMFsdegradationsensitivity fator λ n as λ n = η n min (η) max (η) min (η), (3) where n=,,...,n. (5) All IMFs are sorted in aordane with the sensitive fator from large to small order, and the differene of the sensitivity fator of two adjaent IMFs is alulated. The index of the orresponding maximum differene n is found. Then the former n IMFs are degradation feature sensitive IMF omponents. IMF sensitivity fator not only onsiders the similarity between IMF and the signal itself but also takes into aount thesimilaritybetweentheimfandthenormalsignal.the IMFs ontaining more fault information are more similar to the origin signal. The information that is not related to thefaultinimfanbeevaluatedthroughalulatingthe orrelation oeffiient between the IMF omponent and the normal signal. Consequently, the method is more aurate than only alulating the orrelation oeffiient between IMF omponent and origin signal in seleting the sensitive IMF omponents.

3 International Journal of Rotating Mahinery 3 Table:Hybriddomainfeatures. Peak value x p = x i max Root mean square amplitude Absolute mean value Root mean square value Peak index Waveform index Clearane fator Center of gravity frequeny N X r =( x / N i ) μ x = N i= N i= X rms = N C= X p X rms I f = X p μ x CL f = X p X r x i N x i i= FC = N i= x i x i π N i= x i N i= Mean square frequeny MSF = x i 4π N i= x i Root mean square frequeny RMSF = MSF Frequeny variane VF = MSF (FC) Permutation entropy H p = k j= P j ln(p j ) ln(m!) 3. DET Method and Degradation Feature Extration 3.. Charateristi Parameters of Motor Bearing. IMF sensitive omponent is regarded as the researh objet after EEMD deomposition. harateristi parameters existing in the time domain, the frequeny domain, and the time-frequeny domain are extrated from eah IMF omponent, whih are onduive to status identifiation for motor bearings. Time domain parameters inlude peak value, root mean square amplitude, absolute mean value, root mean square value, peak index, waveform index, and learane fator. Frequeny domain parameters inlude enter of gravity frequeny, mean square frequeny, root mean square frequeny, and frequeny variane. Permutation entropy (PE) is determined as a harateristi parameter based on omplexity, whose embedding dimension m = 6 and the time delay τ = [4]. This paper proposes to use hybrid features displayed in Table. x i represents the vibration signal. x i = (x i x i ) F s ; F s represents sampling frequeny. N represents sampling points. P j is used to denote the probability distribution of eah symbol sequenes. 3.. Calulation of Distane Evaluation Fator. DET is a kind of feature evaluation tehnology based on inner lass distane [5]. The evaluation priniple an be desribed as that the same kind of inner harateristi distane is the smallest, and the distane between two different lasses is the biggest. Features omplying with this priniple are onsidered to be sensitive features. The speifi implementation proess of the evaluation method is as follows [5]. Assume there are a total of N speies of degradation states of the motor bearing and M samples are seleted for eah state and C parameters are extrated from eah sample. An obtained harateristi matrix is {f n,m,, n=,,...,n; m=,,...,m; =,,...,C}. f n,m, represents th harateristi of mth sample of nth degradation state. () The inner lass average Eulidean distane of the motorbearinginthesamestateisalulated.(distaneis Eulidean distane without speial instrutions.) D n, = M (M ) M l,m= l=m f m,n, f l,n,. (4) The average distane between the lasses of N speies is obtained. D (w) = N N n= D n,. (5) () The differene fator of the inner lass distane in the same ondition is defined by analyzing the frequeny spetrum of the vibration signal. ] (w) = max (D n,) min (D n, ). (6) (3) The average value of eah feature of all samples in the same lass is alulated. u n, = M f M m,n,. (7) m= The average distane between different states of eah feature is obtained. D (b) = N (N ) N n,e= n=e u n, u e,. (8) (4) The status of the motor bearing is examined by the variation of the harateristi parameter. The differene between lass distanes is defined as follows: ] (b) = max ( u n, u e, ) min ( u n, u e, ), (9) where n, e=,,...,n; n =e. (5) Weighting fators are alulated. λ = ] (w) / max (] (w) )+] (b) / max (] (b) ). () (6) The ratio of all parameters average distane in different states and average distane in the same state are alulated onsidering the influene of weighting fators. D (b) α =λ D (w). ()

4 4 International Journal of Rotating Mahinery (7) The results obtained are normalized. β = α max (α ). () β is the distane evaluation fator of harateristi parameters. It reats to the sensitivity of eah harateristi parameter diretly Seletion of Sensitive Degradation Features Based on Distane Evaluation Fator. It is a knotty problem how to make use of the distane evaluation fator to selet the sensitive degradation features to redue the reognition time. Currently, there is no aepted standard for the seletion of thresholds [6]. Quite a few sholars have been inspired by the Wrapper method. They used the lassifier to determine the threshold value of the method to sreen the sensitive features [5]. However, the method has the shortomings of high time omplexity, large alulation, and poor generalization ability. In addition, a number of tests are needed [9]. The differential method is utilized to sreen sensitive degradation harateristis in this paper. Compared with the method of determining the threshold value aording to the reognition rate of lassifier, the alulation is smaller, and the degradation features an be seleted without the intervention of the lassifier. Firstly, the harateristi parameters are extrated as degradation state and the operating ondition (five kinds of degradation States, four operating onditions) synhronously hange and the operating onditions (four onditions, the inner rae degradation.7 inh) barely hange. The distane evaluation fator of eah parameter in thetwoasesisdeterminedbyalulatingthedistaneevaluation fator. The differene between the distane evaluation fator of the degradation state and the operating ondition atthesametimeissubtratedfromthedistaneevaluation fator when the ondition is hanged. The differene between distane evaluation fator for degradation state and operating ondition simultaneous hanging and distane evaluation fator for only operating onditions hanging is obtained. For D kinds of different onditions, β d aording to the above method an be obtained. The differene between β and β d is χ. The harateristi parameters are more sensitive to the hange of the state of the motor bearing as χ is greater. The harateristi parameters of χ > are seleted as the sensitive parameters that exlude the influene of operating ondition Support Vetor Mahine (SVM). Support vetor mahine (SVM) is an effetive omputational learning method presented by Vapnik [7], whih speializes for a smaller number of samples for training. The basi priniple of SVM an be illustrated in a two-dimensional way as represented in Figure. It demonstrates the lassifiation of a series of points for two different lasses, Class and Class. The SVM tries to alulate the distane between the boundary and the nearest data point in eah lass maximal. The nearest data points are regarded as support vetors. SVM lassifiation funtion output is a linear ombination of intermediate nodes, and eah intermediate node Maximal margin Class H H Class Support vetors H The optimal separating hyper plane Figure : The optimized separating hyper plane in lassifiation. b Input vetors x φ(x ) φ(x ) φ(x i ) K(x, x ) K(x, x ) K(x,x i ) α α α i f(x) Output Weights Figure : The arhiteture of SVM. Mapped vetors Inner produt orresponds to a support vetor. The arhiteture of SVM is manifested in Figure. Consider a training sample set, S = {(x i,y i ) x i R N, y i {,}, i =,,...,l},wherex i is an input vetor and y i is a label of x i. The general form of linear disriminant funtion in n-dimensional spae is g(x) = (w x) + b. The data in the set an be orretly lassified by the optimized hyperplane. SVM turns to the following dual optimization problem for lassifiation problem: minimize n ζ i i= wt w+c subjet to y i [(w x i +b)]+ζ i, (3) where w and b represent undetermined oeffiient. C represents penalty fator. ζ i represents relaxation fator. The final result is a disrimination funtion f(x) onveniently expressed as a funtion of the data in the lower dimensional feature spae: f (x) = sgn ( n i= α i y i K(x i,x)+b). (4) Kernel funtion an avoid omplex operation in high dimensional feature spae. The most widely adopted kernel

5 International Journal of Rotating Mahinery 5 Input x SVM No SVM No SVM k No f(x) =? f(x) =? f(x) =? Yes Yes Yes Fault type Fault type Fault type k Fault type k + Figure 3: One to others multilass SVM lassifier. The original vibration signal is deomposed by EEMD when the degradation state and the operating ondition synhronously hange The sensitive omponents are seleted as the analysis objet and the hybrid domain feature extration is ahieved The distane evaluation fator of eah feature is alulated through DET method The distane evaluation fator of eah feature is alulated only when the operating ondition hanges The sensitive features is seleted by the differential method and they form a degradation feature matrix The validity of the method is verified by the reognition auray of SVM Figure 4: Motor bearing degradation feature extration proess based on DET. funtion is the radial basis funtion (RBF), and it is defined as follows: K(x,x i )=exp ( x x i σ ). (5) SVM is mainly used as one to one binary lassifier and one to others multilass lassifier. SVM is utilized to identify the different degradation states of the motor bearings. Aordingly, multilass lassifier is determined in this paper. A multilass lassifier onsists of an oean of binary lassifier. The proess of the multilass SVM is desribed in Figure Hybrid Domain Degradation Feature Extration Method Based on DET. The proess of motor bearing degradation feature extration based on DET is shown in Figure 4. () The original vibration signal is deomposed by EEMD when the degradation states and the operating onditions synhronously hange, and the sensitive IMF omponents are seleted as the analysis objet aording to the similarity evaluation algorithm. () For eah sensitive IMF omponent, harateristi parameters are extrated to form a hybrid domain feature matrix. (3) The distane evaluation fators of harateristi parameters are obtained aording to the DET method. (4) Repeat steps () (3) in a speial degradation state, only when the operating onditions hange. (5) Aording to the differene method, the sensitive features are seleted to form the degradation feature matrix. (6) Finally, the effetiveness of the proposed method is verified by SVM. 4. Experimental Results 4.. Experiment Data Soures. The data of deep groove ball bearings of 65-RS JEM SKF from the bearing data enter of Case Western Reserve University is used as the experimental raw data in our study [8]. The test bed is manifested in Figure 5. The test rig onsists of a Hp (horsepower) motor, a torque onverter, a dynamometer, and the ontrol iruit. The vibration signals, inluding the normal, inner rae fault, outer rae fault, and rolling element fault signal, are olleted by an aeleration sensor installed on the magneti base shell with a sampling frequeny of khz, and sampling points are 4. In this paper, the degradation states of inner rae serve as an example for analysis, and four operating onditions exist in eah state. For eah fault mode, data are olleted under four operating onditions orresponding to motor speeds of 73, 75, 77, and 797 r/min. Inner rae fault diameter is.7,.4,.,and.8.thenormalstateanbeseenasa speial degradation state, and there are five kinds of degradation states. The vibration signal of the five degradation states and their frequeny spetrum are shown in Figure Degradation Feature Extration Based on DET. First of all, it is analyzed under the ondition of five kinds of degradation states, and four kinds of working onditions exist in eah state. The olleted signal is deomposed by EEMD method, and the sensitive omponents are seleted based on the similarity assessment of sensitive IMF seletion algorithm. The results are shown in Figure 7. Eah signal an be deomposed to get the first two IMF omponents for the sensitive IMF omponents. The sensitive omponents of the five degradation states are manifested in Figure 8. Feature extration is performed by seleting the IMF omponents whih ontain the main degradation state information. In this paper, mixed domain parameters of the IMF omponents are extrated as the harateristi value, inluding peak value, root mean square amplitude, absolute mean value, root mean square value, peak index, waveform index, learane fator, enter of gravity frequeny, mean square frequeny, root mean square frequeny, frequeny variane, and permutation entropy. Therefore, 4 harateristi values an be extrated for eah signal. The harateristis of the IMF omponent values are t t, and harateristis of IMF omponent values are t 3 t 4. Then, the degradation features are extrated

6 6 International Journal of Rotating Mahinery Housing Aelerometer Magneti base Drive end bearing Torque transduer Motor Dynamometer Figure 5: The bearing experimental system Normal signal Frequeny (Hz) Inner rae degradation diameter Frequeny (Hz) 3 Inner rae degradation diameter Frequeny (Hz) Inner rae degradation diameter Frequeny (Hz) 4 4 Inner rae degradation diameter Inner rae degradation diameter Frequeny (Hz) Figure 6: Time domain and frequeny spetrum of five-degradation state vibration signal of the motor bearing.

7 International Journal of Rotating Mahinery Sensitive fator differene Sensitive fator differene IMF omponent IMF omponent (a) Inner rae degradation diameter.7 (b) Inner rae degradation diameter Sensitive fator differene Sensitive fator differene IMF omponent IMF omponent () Inner rae degradation diameter. (d) Inner rae degradation diameter.8 Figure 7: The sensitive IMF omponents based on similarity evaluation. through the DET method. Before feature seletion, the sensitivity to lass of eah harateristi parameter that is the distane evaluation fator β has been alulated. In a speifi degradation state, the distane evaluation fator β d is obtained in aordane with the same method only when four kinds of operating onditions hange. The differenes between β and β d named χ are demonstrated in Figure 9. The positive diretion of y-axis indiates that it is sensitive to the hange of degradation state and is not sensitive to the hange of operating ondition. It is ontrary to the negative diretion of y-axis. The sensitive harateristi values are seleted by the differential method, and the degradation harateristi matrix is further formed. Distane evaluation fators and their differenes are displayed in Table. The seleted feature parameters, that is, [t,t 8,t 7,t 6, t 9,t 3,t 4,t 9,t,t 3 ],anbelearlyseenaordingtofigure 9. Degradation feature is redued from 4 dimensions to dimensions and the physial meaning of the original feature is maintained The Effetiveness of the Degradation Feature Extration. In order to illustrate the effetiveness of the proposed method, SVM method is used to verify the effetiveness of the degradation feature extration. The reognition auray of SVM output and the reognition time of SVM are used as the basis for judgment [4]. Fifty groups of samples are trained for eah degenerate state, and fifty groups of samples are tested. The RBF kernel funtion is seleted as SVM kernel funtion, and its lassifiation performane is better than other types of nulear funtion [9]. The penalty parameter C and kernel funtion parameter g are seleted and the partile swarm optimization (PSO) algorithm is used to optimize theproess[3].thenumberofpartilesissettoper degenerate state. Iteration maximal number is set to. The range of C is set from. to, and the range of g is set from. to. The highest reognition auray is taken as the optimization target. The fitness urve of parameter optimization is manifested in Figure (sine the iteration of the twelfth generation has been onvergent, only the first fifty generations are listed). The time of PSO searhing for

8 8 International Journal of Rotating Mahinery Normal signal IMF Inner rae degradation diameter.7 IMF Inner rae degradation diameter.4 IMF Inner rae degradation diameter. IMF Figure 8: Continued. Normal signal IMF Inner rae degradation diameter.7 IMF Inner rae degradation diameter.4 IMF Inner rae degradation diameter. IMF

9 International Journal of Rotating Mahinery 9 4 Inner rae degradation diameter.8 IMF.5 Inner rae degradation diameter.8 IMF Figure 8: The sensitive IMF omponents of the five degradation states. Sensitivity Charateristi parameters Figure 9: Sensitivity of eah harateristi parameter. optimized parameter is.8546s (operating platform for Rb Matlab; the main omputer is onfigured as follows: CPU Intel(R) Core(TM) i GHz, memory 4 G). The result of partile seeking optimal parameters in twodimensional spae is that the optimal penalty parameter C= 4 and the kernel funtion parameter g =.5. The identifiation result is demonstrated in Figure. Test sample labels,, 3, 4, and 5, respetively, represent five degradation states. The final reognition rate reahes 99.6% and the reognition time is s. The results indiate that most of the test data are identified orretly. The effetivenessoftheproposedmethodisverified Comparative Analysis. In order to further verify the effet of the degradation feature extration based on DET, we use the single domain feature extration and high dimension feature extration to ompare the proposed feature extration. The result of degradation state reognition rate and the reognition time is displayed in Table 3. Table : Distane evaluation fators and their differenes between β and β d. Feature parameter β β d χ t t t e t e 4 t t t t t t t t t t t t t t t t t t t t Firstly, single domain feature extration is performed for sensitive IMF omponents. Aording to Table, the root mean square frequeny of IMF is the most sensitive. This feature is seleted as the degradation feature, and the effetiveness of the feature extration is verified by SVM. The degradationstatereognitionresultisshowninfigure. Seondly, the feature extration of high dimension degradation is performed for sensitive IMF omponents, that is,

10 International Journal of Rotating Mahinery Method Table 3: Reognition rate and reognition time. Reognition rate Reognition time The single domain feature extration 98.4% s High dimension feature extration 99.% s The proposed method 99.6% s Test set sample label Sample Fitness Test set sample label Generations of evolution PSO Figure : The fitness urve of parameter optimization Sample Raw data SVM reognition result Figure : Degradation state reognition results. sreening sensitive features without the method of DET. 4 harateristis are used to form a degeneration feature matrix, andsvmisusedtoverifytheeffetivenessofthefeature extration. The reognition rate is manifested in Figure 3. Through the omparative analysis, we an draw the following onlusions: the proposed method an extrat the degradation state information of motor bearings more effetively. The redundant features of the motor bearing degradation state an be removed, and the degradation feature an be extrated more aurately. Raw data SVM reognition result Figure : Degradation state reognition results for single domain degradation feature. Test set sample label Sample Raw data SVM reognition result Figure 3: Degradation state reognition results for high dimension degradation feature. 5. Conlusions In this paper, a hybrid domain degradation feature extration method based on DET is proposed to extrat the degradation harateristis. Finally, the reognition rate of SVM output demonstrates the validity and auray of the feature extrationmethod.throughtheanalysis,weangetthefollowing onlusions. () The hybrid domain feature is more omprehensive thanthesingledomainfeature,whihanrefletthedegradation state of the bearing aurately. The proposed method an be overwhelmingly aurate to selet the appropriate harateristi parameters of the motor bearing degradation state. It an ahieve the degradation feature extration more aurately. () The DET method an remove redundant features and unrelated features effetively, whih an maintain the physial meaning of the original feature parameters and shorten the reognition time. (3) The sensitive feature seletion is related to the quality of the feature extration. The degradation feature is extrated through the proposed method, whih is applied to the atual vibration signal of the motor bearing. The reognition rate

11 International Journal of Rotating Mahinery an reah 99.6%, and the result indiates the validity and feasibility of the proposed method. Competing Interests The authors delare that they have no ompeting interests. Aknowledgments The authors are grateful to Case Western Reserve University. This projet is supported by National Natural Siene Foundation of China (Grant no ). Referenes []Z.Li,J.Zhu,X.Shen,C.Zhang,andJ.Guo, Faultdiagnosis of motor bearing based on the Bayesian network, Proedia Engineering,vol.6,pp.8 6,. [] B.Wang,H.-R.Li,andB.-H.Xu, Motorbearingforeastfeature extrating and degradation status identifiation based on multisale morphologial deomposition spetral entropy, Journal of Vibration and Shok,vol.3,no.,pp.4 39,3. [3] J. Yang and M. Zhao, Fault diagnosis of tration motor bearings using modified bispetrum and empirial mode deomposition, Proeedings of the Chinese Soiety of Eletrial Engineering, vol.3,no.8,pp.6,. [4] P. Konar and P. Chattopadhyay, Bearing fault detetion of indution motor using wavelet and Support Vetor Mahines (SVMs), Applied Soft Computing, vol., no. 6, pp. 43 4,. [5] D.Zhu,Q.Gao,D.Sun,Y.Lu,andS.Peng, Adetetionmethod for bearing faults using null spae pursuit and S transform, Signal Proessing,vol.96,no.5,pp.8 89,4. [6] S.Hong,Z.Zhou,E.Zio,andK.Hong, Conditionassessment for the performane degradation of bearing based on a ombinatorial feature extration method, DigitalSignalProessing, vol. 7, pp , 4. [7] B. Wang, H.-R. Li, Q.-H. Chen, and B.-H. Xu, Rolling bearing performane degradative state reognition based on mathematial morphologial fratal dimension and fuzzy enter means, Ata Armamentarii,vol.36,no.,pp.98 99,5. [8]H.Jiang,C.Li,andH.Li, AnimprovedEEMDwithmultiwavelet paket for rotating mahinery multi-fault diagnosis, Mehanial Systems and Signal Proessing,vol.36,no.,pp.5 39, 3. [9]N.Mariyappa,S.Sengottuvel,R.Pateletal., Denoisingof multihannel MCG data by the ombination of EEMD and ICA and its effet on the pseudo urrent density maps, Biomedial Signal Proessing and Control,vol.8,pp.4 3,5. [] M. Žvokelj, S. Zupan, and I. Prebil, EEMD-based multisale ICA method for slewing bearing fault detetion and diagnosis, Journal of Sound and Vibration,vol.37,pp ,6. [] Y. Lei, Z. He, and Y. Zi, EEMD method and WNN for fault diagnosisofloomotiverollerbearings, Expert Systems with Appliations, vol. 38, no. 6, pp ,. [] H.-M. Dai, A.-Q. Xu, W.-F. Li, and W.-C. Sun, Fault diagnosis of rolling bearings in mixed domain based on WMRMR, Journal of Vibration and Shok,vol.34,no.9,pp.57 6,5. [3] J.-B. Yu, Bearing performane degradation assessment using loality preserving projetions, Expert Systems with Appliations,vol.38,no.6,pp ,. [4] F. Chen, B. Tang, T. Song, and L. Li, Multi-fault diagnosis study on roller bearing based on multi-kernel support vetor mahine with haoti partile swarm optimization, Measurement: Journal of the International Measurement Confederation,vol.47,no., pp , 4. [5] Y.Lei,Z.He,andY.Zi, Faultdiagnosisbasedonnovelhybrid intelligent model, Chinese Journal of Mehanial Engineering, vol.44,no.7,pp. 7,8. [6]J.BenAli,L.Saidi,A.Mouelhi,B.Chebel-Morello,andF. Fnaieh, Linear feature seletion and lassifiation using PNN and SFAM neural networks for a nearly online diagnosis of bearing naturally progressing degradations, Engineering Appliations of Artifiial Intelligene,vol.4,pp.67 8,5. [7] S. M. Vieira, J. M. Sousa, and U. Kaymak, Fuzzy riteria for feature seletion, Fuzzy Sets and Systems. An International Journal in Information Siene and Engineering, vol.89,pp. 8,. [8] S. Mitra, P. P. Kundu, and W. Pedryz, Feature seletion using strutural similarity, Information Sienes, vol. 98, pp. 48 6,. [9] X. Yao, X. D. Wang, Y. X. Zhang, and W. Quan, Summary of feature seletion algorithms, ControlandDeision.Kongzhiyu Juee,vol.7,no.,pp.6 66,9,. []T.Wang,M.Zhang,Q.Yu,andH.Zhang, Comparingthe appliations of EMD and EEMD on time-frequeny analysis of seismi signal, Journal of Applied Geophysis,vol.83,no.6,pp. 9 34,. [] Z. H. Wu and N. E. Huang, Ensemble empirial mode deomposition-a noise assisted data analysis method, Advanes in Adaptive Data Analysis,vol.,no.,pp. 4,9. [] J. Dybała and R. Zimroz, Rolling bearing diagnosing method based on empirial mode deomposition of mahine vibration signal, Applied Aoustis,vol.77,pp.95 3,4. [3] Y. Lei, Mahinery fault diagnosis based on improved Hilbert- Huang transform, Journal of Mehanial Engineering, vol. 47, no.5,pp.7 77,. [4] J.-D. Zheng, J.-S. Cheng, and Y. Yang, A rolling bearing fault diagnosis method based on LCD and permutation entropy, Journal of Vibration, Measurement & Diagnosis, vol.34,no.5, pp.8 86,4. [5] Y. Yang, H.-Y. Pan, and J.-S. Cheng, The rolling bearing fault diagnosis method based on the feature seletion and RRVPMCD, Journal of Vibration Engineering,vol.7,no.4,pp , 4. [6] J. Xie, J. Lei, W. Xie et al., Hybrid feature seletion methods basedond-soreandsupportvetormahine, Journal of Computer Appliations,vol.3,no.,pp ,. [7] V. N. Vapnik, The Nature of Statistial Learning Theory,Springer, New York, NY, USA, 995. [8] K. A. Loparo, Case Western Reserve University Bearing Data Center. [EB/OL], pages/download-data-file. [9] K. Zhu, X. Song, and D. Xue, A roller bearing fault diagnosis method based on hierarhial entropy and support vetor mahine with partile swarm optimization algorithm, Measurement: Journal of the International Measurement Confederation,vol.47,no.,pp ,4. [3] Z. Liu, H. Cao, X. Chen, Z. He, and Z. Shen, Multi-fault lassifiation based on wavelet SVM with PSO algorithm to analyze vibration signals from rolling element bearings, Neuroomputing,vol.99,pp.399 4,3.

12 International Journal of Rotating Mahinery Engineering Journal of Volume 4 The Sientifi World Journal Volume 4 International Journal of Distributed Sensor Networks Journal of Sensors Volume 4 Volume 4 Volume 4 Journal of Control Siene and Engineering Advanes in Civil Engineering Volume 4 Volume 4 Submit your manusripts at Journal of Journal of Eletrial and Computer Engineering Robotis Volume 4 Volume 4 VLSI Design Advanes in OptoEletronis International Journal of Navigation and Observation Volume 4 Chemial Engineering Volume 4 Volume 4 Ative and Passive Eletroni Components Antennas and Propagation Aerospae Engineering Volume 4 Volume 4 Volume 4 International Journal of International Journal of International Journal of Modelling & Simulation in Engineering Volume 4 Volume 4 Shok and Vibration Volume 4 Advanes in Aoustis and Vibration Volume 4