CitationsSee all >18 References
8.67Institut FEMTO-ST28.02?cole Nationale Supérieure de Mécanique et des Microtechniques+ 323.71Ecole Nationale d'Ingénieur de Tarbes13.74?cole Nationale Supérieure de Mécanique et des MicrotechniquesShow
more authorsAbstractThis paper deals with the presentation of an experimental
platform called PRONOSTIA, which enables testing,
verifying and validating methods related to bearing health
assessment, diagnostic and prognostic. The choice of bearings
is justified by the fact that most of failures of rotating machines
are related to these components. Therefore, bearings can be
considered as critical as their failure significantly decreases
availability and security of machines.
The main objective of PRONOSTIA is to provide real data related
to accelerated degradation of bearings performed under constant
and/or variable operating conditions, which are online controlled.
The operating conditions are characterized by two sensors:
a rotating speed sensor and a force sensor. In PRONOSTIA
platform, the bearing’s health monitoring is ensured by gathering
online two types of signals: temperature and vibration (horizontal
and vertical accelerometers). Furthermore, the data are recorded
with a specific sampling frequency which allows catching all the
frequency spectrum of the bearing during its whole degradation.
Finally, the monitoring data provided by the sensors can be used
for further processing in order to extract relevant features and
continuously assess the health condition of the bearing.
During the PHM conference, a “IEEE PHM 2012 Prognostic
Challenge” is organized. For this purpose, a web link to the
degradation data is provided to the competitors to allow them
testing and verifying their prognostic methods. The results of
each method can then be evaluated regarding its capability
to accurately estimate the remaining useful life of the tested
bearings.Discover the world's research13+ million members100+ million publications700k+ research projects
PRONOSTIA: An Experimental Platform forBearings Accelerated Degradation TestsP. Nectoux, R. Gouriveau, K. Medjaher, E. Ramasso, B. Morello, N. Zerhouni, C. VarnierFEMTO-ST Institute, AS2M department, UMR CNRS 6174 - UFC / ENSMM / UTBM25000 Besan?on, FranceEmail: ieee-2012-PHM-challenge@femto-st.frAbstract—This paper deals with the presentation of an ex-perimental platform called PRONOSTIA, which enables testing,verifying and validating methods related to bearing healthassessment, diagnostic and prognostic. The choice of bearingsis justified by the fact that most of failures of rotating machinesare related to these components. Therefore, bearings can beconsidered as critical as their failure significantly decreasesavailability and security of machines.The main objective of PRONOSTIA is to provide real data relatedto accelerated degradation of bearings performed under constantand/or variable operating conditions, which are online controlled.The operating conditions are characterized by two sensors:a rotating speed sensor and a force sensor. In PRONOSTIAplatform, the bearing’s health monitoring is ensured by gatheringonline two types of signals: temperature and vibration (horizontaland vertical accelerometers). Furthermore, the data are recordedwith a specific sampling frequency which allows catching all thefrequency spectrum of the bearing during its whole degradation.Finally, the monitoring data provided by the sensors can be usedfor further processing in order to extract relevant features andcontinuously assess the health condition of the bearing.During the PHM conference, a “IEEE PHM 2012 PrognosticChallenge” is organized. For this purpose, a web link to thedegradation data is provided to the competitors to allow themtesting and verifying their prognostic methods. The results ofeach method can then be evaluated regarding its capabilityto accurately estimate the remaining useful life of the testedbearings.Index Terms—Condition monitoring, Fault detection, Faultdiagnostic, Failure prognostic, Condition-Based Maintenance,Predictive maintenance.I. INTRODUCTIONTo remain competitive, industrial companies must con-tinuously keep their production means in good operatingconditions by improving their availability, reliability, secu-rity while reducing their maintenance costs. One of possiblesolutions which allows satisfying the above requirements isthe implementation of appropriate maintenance strategies. Inthis domain the Condition-Based Maintenance (CBM) andPredictive Maintenance (PM) are the most efficient ones[1]–[3], because they allow optimizing the maintenance byanticipating the failure’s occurrence. Indeed, contrary to atraditional corrective maintenance, where the interventions aredone after the occurrence of the failure, in a CBM (or aPM), the interventions are done according to the observed orestimated health condition of the equipment.Generally, a CBM system is seen as the integration of sevenlayers [4]: sensors, signal processing, condition monitoring(or fault detection), health assessment (or fault diagnostic),prognostic, decision support and finally presentation layers.Among these activities, failure prognostic is considered asthe most recent one, with an increasing research as well asindustrial interest. The increasing interest accorded to failureprognostic has led to numerous methods, tools and applicationsduring the last decade. According to the reported literature,failure prognostic methods can be classified into three mainapproaches: model-based, data-driven and hybrid approaches[1], [4], [5].Model-based prognostic approach relies on the use of ananalytical model (set of algebraic or differential equations) torepresent the behavior of the system including its degradation.The advantage of this approach is that it provides preciseresults. However, its drawback dwells in the fact that realsystems are often nonlinear and the degradation mechanismsare generally stochastic and difficult to obtain in the form ofanalytical models.Data-driven prognostic approach aims at transforming themonitoring and operating data into relevant information andbehavior models of the system including its degradation. Thisapproach uses artificial intelligence tools and/or statisticalmethods to learn the degradation model and to predict theRemaining Useful Life (RUL) of the equipment. The data-driven approach can be used in cases where getting monitoringdata and processing them is easier than constructing physicaland analytical behavior models.Hybrid prognostic methods combine both model-based anddata-driven approaches and thus take benefit and drawbackfrom both of them.In practice, tests and verifications of fault detection andisolation (or fault diagnostic) methods are easy to perform,because the faults can be easily simulated or introduced onthe real industrial system. However, this is not the case forprognostic methods where the fault is generally a consequenceof a long and slow degradation of one or more components ofthe system. Thus, to test these methods, it is necessary to create(or initiate) the degradation through accelerated degradationtests of physical components. For this purpose, researchershave made their own experimental platforms, but only a fewnumber of these platforms are opened to external researchersto provide them with real monitoring data [6].This paper aims at presenting a new experimental platform,called PRONOSTIA, related to bearings’ degradation tests.hal-, version 1 - 20 Jul 2012Author manuscript, published in &IEEE International Conference on Prognostics and Health Management, PHM&12., Denver,Colorado : United States (2012)&
This platform comes to complete the list of existing ones andwill be a source of experimental data acquired for constantand/or variable operating conditions, enabling therefore theverification of condition monitoring, fault detection, faultdiagnostic and prognostic approaches. In this sens, an IEEEPHM 2012 Prognostic Challenge is organized during the 2012IEEE PHM conference, which took place in Denver. Thus, inaddition to the presentation of PRONOSTIA, this paper givesdetails on the organized PHM challenge (who and how toparticipate, the related data, the requested results, etc.).The paper is organized as follows: after the introduction,section 2 gives a brief state of the art on the existing experi-mental platforms reported in the literature, the PRONOSTIAplatform is presented in section 3, the acquired data andsome experimental results are given in section 4, the IEEEPHM 2012 Prognostic Challenge is explained in section 5 andfinally, section 6 concludes the paper.II. STATE OF THE ART ON EXPERIMENTAL PLATFORMSIn order to test and verify the prognostic methods developedand published in the literature, dedicated test beds and plat-forms have been designed and realized by several laboratoriesover the world. Most of these experimental systems concernspecific physical components, such as bearings, gears, pumps,etc. The following paragraphs summarize the experimentalplatforms which are yet published. Note that not all of thepublished test beds provide experiment data for external users,but only some of them [6].A. Overview of test bedsThe following paragraphs present an overview of experi-mental platforms reported in the literature and related to crit-ical physical components such as gearboxes, pumps, pinions,etc. The test beds related to bearings will be presented in thesubsection II-B.A test bed related to a gearbox and a pinion gear has been usedin [7]. In this application, a spiral bevel pinion was seeded withtwo electrical discharge machine (EDM) notches (heel and toe)on the drive side of one of the pinion gear teeth to artificiallyaccelerate tooth root cracking. Several accelerometers werethen placed on the gearbox with a health and usage monitoringsystem (HUMS) used to generate the vibration features.In the case of machining tools, a milling data set experimentsrelated to milling machine for different speeds, feeds and depthof cut can be found in [6]. In the same way, an experimentalplatform has been developed by SIMTech of Singapore [8]to provide data during the PHM challenge organized in 2010by the PHM society. In [9] the authors used an experimentalsetup related to drilling life tests to verify their method. Thetests were conducted on a MAHO 700S machine, which isa computer numerical controlled (CNC) five axis machiningcenter, with movement in three perpendicular axes and arotary/tilt table. Finally, in [10] a method is proposed toestimate the tool wear of a turning process over a wide rangeof cutting conditions. The developments were validated thanksto experiments conducted on a conventional lathe TUD-50.Concerning the pumps, an experimental setup has been usedin [11] to evaluate the performance of a developed HiddenSemi-Markov Model method for equipment health prognostic.The experimental setup consisted in a real hydraulic pump.During the experiments, long-term wear test experiments wereconducted at a research laboratory facility. Three pumps werethen worn by running them using oil containing dust.Finally, data set experiments related to charging and discharg-ing of Li-Ion batteries can be found in [6]. The records concernthe impedance as the damage criterion and the data set wasprovided by the Prognostics Center of Excellence at NASA.The same center proposed preliminary data from thermal over-stress accelerated aging for six devices.Note that this overview is obviously not exhaustive, butenables to see that real systems are required to test and verifyPHM algorithms. Also, the variety of presented applicationsreveals that most PHM tools are application-based. Thus,further developments to face this aspect are still required.Among the works of the PHM field, bearings failures analysisbenefits from a great interest and this is the point addressedin the following of the paper.B. Bearings test bedsNumerous prognostic methods proposed in the literaturewere tested on the degradation of bearings. Thus, in [12] theauthors have used a test bed related to bearings’ degradationsto test and verify their fault detection and diagnostic method.Similarly, a bearing test bed is proposed in [13] to detectdefects on the balls, the inner and outer raceways of bearings.In this application, the defects were induced by means ofelectrical discharge machine. In the same way, a diagnosticmethod has been proposed in [14] were three accelerometerswere used to measure the vibration on the tested bearings.The difference with the above applications is that in this onea load is applied on the bearing to accelerate its degradation.Also, in [15] a dedicated test bed has been used to performfailure prognostic on bearings. The particularity of the usedapplication is that the bearings’ degradations were obtainedafter several days (around 50 days) and thus the amount of datato process was considerable. An experimental platform wasused in [16] to simultaneously degrade four bearings, whereasin [17] a test bed for aerospace industry with special bearingstested without lubrication was used. Finally, experimental datarelated to bearings can be found in the NASA data repository[6]. These data are provided by the Center for IntelligentMaintenance Systems (IMS).In the particular case of bearings test beds, and comparedto those proposed in the literature, the data provided bythe PRONOSTIA experimental platform are different in thesense that they correspond to “normally”degraded bearings.This means that the defects were not initially initiated onthe bearings and that each degraded bearing contains almostall the types of defects (balls, rings and cage). The acquiredexperimental data can then be used for fault detection, diag-nostic and prognostic. Furthermore, even if the data presentedin this paper concern at this time only constant operatinghal-, version 1 - 20 Jul 2012
Bearing tested Accelerometers TorquemeterSpeed sensorAC MotorNI cDAQ cards Pressure regulatorSpeed reducerForce sensorPlatinum RTDCouplingCylinder Pressure Fig. 1. Overview of PRONOSTIA.conditions for each realized experiment, the current design ofPRONOSTIA allows us in the future to provide data relatedto bearings degraded under variable operating conditions.III. THE PRONOSTIA PLATFORMPRONOSTIA is an experimentation platform (Fig. 1) ded-icated to test and validate bearings fault detection, diagnosticand prognostic approaches. The platform has been designedand realized at AS2M1department of FEMTO-ST2institute.The main objective of PRONOSTIA is to provide real exper-imental data that characterize the degradation of ball bearingsalong their whole operational life (until their total failure).This experimental platform allows to conduct bearings’ degra-dations in only few hours, and thus it is possible to get signif-icant number of experiments within a week. PRONOSTIA iscomposed of three main parts: a rotating part, a degradationgeneration part (with a radial force applied on the testedbearing) and a measurement part, which are detailed hereafter.A. The rotating partThis part includes the asynchronous motor with a gearboxand its two shafts. The first shaft is near to the motor andthe second shaft is placed at the ride side of the incrementalencoder.The asynchronous motor is the actuator that allows the bearingto rotate through a system of gearing and different couplings.The motor has a power equal to 250 W and transmits therotating motion through a gearbox, which allows the motorto reach its rated speed of 2830 rpm, so that it can deliverits rated torque while maintaining the speed of the secondary1Automatic control and Micro-Mechatronic Systems2Franche-Comté Electronics, Mechanics, Thermal Processing, Optics -Science and Technologyshaft to a speed less than 2000 rpm. The gearbox is home-made and consists of two pulleys bound by a timing belt, itselfheld by a turnbuckle. Compliant and rigid shaft couplings areused to create connections for the transmission of the rotatingmotion produced by the motor to the shaft support bearing.The bearing support shaft (Fig. 2) leads the bearing throughits inner race. This one is kept fixed to the shaft with ashoulder on the right hand and a threaded locking ring onthe left hand. The shaft which is made of one piece is heldby two pillow blocks and their large gearings. Two clampingsallow the longitudinal blocking of the shaft between the twopillow blocks. The human machine interface of PRONOSTIAallows the operator to set the speed, to select the direction ofthe motor’s rotation and to display the monitoring parameterssuch as the motor’s instantaneous temperature expressed inpercentage of the maximum temperature of use. The wholedriving chain of the motor includes the human interfacemachine and a frequency converter, which are both connectedwith a Profibus-DP link to an industrial programmable logiccontroller.Fig. 2. Shaft support bearing.hal-, version 1 - 20 Jul 2012
B. Generation of the radial forceThis part has all its components, except the proportionalregulator, grouped in a unique and same aluminum plate andis partially isolated from the instrumentation part by a thinlayer of polymer. The aluminum plate supports a pneumaticjack (Fig. 3), a vertical axis and its lever arm, a force sensor, aclamping ring of the test bearing, a support test bearing shaft,two pillow blocks and their large oversized bearings.The radial force applied on the test ball bearing constitutesthe heart of the global system. In fact, the radial force reducesthe bearing’s life duration by setting its value up to thebearing’s maximum dynamic load which is 4000 N. Thisload is generated by a force actuator, which consists in apneumatic jack, where the supply pressure is delivered by adigital electro-pneumatic regulator.ProportionalregulatorPneumaticjackFig. 3. The pneumatic jack, the pressure regulator and the radial forcegeneration.The force issued from the pneumatic jack is indirectlyapplied on the external ring of the test ball bearing. This forceis first amplified by a rotating lever arm, then transmitted tothe test bearing through its clamping ring (Fig. 3).C. Measurements partThe Bearing’s operation conditions are determined byinstantaneous measures of the radial force applied on thebearing, the rotation speed of the shaft handling the bearingand of the torque inflicted to the bearing. Each of these threeanalog measures is acquired at a frequency equal to 100 Hz.The characterization of the bearing’s degradation is basedon two data types of sensors: vibration and temperature(Fig. 4). The vibration sensors consists of two miniatureaccelerometers positioned at 90o the first isplaced on the vertical axis and the second is placed on thehorizontal axis. The two accelerometers are placed radiallyon the external race of the bearing. The temperature sensor isan RTD (Resistance Temperature Detector) platinum PT100(1/3 DIN class) probe, which is place inside a hole close tothe external bearing’s ring. The acceleration measures aresampled at 25.6 kHz and the temperature ones are sampledat 10 Hz.Fig. 4. The accelerometers and temperature sensor.The data acquisition system is based on a four slot chassis,which includes three I/O modules. It aggregates the dataissued from the whole sensors and transmits them throughan USB 2.0 link to the central unit in charge of real time datavisualization and storage.Finally, the visualization of the monitoring data is done by aspecific application implemented under Labview environmentand installed on a dedicated computer (Fig. 5).Fig. 5. Data visualization and Human Interface Machine.This application allows the operator to visualize raw signalsfrom different sensors. The acquired data are formatted, timestamped and stored locally in the computer under differentfiles. The data can then be used for offline or online pro-cessing in order to continuously assess the health conditionof the test bearing (fault detection, diagnostic and prognosticapplications). Figure 6 depicts an example of what one canobserve on the ball bearing components before and after anexperiment.IV. EXPERIMENTAL RESULTSA. Degradation patternsDepending on various factors, the degradation may bedifferent for distinct bearings. Assuming that no other in-formation is available about the other components than therotating system and that load and speed are constant, one hasto use only the data collected via the sensors located aroundthe bearings. Moreover, nothing is known about the natureand the origin of the degradation (balls, inner or outer races,hal-, version 1 - 20 Jul 2012
Fig. 6. Normal and degraded bearing.cage...) therefore data-driven techniques has to be applied.According to the bearing and to the way the degradationevolves, the fault modes can be slightly different for distinctbearings. As a result, the degradation “patterns” may haveparticular characteristics as illustrated in the following. In thispaper, we considered several experiments for which differentfeatures (health index) were extracted. The choice of featuresis typical of bearings diagnostics and prognostics [18]. Figure7 shows a vibration raw signal taken from PRONOSTIAduring a whole experiment.x 106-50-40-30-20-1001020304050Fig. 7. A vibration raw signal.B. The ideal degradationAs a first example, consider an experiment where severalfeatures (and health index) agree about the degradation andwhere the patterns depicted are very typical. Figure 8 il-lustrates the evolution of the power spectrum density (PSD)computed on the horizontal accelerometer sensor. The evolu-tion is mainly monotonic increasing and represents an idealcase where a prediction model can be used with some easythresholds for RUL estimation. The K-factor computed on thevertical accelerometer signal also depicts a slowly degradationwith almost no noise (Fig. 9).Fig. 8. Power spectrum density (PSD) computed on the data of the horizontalaccelerometer.0 5000 0200400600Horizontal sensor K factor0 5000 0200400600800Vertical sensor K factorFig. 9. K-factor computed on the data of the vertical and horizontalaccelerometers.C. Sudden degradationsIn some cases, the degradation appears suddenly and doesnot depict a slow monotonic behavior. In this case, finding aprediction model is much more difficult based on those typicalhealth index such as PSD (Fig. 10) or the K-factor (Fig. 11).Therefore, other features have to be imagined.D. Theoretical models mismatchApplying prediction models based on life duration is notrelevant. Indeed, the degradation of bearings considered inPRONOSTIA depict very different behaviors leading to verydifferent experiment duration (until the fault). Moreover, thetheoretical models based on frequency signatures to detectbearings’ faults (such as the inner and outer races and the cagefaults) do not work with the data provided by PRONOSTIA.This is because the frequency signatures are difficult to obtaindue to the fact that the degradation may concern at a same timehal-, version 1 - 20 Jul 2012
Fig. 10. Power spectrum density (PSD) computed on the data of thehorizontal accelerometer.0 00 00 00051015Horizontal sensor K factor0 00 00 000204060Vertical sensor K factorFig. 11. K-factor computed on the data of the vertical and horizontalaccelerometers.all the components of the test bearing. Finally, the existingreliability laws for bearings’ life duration, such as the L10,donot give same results than those obtained by the experiments(theoretical estimated life durations are different from thosegiven by the experiments).E. Level of noiseThe level of noise is not controlled and depends on thedegradation process. In figures 8 to 11 the level of noisepresented different values. There are some cases where thislevel can be high, possibly explained by the interactions withother parts of the rotating system. Figures 12 and 13 depictthe crest factor and the PSD for an experiment where the levelof noise is particularly high.Figure 14 shows an example of a Wavelet Packet Decom-position (WPD) performed on vibration data provided by anaccelerometer. A WPD is a time-frequency technique which0 0.5 1 1.5 2 2.5x 104051015Horizontal sensor crest factor0 0.5 1 1.5 2 2.5x 10405101520Vertical sensor crest factorFig. 12. Crest factor computed on the data of the horizontal and verticalaccelerometers.Fig. 13. Power spectrum density (PSD) computed on the data of the verticalaccelerometer.permits to adjust the size of the temporal window according tothe analyzed frequencies. A WPD has two parameters: a scaleparameter afor the frequency and a translation parameter bforthe time [19]. By using a WPD, the original vibration signalscan be decomposed into several levels and the energy of eachlevel can be calculated.V. IEEE PHM 2012 PROGNOSTIC CHALLENGEA. Outline of the challengeThe IEEE Reliability Society and Femto-st Institute orga-nize the IEEE PHM 2012 Prognostic Challenge. The challengeis focused on prognostics of the remaining useful life (RUL) ofbearings, a critical problem since most of failures of rotatingmachines are related to these components, strongly affectingavailability, security and cost effectiveness of mechanical orpower industries.hal-, version 1 - 20 Jul 2012
0 0.5 1 1.5 2 2.5x 104051015202530354045Time (seg)EnergyHorizontal sensor energy (wavelet)
Packet 1Packet 2Packet 3Packet 4Packet 5Packet 6Packet 7Packet 8Fig. 14. Wavelet Packet Decomposition (WPD) performed on the data ofthe horizontal accelerometer.Challenge datasets are provided by Femto-st Institute. Ex-periments were carried out on the platform (PRONOSTIA)presented in this paper. The challenge is open to all potentialconference attendees. Both Academic (from University) andProfessional teams (from Industry) are encouraged to enter.The two top scoring participants from both categories willbe distinguished and invited to present their results at aspecial session of the 2012 IEEE International Conference onPrognostics and Health Management.B. Aims of the challengeAs for the PHM challenge, 3 different loads were consid-ered:oFirst operating conditions: 1800 rpm and 4000 N;oSecond operating conditions: 1650 rpm and 4200 N;oThird operating conditions: 1500 rpm and 5000 N.Participants are provided with 6 run-to-failure datasets in orderto build their prognostics models, and are asked to estimateaccurately the RUL of 11 remaining bearings (see Table I).Monitoring data of the 11 test bearings are truncated so thatparticipants are supposed to predict the remaining life, andthereby perform RUL estimates. Also, no assumption on thetype of failure to be occurred is given.The learning set is quite small while the spread of thelife duration of all bearings is very wide (from 1h to 7h).Performing good estimates is thereby quite difficult and thismakes the challenge more exciting. Note also that, as statedin IV-D, there is a mismatch between the experiments and thetheoretical framework (L10 law, BPFI, BPFE, etc.).C. Scoring of the challengeParticipants are scored based on their RUL results that areconverted into percent errors of predictions. Let note?RULiand ActRU Lirespectively the remaining useful life of thebearing estimated by a participant, and the actual RUL toTABLE IDATASETS OF IEEE PHM 2012 PROGNOSTIC CHALLENGEData setsOperating ConditionsConditions 1 Conditions 2 Conditions 3Learning set Bearing1_1 Bearing2_1 Bearing3_1Bearing1_2 Bearing2_2 Bearing3_2Test setBearing1_3 Bearing2_3 Bearing3_3Bearing1_4 Bearing2_4Bearing1_5 Bearing2_5Bearing1_6 Bearing2_6Bearing1_7 Bearing2_7be predicted (where istates for the test bearings defined inTable I). The percent error on experiment iis defined by:%Eri= 100 ×ActRU Li-?RULiActRU Li(1)Underestimates and overestimates will not be considered inthe same manner: good performance of estimates relates toearly predictions of RUL (i.e. cases where %Eri&0), withdeduction to early removal, and more severe deductions forRUL estimates that exceed actual component RUL (i.e. caseswhere %Eri&0). The score of accuracy of a RUL estimatesfor experiment iis thereby defined as follows:Ai=?exp-ln(0.5).(Eri/5) if Eri≤0exp+ln(0.5).(Eri/20) if Eri&0(2)Figure 15 depicts the evolution of this scoring function.-50 -40 -30 -20 -10 010 20 30 40 5000.250.50.751X: -10Y: 0.25%EriX: 20Y: 0.5Ailate predictions early predictionsFig. 15. Scoring function of a RUL estimates according to its percent errorThe final score of all RUL estimates will be defined as beingthe mean of all experiment’s score:Score =11111?i=1(Ai)(3)More details on the objective of the challenge, the scoring ofresults, the application form, and obviously on the provideddatasets, can be found on the home page of the challenge:http://www.femto-st.fr/ieee-PHM2012-data-challenge.hal-, version 1 - 20 Jul 2012
VI. CONCLUSIONA new experimental platform, called PRONOSTIA, is pre-sented in this paper. Its main purpose is to provide exper-imental data related to bearings’ degradations. These datacan then be used to test and verify research methods inthe following fields: condition monitoring, fault detection,diagnostic and prognostic. The particularity of this platform isthat the bearing’s degradation can be realized under constant aswell as variable operating conditions and the data are acquiredthroughout the whole duration of each experiment.Three sets of experimental data realized under three differentoperating conditions are provided to researchers in order to testtheir methods for the prediction of the remaining useful lifeof the degraded bearings. These tests are organized in a formof a challenge where the results of the proposed methods arethen assessed and compared. Moreover, the experimental dataremain stored in the indicated website to allow verificationsby researchers working in the PHM field over the world.Finally, it can be noted that in the three sets of experimentaldata provided for the organized challenge, each set containsa given number of historical data realized in same operatingconditions. So, to be close to the real industrial applications,further works concern online variations of operating conditionswithin the same experiment.REFERENCES[1] A. K. Jardine, D. Lin, and D. Banjevic, “A review on machinery di-agnostics and prognostics implementing condition-based maintenance,”Mechanical Systems and Signal Processing, vol. 20, no. 7, pp. 1483 –.[2] A. Heng, A. C. Tan, J. Mathew, N. Montgomery, D. Banjevic, and A. K.Jardine, “Intelligent condition-based prediction of machinery reliability,”Mechanical Systems and Signal Processing, vol. 23, no. 5, pp. 1600 –.[3] G. Vachtsevanos, F. L. Lewis, M. Roemer, A. Hess, and B. Wu,Intelligent fault diagnosis and prognosis for engineering systems. Wiley,2006.[4] M. Lebold and M. Thurston, “Open standards for condition-basedmaintenance and prognostic systems,” in Maintenance and ReliabilityConference (MARCON), 2001.[5] A. Heng, S. Zhang, A. C. Tan, and J. Mathew, “Rotating machineryprognostics: State of the art, challenges and opportunities,” MechanicalSystems and Signal Processing, vol. 23, no. 3, pp. 724 – 739, 2009.[6] NSF I/UCRC Center for Intelligent Maintenance Sys-tems, “Prognostic data repository: Bearing data set,” inhttp://ti.arc.nasa.gov/tech/dash/pcoe/prognostic-data-repository/, visitedin January 2012.[7] G. Kacprzynski, A. Sarlashkar, M. Roemer, A. Hess, and W. Hardman,“Predicting remaining gear life by fusing diagnostics and physics offailure models,” JOM Journal, vol. 56, no. 3, pp. 29–35, 2004.[8] J. Zhou, X. Li, O. P. Gan, S. Han, and W. K. Ng, “Genetic algorithmsfor feature subset selection in equipment fault diagnostics,” Journal ofEngineering asset management, vol. 10, pp. , 2006.[9] H. M. Ertunc and C. Oysu, “Drill wear monitoring using cutting forcesignals,” Mechatronics, vol. 14, pp. 533–548, 2004.[10] Q. Ren, M. Balazinski, L. Baron, and K. Jemielniak, “Tsk fuzzymodeling for tool wear condition in turning processes: An experimentalstudy,” Engineering Applications of Artificial Intelligence, vol. 24, no. 2,pp. 260 – 265, 2001.[11] Y. Peng and MingDong, “A prognosis method using age-dependent hid-den semi-markov model for equipment health prediction,” MechanicalSystems and Signal Processing, vol. 25, pp. 237–252, 2011.[12] M. Subrahmanyam and C. Sujatha, “Using neural networks for thediagnosis of localized defects in ball bearings,” Tribology International,vol. 30, no. 10, pp. 739–752, 1997.[13] Y.-T. Sheen, “On the study of applying morlet wavelet to the hilberttransform for the envelope detection of bearing vibrations,” MechanicalSystems and Signal Processing, vol. 23, no. 5, pp. , 2009.[14] R. Yan and R. X. Gao, “Multi-scale enveloping spectrogram for vibrationanalysis in bearing defect diagnosis,” Tribology International, vol. 42,no. 2, pp. 293–302, 2009.[15] H. Ocak, K. A. Loparo, and F. M. Discenzo, “Online tracking of bearingwear using wavelet packet decomposition and probabilistic modeling: Amethod for bearing prognostics,” Journal of sound and vibration, vol.302, pp. 951–961, 2007.[16] J. Lee, J. Ni, D. Djurdjanovic, H. Qiu, and H. Liao, “Intelligentprognostics tools and e-maintenance,” Computers in Industry, vol. 57,no. 6, pp. 476–489, 2006.[17] C. Dellacorte, V. Lukaszewicz, M. Valco, K. C. Radil, and H. Heshmat,“Performance and durability of high temperature foil air bearings foroil-free turbomachinery,” Tribology transactions, vol. 43, no. 4, pp. 774–780, 2000.[18] W. Yan, H. Qiu, and N. Iyer, “Feature extraction for bearing prognosticsand health management (PH M)-a survey,” Air Force Research Labora-tory, Tech. Rep., 2008.[19] J. Zarei and J. Poshtan, “Bearing fault detection using wavelet packettransform of induction motor stator current,” Tribology International,vol. 40, no. 5, pp. 763 – 769, 2007.hal-, version 1 - 20 Jul 2012
CitationsCitations79ReferencesReferences18Works within this field are mainly focused on the assessment of faults in ball, inner and outer races with one point defect of the wear process is also studied. The Bearing Data Center of Case Western Reserve University is the most common database used for testing the approaches, and recently some works use PRONOSTIA for testing health indicator construction on bearing degradation datasets[101]. Cococcioni et al.[102]presented a general overview on the use of some learning- statistical classifiers such as neural networks are used for identifying several levels of damages in bearings. ABSTRACT: Health condition monitoring of rotating machinery is a crucial task to guarantee reliability in industrial processes. In particular, bearings are mechanical components used in most rotating devices and they represent the main source of fault reason for which research activities on detecting and diagnosing their faults have increased. Fault detection aims at identifying whether the device is or not in a fault condition, and diagnosis is commonly oriented towards identifying the fault mode of the device, after detection. An important step after fault detection and diagnosis is the analysis of the magnitude or the degradation level of the fault, because this represents a support to the decision-making process in condition based-maintenance. However, no extensive works are devoted to analyse this problem, or some works tackle it from the fault diagnosis point of view. In a rough manner, fault severity is associated with the magnitude of the fault. In bearings, fault severity can be related to the physical size of fault or a general degradation of the component. Due to literature regarding the severity assessment of bearing damages is limited, this paper aims at discussing the recent methods and techniques used to achieve the fault severity evaluation in the main components of the rolling bearings, such as inner race, outer race, and ball. The review is mainly focused on data-driven approaches such as signal processing for extracting the proper fault signatures associated with the damage degradation, and learning approaches that are used to identify degradation patterns with regards to health conditions. Finally, new challenges are highlighted in order to develop new contributions in this field.Article · Jan 2018 +1 more author...Fannia Karolina PachecoThe selected data concern the outer race fault[27]that is naturally generated rather than artificially created. Based on the parameters of this test bearing[26], the characteristic frequency for the outer race fault (f BPFO ) is 5.6113 ? f n , where f n is the shaft rotating frequency (i.e., 25 Hz for the test data). Figs. ABSTRACT: Local mean decomposition (LMD) is a promising approach to implement time-frequency representation (TFR) for multicomponent amplitude-modulated (AM) and frequency-modulated (FM) however, its performance usually suffers from end effect and mode mixing problems. To address this issue, this paper proposes a novel comprehensive scheme to improve LMD performance. The novel scheme can automatically determine the fix subset size of the moving average algorithm and the optimal number of sifting iterations in a sifting process. Extensive simulations have been explored for multicomponent AM-FM signal analysis by means of TFR with the improved LMD. Moreover, the improved LMD shows potential application in bearing fault diagnosis in conjunction with the well-known fast kurtogram.Article · Oct 2017 Vibration signals collected by the two accelerometers were sampled every 10 s, and the duration of the sampling lasted 0.1 s with a sampling frequency 25.6 kHz. The detailed information about the platform and experiments can be found in[35]. In the experiments, there were three different operating conditions: the first one (1800 rpm and 40 0 0 N), the second one (1650 rpm and 4200 N) and the third one (1500 rpm and 50 0 0 N). ABSTRACT: In data-driven prognostic methods, prediction accuracy of bearing remaining useful life (RUL) mainly depends on the performance of bearing health indicators, which are usually fused from some statistical features extracted from vibration signals. However, many existing bearing health indicators have the following two shortcomings: (1) many statistical features do not have equal contribution to construction of health indicators since the ranges of these statistical fe (2) it is difficult to determine a failure threshold since health indicators of different machines are generally different at a failure time. To overcome these drawbacks, a recurrent neural network based health indicator (RNN-HI) for RUL prediction of bearings is proposed in this paper. Firstly, six related-similarity features are proposed to be combined with eight classical time-frequency features so as to form an original feature set. Then, with monotonicity and correlation metrics, the most sensitive features are selected from the original feature set. Finally, these selected features are fed into a recurrent neural network to construct the RNN-HI. The performance of the RNN-HI is verified by two bearing data sets collected from experiments and an industrial field. The results show that the RNN-HI obtains fairly high monotonicity and correlation values and it is beneficial to bearing RUL prediction. In addition, it is experimentally demonstrated that the proposed RNN-HI is able to achieve better performance than a self organization map based method.Article · Feb 2017 +1 more author...Some experimental platforms have been proposed and used to provide data for RUL calculations. A bearing accelerated degradation test bed has been developed at femto Nectoux et al. (2012), the test bed PRONOSTIA was aimed at validating methods related to bearing health assessment, this experimental platform provided 3 test sets for the PHM challenge on RUL predictions. A bearing test bed is proposed Yan et al. (2009) Vibration readings were collected using 3 accelerometers for the accelerated degradation test, and a new vibration signal analysis method was developed to extract wear related features. ABSTRACT: This paper proposes a general deterioration model for a multi-component system. The deterioration process of a component depends upon the operational conditions, the component’s own state, and also the state of other components. An experimental platform that aims to provide more insight into the true nature of degradation of multi-component systems is also described. Some preliminary experimental results demonstrate the feasibility and advantages of the proposed deterioration model for describing highly stochastic degradation processes in industrial engineering.Conference Paper · Dec 2016 · Shock and VibrationExperimental datasets from PRONOSTIA [19] are used to further demonstrate the effectiveness of the proposed RUL estimation approach. Seventeen groups of life cycle vibration data are generated in this platform. ABSTRACT: Predicting the remaining useful life (RUL) of critical subassemblies can provide an advanced maintenance strategy for wind turbines installed in remote regions. This paper proposes a novel prognostic approach to predict the RUL of bearings in a wind turbine gearbox. An artificial neural network (NN) is used to train data-driven models and to predict short-term tendencies of feature series. By combining the predicted and training features, a polynomial curve reflecting the long-term degradation process of bearings is fitted. Through solving the intersection between the fitted curve and the pre-defined threshold, the RUL can be deduced. The presented approach is validated by an operating wind turbine with a faulty bearing in the gearbox. Full-text · Article · Dec 2016 +1 more author...As a solution, signal processing techniques are usually utilized to extract some features from the vibration data [5]. It is not uncommon that quite a few features can be extracted in the time, frequency, and timefrequency domains as well as via information entropy analysis [3]. However, since not all features are required in describing the product's degradation process, it is necessary to perform dimension reduction for the purpose of degradation modeling. ABSTRACT: Accelerated degradation testing (ADT) has been widely used for reliability prediction of highly reliable products. In many applications, ADT data consists of multiple degradation-related features, and these features are usually dependent. When dealing with such ADT data, it is important to fully utilize the multiple degradation features and take into account their inherent dependency. This paper proposes a novel reliability-assessment method that combines Brownian motion and copulas to model ADT data obtained from vibration signals. In particular, degradation feature extraction is first carried out using the raw vibration signals, and a feature selection method quantifying feature properties, such as trendability, monotonicity, and robustness, is adopted to determine the most suitable degradation features. Then, a multivariate s-dependent ADT model is developed, where a Brownian motion is used to depict the degradation path of each degradation feature and a copula function is employed to describe the dependence among these degradation features. Finally, the proposed ADT model is demonstrated using the vibration-based ADT data for an electric motor. Full-text · Article · Oct 2016 ProjectThe goal is to study convex optimisation-based
methods for Pronostics and Health Management. Project[...]This project is about Pronostics, Diagnostics and anomaly detection
of roller bearing machine.
We are focusing about Relevant fault feature identification, since the tasks cited above are based m…& ProjectThis project aims at creating a startup that proposes software solutions and services linked to Prognostics and Health Management.
For more information, please contact us at propiceindustries@gmail…& ProjectPHM Prognostics and Health Management Data provided are for informational purposes only. Although carefully collected, accuracy cannot be guaranteed. Publisher conditions are provided by RoMEO. Differing provisions from the publisher's actual policy or licence agreement may be applicable.This publication is from a journal that may support self archiving.Last Updated: 31 Aug 17