Fault Diagnosis in Rolling Element Bearing Using Filtered Vibration and Acoustic Signal

Transcription

1 Volume 8 o , ISS: (prited versio); ISS: (o-lie versio) url: ijpam.eu Fault Diagosis i Rollig Elemet Usig Filtered Vibratio ad Acoustic Sigal Sudarsa Sahoo, J. K. Das School of Electroics Egieerig KIIT Uiversity Bhubaeswar, Idia sudarsa_iisc@yahoo.i; jite.cuttack@gmail.com Bapi Debath Departmet of Electroics & Istrumetatio Egieerig IT Silchar Silchar, Idia sura.cuttack@gmail.com Abstract- The defects preset i the rollig elemet bearigs may affect the performace of a machiery ad may reduces its efficiecy. So early detectio of the faults i the rollig elemet bearigs is very essetial. The vibratio or acoustic sigature geerated from the rollig elemet bearigs may be used as the measurig parameters for the fault diagosis. The rollig elemet bearig has its ow sigature(vibratio ad acoustic sigal) i its healthy coditio ad whe a defect occurs i it the its vibratio ad acoustic sigatures get chaged. The vibratio ad acoustic sigatures i healthy ad defective coditios are compared i time, frequecy ad timefrequecy domai to detect the fault. Though either vibratio or acoustic sigature aloe ca be used for the fault detectio but i this work both the vibratio ad acoustic sigatures are used for the fault detectio i the bearigs. I this work iitially the statistical aalysis is doe o the acquired vibratio ad acoustic sigature to detect the fault. The frequecy aalysis ad time-frequecy aalysis is doe for the fault detectio. FFT is used for frequecy aalysis ad wavelet trasform is used for time-frequecy aalysis. But the vibratio ad acoustic sigal used for the fault aalysis may gets corrupted by the oise ad i geeral is oisy. So before its aalysis the oise must be removed from this oisy vibratio ad acoustic sigals. I this work at the pre processig stage adaptive oise cacellatio is used to filter the oise from the acquired vibratio ad acoustic sigal ad hece to improve its sigal to oise ratio. I this work a experimetal set-up is developed to acquire the measured vibratio ad acoustic sigals ad the the experimet is carried out i two stages. I first stage adaptive oise cacellatio(ac) techique is implemeted o the acquired oisy vibratio ad acoustic sigals to remove the oise ad hece to ehace the SR. The i the secod stage three differet aalysis are doe o the AC filtered vibratio ad acoustic sigals to diagose the defect preset i the bearigs. I this experimet oe healthy bearig ad two differet defective bearigs are used for the experimetal work. The measured vibratio ad acoustic sigals are acquired from the set-up by moutig the bearigs oe by oe ad by usig the advaced istrumetatio system. Keywords- AC; SR; statistical aalysis; ORF; wavelet trasform. I. ITRODUCTIO Every rotatig elemet bearigs carries its ow vibratio ad acoustic sigature i its healthy coditio. But whe a defect occurs i it the a chage appears i its vibratio ad acoustic sigatures. This chage ca be observed i time ad frequecy aalysis. But the oise geerated by other elemets of the machiery may affect the aalysis process used to diagose the faults. So for this reaso the oise should be reduced from the acquired measured sigal before its aalysis. But as the oise is o-statioary i ature so, the adaptive oise cacellatio techique is used for the oise removal. So may work ca be foud i literature for the fault detectio ad diagosis i rollig elemet bearigs. P.K. Kakar et al. shows the use of CWT for the fault detectio i the bearigs[], [2]. I aother work P. Shakya et al. use the vibratio sigal as the measurig parameter to diagose the defect i rollig elemet bearigs[3]. V.Shamukha Priya et al. shows the applicatio of wavelet decompositio for bearig health coditio moitorig[4]. I literature it is also foud the use of wavelet trasform i machie coditio moitorig[5]. Sukhjeet sigh et al. use the curret sigature as the measurig parameter to diagose the fault[6]. It is also foud the use of adaptive morlet wavelet for the feature extractio i wid turbie[7]. J.I Taylor shows the aalysis of vibratio sigal i feature extractio[8]. It ca be foud from [9], [0] the details about wavelet aalysis for feature extractio i diagosig the faults. aother dimesio of this work is to filter the oise from the acquired oisy sigal to improve its SR before its aalysis to diagose the fault. Adaptive oise cacellatio techique is used for the oise filtratio. C.W. Liao et al. shows the applicatio of FIR based adaptive algorithm to improve the 95

2 AC techique[]. The LMS algorithm is used i adaptive filter for the error covergece[2]. so i this work at the pre-processig stage the AC oise cacellatio techique is employed to filter the oise. The feature is extracted from the filtered(oise free) vibratio ad acoustic sigals, to diagose the fault i the bearigs. I this work oe healthy bearig ad two defective bearigs are used for the experimetal work. II. PERFORMACE COMPARISO AD SELECTIO OF ADAPTIVE OISE CACELLATIO ALGORITHM I this sectio three adaptive oise cacellatio(ac) techiques are employed o the oisy vibratio sigal acquired from the experimetal set-up. The the performace of the three AC algorithms are compared based o the sigal to oise ratio(sr) ad mea square error(mse). The the best AC algorithm is selected for the pre-processig of all the acquired vibratio ad acoustic sigals from the experimetal set-up. The three AC algorithms tested are LMS, wavelet ad EMD. The priciple of AC de-oisig process is show i Fig.. I this process the oise is removed from the oisy vibratio sigal acquired from the defective bearig by usig the healthy bearig vibratio sigal as the referece sigal. From the compariso table EMD is foud the best amog the three AC techiques. So EMD algorithm is used at the pre-processig stage to remove the oise i all the experimetally acquired vibratio ad acoustic sigals before its aalysis to diagose the faults. III. EXPERIMETAL WORK To carry out the proposed experimet a experimetal setup is developed. The set-up cosists of a motor to drive the bearigs. I this work three bearigs of similar specificatio are take. The bearigs cosists of 8 umbers of balls ad havig ball diameter of 6.74 mm ad pitch diameter of 30 mm. Out of the three bearigs oe is a healthy bearig ad rest two are defective bearigs(type- defect ad Type-2 defect bearigs). A advaced istrumetatio system is used alog with this motor bearig arragemet for the acquisitio of the vibratio ad acoustic data. To acquire vibratio data a vibratio sesor (PCB 325c-03 accelerometer), a 4 chael DAQ card(i-9234) ad a PC with LabVIEW software is used. To acquire soud data a Brüel & Kjær 2 chael hadheld Soud aalyser is used. The experimetal set-up ad its model is show i Fig. 3 ad Fig. 4 respectively. Fig.. Schematic of AC de-oisig process Fig. 3. The set-up ad the istrumetatio system for the Experimet The filtered vibratio sigal before ad after implemetatio of the three AC techiques is show i Fig. 2 ad their performace compariso is show i Table I. Fig. 4. The model for the Experimetal set-up Fig. 2. Vibratio sigal before ad after implemetatio of the three AC techiques. TABLE I. COMPARISO OF AC TECHIQUES AC SR MSE Techiques LMS EMD Wavelet The healthy bearig, Type- defect bearig ad Type-2 defect bearigs used i the experimet are show i Fig. 5, Fig. 6 ad Fig. 7 respectively. To acquire the data iitially the healthy bearig is mouted o the set-up ad the correspodig vibratio ad acoustic data is acquired. The the Type- ad Type-2 defect bearig is mouted oe by oe ad the correspodig vibratio ad acoustic data is acquired. 96

3 Fig. 5. The healthy bearig Fig. 9. Filtered Vibratio sigal from Type- defect bearig Fig. 6. The Type- defect bearig Fig. 0. Filtered Vibratio sigal from Type-2 defect bearig The filtered acoustic sigal from the healthy, Type- ad Type-2 defect bearig is show i Fig., Fig. 2 ad Fig. 3 respectively. Fig.. Filtered acoustic sigal from healthy bearig Fig. 7. The type-2 defect bearig IV. RESULTS AD DISCUSSIO I this sectio iitially the statistical aalysis ad the the frequecy(fft aalysis) ad the time-frequecy (wavelet aalysis) is doe o the filtered vibratio ad acoustic sigals. I the aalysis the compariso of the sigals from the defective bearigs is made with the sigals from the healthy bearigs to diagose the fault. The filtered vibratio data from the healthy, Type- defect ad Type-2 defect bearig is show i Fig. 8, Fig. 9 ad Fig. 0 respectively. Fig. 2. Filtered acoustic sigal from Type- defect bearig Fig. 8. Filtered Vibratio sigal from healthy bearig Fig. 3. Filtered acoustic sigal from Type-2 defect bearig 97

4 A. Statistical Aalysis The statistical parameters are computed from the filtered vibratio ad acoustic sigals acquired from the healthy ad defective bearigs by usig the mathematical formulas. RMS = i= (X i μ) 2 () Mea = μ = i= (2) Peak Level = P v = [Max X i Mi X i ] (3) X i 2 Crest Factor = C = P v Skewess = S = Kurtosis = K = variace = σ 2 = RMS i= (X i μ ) 3 RMS 3 (5) i= (X i μ ) 4 (4) RMS 4 (6) i= (X i μ ) 2 Stadard Deviatio = σ = Clearace Factor = CI f = Impulse Factor = I f = Shape Factor = S f = 2 (7) X i= i μ 2 (8) P v X 2 (9) i= i P v (0) X i= i RMS () i= X i TABLE II. STATISTICAL PARAMETER COMPARISO FOR VIBRATIO SIGAL Sl. o Static Parameters Healthy Type-I Defect Type-II Defect Root Mea Square (RMS) 2 Mea Mea Peak Value Crest Factor Skewess Kurtosis Variace Stadard Deviatio Clearace Factor Impulse Factor Shape Factor B. Frequecy Aalysis The Fast Fourier Trasform(FFT) of the acquired vibratio ad acoustic sigal is achieved ad the frequecy spectrum is compared. The frequecy spectrum compariso of the vibratio sigal from the healthy bearig (B) ad Type- defect bearig(b2) is show i Fig. 4. ad compariso betwee healthy bearig (B) ad Type-2 defect bearig (B3) is show i Fig Peak Value Crest Factor Skewess Kurtosis Variace Fig. 4. FFT compariso of vibratio sigal from the healthy bearig(b) ad Type- defect bearig(b2) 8 Stadard Deviatio Clearace Factor Impulse Factor Shape Factor The statistical parameters are computed from the vibratio ad acoustic sigals ad is tabulated i Table II ad Table III respectively. From the compariso tables it is foud that the statistical parameter values of both vibratio ad acoustic sigal data i the case of defective bearigs chages as compare to the healthy bearig. Which detects the fault preset i the defective bearigs. TABLE III. STATISTICAL PARAMETER COMPARISO FOR ACOUSTIC SIGAL Sl. o Static Parameters Healthy Type-I Defect Type-II Defect Root Mea Square (RMS) Fig. 5. FFT compariso of vibratio sigal from healthy bearig(b) ad Type-2 defect bearig(b3) The the bearig characteristic frequecy also kow as the outer race defect frequecy(ordf) is calculated based o the geometric cofiguratio of the bearig by usig the formula i equatio. ORDF outer race fault = ω 2 d cos α () D The outer race defect frequecy is calculated as The the FFT compariso of the vibratio sigal from the healthy ad defective bearigs at ORDF is plotted ad is show i Fig. 6 ad Fig

5 . Fig. 6. FFT compariso of the vibratio sigal from the healthy bearig (B) ad Type- defect bearig(b2) at ORDF Fig. 20. FFT compariso of the acoustic sigal from the healthy bearig (B) ad Type- defect bearig(b2) at ORDF Fig. 7. FFT compariso of the vibratio sigal from the healthy bearig (B) ad Type-2 defect bearig(b3) at ORDF. Similarly the FFT compariso of acoustic sigal from the healthy ad defective bearig is show i Fig. 8 ad Fig. 9 ad the compariso at ORDF is show i Fig. 20 ad Fig. 2. The figures idicate the chage i spectrum whe defects preset i the bearigs. Fig. 8. FFT compariso of acoustic sigal from healthy bearig(b) ad Type- defect bearig(b2) Fig. 2. FFT compariso of the acoustic sigal from the healthy bearig (B) ad Type-2 defect bearig(b3) at ORDF C. Time-Frequecy Aalysis This aalysis shows the iformatio both i time ad frequecy scale. For this wavelet aalysis Morlet wavelet is used as the mother wavelet. 2D scalograms for the vibratio sigal data is determied based o the morlet wavelet ad are plotted. A differece is clearly visible aroud the scale value of equivalet BCF for the healthy, type- defect ad type-2 defect bearigs as show i Fig. 22 to Fig. 24 respectively. Similarly the 2D scalograms for the acoustic sigal data are determied ad plotted i Fig. 25 to Fig. 27. Here also the differece is clearly visible for the healthy ad defective bearigs. Fig. 9. FFT compariso of acoustic sigal from healthy bearig(b) ad Type-2 defect bearig(b3) Fig. 22. Scalogram for the vibratio sigal data from the healthy bearig 99

6 Fig. 23. Scalogram for the vibratio sigal data from the Type- defect bearig. Fig. 27. Scalogram for the Acoustic sigal data from the Type-2 defect bearig. V. COCLUSIO Fig. 24. Scalogram for the vibratio sigal data from the Type-2 defect bearig. Fig. 25. Scalogram for the Acoustic sigal data from the healthy bearig. I the implemetatio of the preset experimetal work the adaptive oise cacellatio is used at the preprocessig stage to improve the sigal to oise ratio which helps i better aalysis ad diagosis of the defect i the bearigs. Though vibratio or acoustic sigal aloe ca give iformatio about the defect but i this work both vibratio ad acoustic sigals are used to diagose the fault i the bearigs so that if oe sigal aalysis fails to detect the fault the the other sigal aalysis will diagose the fault. Iitially i time aalysis the statistical data compariso idicate the defect. I frequecy aalysis the FFT compariso at the ORDF shows a clear idicatio of the defect. Wavelet aalysis give the iformatio both i time ad frequecy scale. The precise iformatio about the defect i the bearigs is foud from the 2D scalograms of the wavelet aalysis. So the preset work is a better ad reliable techique to detect the fault i bearigs. REFERECES [] P.K. Kakar, S.C. Sharma, S.P. Harsha, Fault diagosis of ball bearigs usig cotiuous wavelet trasform, Appl. Soft Comput. (20) [2] P.K. Kakar, Satish C. Sharma, S.P. Harsha, Rollig elemet bearig fault diagosis usig wavelet trasform, eurocomputig, volume 74, issue 0, May 20, pages [3] P Shakya, A K Darpe ad M S Kulkari, Vibratio-based fault diagosis i rollig elemet bearigs: rakig of various time, frequecy ad time-frequecy domai data-based damage idetificatio parameters, The Iteratioal Joural of Coditio Moitorig, Volume 3, Issue 2, October 203. Fig. 26. Scalogram for the Acoustic sigal data from the Type- defect bearig. [4] V. Shamukha Priya, P. Mahalakshmi ad V. P. S. aidu, Health Coditio Moitorig: Wavelet Decompositio, Idia Joural of Sciece ad Techology, Vol 8(26), IPL0569, October 205. [5] Z.K. Peg, F.L. Chu, Applicatio of the wavelet trasform i machie coditio moitorig ad fault diagostics: a review with bibliography, Mechaical Systems ad Sigal Processig, Volume 8, Issue 2, March 2004, Pages [6] Sukhjeet Sigh, Amit Kumar, avi Kumar, Motor Curret Sigature Aalysis for Fault Detectio i Mechaical Systems, Procedia Materials Sciece 6 ( 204 )

7 [7] Y. Jiag, B. Tag, Y. Qi, W. Liu, Feature extractio method of wid turbie based o adaptive Morlet wavelet ad SVD, Reew. Eerg. 36 (20) [8] J.I. Taylor, The vibratio aalysis had book, 2d ed., USA: Vibratio Cosultats, [9] S. Mallat, A wavelet tour of sigal processig, st ed., Academic Press, [0] M. Misiti, Y. Misiti, G. Oppeheim, J.M.Poggi, Wavelets ad their Applicatios. 3rd ed., Frace: Hermes Sciece, [] C.W. Liao, J.Y. Li, ew FIR filter-based adaptive algorithms icorporatig with commutatio error to improve active oise cotrol performace, Automatica 43 (2007) [2] M.I. Troparevsky, C.E. D Attellis, O the covergece of the LMS algorithm i adaptive filterig, Sigal Process. 84 (2004)

8 02